- Email: [email protected]
What installed mobile applications tell about their owners and how they affect users’ download behavior
Accepted Manuscript What installed mobile applications tell about their owners and how they affect users’ download behavior Perin Unal, Tugba Taskaya Temizel, P. Erhan Eren PII: DOI: Reference:
S0736-5853(17)30049-7 http://dx.doi.org/10.1016/j.tele.2017.05.005 TELE 948
To appear in:
Telematics and Informatics
Received Date: Accepted Date:
10 April 2017 10 May 2017
Please cite this article as: Unal, P., Temizel, T.T., Eren, P.E., What installed mobile applications tell about their owners and how they affect users’ download behavior, Telematics and Informatics (2017), doi: http://dx.doi.org/ 10.1016/j.tele.2017.05.005
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
May, 11, 2017
What installed mobile applications tell about their owners and how they affect users’ download behavior Perin Unal 1, Tugba Taskaya Temizel1, P. Erhan Eren1 1
Department of Information Systems, Informatics Institute, Middle East Technical University, Ankara, Turkey
Corresponding Author: Perin UNAL Department of Information Systems Informatics Institute, Middle East Technical University Universiteler Mahallesi, Dumlupınar Bulvarı, No:1, 06800 Çankaya/Ankara/TURKEY Mobile Phone: +90 532 4155783 Fax Number: +90 312 5921848 E-mail: [email protected]
What installed mobile applications tell about their owners and how they affect users’ download behavior Abstract. The rapid growth in the mobile application market presents a significant challenge to find interesting and relevant applications for users. An experimental study was conducted through the use of a specifically designed mobile application, on users’ mobile phones. The goals were; first, to learn about the users’ personality and the applications they downloaded to their mobile phones, second to recommend applications to users via notifications through the use of experimental mobile application and learn about user behavior in mobile environment. The question of how the personality features of users affect their compliance to recommendations is explored in this study. It is found that conscientiousness is positively related with accepting recommended applications and being agreeable is related with the preference for the applications of editor’s choice. Furthermore, in this study, applications owned by the user and the composition of applications under categories and their relation with personality features are explored. It is shown that the number of user owned applications and their category differ according to gender and personality. Having similar applications and the number of applications owned under specific categories increase the probability of accepting recommended applications. Keywords. Mobile phone use, mobile application use, user profiling, mobile application recommendation, recommender systems
1.
Introduction
Mobile users predominantly prefer to use mobile applications rather than browsers to access internet services. Application markets have grown rapidly as a result of vesting user interest in mobile applications. Mobile application recommendation websites and services fulfill the growing need to filter, rank and recommend the best applications from the hundreds of thousands available. Some of these sites operate in the official application marketplaces such as the Genius of iTunes App Store and the recommendations in Google Play and others on marketplaces like Amazon Appstore and Yandex. In all marketplaces, implementations of consensus and authority influence strategies are predominantly used on the basis of the most popular applications, most downloaded ones, editor’s choice, applications of the day and similar ones. Recommender systems that are used for mobile applications may make use of user profiles. User’s gender, personality features, the number of applications owned by the user and the composition of the applications under categories can be used as a source for understanding user profiles and preferences. This information can be processed and integrated into the recommender systems to address users’ needs and preferences and build personalized relevant outputs. Understanding user preferences is a complex issue and has been the subject of numerous studies in the social and psychological sciences. Personality has fundamental dimensions referred to as traits and it has been shown that people’s behavior can be explained by their underlying personality traits, which reflect the enduring dispositions of their nature (Costa and McCrae, 1992). There are various methodologies in the literature that identify and classify the personality traits but the most widely used and accepted model is the Big Five personality framework which defines the traits as extraversion, agreeableness, conscientiousness, neuroticism and openness (McCrae and John, 1992). Openness to Experience refers to being curious, intelligent and imaginative, artistic, sophisticated in taste and appreciates diverse views, ideas, and experiences (Golbeck et al., 2011b). The conscientious personality is characterized by self-control, reflected in a need for achievement, order, and persistence (Costa and McCrae, 1992). Conscientious individuals are extremely reliable and tend to be high achievers, hard workers, and planners (Golbeck et al., 2011a). Extraversion is related with energy and enthusiasm (John and Srivastava, 1999). Extraverts are friendly and energetic, and they draw inspiration from social situations; neuroticism refers to being anxious, insecure, sensitive, moody, tense whereas agreeableness means being cooperative, helpful, nurturing, optimistic and trusting of others (Golbeck et al., 2011a).
People who score high on agreeableness are sympathetic, good natured, cooperative and forgiving (McElroy et al., 2007). Personality is an important determinant of mobile phone use. As mobile phones mediate social interactions, the use of mobile phone reflects the individual’s personality (Butt and Phillips, 2008). Prior research has shown that personality is linked to user interface design (Lee et al., 2012), preferences for mobile advertisements, and acceptance of the adaptability dimensions of mobile systems (Graziola et al., 2005). Wolfradt and Doll (2001) found that personality traits influence the motives of media use. The relation between personality and social media use has been another area of interest for researchers. Almost all previous studies have reported the existence of a correlation between personality traits and the high or low significance of social media use (Chen and Marcus, 2012; Golbeck et al., 2011a). Implications of personality on mobile phone use for voice calls and SMS messaging constitute another area of research. Bianchi and Phillips (2005) investigated the effect of the Big Five personality traits on the smartphones’ ownership and use. Butt and Phillips (2008) carried out a study with 120 mobile phone owners to investigate the effect of Big-Five personality traits on their use of mobile phones. Chittaranjan et al. (2013) showed that personality traits significantly affected mobile phone use. The studies in the literature show that there is a strong relationship between personality features and mobile phone use.
2.
Theoretical background and hypotheses
Despite the data in the social science literature, the relationship between mobile application use and personality features is a relatively unexplored field. Chittaranjan et al. (2013) investigated this relationship by examining application logs obtained from Nokia N95 phones in the period from October 2009 to February 2011. In their study, only basic applications that are mostly preinstalled in Nokia N95 phones such as calendar, office, camera, video/audio/music, YouTube, internet, mail, chat, SMS, maps, games were examined. The relation between mobile applications and the Big Five personality traits were investigated in their study. As a brief summary of their regression tests, they found that: (1) Office app was more likely to be used by conscientious participants, (2) the Internet by introverts and disagreeable users (3) Mail app by disagreeable and conscientious users (4), the Video/Audio/Music apps by less conscientious more open participants (5), Youtube by extraverts and non-conscientious participants (6) and the Calendar app, by disagreeable participants. Tan and Yang (2014) found that individuals high in extraversion tend to more frequently use internet applications in transaction, social networking, finance, games and online friends categories. Finance applications are more likely to be used by conscientious and open users, entertainment applications are more likely to be used by users who score high in openness and transaction, social networking, games and online friends categories are more likely to be used by users scoring high in neuroticism. The results revealed by Tan and Yang (2014) should be tackled cautiously since they obtained the users’ assessments of internet applications used in web environment. Therefore, the term application in their study does not refer to mobile applications. In another study, Lane and Manner (2012) measured the importance of the six categories of smartphone applications, namely, communications, games, multimedia, productivity, travel, and utilities. The extraversion trait in users was found to be giving greater importance on games and less importance to productivity applications. For neurotic users, travel applications were more important. Users scoring low in conscientiousness indicated that communication, productivity, and utilities applications were less important to them. Recent studies explored the relationship between installed mobile applications and user attributes such as gender, religion, country and language (Seneviratne et al., 2014a, b). Using machine-learning algorithms, it was shown that user’s gender can be predicted with 70% accuracy and most personality traits with over 90% precision (Seneviratne et al., 2014a, b). It was also shown that smartphone users form clusters based on the characteristic of the apps they install. Kim et al. (2015) engaged in research on a large, diverse sample, and demonstrated that sociodemographics (e.g., gender, age, education, and income) were major predictors of smartphone and application use, but personality traits were also useful to provide additional information. They found that females tended towards a greater use of e-commerce applications, and relational applica-
tions. Extraverts were associated with decreased literacy application use and increased relational application use and conscientiousness was associated with decreased e-commerce application use. From the discussion in the literature, it is known that personality traits affect the use of mobile applications; however, the relation between personality features and users’ acceptance of recommended mobile applications has not been discussed. Thus, the following research question is posed: RQ1 Is there a relation between personality features and acceptance of recommended applications? Although they offer a promising field of research, none of the previous studies has investigated the influence strategies employed or that can be employed in the context of mobile application recommendations. Cialdini (2001) identified authority as an influence strategy, which implies that when a request or statement is made by a legitimate authority, people are more inclined to comply with it or find the information credible. Consensus is another influence strategy that stresses that people do as other people do. Headings that employ the influence strategies of consensus or authority such as the most popular, most downloaded, and editor’s choice are widely used in mobile interactions; however, in the literature, there is only limited information about the users’ perception of these influence strategies (Unal et al., 2014). The effects of influence strategies have been explored in different domains such as health promotion (Kaptein et al., 2010) and eating habits (Kaptein et al., 2012). The results show that different influence strategies are effective in modifying user behavior depending on individual differences. Therefore, the second research question is formulated as follows: RQ2 Is there a relation between personality features including user owned application number and influence strategy that is effective in users’ preferences? Applications may be recommended to the users based on the similarity of the installed applications to the applications in the market. Similarity can be either inferred by explicit information such as application categories, features, ratings, and descriptions or by implicit information that can be deduced form the installed apps on users’ phone and logs of application usage history. All applications in the mobile application markets are labeled by a category and organized under a category tree. Liu et al. (2016) developed a structural user choice model to learn user preferences by leveraging the tree hierarchy of application. In the category tree, internal nodes represent categories or subcategories and leaf nodes represent applications, and the users should first select the category which represents the functionality of the application. The procedure followed by the authors outperformed the state-of-the-art Top-N recommendation methods by a significant margin. In another study, the mobile application recommendations are analyzed to address the research question of whether mobile users prefer viewing or installing similar applications to the ones they own (Xia et al., 2013). The authors generated mobile application recommendations by analyzing both the metadata and measuring the similarity between applications in the same category, using real data on a large scale. The study showed that the mobile users are looking for applications similar to the ones that they own; users tend to install applications that are more similar for free applications whereas for paid applications, they prefer to install applications that are less similar. Having similar applications can have a two-fold consequence on the users’ decisions. The user may not download the recommended application if they have a similar one. This may be especially true for applications in the tools category, such as calendar and torch in which the features of the apps are known and well-defined. Only novel applications with user-friendly interfaces and attractive features may appeal to the user and convince them to download the recommended application. and showed that their approach can recommend highly novel serendipitous applications and reduce overpersonalization The other possible consequence is that users have a general tendency to download applications in specific categories for which the user has an interest and have similar applications as reported by Xia et al. (2013). As a result, it is expected that user responses to mobile application recommendations differ according to their needs and preferences depending on the application and its category. However, although the information about the number and category of applications owned by users can provide valuable insights into user behavior in mobile environment, they have not been discussed in previous studies. Thus, the following research questions are suggested:
RQ3 Do the personality features affect the number of user owned applications in each category? RQ4 Are people with more applications in a category more eager to download new applications due to their interest in the category or are they more likely to download more new applications since they have already downloaded similar applications? RQ5 Do applications in specific categories have an impact on users’ download behavior related to recommended applications? One of the main contributions of this study is that a user based novel experimental context is used to address the individual differences in recommended application download behavior with a specially designed mobile application that recommends applications in Google Play Store. Second, the effects of influence strategies in mobile environment are explored for the first time in the literature, which may be of help for tailored and personalized recommendations. Third, the portfolio of user-owned mobile applications, application categories, and their relationship with personality features and download behavior are explored for the first time. There is a vast demand for personalized application recommendations; therefore, we believe that this study will greatly contribute to the literature in terms of addressing the lack of knowledge regarding user profiles and preferences in mobile application recommendations. The remainder of this paper is organized as follows. In section 3 the design of the experiment and methodology is given. In section 4 the results and discussion are provided which is followed by conclusion and future work in section 5.
3.
Research methodology
The experimental study was conducted through the use of the specifically designed mobile application, on users’ mobile phones. The goals were; first, to learn about the users’ personality and the applications they downloaded to their mobile phones, second to recommend applications to users via notifications through the use of the mobile application and learn about user behavior in mobile environment. 3.1.
User context
User context is explored by collecting data on personality features such as gender, and Big Five personality traits. One of the most widely used instruments in Big Five framework is the 44-item BigFive Inventory (BFI) provided by John and Srivastava (1999). In the current study, the Turkish version of the 44 items BFI was administered based on a scale, in which 1 denotes strong disagreement and 7 denotes strong agreement. It is found that the BFI five constructs are normally distributed. Cronbach’s Alpha values for personality traits were found to be acceptable with values of 0.81 for extraversion, .67 for agreeableness, .91 for conscientiousness, .90 for neuroticism and .91 for openness to experience. 3.2.
Experimental design
An experimental design was devised to test the impact of mobile application recommendations and users’ choice in headings under which recommendations are presented. Prior to the experiment, the participants were informed by email that the purpose of the study was to recommend mobile application and measure their involvement or interest in mobile applications. In addition, users were informed that they would be required to complete two questionnaires each would take five to six minutes, and the results of the questionnaires would only be used for research and the questions would not require any responses containing personal information. The consent of the users was obtained regarding the information that will be accessed by the application that they installed. Although it was not compulsory, permission was requested to access the users’ installed applications. To increase participation the participant users were offered a choice of two gifts for their participation, either an 8 $ cinema ticket as a gratis or donation for a sapling to be planted on their behalf.
Eight applications which might be of interest to the participants were chosen from the major application categories of productivity, games, music & audio, entertainment, tools, books & references, and health & fitness. Although recommendations can be highly accurate but may be useless, e.g., suggesting bananas to customers in grocery stores (Ziegler et al., 2005) thus the novelty of the applications and their competency to address the participants’ needs and interest were the factors that were considered. Another factor was whether the applications were free or paid. For this study, only free applications were chosen to achieve consistency and comparability between applications. A pre-test was conducted to establish content validity in terms of product involvement and improve the questions, format and scales. For pre-test, a total of 10 people tested the applications and instruments in the field and their feedback was incorporated into the final revision of the application. The participants were presented with mobile application introductions on separate screens and were expected to process them one by one as they received the notifications. The order of the eight applications was the same for all users who downloaded the application. The notifications were sent in 5minute intervals and succeeding notification was not sent until the user processes the current notification. The application introduction was given in 3 or 4 sentences as presented in the summaries of application introductions in application markets such as iTunes App Store or Google Play. The arguments contained in the messages were selected by undertaking a preliminary study on mobile application recommender systems and mobile application advertisements. For each application introduction, the user was presented with four options to choose from which contained the choices available in the application markets (most popular and downloaded apps and editor’s choice apps) are presented to the user in the same order. The participants were given the choice to either select one of the recommended applications as given in three hyperlinks or not to download. Three hyperlinks were derived from reallife conditions present in the Google Play Store. An example of one of the applications, a voice recorder, is given below as an example. The screenshot of the application in Turkish is given in Figure 1.a. Voice Recorder Voice Recorder is a mobile application to record voices. You can use this application to record your classes, memos, greeting messages or other events. With 14 distinct sound effects, you can add special effects, alter the tempo and convert your recordings to different formats. You can upload your recordings to Dropbox or Google Drive and send/share them whenever you want. “Below are the three recommendations for a voice recorder app. Please choose the appropriate one for you. You can download the application to your phone by clicking the link: • • • •
From the most popular applications list: https://play.google.com/store/apps/details?id=com.andrwq.recorder From the editor’s choice list: https://play.google.com/store/apps/details?id=com.coffeebeanventures.easyvoicerecorder From the most downloaded applications list: https://play.google.com/store/apps/details?id=com.coffeebeanventures.easyvoicerecorder I do not want to install this application.”
The screenshot of the recommendations page in Turkish is given in Figure 1.b. When the participants clicked one of the first three links, they were forwarded to Google Play Store and presented with the application that was in the market with its real name and introduction page. An example screenshot is given in Figure 1.c. After examining the application in Google Play Store, they either downloaded or ended the evaluation phase for the relevant application. Whether the user downloaded the relevant application was tracked by the specifically designed mobile application used in the experiment. After the preliminary evaluation ended, the participants moved to the next stage via a notification.
1.a
1.b
1.c
Fig. 1. Screenshots of mobile application recommendation app used in experiments
3.3.
Participants
The empirical data was collected in three chunks, using a mobile application which was e-mailed to the undergraduate and graduate university student lists of a well-known university in Turkey. Of the invited students, 171 participants (91 male and 80 female) completed the experiments with the average age being 21.7. 3.4.
Statistical tests
The normality of data was checked with the Kolmogorov-Smirnov Test and Shapiro-Wilk Test using SPSS. If the data is normally distributed parametric tests are used, if not as in our case, nonparametric tests are used. The Kruskal–Wallis test and Jonckheere-Terpstra tests are non-parametric exact tests used for k independent samples. These two tests are extension of the Mann–Whitney test for more than two independent samples, so when used for two variables it gives the same results as Mann–Whitney test (Sheskin, 2003) and that is the way it is used for gender tests in this study. They are the nonparametric alternatives to the ANOVA test. The Jonckheere–Terpstra test is a variation of Kruskal–Wallis test that can be used when the variables are ordered. When the alternatives to the null hypothesis of equality of the k populations are ordered, it is appropriate to use Jonckheere-Terpstra test, which addresses our data. These tests are applicable both to data arising from nonparametric continuous univariateresponse models and to data that emerge from categorical-response models (Mehta and Patel, 2011). To use the Pearson r correlation, both variables should be normally distributed and the data must possess the properties of magnitude and equal interval between adjacent units. Kendall's tau-b is a nonparametric measure of association for ordinal data. For convenience of the data set used in the current study Kendall's tau-b correlation coefficient was applied to ordinal data. To identify univariate outliers, all of the scores for a variable are converted to standard scores. A cut-off of z ± 3.29, p < 0.001 was employed in identifying univariate outliers. The Mahalanobis distance test was used to screen multivariate outliers. All the results were reviewed by visual inspection, which revealed that most of the outliers stemmed from null values and/or giving the highest or lowest score to all questions.
4.
Results and discussion
To identify user characteristics only applications installed by the user were used. The pre-installed applications by device providers were not considered in the current study. This is in line with previous study (Seneviratne et al., 2014b) which assumes that the applications that a user has installed are potentially good indicators of their life style and interests. The categories of the applications installed by the users were identified one by one, using Google Play Store search. The 27 predefined categories of Google Play Store were used to sort and label applications where games category is an aggregate of subcategories. If the application could not be found in Google Play Store, Google was searched to find alternative application markets such as bestappsmarket.com, getandroidapp.org, and tamindir.com. If the category could not be found due to the removal of application from the market, then the instance was eliminated from the sample. All user owned applications were scanned to find whether the users already had the recommended application. The number of users that already owned the applications in different categories was; 10 in tools category; 3 in reference; 2 in games; 4 in books; 5 in entertainment; 3 in health & fitness, and 5 in productivity. These instances were eliminated from the sample. After eliminating outliers, a sample size of 158 was used for statistical tests among 171 completed. The tests based on BFI were conducted on the scores of 139 participants as participation was lower in this section. 4.1.
Personality features and user responses to application recommendations
User responses refer to the behavior of users when they are recommended mobile applications. They either ignored the recommendation or chose one of the four options, namely most popular, most downloaded, labeled as editor's choice and lastly not to install the application. When a recommendation was clicked, the next behavior of the user was whether to download the application. The descriptive statistics for personality features are given in Table 1 with their pre-processed values and normality tests are presented in Table 2. Normality tests for personality traits and the number of applications downloaded show that the data is not normally distributed. Since the assumption of normality was not satisfied, non-parametric tests were used. To determine the convenience of the data set, the Kendall's tau-b Correlation Coefficient was performed. There were 8 recommended applications; therefore, the maximum value was 8 and the minimum value was 0. The mean of number of applications recommended was 1.8, which corresponds to 22.6 % download rate of the recommended applications. In the AppJoy system, which makes personalized application recommendations based on installation and usage records, 20.6% of the recommended applications were installed by the users (Yan and Chen, 2011) which supports our results. As Table 3 depicts, conscientiousness was significantly correlated with the number of applications downloaded. Conscientious people are described by competence, achievement, self-discipline and dutifulness (Anastasi and Urbina, 1997), and these characteristics may be effective on their compliance with recommended applications. Other personality features were not found to be correlated with the number of applications downloaded. According to the test results, no difference was found between men and women in terms of the links they clicked and the number of applications they downloaded. Among the personality features, being agreeable was found to be significantly correlated with the preference for the links to the editor’s choice (Kendall's tau-b = 0.225 and p = 0.003). This may be an indication of the effect of authority figures on more agreeable individuals. This reflects the characteristics of agreeable people that refer to individuals who have cooperative values (Zhao and Seibert, 2006) and who comply with authority (Karim et al., 2009).
Table 1 Descriptive statistics for personality features and number of applications downloaded
N Min Statistic Statistic Extraversion 139 2 Agreeableness 139 3 Conscientiousness 139 2 Neuroticism 139 2 Openness 139 3 No of Apps Downl. 158 0
Max Statistic 7 7 7 6 7 8
Mean Statistic 4.53 5.02 4.68 3.68 5.10 1.81
Std. Dev. Statistic 1.040 .770 .837 1.012 .825 2.069
Skewness Statistic Std.Error -.124 .206 -.087 .206 -.046 .206 .254 .206 -.102 .206 1.201 .193
Table 2 Tests of normality for personality features and number of applications downloaded
Kolmogorov-Smirnova Statistic df Sig. .258 135 .000 .268 135 .000 .245 135 .000 .247 135 .000 .237 135 .000 .240 135 .000
Tests of Normality
Extraversion Agreeableness Conscientiousness Neuroticism Openness No of Apps Downloaded
Shapiro-Wilk Statistic df .806 135 .802 135 .808 135 .808 135 .809 135 .819 135
Sig. .000 .000 .000 .000 .000 .000
Table 3 Correlations between personality features and number of applications downloaded
Kendall's tau-b Correlations Gender (1)
Extraversion (2) Agreeableness (3)
Conscientious (4) Neuroticism (5) Openness (6)
No of Apps Downloaded (7)
4.2.
Cor. Coeff. Sig. (2-tailed) Cor. Coeff. Sig. (2-tailed) Cor. Coeff. Sig. (2-tailed) Cor. Coeff. Sig. (2-tailed) Cor. Coeff. Sig. (2-tailed) Cor. Coeff. Sig. (2-tailed) Cor. Coeff. Sig. (2-tailed)
2 .028 .689
3 -.002 .980 .193 .001
4 -.144 .042 .154 .009 .341 .000
5 .221 .002 -.325 .000 -.320 .000 -.286 .000
6 .069 .331 .189 .002 .211 .000 .096 .104 -.132 .026
7 .065 .362 .033 .598 .110 .081 .153 .015 -.065 .305 .050 .431
Number of user owned applications and user responses to application recommendations
The descriptive statistics for user responses by recommended application categories are given in Table 4 and normality tests are presented in Table 5. The normality of data was checked for all test
variables with the Kolmogorov-Smirnov Test and Shapiro-Wilk Test using SPSS. Since the assumption of normality was not satisfied, non-parametric tests were used. The Kendall’s tau-b correlation tests were conducted to measure the association between user responses and number of recommended application downloads. Table 4 shows that the number of user owned applications ranged from 8 to 118 with a mean of 41 applications per user. In the study by Liao et al. (2013), the average number of applications per user was 53 for a small dataset collected from 15 participants and 41 for the large dataset collected from 80 participants. Table 4 also shows that the most popular application link was more preferred than the links to the editor’s choice and the most downloaded apps. As shown in Table 6, it was found that the preference for the most popular apps and editor’s choice apps was significantly related to the number of applications downloaded. This result shows that people who choose most popular apps and editor’s choice apps are more likely to comply with recommended mobile applications. However, the most downloaded link option was not found to be significant for application downloads. This result is important since the same application was accessible through the most downloaded link and editor’s choice link. This indicates that in addition to the decision to visit the introduction page, the link click label is important for download decision. Another important result was that there is a negative correlation between choosing the most popular app link and the editor’s choice link. We can therefore conclude that users who prefer most popular applications are less likely to prefer editor’s choice applications, which further indicates that people complying consensus authority strategies are less likely to comply authority influence strategies. Furthermore, those having more applications are found to be more likely to choose the most downloaded app recommendation. It is found that the most downloaded link option is more preferable for users that have higher number of applications. This may indicate that users having higher number of applications believe that most downloaded applications address their interests in terms of variety and it is worth trying more applications. This may also show their curiosity about an application accessible by the link, which is not available in the application markets yet.
Table 4 Descriptive statistics for user responses and number of applications downloaded
N Min Max Mean Std. Dev. Skewness Statistic Statistic Statistic Statistic Statistic Statistic Std.Error No of User Owned Apps 158 8 118 41.367 24.040 .960 .193 No of Most Popular Links 158 0 8 1.614 2.273 1.627 .193 No of Editor's Choice Links 158 0 8 1.228 2.044 2.063 .193 No of Most Downl. Links 158 0 8 1.127 1.758 1.947 .193 No of Apps Downloaded 158 0 8 1.810 2.069 1.201 .193
Table 5 Tests of normality for user responses and number of applications downloaded
Kolmogorov-Smirnova Shapiro-Wilk StaStatistic df Sig. tistic df No of User Owned Apps .116 157 .000 .904 157 No of Most Popular Links .274 157 .000 .736 157 No of Editor's Choice Links .290 157 .000 .650 157 No of Most Downloaded Links .281 157 .000 .691 157 No of Apps Downloaded .243 157 .000 .800 157
Sig. .000 .000 .000 .000 .000
Table 6 Correlations between user responses and number of applications downloaded
Kendall's tau_b Correlations No of Apps Downloaded (1) No of User Owned Apps (2) No of Most Popular Links (3)
No of Editor's Choice Links (4) No of Most Downl. Links (5)
4.3.
Cor. Sig. Cor. Sig. Cor. Sig. Cor. Sig. Cor. Sig.
2 .040 .499
3 .248 .000 -.021 .726
4 .227 .001 .098 .107 -.134 .045
5 .093 .158 .128 .036 -.072 .280 -.016 .809
Number of user owned applications in each category
The number of user owned applications and the categories of applications for the sample of the current study are given in Figure 2. This figure shows that the applications were presented under 25 categories. Among the 27 categories defined by Google Play Store, the categories of live wallpaper and widget were not used in the current sample. Tools, games, social, communication, and productivity were found to be the most downloaded categories in a descending order. Similarly, in Google Play Store, the most downloaded categories were games, communication, tools, entertainment, and social in a descending order (App Annie, 2015).
Cartoon Libraries & Demo Medicine Weather Life style Sports Transportation News and Magazines Personalization Health and Fitness Shopping Media and Video Business Photography Finance Books & Reference Travel & Local Education Entertainment Music and Audio Productivity Communication Social Games Tools 0
100
200
300
400
500
600
700
800
900
Fig. 2. Number of user owned applications for Google Play Store categories for the sample
4.4.
Personality features and the number of user owned applications in each category
The Jonckheere–Terpstra test was implemented to analyze the personality features and the number of category based applications owned by users. This test was appropriate since there were 27 categories, where the users could have applications from zero to n number of applications in an ordered scheme. In order to compare the means, the categorical values of the personality features were used in this test. To compare the groups with considerable difference in their BFI scores, we used quartiles for categorical variables (Kaptein et al., 2010). For the BFI variables, 1 was considered ‘low’ and consisted of the first quartile of the sample, 3 referred to ‘high’ with a fourth quartile and 2 represented ‘moderate’ with the second and third quartiles. In Table 7, significant values having p < 0.05 are ranked by the absolute value of Kendall's Tau coefficient. The number of user owned applications was found to be significantly related to gender being higher for the male participants (Table 7). The mean scores for the number of user owned applications were 31 for females and 47 for males. This is in agreement with the results of previous research by Anckar and D’Incau (2002), who found that male users tended to purchase more applications than female users and they were more likely to spend more money through mobile commerce applications. Similarly, Seneviratne et al. (2014a) found that male users tended to purchase more apps than female users and were more likely to spend more money on mobile applications. When the number of applications for each category was examined, it was found that the photography category was positively correlated with female users. Media and video, sports, finance, tools, business, productivity, entertainment, travel & local, shopping, transportation, and communication categories were positively correlated with male users in a descending order of magnitude. Since men had a significantly higher number of applications than women, it can be concluded that men are more prone to download mobile applications. As Kim et al. (2015) demonstrated sociodemographics such as gender were major predictors of smartphone and application use followed by personality traits. The number of user owned mobile applications does not significantly differ for other personality features, namely, Big Five personality features. Extraversion is negatively correlated with education which is similar to the previous study by Kim et al. (2015), who found that extraverts were associated with the decreased use of literacy applications. Agreeableness is positively correlated with sports. Conscientiousness is correlated with books & reference and news & magazines categories. Similarly, Chittaranjan et al. (2013) found that office and mail applications were more likely to be used by conscientious participants. Neuroticism is negatively correlated with entertainment and media & video in our study. Chittaranjan et al. (2013) found that neuroticism was positively correlated with the use of office and calendar applications. Having negative relation with entertainment and media & video may imply positive relation with literacy applications. In our study, openness was found to be positively correlated with the social application category and negatively with sports, media & video and books & reference categories. In our previous study, we found that people who had a high level of innovativeness spent more time on communicating with others (Unal et al., 2016). It can be concluded that these people also have higher scores in openness due to their inclination to and willingness for new experience. In another study, the openness trait was found to be negatively correlated with the use of office and calendar applications (Chittaranjan et al., 2013). Tan and Yang (2014) found that entertainment applications were more likely to be used by users who scored high in openness. Positively correlation with the social application category may imply open-mindedness to new social interactions.
Table 7 Jonckheere–Terpstra Test results between personality features and number of applications by category
JT
Kendall's
p
Gender No.of Media and Video Apps Owned Sports No.of Finance Apps Owned No.of Tools Apps Owned No.of Business Apps Owned No.of Productivity Apps Owned No.of Entertainment Apps Owned No.of Travel & Local Apps Owned No.of Photography Apps Owned No.of Shopping Apps Owned No.of Transportation Apps Owned No.of Communication Apps Owned Total No.of User Owned Apps
-5.303 -4.716 -3.841 -4.050 -3.635 -3.608 -2.589 -2.366 2.339 -2.177 -2.147 -2.010 -4.094
-.392 -.355 -.278 -.277 -.266 -.248 -.183 -.171 .171 -.161 -.161 -.141 -.269
.000 .000 .000 .000 .000 .000 .010 .018 .019 .030 .032 .044 .000
Extraversion No.of Education Apps Owned
-2.076
-.150
.038
Agreeableness No.of Sports Apps Owned
2.064
.158
.039
Conscientiousness No.of Books & Reference Apps Owned No.of News and Magazines Apps Owned
2.098 1.878
.151 .141
.036 .060
Neuroticism No.of Entertainment Apps Owned No.of Media and Video Apps Owned
-2.025 -1.925
-.145 -.144
.043 .054
Openness -2.330 -.178 No.of Sports Apps Owned -2.292 -.171 No.of Media and Video Apps Owned -2.020 -.145 No.of Books & Reference Apps Owned 1.984 .139 No.of Social Apps Owned * p < 0.05, ranked by the absolute value of Kendall's tau-b coefficient
4.5.
.020 .022 .043 .047
Download of recommended applications and the number of user owned applications in each category
The Jonckheere–Terpstra test was applied to analyze the download behavior and number of category based applications owned by users. In Table 9, significant values having p < 0.05 are ranked by absolute value of Kendall's Tau coefficient.
User with applications in the recommended category are more likely to download the recommended application: Table 8 reveals that for health and fitness, books & references, music & audio, entertainment and games categories, recommended application downloads are significantly related with owning applications in the same category. For health and fitness and books & reference categories, the correlation coefficients were 0.46 and 0.293, respectively. This indicates that having an application in a category increases the probability of having other applications in the same category significantly. This is in line with the observation made by Yin et al. (2013) that some users download multiple similar applications. The number of recommended applications downloaded is positively correlated with having applications in books & reference, education, health and fitness and games categories as can be seen in Table 9. It was found that books & reference is the key category for the users’ download behavior with a correlation coefficient of 0.328. Since conscientiousness is related with the number of application downloads, it is not surprising that having applications in categories such as education and books & reference, which reflect the characteristics of conscientious people (Costa and McCrae, 1992), is significantly related with the download behavior. Table 8 Jonckheere–Terpstra Test results between download behavior and the number of applications present in the same category*
Number of Apps Downloaded in the Same Category No of Health and Fitness Apps Owned No of Books & Reference (Reference) Apps Owned No of Music and Audio Apps Owned No of Entertainment Apps Owned No of Games Apps Owned
JT
Kendall's
p
6.175 4.116 2.714 2.670 2.409
.460 .293 .192 .189 .164
.000 .000 .007 .008 .016
* p < 0.05, ranked by the absolute value of Kendall's tau-b coefficient Table 9 Jonckheere–Terpstra Test results between the number of applications downloaded and the number of applications in the same category*
JT
Kendall's
Number of Apps Downloaded 5.157 .328 No of Books & Reference Apps Owned 3.600 .231 No of Education Apps Owned 3.452 .229 No of Health and Fitness Apps Owned 2.170 .132 No of Games Apps Owned * p < 0.05, ranked by the absolute value of Kendall's tau-b coefficient
5.
p .000 .000 .001 .030
Conclusion and future work
In this study, individual differences and the effects of the Big Five personality features on the users’ compliance with the mobile application recommendations were explored. It was found that intention and download behavior towards the recommended applications differed according to personality features. Conscientiousness was found to be positively related with the tendency to accept recommended applications. In almost all application stores the mobile applications are grouped under headings that employ consensus or authority influence strategies such as the most popular, most downloaded, editor’s choice or applications of the day. Users’ perceptions of such influence strategies and underlying factors that lie beyond the users’ preferences are examined in this paper. It was shown that people having more applications are more likely to accept the recommendation of the most downloaded application. Most popular apps and editor’s choice apps were significantly related to the number of applications down-
loaded to the user’s mobile phone and being agreeable was significantly related with a preference for the editor’s choice link. Another important result was that there was a negative correlation between choosing the most popular app link and the editor’s choice link. Furthermore, it was shown that there was a relation between personality features and number of applications owned by the user in total and in each application category. The number of user owned applications was significantly higher for males. The number of applications owned in the photography category was positively correlated with females whereas the numbers of applications in the categories of media & video, sports, finance, tools, business, productivity, entertainment, travel & local, shopping, transportation, and communication were positively correlated in descending order of magnitude with male. The Big Five personality features and number of applications owned by the user in the application categories are shown to be related. Extraversion was negatively correlated with number of applications owned in the education category and agreeableness was positively correlated with sports. Conscientiousness was correlated with the categories of books & reference and news & magazines. Neuroticism was negatively correlated with entertainment and media & video. Openness was positively correlated with social application category and negatively with sports, media & video and books & reference categories. The number of applications downloaded was positively correlated with having applications in books & references, education, health and fitness and games categories. Another generalization that can be made is that having applications in the same category increases the probability of accepting more application recommendations in the same category. 5.1.
Limitations
The major limitation of the study reported in this paper was the use of a convenience sample from graduate and undergraduate university students since it does not completely represent the general consumer population who are likely to be less familiar with mobile applications. The students were from one of the top universities in Turkey, which implies that the participant group has a high level of literacy and innovativeness. In the experimental study on a mobile platform, log data was used to extract user profiles through the list of installed applications; however, this was only a snapshot of the applications the participants owned. This was not longitudinal research following users after downloading and during their use of the recommended applications over time. Therefore, the usage time of applications, the frequency of application use and uninstall behavior that occurred after the download of the applications, were not fully analyzed in the current research. Another limitation of our study is that due to the novel extensions in scope, some of the findings of our study could not directly be compared with the findings in the literature. It will be beneficial to conduct qualitative research to explore the findings of this study. 5.2.
Future work
The main contribution of this paper is to ground the recommender systems’ design in an understanding of users’ needs and to provide personalization means to better address users’ individual differences. It can also be used to tailor for specific needs and specific populations and make predictions based on the available data. Mobile technology products should avoid the one-size-fits-all approach to meet different needs of individuals. The findings of this study provide valuable information to improve personalization and recommendation services of mobile applications. People with similar personality features may be used for identifying and clustering people with similar tastes and preferences. Understanding the patterns of mobile phone use is significant in revealing user preferences in order to improve features of mobile phones and personalized mobile services. It can also help tailor specific needs and particular populations and make predictions based on the available data. The identification of users who are innovators, early-adopters, leader users or have a large number of friends is also important to determine the primary segment that should be targeted by marketing practitioners. The findings in this study provide useful insights for mobile technology designers, especially those planning for specific populations. Stakeholders including phone operators, service providers, application developers and advertising compa-
nies can benefit from the results to further explore the behaviors of mobile technology users from the perspective of consumers as well as other providers and advertisers. Although the findings of this study are interesting and significant, there are still other areas for future exploration. Different personality features such as trust and privacy concerns, well-being, mood and social network characteristics can be further analyzed that may contribute to an understanding of user preferences regarding the use of mobile phones and applications. It is shown that mobile application usage is denominated by contextual factors such as location, time of the day, activity and people around which is usually determined by the Bluetooth feature. Most probably, the users’ acceptance of mobile application recommendations depends on the contextual environment. Further research can tackle the effects of contextual factors on mobile phone and application use and recommendation acceptance. Using contextual information like time and location has the potential to be used in offering recommendations to new users who do not yet have a profile or history, hence addressing the coldstart problem.
References App Annie, 2015. App Annie Index - Market Q1 2014: Revenue soars in the United States and China. From
Liao, Z.-X., Pan, Y.-C., Peng, W.-C., Lei, P.-R., 2013. On mining mobile apps usage behavior for predicting apps usage in smartphones, in: Proceedings of the 22nd ACM International Conference on Information & Knowledge Management. ACM, New York, pp. 609-618. Liu, B., Wu, Y., Gong, N.Z., Wu, J., Xiong, H., Ester, M., 2016. Structural analysis of user choices for mobile app recommendation. ACM Trans. Knowl. Discov. Data 11(2), Article 17. doi:10.1145/2983533. Mehta, C.R., Patel, N.R., 2011. IBM SPSS exact tests. IBM Corporation, Cambridge, MA. McCrae, R.R., John, O.P., 1992. An introduction to the Five-Factor model and its applications. Journal of Personality 60, 175-215. McElroy, J.C., Hendrickson, A.R., Townsend, A.M., DeMarie, S.M., 2007. Dispositional factors in internet use: personality versus cognitive style. MIS Quarterly 31, 809-820. Seneviratne, S., Seneviratne, A., Mohapatra, P., Mahanti, A., 2014a. Predicting user traits from a snapshot of apps installed on a smartphone. ACM SIGMOBILE Mobile Computing and Communications Review 18, 1-8. Seneviratne, S., Seneviratne, A., Mohapatra, P., Mahanti, A., 2014b. Your installed apps reveal your gender and more!, in: ACM MobiCom Workshop on Security and Privacy in Mobile Environments (SPME '14), ACM. Sheskin, D.J., 2003. Handbook of parametric and nonparametric statistical procedures, 5th edition. Chapman and Hall/CRC. Tan, W.K., Yang, C.Y., 2014. Internet applications use and personality. Telematics and Informatics 31, 27-38. Unal, P., Temizel, T.T., Eren, P.E., 2014. An exploratory study on the outcomes of influence strategies in mobile application recommendations, in: 2nd International Workshop on Behavior Change Support Systems (BCSS 2014), pp. 27-40. Unal, P., Temizel, T.T., Eren, P.E., 2016. Impact of ındividual differences on the use of mobile phones and applications, in: International Conference on Mobile Web and Information Systems. Springer International Publishing, pp. 379-392. Wolfradt, U., Doll, J., 2001. Motives of adolescents to use the internet as a function of personality traits, personal and social factors. Journal of Educational Computing Research 24(1), 13-27. doi: 10.2190/ANPM-LN97-AUT2-D2EJ. Xia X., Wang X., Zhou X., 2013. Evolving recommender system for mobile apps: A diversity measurement approach. The Smart Computing Review 3, 139-154. Yan, B., Chen, G., 2011. AppJoy: Personalized mobile application discovery, in: Proceedings of the 9th International Conference on Mobile Systems, Applications, and Services (MobiSys). ACM, New York, pp. 113-126. Yin, P., Luo, P., Lee, W.-C., Wang, M., 2013. App recommendation: a contest between satisfaction and temptation, in: Proceedings of the 6th International Conference on Web Search and Data Mining. ACM, New York, pp. 395-404. Zaichkowsky, J.L., 1985. Measuring the involvement construct. Journal of Consumer Marketing Research 12, 341-352. Zhao, H., Seibert, S.E., 2006. The Big Five personality dimensions and entrepreneurial status: A meta-analytical review. Journal of Applied Psychology 91, 259-271. Ziegler, C.N., McNee, S.M., Konstan, J.A., Lausen, G., 2005. Improving recommendation lists through topic diversification, in: Proceedings of the 14th İnternational Conference on World Wide Web. ACM, New York, pp. 22-32.
Highlights Number of user owned apps and their category differ with gender and personality Having similar apps increases the probability of accepting recommended applications Number of apps owned in some categories implies higher acceptance of recommended apps Conscientiousness is positively related with accepting recommended applications Being agreeable is related with editor’s choice application preference
S0736-5853(17)30049-7 http://dx.doi.org/10.1016/j.tele.2017.05.005 TELE 948
To appear in:
Telematics and Informatics
Received Date: Accepted Date:
10 April 2017 10 May 2017
Please cite this article as: Unal, P., Temizel, T.T., Eren, P.E., What installed mobile applications tell about their owners and how they affect users’ download behavior, Telematics and Informatics (2017), doi: http://dx.doi.org/ 10.1016/j.tele.2017.05.005
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
May, 11, 2017
What installed mobile applications tell about their owners and how they affect users’ download behavior Perin Unal 1, Tugba Taskaya Temizel1, P. Erhan Eren1 1
Department of Information Systems, Informatics Institute, Middle East Technical University, Ankara, Turkey
Corresponding Author: Perin UNAL Department of Information Systems Informatics Institute, Middle East Technical University Universiteler Mahallesi, Dumlupınar Bulvarı, No:1, 06800 Çankaya/Ankara/TURKEY Mobile Phone: +90 532 4155783 Fax Number: +90 312 5921848 E-mail: [email protected]
What installed mobile applications tell about their owners and how they affect users’ download behavior Abstract. The rapid growth in the mobile application market presents a significant challenge to find interesting and relevant applications for users. An experimental study was conducted through the use of a specifically designed mobile application, on users’ mobile phones. The goals were; first, to learn about the users’ personality and the applications they downloaded to their mobile phones, second to recommend applications to users via notifications through the use of experimental mobile application and learn about user behavior in mobile environment. The question of how the personality features of users affect their compliance to recommendations is explored in this study. It is found that conscientiousness is positively related with accepting recommended applications and being agreeable is related with the preference for the applications of editor’s choice. Furthermore, in this study, applications owned by the user and the composition of applications under categories and their relation with personality features are explored. It is shown that the number of user owned applications and their category differ according to gender and personality. Having similar applications and the number of applications owned under specific categories increase the probability of accepting recommended applications. Keywords. Mobile phone use, mobile application use, user profiling, mobile application recommendation, recommender systems
1.
Introduction
Mobile users predominantly prefer to use mobile applications rather than browsers to access internet services. Application markets have grown rapidly as a result of vesting user interest in mobile applications. Mobile application recommendation websites and services fulfill the growing need to filter, rank and recommend the best applications from the hundreds of thousands available. Some of these sites operate in the official application marketplaces such as the Genius of iTunes App Store and the recommendations in Google Play and others on marketplaces like Amazon Appstore and Yandex. In all marketplaces, implementations of consensus and authority influence strategies are predominantly used on the basis of the most popular applications, most downloaded ones, editor’s choice, applications of the day and similar ones. Recommender systems that are used for mobile applications may make use of user profiles. User’s gender, personality features, the number of applications owned by the user and the composition of the applications under categories can be used as a source for understanding user profiles and preferences. This information can be processed and integrated into the recommender systems to address users’ needs and preferences and build personalized relevant outputs. Understanding user preferences is a complex issue and has been the subject of numerous studies in the social and psychological sciences. Personality has fundamental dimensions referred to as traits and it has been shown that people’s behavior can be explained by their underlying personality traits, which reflect the enduring dispositions of their nature (Costa and McCrae, 1992). There are various methodologies in the literature that identify and classify the personality traits but the most widely used and accepted model is the Big Five personality framework which defines the traits as extraversion, agreeableness, conscientiousness, neuroticism and openness (McCrae and John, 1992). Openness to Experience refers to being curious, intelligent and imaginative, artistic, sophisticated in taste and appreciates diverse views, ideas, and experiences (Golbeck et al., 2011b). The conscientious personality is characterized by self-control, reflected in a need for achievement, order, and persistence (Costa and McCrae, 1992). Conscientious individuals are extremely reliable and tend to be high achievers, hard workers, and planners (Golbeck et al., 2011a). Extraversion is related with energy and enthusiasm (John and Srivastava, 1999). Extraverts are friendly and energetic, and they draw inspiration from social situations; neuroticism refers to being anxious, insecure, sensitive, moody, tense whereas agreeableness means being cooperative, helpful, nurturing, optimistic and trusting of others (Golbeck et al., 2011a).
People who score high on agreeableness are sympathetic, good natured, cooperative and forgiving (McElroy et al., 2007). Personality is an important determinant of mobile phone use. As mobile phones mediate social interactions, the use of mobile phone reflects the individual’s personality (Butt and Phillips, 2008). Prior research has shown that personality is linked to user interface design (Lee et al., 2012), preferences for mobile advertisements, and acceptance of the adaptability dimensions of mobile systems (Graziola et al., 2005). Wolfradt and Doll (2001) found that personality traits influence the motives of media use. The relation between personality and social media use has been another area of interest for researchers. Almost all previous studies have reported the existence of a correlation between personality traits and the high or low significance of social media use (Chen and Marcus, 2012; Golbeck et al., 2011a). Implications of personality on mobile phone use for voice calls and SMS messaging constitute another area of research. Bianchi and Phillips (2005) investigated the effect of the Big Five personality traits on the smartphones’ ownership and use. Butt and Phillips (2008) carried out a study with 120 mobile phone owners to investigate the effect of Big-Five personality traits on their use of mobile phones. Chittaranjan et al. (2013) showed that personality traits significantly affected mobile phone use. The studies in the literature show that there is a strong relationship between personality features and mobile phone use.
2.
Theoretical background and hypotheses
Despite the data in the social science literature, the relationship between mobile application use and personality features is a relatively unexplored field. Chittaranjan et al. (2013) investigated this relationship by examining application logs obtained from Nokia N95 phones in the period from October 2009 to February 2011. In their study, only basic applications that are mostly preinstalled in Nokia N95 phones such as calendar, office, camera, video/audio/music, YouTube, internet, mail, chat, SMS, maps, games were examined. The relation between mobile applications and the Big Five personality traits were investigated in their study. As a brief summary of their regression tests, they found that: (1) Office app was more likely to be used by conscientious participants, (2) the Internet by introverts and disagreeable users (3) Mail app by disagreeable and conscientious users (4), the Video/Audio/Music apps by less conscientious more open participants (5), Youtube by extraverts and non-conscientious participants (6) and the Calendar app, by disagreeable participants. Tan and Yang (2014) found that individuals high in extraversion tend to more frequently use internet applications in transaction, social networking, finance, games and online friends categories. Finance applications are more likely to be used by conscientious and open users, entertainment applications are more likely to be used by users who score high in openness and transaction, social networking, games and online friends categories are more likely to be used by users scoring high in neuroticism. The results revealed by Tan and Yang (2014) should be tackled cautiously since they obtained the users’ assessments of internet applications used in web environment. Therefore, the term application in their study does not refer to mobile applications. In another study, Lane and Manner (2012) measured the importance of the six categories of smartphone applications, namely, communications, games, multimedia, productivity, travel, and utilities. The extraversion trait in users was found to be giving greater importance on games and less importance to productivity applications. For neurotic users, travel applications were more important. Users scoring low in conscientiousness indicated that communication, productivity, and utilities applications were less important to them. Recent studies explored the relationship between installed mobile applications and user attributes such as gender, religion, country and language (Seneviratne et al., 2014a, b). Using machine-learning algorithms, it was shown that user’s gender can be predicted with 70% accuracy and most personality traits with over 90% precision (Seneviratne et al., 2014a, b). It was also shown that smartphone users form clusters based on the characteristic of the apps they install. Kim et al. (2015) engaged in research on a large, diverse sample, and demonstrated that sociodemographics (e.g., gender, age, education, and income) were major predictors of smartphone and application use, but personality traits were also useful to provide additional information. They found that females tended towards a greater use of e-commerce applications, and relational applica-
tions. Extraverts were associated with decreased literacy application use and increased relational application use and conscientiousness was associated with decreased e-commerce application use. From the discussion in the literature, it is known that personality traits affect the use of mobile applications; however, the relation between personality features and users’ acceptance of recommended mobile applications has not been discussed. Thus, the following research question is posed: RQ1 Is there a relation between personality features and acceptance of recommended applications? Although they offer a promising field of research, none of the previous studies has investigated the influence strategies employed or that can be employed in the context of mobile application recommendations. Cialdini (2001) identified authority as an influence strategy, which implies that when a request or statement is made by a legitimate authority, people are more inclined to comply with it or find the information credible. Consensus is another influence strategy that stresses that people do as other people do. Headings that employ the influence strategies of consensus or authority such as the most popular, most downloaded, and editor’s choice are widely used in mobile interactions; however, in the literature, there is only limited information about the users’ perception of these influence strategies (Unal et al., 2014). The effects of influence strategies have been explored in different domains such as health promotion (Kaptein et al., 2010) and eating habits (Kaptein et al., 2012). The results show that different influence strategies are effective in modifying user behavior depending on individual differences. Therefore, the second research question is formulated as follows: RQ2 Is there a relation between personality features including user owned application number and influence strategy that is effective in users’ preferences? Applications may be recommended to the users based on the similarity of the installed applications to the applications in the market. Similarity can be either inferred by explicit information such as application categories, features, ratings, and descriptions or by implicit information that can be deduced form the installed apps on users’ phone and logs of application usage history. All applications in the mobile application markets are labeled by a category and organized under a category tree. Liu et al. (2016) developed a structural user choice model to learn user preferences by leveraging the tree hierarchy of application. In the category tree, internal nodes represent categories or subcategories and leaf nodes represent applications, and the users should first select the category which represents the functionality of the application. The procedure followed by the authors outperformed the state-of-the-art Top-N recommendation methods by a significant margin. In another study, the mobile application recommendations are analyzed to address the research question of whether mobile users prefer viewing or installing similar applications to the ones they own (Xia et al., 2013). The authors generated mobile application recommendations by analyzing both the metadata and measuring the similarity between applications in the same category, using real data on a large scale. The study showed that the mobile users are looking for applications similar to the ones that they own; users tend to install applications that are more similar for free applications whereas for paid applications, they prefer to install applications that are less similar. Having similar applications can have a two-fold consequence on the users’ decisions. The user may not download the recommended application if they have a similar one. This may be especially true for applications in the tools category, such as calendar and torch in which the features of the apps are known and well-defined. Only novel applications with user-friendly interfaces and attractive features may appeal to the user and convince them to download the recommended application. and showed that their approach can recommend highly novel serendipitous applications and reduce overpersonalization The other possible consequence is that users have a general tendency to download applications in specific categories for which the user has an interest and have similar applications as reported by Xia et al. (2013). As a result, it is expected that user responses to mobile application recommendations differ according to their needs and preferences depending on the application and its category. However, although the information about the number and category of applications owned by users can provide valuable insights into user behavior in mobile environment, they have not been discussed in previous studies. Thus, the following research questions are suggested:
RQ3 Do the personality features affect the number of user owned applications in each category? RQ4 Are people with more applications in a category more eager to download new applications due to their interest in the category or are they more likely to download more new applications since they have already downloaded similar applications? RQ5 Do applications in specific categories have an impact on users’ download behavior related to recommended applications? One of the main contributions of this study is that a user based novel experimental context is used to address the individual differences in recommended application download behavior with a specially designed mobile application that recommends applications in Google Play Store. Second, the effects of influence strategies in mobile environment are explored for the first time in the literature, which may be of help for tailored and personalized recommendations. Third, the portfolio of user-owned mobile applications, application categories, and their relationship with personality features and download behavior are explored for the first time. There is a vast demand for personalized application recommendations; therefore, we believe that this study will greatly contribute to the literature in terms of addressing the lack of knowledge regarding user profiles and preferences in mobile application recommendations. The remainder of this paper is organized as follows. In section 3 the design of the experiment and methodology is given. In section 4 the results and discussion are provided which is followed by conclusion and future work in section 5.
3.
Research methodology
The experimental study was conducted through the use of the specifically designed mobile application, on users’ mobile phones. The goals were; first, to learn about the users’ personality and the applications they downloaded to their mobile phones, second to recommend applications to users via notifications through the use of the mobile application and learn about user behavior in mobile environment. 3.1.
User context
User context is explored by collecting data on personality features such as gender, and Big Five personality traits. One of the most widely used instruments in Big Five framework is the 44-item BigFive Inventory (BFI) provided by John and Srivastava (1999). In the current study, the Turkish version of the 44 items BFI was administered based on a scale, in which 1 denotes strong disagreement and 7 denotes strong agreement. It is found that the BFI five constructs are normally distributed. Cronbach’s Alpha values for personality traits were found to be acceptable with values of 0.81 for extraversion, .67 for agreeableness, .91 for conscientiousness, .90 for neuroticism and .91 for openness to experience. 3.2.
Experimental design
An experimental design was devised to test the impact of mobile application recommendations and users’ choice in headings under which recommendations are presented. Prior to the experiment, the participants were informed by email that the purpose of the study was to recommend mobile application and measure their involvement or interest in mobile applications. In addition, users were informed that they would be required to complete two questionnaires each would take five to six minutes, and the results of the questionnaires would only be used for research and the questions would not require any responses containing personal information. The consent of the users was obtained regarding the information that will be accessed by the application that they installed. Although it was not compulsory, permission was requested to access the users’ installed applications. To increase participation the participant users were offered a choice of two gifts for their participation, either an 8 $ cinema ticket as a gratis or donation for a sapling to be planted on their behalf.
Eight applications which might be of interest to the participants were chosen from the major application categories of productivity, games, music & audio, entertainment, tools, books & references, and health & fitness. Although recommendations can be highly accurate but may be useless, e.g., suggesting bananas to customers in grocery stores (Ziegler et al., 2005) thus the novelty of the applications and their competency to address the participants’ needs and interest were the factors that were considered. Another factor was whether the applications were free or paid. For this study, only free applications were chosen to achieve consistency and comparability between applications. A pre-test was conducted to establish content validity in terms of product involvement and improve the questions, format and scales. For pre-test, a total of 10 people tested the applications and instruments in the field and their feedback was incorporated into the final revision of the application. The participants were presented with mobile application introductions on separate screens and were expected to process them one by one as they received the notifications. The order of the eight applications was the same for all users who downloaded the application. The notifications were sent in 5minute intervals and succeeding notification was not sent until the user processes the current notification. The application introduction was given in 3 or 4 sentences as presented in the summaries of application introductions in application markets such as iTunes App Store or Google Play. The arguments contained in the messages were selected by undertaking a preliminary study on mobile application recommender systems and mobile application advertisements. For each application introduction, the user was presented with four options to choose from which contained the choices available in the application markets (most popular and downloaded apps and editor’s choice apps) are presented to the user in the same order. The participants were given the choice to either select one of the recommended applications as given in three hyperlinks or not to download. Three hyperlinks were derived from reallife conditions present in the Google Play Store. An example of one of the applications, a voice recorder, is given below as an example. The screenshot of the application in Turkish is given in Figure 1.a. Voice Recorder Voice Recorder is a mobile application to record voices. You can use this application to record your classes, memos, greeting messages or other events. With 14 distinct sound effects, you can add special effects, alter the tempo and convert your recordings to different formats. You can upload your recordings to Dropbox or Google Drive and send/share them whenever you want. “Below are the three recommendations for a voice recorder app. Please choose the appropriate one for you. You can download the application to your phone by clicking the link: • • • •
From the most popular applications list: https://play.google.com/store/apps/details?id=com.andrwq.recorder From the editor’s choice list: https://play.google.com/store/apps/details?id=com.coffeebeanventures.easyvoicerecorder From the most downloaded applications list: https://play.google.com/store/apps/details?id=com.coffeebeanventures.easyvoicerecorder I do not want to install this application.”
The screenshot of the recommendations page in Turkish is given in Figure 1.b. When the participants clicked one of the first three links, they were forwarded to Google Play Store and presented with the application that was in the market with its real name and introduction page. An example screenshot is given in Figure 1.c. After examining the application in Google Play Store, they either downloaded or ended the evaluation phase for the relevant application. Whether the user downloaded the relevant application was tracked by the specifically designed mobile application used in the experiment. After the preliminary evaluation ended, the participants moved to the next stage via a notification.
1.a
1.b
1.c
Fig. 1. Screenshots of mobile application recommendation app used in experiments
3.3.
Participants
The empirical data was collected in three chunks, using a mobile application which was e-mailed to the undergraduate and graduate university student lists of a well-known university in Turkey. Of the invited students, 171 participants (91 male and 80 female) completed the experiments with the average age being 21.7. 3.4.
Statistical tests
The normality of data was checked with the Kolmogorov-Smirnov Test and Shapiro-Wilk Test using SPSS. If the data is normally distributed parametric tests are used, if not as in our case, nonparametric tests are used. The Kruskal–Wallis test and Jonckheere-Terpstra tests are non-parametric exact tests used for k independent samples. These two tests are extension of the Mann–Whitney test for more than two independent samples, so when used for two variables it gives the same results as Mann–Whitney test (Sheskin, 2003) and that is the way it is used for gender tests in this study. They are the nonparametric alternatives to the ANOVA test. The Jonckheere–Terpstra test is a variation of Kruskal–Wallis test that can be used when the variables are ordered. When the alternatives to the null hypothesis of equality of the k populations are ordered, it is appropriate to use Jonckheere-Terpstra test, which addresses our data. These tests are applicable both to data arising from nonparametric continuous univariateresponse models and to data that emerge from categorical-response models (Mehta and Patel, 2011). To use the Pearson r correlation, both variables should be normally distributed and the data must possess the properties of magnitude and equal interval between adjacent units. Kendall's tau-b is a nonparametric measure of association for ordinal data. For convenience of the data set used in the current study Kendall's tau-b correlation coefficient was applied to ordinal data. To identify univariate outliers, all of the scores for a variable are converted to standard scores. A cut-off of z ± 3.29, p < 0.001 was employed in identifying univariate outliers. The Mahalanobis distance test was used to screen multivariate outliers. All the results were reviewed by visual inspection, which revealed that most of the outliers stemmed from null values and/or giving the highest or lowest score to all questions.
4.
Results and discussion
To identify user characteristics only applications installed by the user were used. The pre-installed applications by device providers were not considered in the current study. This is in line with previous study (Seneviratne et al., 2014b) which assumes that the applications that a user has installed are potentially good indicators of their life style and interests. The categories of the applications installed by the users were identified one by one, using Google Play Store search. The 27 predefined categories of Google Play Store were used to sort and label applications where games category is an aggregate of subcategories. If the application could not be found in Google Play Store, Google was searched to find alternative application markets such as bestappsmarket.com, getandroidapp.org, and tamindir.com. If the category could not be found due to the removal of application from the market, then the instance was eliminated from the sample. All user owned applications were scanned to find whether the users already had the recommended application. The number of users that already owned the applications in different categories was; 10 in tools category; 3 in reference; 2 in games; 4 in books; 5 in entertainment; 3 in health & fitness, and 5 in productivity. These instances were eliminated from the sample. After eliminating outliers, a sample size of 158 was used for statistical tests among 171 completed. The tests based on BFI were conducted on the scores of 139 participants as participation was lower in this section. 4.1.
Personality features and user responses to application recommendations
User responses refer to the behavior of users when they are recommended mobile applications. They either ignored the recommendation or chose one of the four options, namely most popular, most downloaded, labeled as editor's choice and lastly not to install the application. When a recommendation was clicked, the next behavior of the user was whether to download the application. The descriptive statistics for personality features are given in Table 1 with their pre-processed values and normality tests are presented in Table 2. Normality tests for personality traits and the number of applications downloaded show that the data is not normally distributed. Since the assumption of normality was not satisfied, non-parametric tests were used. To determine the convenience of the data set, the Kendall's tau-b Correlation Coefficient was performed. There were 8 recommended applications; therefore, the maximum value was 8 and the minimum value was 0. The mean of number of applications recommended was 1.8, which corresponds to 22.6 % download rate of the recommended applications. In the AppJoy system, which makes personalized application recommendations based on installation and usage records, 20.6% of the recommended applications were installed by the users (Yan and Chen, 2011) which supports our results. As Table 3 depicts, conscientiousness was significantly correlated with the number of applications downloaded. Conscientious people are described by competence, achievement, self-discipline and dutifulness (Anastasi and Urbina, 1997), and these characteristics may be effective on their compliance with recommended applications. Other personality features were not found to be correlated with the number of applications downloaded. According to the test results, no difference was found between men and women in terms of the links they clicked and the number of applications they downloaded. Among the personality features, being agreeable was found to be significantly correlated with the preference for the links to the editor’s choice (Kendall's tau-b = 0.225 and p = 0.003). This may be an indication of the effect of authority figures on more agreeable individuals. This reflects the characteristics of agreeable people that refer to individuals who have cooperative values (Zhao and Seibert, 2006) and who comply with authority (Karim et al., 2009).
Table 1 Descriptive statistics for personality features and number of applications downloaded
N Min Statistic Statistic Extraversion 139 2 Agreeableness 139 3 Conscientiousness 139 2 Neuroticism 139 2 Openness 139 3 No of Apps Downl. 158 0
Max Statistic 7 7 7 6 7 8
Mean Statistic 4.53 5.02 4.68 3.68 5.10 1.81
Std. Dev. Statistic 1.040 .770 .837 1.012 .825 2.069
Skewness Statistic Std.Error -.124 .206 -.087 .206 -.046 .206 .254 .206 -.102 .206 1.201 .193
Table 2 Tests of normality for personality features and number of applications downloaded
Kolmogorov-Smirnova Statistic df Sig. .258 135 .000 .268 135 .000 .245 135 .000 .247 135 .000 .237 135 .000 .240 135 .000
Tests of Normality
Extraversion Agreeableness Conscientiousness Neuroticism Openness No of Apps Downloaded
Shapiro-Wilk Statistic df .806 135 .802 135 .808 135 .808 135 .809 135 .819 135
Sig. .000 .000 .000 .000 .000 .000
Table 3 Correlations between personality features and number of applications downloaded
Kendall's tau-b Correlations Gender (1)
Extraversion (2) Agreeableness (3)
Conscientious (4) Neuroticism (5) Openness (6)
No of Apps Downloaded (7)
4.2.
Cor. Coeff. Sig. (2-tailed) Cor. Coeff. Sig. (2-tailed) Cor. Coeff. Sig. (2-tailed) Cor. Coeff. Sig. (2-tailed) Cor. Coeff. Sig. (2-tailed) Cor. Coeff. Sig. (2-tailed) Cor. Coeff. Sig. (2-tailed)
2 .028 .689
3 -.002 .980 .193 .001
4 -.144 .042 .154 .009 .341 .000
5 .221 .002 -.325 .000 -.320 .000 -.286 .000
6 .069 .331 .189 .002 .211 .000 .096 .104 -.132 .026
7 .065 .362 .033 .598 .110 .081 .153 .015 -.065 .305 .050 .431
Number of user owned applications and user responses to application recommendations
The descriptive statistics for user responses by recommended application categories are given in Table 4 and normality tests are presented in Table 5. The normality of data was checked for all test
variables with the Kolmogorov-Smirnov Test and Shapiro-Wilk Test using SPSS. Since the assumption of normality was not satisfied, non-parametric tests were used. The Kendall’s tau-b correlation tests were conducted to measure the association between user responses and number of recommended application downloads. Table 4 shows that the number of user owned applications ranged from 8 to 118 with a mean of 41 applications per user. In the study by Liao et al. (2013), the average number of applications per user was 53 for a small dataset collected from 15 participants and 41 for the large dataset collected from 80 participants. Table 4 also shows that the most popular application link was more preferred than the links to the editor’s choice and the most downloaded apps. As shown in Table 6, it was found that the preference for the most popular apps and editor’s choice apps was significantly related to the number of applications downloaded. This result shows that people who choose most popular apps and editor’s choice apps are more likely to comply with recommended mobile applications. However, the most downloaded link option was not found to be significant for application downloads. This result is important since the same application was accessible through the most downloaded link and editor’s choice link. This indicates that in addition to the decision to visit the introduction page, the link click label is important for download decision. Another important result was that there is a negative correlation between choosing the most popular app link and the editor’s choice link. We can therefore conclude that users who prefer most popular applications are less likely to prefer editor’s choice applications, which further indicates that people complying consensus authority strategies are less likely to comply authority influence strategies. Furthermore, those having more applications are found to be more likely to choose the most downloaded app recommendation. It is found that the most downloaded link option is more preferable for users that have higher number of applications. This may indicate that users having higher number of applications believe that most downloaded applications address their interests in terms of variety and it is worth trying more applications. This may also show their curiosity about an application accessible by the link, which is not available in the application markets yet.
Table 4 Descriptive statistics for user responses and number of applications downloaded
N Min Max Mean Std. Dev. Skewness Statistic Statistic Statistic Statistic Statistic Statistic Std.Error No of User Owned Apps 158 8 118 41.367 24.040 .960 .193 No of Most Popular Links 158 0 8 1.614 2.273 1.627 .193 No of Editor's Choice Links 158 0 8 1.228 2.044 2.063 .193 No of Most Downl. Links 158 0 8 1.127 1.758 1.947 .193 No of Apps Downloaded 158 0 8 1.810 2.069 1.201 .193
Table 5 Tests of normality for user responses and number of applications downloaded
Kolmogorov-Smirnova Shapiro-Wilk StaStatistic df Sig. tistic df No of User Owned Apps .116 157 .000 .904 157 No of Most Popular Links .274 157 .000 .736 157 No of Editor's Choice Links .290 157 .000 .650 157 No of Most Downloaded Links .281 157 .000 .691 157 No of Apps Downloaded .243 157 .000 .800 157
Sig. .000 .000 .000 .000 .000
Table 6 Correlations between user responses and number of applications downloaded
Kendall's tau_b Correlations No of Apps Downloaded (1) No of User Owned Apps (2) No of Most Popular Links (3)
No of Editor's Choice Links (4) No of Most Downl. Links (5)
4.3.
Cor. Sig. Cor. Sig. Cor. Sig. Cor. Sig. Cor. Sig.
2 .040 .499
3 .248 .000 -.021 .726
4 .227 .001 .098 .107 -.134 .045
5 .093 .158 .128 .036 -.072 .280 -.016 .809
Number of user owned applications in each category
The number of user owned applications and the categories of applications for the sample of the current study are given in Figure 2. This figure shows that the applications were presented under 25 categories. Among the 27 categories defined by Google Play Store, the categories of live wallpaper and widget were not used in the current sample. Tools, games, social, communication, and productivity were found to be the most downloaded categories in a descending order. Similarly, in Google Play Store, the most downloaded categories were games, communication, tools, entertainment, and social in a descending order (App Annie, 2015).
Cartoon Libraries & Demo Medicine Weather Life style Sports Transportation News and Magazines Personalization Health and Fitness Shopping Media and Video Business Photography Finance Books & Reference Travel & Local Education Entertainment Music and Audio Productivity Communication Social Games Tools 0
100
200
300
400
500
600
700
800
900
Fig. 2. Number of user owned applications for Google Play Store categories for the sample
4.4.
Personality features and the number of user owned applications in each category
The Jonckheere–Terpstra test was implemented to analyze the personality features and the number of category based applications owned by users. This test was appropriate since there were 27 categories, where the users could have applications from zero to n number of applications in an ordered scheme. In order to compare the means, the categorical values of the personality features were used in this test. To compare the groups with considerable difference in their BFI scores, we used quartiles for categorical variables (Kaptein et al., 2010). For the BFI variables, 1 was considered ‘low’ and consisted of the first quartile of the sample, 3 referred to ‘high’ with a fourth quartile and 2 represented ‘moderate’ with the second and third quartiles. In Table 7, significant values having p < 0.05 are ranked by the absolute value of Kendall's Tau coefficient. The number of user owned applications was found to be significantly related to gender being higher for the male participants (Table 7). The mean scores for the number of user owned applications were 31 for females and 47 for males. This is in agreement with the results of previous research by Anckar and D’Incau (2002), who found that male users tended to purchase more applications than female users and they were more likely to spend more money through mobile commerce applications. Similarly, Seneviratne et al. (2014a) found that male users tended to purchase more apps than female users and were more likely to spend more money on mobile applications. When the number of applications for each category was examined, it was found that the photography category was positively correlated with female users. Media and video, sports, finance, tools, business, productivity, entertainment, travel & local, shopping, transportation, and communication categories were positively correlated with male users in a descending order of magnitude. Since men had a significantly higher number of applications than women, it can be concluded that men are more prone to download mobile applications. As Kim et al. (2015) demonstrated sociodemographics such as gender were major predictors of smartphone and application use followed by personality traits. The number of user owned mobile applications does not significantly differ for other personality features, namely, Big Five personality features. Extraversion is negatively correlated with education which is similar to the previous study by Kim et al. (2015), who found that extraverts were associated with the decreased use of literacy applications. Agreeableness is positively correlated with sports. Conscientiousness is correlated with books & reference and news & magazines categories. Similarly, Chittaranjan et al. (2013) found that office and mail applications were more likely to be used by conscientious participants. Neuroticism is negatively correlated with entertainment and media & video in our study. Chittaranjan et al. (2013) found that neuroticism was positively correlated with the use of office and calendar applications. Having negative relation with entertainment and media & video may imply positive relation with literacy applications. In our study, openness was found to be positively correlated with the social application category and negatively with sports, media & video and books & reference categories. In our previous study, we found that people who had a high level of innovativeness spent more time on communicating with others (Unal et al., 2016). It can be concluded that these people also have higher scores in openness due to their inclination to and willingness for new experience. In another study, the openness trait was found to be negatively correlated with the use of office and calendar applications (Chittaranjan et al., 2013). Tan and Yang (2014) found that entertainment applications were more likely to be used by users who scored high in openness. Positively correlation with the social application category may imply open-mindedness to new social interactions.
Table 7 Jonckheere–Terpstra Test results between personality features and number of applications by category
JT
Kendall's
p
Gender No.of Media and Video Apps Owned Sports No.of Finance Apps Owned No.of Tools Apps Owned No.of Business Apps Owned No.of Productivity Apps Owned No.of Entertainment Apps Owned No.of Travel & Local Apps Owned No.of Photography Apps Owned No.of Shopping Apps Owned No.of Transportation Apps Owned No.of Communication Apps Owned Total No.of User Owned Apps
-5.303 -4.716 -3.841 -4.050 -3.635 -3.608 -2.589 -2.366 2.339 -2.177 -2.147 -2.010 -4.094
-.392 -.355 -.278 -.277 -.266 -.248 -.183 -.171 .171 -.161 -.161 -.141 -.269
.000 .000 .000 .000 .000 .000 .010 .018 .019 .030 .032 .044 .000
Extraversion No.of Education Apps Owned
-2.076
-.150
.038
Agreeableness No.of Sports Apps Owned
2.064
.158
.039
Conscientiousness No.of Books & Reference Apps Owned No.of News and Magazines Apps Owned
2.098 1.878
.151 .141
.036 .060
Neuroticism No.of Entertainment Apps Owned No.of Media and Video Apps Owned
-2.025 -1.925
-.145 -.144
.043 .054
Openness -2.330 -.178 No.of Sports Apps Owned -2.292 -.171 No.of Media and Video Apps Owned -2.020 -.145 No.of Books & Reference Apps Owned 1.984 .139 No.of Social Apps Owned * p < 0.05, ranked by the absolute value of Kendall's tau-b coefficient
4.5.
.020 .022 .043 .047
Download of recommended applications and the number of user owned applications in each category
The Jonckheere–Terpstra test was applied to analyze the download behavior and number of category based applications owned by users. In Table 9, significant values having p < 0.05 are ranked by absolute value of Kendall's Tau coefficient.
User with applications in the recommended category are more likely to download the recommended application: Table 8 reveals that for health and fitness, books & references, music & audio, entertainment and games categories, recommended application downloads are significantly related with owning applications in the same category. For health and fitness and books & reference categories, the correlation coefficients were 0.46 and 0.293, respectively. This indicates that having an application in a category increases the probability of having other applications in the same category significantly. This is in line with the observation made by Yin et al. (2013) that some users download multiple similar applications. The number of recommended applications downloaded is positively correlated with having applications in books & reference, education, health and fitness and games categories as can be seen in Table 9. It was found that books & reference is the key category for the users’ download behavior with a correlation coefficient of 0.328. Since conscientiousness is related with the number of application downloads, it is not surprising that having applications in categories such as education and books & reference, which reflect the characteristics of conscientious people (Costa and McCrae, 1992), is significantly related with the download behavior. Table 8 Jonckheere–Terpstra Test results between download behavior and the number of applications present in the same category*
Number of Apps Downloaded in the Same Category No of Health and Fitness Apps Owned No of Books & Reference (Reference) Apps Owned No of Music and Audio Apps Owned No of Entertainment Apps Owned No of Games Apps Owned
JT
Kendall's
p
6.175 4.116 2.714 2.670 2.409
.460 .293 .192 .189 .164
.000 .000 .007 .008 .016
* p < 0.05, ranked by the absolute value of Kendall's tau-b coefficient Table 9 Jonckheere–Terpstra Test results between the number of applications downloaded and the number of applications in the same category*
JT
Kendall's
Number of Apps Downloaded 5.157 .328 No of Books & Reference Apps Owned 3.600 .231 No of Education Apps Owned 3.452 .229 No of Health and Fitness Apps Owned 2.170 .132 No of Games Apps Owned * p < 0.05, ranked by the absolute value of Kendall's tau-b coefficient
5.
p .000 .000 .001 .030
Conclusion and future work
In this study, individual differences and the effects of the Big Five personality features on the users’ compliance with the mobile application recommendations were explored. It was found that intention and download behavior towards the recommended applications differed according to personality features. Conscientiousness was found to be positively related with the tendency to accept recommended applications. In almost all application stores the mobile applications are grouped under headings that employ consensus or authority influence strategies such as the most popular, most downloaded, editor’s choice or applications of the day. Users’ perceptions of such influence strategies and underlying factors that lie beyond the users’ preferences are examined in this paper. It was shown that people having more applications are more likely to accept the recommendation of the most downloaded application. Most popular apps and editor’s choice apps were significantly related to the number of applications down-
loaded to the user’s mobile phone and being agreeable was significantly related with a preference for the editor’s choice link. Another important result was that there was a negative correlation between choosing the most popular app link and the editor’s choice link. Furthermore, it was shown that there was a relation between personality features and number of applications owned by the user in total and in each application category. The number of user owned applications was significantly higher for males. The number of applications owned in the photography category was positively correlated with females whereas the numbers of applications in the categories of media & video, sports, finance, tools, business, productivity, entertainment, travel & local, shopping, transportation, and communication were positively correlated in descending order of magnitude with male. The Big Five personality features and number of applications owned by the user in the application categories are shown to be related. Extraversion was negatively correlated with number of applications owned in the education category and agreeableness was positively correlated with sports. Conscientiousness was correlated with the categories of books & reference and news & magazines. Neuroticism was negatively correlated with entertainment and media & video. Openness was positively correlated with social application category and negatively with sports, media & video and books & reference categories. The number of applications downloaded was positively correlated with having applications in books & references, education, health and fitness and games categories. Another generalization that can be made is that having applications in the same category increases the probability of accepting more application recommendations in the same category. 5.1.
Limitations
The major limitation of the study reported in this paper was the use of a convenience sample from graduate and undergraduate university students since it does not completely represent the general consumer population who are likely to be less familiar with mobile applications. The students were from one of the top universities in Turkey, which implies that the participant group has a high level of literacy and innovativeness. In the experimental study on a mobile platform, log data was used to extract user profiles through the list of installed applications; however, this was only a snapshot of the applications the participants owned. This was not longitudinal research following users after downloading and during their use of the recommended applications over time. Therefore, the usage time of applications, the frequency of application use and uninstall behavior that occurred after the download of the applications, were not fully analyzed in the current research. Another limitation of our study is that due to the novel extensions in scope, some of the findings of our study could not directly be compared with the findings in the literature. It will be beneficial to conduct qualitative research to explore the findings of this study. 5.2.
Future work
The main contribution of this paper is to ground the recommender systems’ design in an understanding of users’ needs and to provide personalization means to better address users’ individual differences. It can also be used to tailor for specific needs and specific populations and make predictions based on the available data. Mobile technology products should avoid the one-size-fits-all approach to meet different needs of individuals. The findings of this study provide valuable information to improve personalization and recommendation services of mobile applications. People with similar personality features may be used for identifying and clustering people with similar tastes and preferences. Understanding the patterns of mobile phone use is significant in revealing user preferences in order to improve features of mobile phones and personalized mobile services. It can also help tailor specific needs and particular populations and make predictions based on the available data. The identification of users who are innovators, early-adopters, leader users or have a large number of friends is also important to determine the primary segment that should be targeted by marketing practitioners. The findings in this study provide useful insights for mobile technology designers, especially those planning for specific populations. Stakeholders including phone operators, service providers, application developers and advertising compa-
nies can benefit from the results to further explore the behaviors of mobile technology users from the perspective of consumers as well as other providers and advertisers. Although the findings of this study are interesting and significant, there are still other areas for future exploration. Different personality features such as trust and privacy concerns, well-being, mood and social network characteristics can be further analyzed that may contribute to an understanding of user preferences regarding the use of mobile phones and applications. It is shown that mobile application usage is denominated by contextual factors such as location, time of the day, activity and people around which is usually determined by the Bluetooth feature. Most probably, the users’ acceptance of mobile application recommendations depends on the contextual environment. Further research can tackle the effects of contextual factors on mobile phone and application use and recommendation acceptance. Using contextual information like time and location has the potential to be used in offering recommendations to new users who do not yet have a profile or history, hence addressing the coldstart problem.
References App Annie, 2015. App Annie Index - Market Q1 2014: Revenue soars in the United States and China. From
Liao, Z.-X., Pan, Y.-C., Peng, W.-C., Lei, P.-R., 2013. On mining mobile apps usage behavior for predicting apps usage in smartphones, in: Proceedings of the 22nd ACM International Conference on Information & Knowledge Management. ACM, New York, pp. 609-618. Liu, B., Wu, Y., Gong, N.Z., Wu, J., Xiong, H., Ester, M., 2016. Structural analysis of user choices for mobile app recommendation. ACM Trans. Knowl. Discov. Data 11(2), Article 17. doi:10.1145/2983533. Mehta, C.R., Patel, N.R., 2011. IBM SPSS exact tests. IBM Corporation, Cambridge, MA. McCrae, R.R., John, O.P., 1992. An introduction to the Five-Factor model and its applications. Journal of Personality 60, 175-215. McElroy, J.C., Hendrickson, A.R., Townsend, A.M., DeMarie, S.M., 2007. Dispositional factors in internet use: personality versus cognitive style. MIS Quarterly 31, 809-820. Seneviratne, S., Seneviratne, A., Mohapatra, P., Mahanti, A., 2014a. Predicting user traits from a snapshot of apps installed on a smartphone. ACM SIGMOBILE Mobile Computing and Communications Review 18, 1-8. Seneviratne, S., Seneviratne, A., Mohapatra, P., Mahanti, A., 2014b. Your installed apps reveal your gender and more!, in: ACM MobiCom Workshop on Security and Privacy in Mobile Environments (SPME '14), ACM. Sheskin, D.J., 2003. Handbook of parametric and nonparametric statistical procedures, 5th edition. Chapman and Hall/CRC. Tan, W.K., Yang, C.Y., 2014. Internet applications use and personality. Telematics and Informatics 31, 27-38. Unal, P., Temizel, T.T., Eren, P.E., 2014. An exploratory study on the outcomes of influence strategies in mobile application recommendations, in: 2nd International Workshop on Behavior Change Support Systems (BCSS 2014), pp. 27-40. Unal, P., Temizel, T.T., Eren, P.E., 2016. Impact of ındividual differences on the use of mobile phones and applications, in: International Conference on Mobile Web and Information Systems. Springer International Publishing, pp. 379-392. Wolfradt, U., Doll, J., 2001. Motives of adolescents to use the internet as a function of personality traits, personal and social factors. Journal of Educational Computing Research 24(1), 13-27. doi: 10.2190/ANPM-LN97-AUT2-D2EJ. Xia X., Wang X., Zhou X., 2013. Evolving recommender system for mobile apps: A diversity measurement approach. The Smart Computing Review 3, 139-154. Yan, B., Chen, G., 2011. AppJoy: Personalized mobile application discovery, in: Proceedings of the 9th International Conference on Mobile Systems, Applications, and Services (MobiSys). ACM, New York, pp. 113-126. Yin, P., Luo, P., Lee, W.-C., Wang, M., 2013. App recommendation: a contest between satisfaction and temptation, in: Proceedings of the 6th International Conference on Web Search and Data Mining. ACM, New York, pp. 395-404. Zaichkowsky, J.L., 1985. Measuring the involvement construct. Journal of Consumer Marketing Research 12, 341-352. Zhao, H., Seibert, S.E., 2006. The Big Five personality dimensions and entrepreneurial status: A meta-analytical review. Journal of Applied Psychology 91, 259-271. Ziegler, C.N., McNee, S.M., Konstan, J.A., Lausen, G., 2005. Improving recommendation lists through topic diversification, in: Proceedings of the 14th İnternational Conference on World Wide Web. ACM, New York, pp. 22-32.
Highlights Number of user owned apps and their category differ with gender and personality Having similar apps increases the probability of accepting recommended applications Number of apps owned in some categories implies higher acceptance of recommended apps Conscientiousness is positively related with accepting recommended applications Being agreeable is related with editor’s choice application preference