Improving e-learning communities through optimal composition of multidisciplinary learning groups

Computers in Human Behavior xxx (2013) xxx–xxx

Contents lists available at SciVerse ScienceDirect

Computers in Human Behavior journal homepage: www.elsevier.com/locate/comphumbeh

Improving e-learning communities through optimal composition of multidisciplinary learning groups Maria-Iuliana Dascalu a,⇑, Constanta-Nicoleta Bodea b, Miltiadis Lytras c, Patricia Ordoñez de Pablos d, Alexandru Burlacu a a

Politehnica University of Bucharest, Department of Engineering in Foreign Languages, Splaiul Independentei 313, Bucharest 060042, Romania The Bucharest Academy of Economic Studies, Department of Economic Informatics and Cybernetics, Calea Dorobantßi, 15-17, Sector 1, Bucharest 010552, Romania ELTRUN, The Research Center, Department of Management Science and Technology, Athens University of Economics and Business, Athens, Greece d University of Oviedo, Department of Business Administration, Oviedo, Spain b c

a r t i c l e

i n f o

Article history: Available online xxxx Keywords: Multidisciplinary learning groups Particle Swarm Optimization E-learning communities

a b s t r a c t The current study proposes an intelligent approach to compose optimal learning groups in which the members have different domain backgrounds. The approach is based on a well-known evolutionary algorithm – Particle Swarm Optimization. The authors claim that quantifying various indicators, such as background diversity and similarity between the type of interest of the participants, within a group and between groups can positively impact on building learning groups. The algorithm is integrated in an ontology-based e-learning system, designed to create self-built educating communities, in which a trainees goes through the education process, gains points through achievements and ultimately becomes a trainer. When creating a new account, the newly created trainee is asked to self asses himself by filling out a form. The resulting profile is used to assign the user to the most suitable learning group. We propose to assign him by the following rule: maximizing the diversity within a group (due to the fact that multidisciplinary teams are more challenging) and minimizing the diversity between groups (all the groups should have similar composition), meaning a group will have members with similar interests. The study is presented in the context of group building strategies in adults’ education. Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction The learning communities, defined as groups of people engaged in intellectual interactions for the purpose of learning (Cross, 1998) are becoming more present as e-learning communities (eLCs). This is mainly because of ICTs developments, which enhance the collaboration among participants and stimulate the interactivity. Different eLCs types and characteristics are extensively reported in the literature (Riel & Polin, 2004, Barab, Kling, & Gray, 2004, Moule, 2006). Gannon-Leary and Fontainha (2007) consider that the sense of connectedness, the shared passion and the deep of knowledge derived from ongoing interactions transform the knowledge development process into a continuous, cyclical and fluid one. According to their studies, the main eLCs benefit is the possibility to achieve the collaborative learning, which allows us to obtain a synergy effect. ⇑ Corresponding author. Address: Dorneasca Street, Nr.11, Bl. P79, Scara B, Et. 7, Ap. 66, Bucharest 051715, Romania. Tel.: +40 740917410. E-mail addresses: [email protected] (M.-I. Dascalu), [email protected] (C.-N. Bodea), [email protected] (M. Lytras), [email protected] (P.O. de Pablos), [email protected] (A. Burlacu).

On the eLCs context, the synergy means that two or more discrete learning agents acting together will create a learning result greater than that obtained by acting individual. This is based to the general systemic theory approach and, more specifically on Vygotsky’s (1978) approach about how a person’s learning may be enhanced through the engagement with others. For achieving collaborative learning, incorporation of appropriate pedagogical models, such as team projects and/or case studies, peer reviews. From the technological point of view, group work spaces, discussion forums, email/white boards, chat services, online facilities for sharing documents in projects and case studies, online evaluation of projects and quizzes are some of the solutions adopted by the participants in order to develop and share knowledge. There are several challenges in attaining meaningful collaboration in eLCs. A primer issue is the instruction design and analysis of learners’ (Inaba, Tamura, Ohkubo, Ikeda, & Mizoguchi, 2001) and inter-animation (Trausan-Matu, Stahl, & Sarmiento, 2008). A second issue is optimal group building, as the individuals have to ‘‘negotiate and share meanings relevant to the specific problemsolving task at hand’’ (Stahl, Koschmann, & Suthers, 2006). Adequate group formation means optimal resources (tangible or intangible) sharing, so the chances of learning to be maximal. A third

0747-5632/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.chb.2013.01.022

Please cite this article in press as: Dascalu, M.-I., et al. Improving e-learning communities through optimal composition of multidisciplinary learning groups. Computers in Human Behavior (2013), http://dx.doi.org/10.1016/j.chb.2013.01.022

2

M.-I. Dascalu et al. / Computers in Human Behavior xxx (2013) xxx–xxx

issue is related to ethics. The quality of the information given in learning groups may be a critical issue, because of malicious users. A tutor might establish the correctness of the information, but the tutor has to keep the balance between guidance and freedom given to the peers, so the learning environment to keep its collaborative features. Also, the quality of the information given by certain members may vary in time: ‘‘past experiences not necessary predict future performance’’ (Awerbuch & Kleinberg, 2008). The current study proposes an ontology-based e-learning system, designed to create a self-built eLC, where the trainees can become trainers, after increasing their knowledge. Our main concern was to create appropriate learning groups, as we consider that this could be the main lever to gain knowledge in such an eLC. Particle Swarm Optimization (PSO) was applied to obtain such groups.

2. Collaboration in e-learning communities For achieving collaboration in eLCs, the interactions based on the social presence are essential. Participants need to interact with their peers and want to be perceived as being ‘‘there’’ and being ‘‘real.’’ The social presence influences online interaction and the learning process as well. The quantity or frequency of online participation does not necessarily mean a high social presence; rather, it is the quality of online interactions that make the difference. Perceptions of social presence and the corresponding adjustments are more important than the objective quality of the communication medium. Learner’s perceptions of social presence are related to their satisfaction with the course, the trainer, and learning environment. Social presence is one of the most important instruments for determining the level of interaction and the effectiveness of learning in an online learning environment (Mykota & Duncan, 2007).The measurement of social presence focuses on the observable behaviors used by the students to project themselves as ‘‘real’’ people (Bulu & Yildirim, 2008). In many of eLCs, the learning process is oriented toward professional competencies development, instead of knowledge transfer (Rosu & Dragoi, 2011). eLCs can empower participants to execute and learn actions which are relevant for behavior of competent professionals. The active and experiential learning, based on field observations, real world cases, current events, critical reflections and problem solving are pedagogical approaches often applied in eLCs. According to Dale’s theory (1946), students usually retain more information by what they ‘‘do’’ as opposed to what they hear, read or observe). And, performing professional activities, they gain experience and develop their professional skills. The task and practice based learning communities are defined by Riel and Polin (2004) and they consider that practice based learning community differs from task based community mainly by voluntary participation. Researches regarding the learning communities of practice in many professions were already done, for example in the educators’ learning (Hanewald, 2009), in Ph.D. and master students (Van Brakel, 2010), attorneys’ learning (Hara & Kling, 2002), and project management (Bodea, Mogos, & Dascalu, 2012). The term eLCs is sometimes used interchangeably with communities of practice (CoP). Wenger defines community of practice as ‘‘group of people who share a concern or a passion for something they do and learn how to do it better as they interact regularly’’ (Wenger, 2006). Members of a community of practice (CoP) are engaged in common activities and discussions, sharing knowledge. They are united by confidence, trust and common identity. The interactions act as learning enablers. The CoP members build relationships, interact and thus, learn together. They share experiences, stories, solutions to real problems, in other words, they share practices. Practice is, according to Wenger, about meaning as an experience of everyday life (Wenger, 1998). The primary

interest of a CoP’s members is apprenticeship, which means learning from the other more experimented members of CoP: ‘‘novices learn how to become professionals by being mentored by and appreciated to more experiences mentors’’ (Hara & Kling, 2002), they come into contact the expert ways of knowing, thinking and reasoning (Zimitat, 2007). New learners are learning through interactions with experienced members of the CoP, but the interesting aspect is that these more experienced members also learn by teaching. It is widely accepted that eLCs and CoPs are connected concepts. CoPs is a learning community, but not every learning community is a CoPs. When the learning approach is a combination of experience and theories and techniques exploration, conducted by a problembased learning curriculum (PBL), then the community could act as a CoP (Zimitat, 2007). The members of a learning community are engaged in joint activities and discussions, as CoPs’ members are, too. The common features of learning communities and CoPs are the stress put on domain experience, shared knowledge, shared knowing, the way in which time is managed, the way in which users’ needs are addressed and in which these needs emerge. The main difference between learning communities and CoPs are the formalization degree (Ragan & Tello, 2005): learning communities are formal, institutionally created, and CoPs are informal, self-generated. There are authors who mix the both concepts, of learning communities and communities of practice, when talking about a community in which certain professionals are trained. Baran and Çag˘ıltay (2006) describe the relationship between teachers’ professional development and online communities of practice. They classify the communities of practice in task based learning communities (‘‘produce a product or outcome and their members know each other. These are generally temporary groups whose members try to accomplish well-specified tasks’’), knowledge based learning communities (‘‘compose knowledge based on a specific area. Members of it may or may not know each other personally. There is a long-term commitment to construct knowledge base’’) and practice based learning communities (these communities ‘‘differ from task based community mainly by voluntary participation’’). There are several computer-based tools, especially web-based tools, used to solve, more or less, the above mentioned issues in eLC: forums, wikis, social networks, chat rooms, blogs, or virtual worlds. Some researchers try to add to the social web tools semantic features: Li, Dong, and Huang (2009) added semantics to discussion transcripts from forums, so the user can be guided to wellstructured forum posts, which are highly relevant to their learning demands (Li et al., 2009). Even more, for obtaining the semantic flavor, other try to exploit ontologies: ‘‘ontologies have been successfully applied to solve problems such as: group formation, collaborative learning representation, interaction analysis and patterns and modeling of learner’s development’’ (Isotani, Inaba, Ikeda, & Mizoguchi, 2009). Ontologies are powerful tools for identifying and analyzing eLCs and CoPs, due to their remarkable power in modeling complex structure, such as networks. Ontology-Based Network Analysis (ONA) is becoming very popular, even if the effectiveness of this approach is highly dependent on the ontology content and CoPs’ characteristics. The more expressive the ontology is, the more likely will be to identify interesting and unforeseen connections but, in the same time, the more difficult to deal with the random noise of the representation. Alani, Dasmahapatra, O’Hara, and Shadbolt (2003) developed an ontology-based CoP identifier, named ONTOCOPI. The system is developed using AKT ontology and Protégé 2000. Tifous, El Ghali, Giboin, and DiengKuntz (2007) introduce O’CoP, an ontology dedicated to annotate the CoPs’ resources, built collaboratively from the analysis of several real CoPs. The ontology has more than 800 concepts and 80 relations, it is a 3-layered structure, allowing the reusability. DOG-

Please cite this article in press as: Dascalu, M.-I., et al. Improving e-learning communities through optimal composition of multidisciplinary learning groups. Computers in Human Behavior (2013), http://dx.doi.org/10.1016/j.chb.2013.01.022

M.-I. Dascalu et al. / Computers in Human Behavior xxx (2013) xxx–xxx

MA approach (de Moor, 2005) offers facilities for developing ontology-guided meaning negotiation, considering that in order to address the communication ambiguities a process of meaning negotiation can be successfully applied. Barros, Verdejo, Read, and Mizoguchi (2002) propose an ontology-based model to virtualize the collaborative learning scenarios, centered on Activity Theory. The challenging part is to virtualize the collaborative learning scenario, by constructing intelligent user profiles, in which students’ profiles should be properly reflected. Kumar, Gress, Hadwin, and Winneb (2010) use ontologies to allow students to share their learning process in collaborative learning, thus assessing it.

3. Strategies applied in building optimal learning groups There are many learning theories to support group activities: learners can be helped to ‘‘grasp, model and evaluate personal data’’ (Becks, Reichling, & Wulf, 2003) of their peers, so that they can develop groups based on similar interests, needs and competences or, quite the opposite, based on complementary competences, but common purposes. In our opinion, there are three main ways of group forming, of which two belong in the random group selection and development category and one that could easily be considered scientifically accurate, based on algorithms and criteria established after prior studies (Knowles, Holton, & Swanson, 2005).  Self-grouping, where the educator/trainer does not get involved. Where the freedom to select the learning group partners belongs to the students themselves, such groups present the disadvantage that they are formed randomly, sometimes taking into account proximity, appearances, ‘‘word of mouth’’ regarding academic performance, age, ethnicity, gender, etc. The advantages may be the responsibility to deliver in the collaborative learning process, due to peer pressure and the unknown variables of academic performance of the group members.  Educator’s choice, where the educator assigns the teams, also randomly, in order to counteract reluctance of teaming up by the students.  Learning groups developed by applying algorithms, where the team is created following the application of specific criteria and algorithmic processes, so as to achieve the most balanced learning groups, thus a highly effective learning process for each participant. With regard to the development of learning groups and teams using algorithms, a new challenge surfaces for the educators – how to select the criteria for the group development? Is gender a criterion that contributes to a better dynamic or better communication in the group? What about level of studies, academic performance, background, experience, topics of interest, learning objectives and purely and simply, chemistry? The practicality of the algorithm-based collaborative learning group development is more suited to business/corporate learning and to academic learning, whereas specialized short-term training needs a more individual approach. When building groups, one has to mine learners’ characteristics: Ou and You (2011) identify, as learners’ features, the following: profile, knowledge level, cognitive ability, learning style, motivation/educational purposes, social preferences. Isotani et al. (2009) also consider important: goals, necessary roles within the group, the context – the available tools or learning materials. According to them there are individual goals and group goals. When talking about group goals, four types were identified:

3

knowledge sharing, creating a solution, spreading of a skill and knowledge building through transmission. Roles are one of the main factors that affect the achievement of goals. Isotani et al. (2009) pin point to two different perspectives of building groups: (1) starting from the intended group goals and roles and assigning those roles to real peers or (2) starting from the individual goals and selecting the proper groups for achieving the students’ goals. Both methods have pros and cons. Learners’ features can be modeled with ontologies and the relationship between ontological concepts can be used to identify the optimal group composition. Huang and Wu (2011) use a learner similarity matrix to analyze the learners’ profiles: they took into account the learners’ gender, ability, individual psychological features, ethnicity, but also behavior patterns. For tracking users’ behaviors in a collaborative learning environment, context-aware techniques were used, to trace and interpret users’ interactions. The learner similarity matrix is the input for the clustering algorithm which builds the learning groups. The algorithm of grouping is given by a learning strategy, which can be changed, according to the tutor. The intra-cluster diversity depends on a threshold, which can be also changed (Huang & Wu, 2011). Razmerita and Brun (2011) try to solve the problem of group building in an automated manner by implementing a pilot study before applying a clustering strategy. They analyzed a set of self-formed students’ groups and identified the criteria for group building, in the context of a collaborative learning platform. They discovered that the students tend to organize in groups, which are homogeneous taking into account their knowledge level, but in heterogeneous in terms of the topic of interests. The purpose of a collaborative learning group should be to gather the individuals having the same objectives and to raise the knowledge level of the ones who do not understand the learning materials: these aspects become difficult when all the group members are not well prepared. Slavin (1995) recommends heterogeneous groups, in terms of background, idea, personality ethnicity and gender, but Bekele (2005) says that homogeneous groups are better for achieving specific goals, while the heterogeneous ones tend to me more creative and more oriented towards innovation. Several evolutionary algorithms were applied to form optimal group learning: Bayesian networks (Bekele, 2005), ant colony optimization (Graf & Bekele, 2006), genetic algorithms (Chan et al., 2010) or Particle Swarm Optimization (Lin, Huang, & Cheng, 2010). We concluded that the best approach in our case (making multidisciplinary teams) is given by PSO: this algorithm gives us the possibility of choosing the best combination of criteria, by simply choosing a proper fitness function (Pedersen & Chipper, 2010).

4. The proposed method for multidisiciplinary learning groups composition In order to optimize the group composition when dealing with multidisciplinary teams, we applied a PSO algorithm and model the fitness function of this algorithm. The algorithm aims at optimizing the group building process by:  speeding up the group composition,  increasing the amount of knowledge of all participants at the collaborative learning process. PSO algorithm is a robust stochastic optimization technique, which is inspired from the movement and intelligence of swarms. PSO applies the concept of social interaction to problem solving. It uses a number of particles that constitute a group moving around in the search space looking for the best solution. It imitates the bird

Please cite this article in press as: Dascalu, M.-I., et al. Improving e-learning communities through optimal composition of multidisciplinary learning groups. Computers in Human Behavior (2013), http://dx.doi.org/10.1016/j.chb.2013.01.022

4

M.-I. Dascalu et al. / Computers in Human Behavior xxx (2013) xxx–xxx

from a flock which is nearest to the food. The two variables which are iteratively changed in PSO algorithm are the following ones:  pbest (personal best): each particle keeps track of its coordinates in the solution space which are associated with the best solution (fitness) that has achieved so far by that particle;  gbest (global best): another best value that is tracked by the PSO is the best value obtained so far by any particle in the neighborhood of that particle; Each particle tries to modify its position using the following information: the current positions, the current velocities, the distance between the current position and pbest, the distance between the current position and gbest. The final result of the algorithm is the gbest value. The fitness function has to be customized for each application of the PSO. The updates of the particles’ position and velocity are made using the typical formulas for PSO (Pedersen & Chipper, 2010). Due to the fact that the PSO algorithm has to be adapted to each case in which it is applied, we propose the function (1) for the problem of solving composition of multidisciplinary learning groups:

Minimize f ðPÞ ¼ ð1=D1Þ þ D2 þ ð1=I1Þ þ I2 where P is the representation of a particle in PSO terms; D1 the background diversity within a group; D2 the background diversity between groups; I1 the similarity between the type of interest of the participants within a group and I2 is the similarity between the type of interest of the participants between groups; The first term of the function is described in formula (2):

D1 ¼

N i;j
jSEBi  SEBj j

g¼0 i;j¼0;i–j

where N is the number of groups, ng the number of students in the gth group and SEBi is the self-evaluation of background knowledge of the ith student in the gth group (average). The formula for D2 is similar, the only difference is that we consider the differences between the self-evaluation of background knowledge of each student from a group and the self-evaluations of background knowledge of all the students in the other groups. The third term of the function is described in formula (3):

I1 ¼

N i;j
jSEIi  SEIj j

g¼0 i;j¼0;i–j

where N is the number of groups, ng the number of students in the gth group and SEIi is the self-evaluation of learning interest of the ith student in the gth group (average). The formula for I2 is similar, the only difference is that we consider the differences between the self-evaluation of learning interest of each student from a group and the self-evaluations of learning interest of all the students in the other groups. According to the above fitness function, at each step, we propose to choose the particle which will maximize the diversity within a group (due to the fact that multidisciplinary teams are more challenging) and minimize the diversity between groups (all the groups should have similar composition) and which will choose members with similar interests. For example, all the students which want a higher grade will be in the same group, so they will have the same approach of solving the given task. The representation of particle in PSO terms is the one proposed by Lin et al. (2010): if we have n students and g groups, P = p11 p12. . .p1n p21. . .p2n. . .pjn. . .pgn is a generic representation of the particle P. The membership of a student to a group is coded with 1.

E.g.: there are 12 students which are divided in three groups; each student has a unique identification code (1, 2, 3. . .12).

P ¼ 111100000000 000011001100 000000110011 The above representation means that at a certain step from the PSO algorithm, the first group is built from students 1, 2, 3, 4, the second group from students 5, 6, 9, 10 and the third group from the students 7, 8, 11, 12. The values of D1, D2, I1 and I2 are given as input to the PSO algorithm. The range of valid values is established by mapping the students’ profiles and the random distribution of enrolled students to an established number of possible learning group. The students’ profiles is created when students enroll: they have to selfasses their knowledge and their interests, using a 5-level Likert scale, with respect to a set of concepts. The concepts are taken from the available domain ontologies, which are accessed by our eLC.

5. Integration of the proposed method into an e-learning system for self-built educating communities 5.1. The system architecture The above algorithm is included in an e-learning system which can be used by self-built eLCs. The high-level block diagram of the system is available in Fig. 1. The class diagram representing the system architecture is available in Fig. 2. The system is implemented in Java and designed using design patterns, in order to make it more flexible, organized, and easy to change. The design patterns sued are: Abstract Factory, Command, Controller, Singleton, Observer and Iterator. The Abstract Factory provides an interface that delegates creation calls to one or more concrete classes in order to deliver specific objects. The program uses it to create and instantiate objects of the following types: MultiDisciplinaryGroup, Activity or User. The Singleton design pattern ensures that only one instance of a class is allowed within a system. The class SystemController aims at centralizing the control of the whole system. It is necessary that we have controlled access to a single object. It contains a CommandInvoker object which is called each time a user requests an event to take place. Therefore, each program execution must contain a sole instance of this central class. Working with collections of objects requires the use of the Iterator design pattern. It allows for access to the elements of an aggregate object without allowing access to its underlying representation. The MultiDisciplinaryGroups class contains an ArrayList of MultiDisciplinaryGroup objects. The groups taken from a Database will be used as a collection. In order to iterate through this collection, we use the already implemented Iterator of Java. The Observer design pattern lets one or more objects be notified of state changes in other objects within the system. In the same context of collections, the observer class can be used to keep the Main Graphical User Interface update with a list of multi disciplinary groups. The Command design pattern encapsulates a request allowing it to be treated as an object. This allows the request to be handled in traditionally object based relationships such as queuing and callbacks. It is used when: one needs callback functionality, requests need to be handled at variant times or in variant orders, a history of requests is needed, the invoker should be decoupled from the object handling the invocation. In our case, the Command pattern plays a central role. SystemController delegates requests to the CommandInvoker which executes the do() function of an object function. There are three types of methods according to the user type. As a consequence, we will need three interfaces to connect the front end of the system with the back end:

Please cite this article in press as: Dascalu, M.-I., et al. Improving e-learning communities through optimal composition of multidisciplinary learning groups. Computers in Human Behavior (2013), http://dx.doi.org/10.1016/j.chb.2013.01.022

M.-I. Dascalu et al. / Computers in Human Behavior xxx (2013) xxx–xxx

5

Fig. 1. The block diagram of the e-learning system for self-built eLCs.

 TrainerCommandInterface  GlobalCommandInterface  TraineeCommandInterface Each of the three interfaces handles requests to their subclasses. The Particle Swarm Optimization algorithm is integrated as an implementation of the do() function of the viewRecommendedGroups object, which implements the GlobalCommandInterface. 5.2. The system functionalities The system functionalities are reflected in the sequence diagram from Fig. 3. When a user logs on, he is asked to create a profile, otherwise he logs in with his account name and password. The user is prompted with a Graphical User Interface (GUI) where the execution of commands is possible. Whether a user is also a trainer, a series of commands are enabled or disabled. For example, a trainee not having reached a trainer level, cannot create a group. In either case, the system controller is notified of a command invocation and calls the do method of the specific concrete command class. The functionalities are designed according to their purpose. They are Admin Specific, Global specific, Individual specific, and Trainer specific. The main functionalities are listed below. Individual specific functions: a trainee can:  view the groups he has joined,  do the necessary activities posten in a certain group that he belongs to,  enroll to a new group. Trainer specific functions: a trainer can:     

create a group, manage the group, create an activity in a specified group, verify response correctness of certain activity types, invite another trainer for collaborative teaching. Global specific functions: both trainer/trainee can:

 view his lifeline (a chronological representation of how he obtained proficiency in certain domains),  view profile,  modify system settings such as selecting the server,

    

view all users on the server, view the profile of other users, view available groups (must meet requirements), fill out the self assessment form when creating a new profile, view recommended groups according to domain proficiency and personal interests. The Particle Swarm Optimization algorithm is included here.

Admin specific functions: work with the ontological repository (add/delete a domain ontology or parts of a domain ontology), by calling the ontological exploitation service. 5.3. A scenario for group building 5.3.1. Scenario description Michael decides to create a new account on our system and begins his e-learning journey. The journey has two phases: the one where a user is a trainee and the one where he becomes a trainer. Michael has a particular keen interest in advanced mathematics, calculus and code optimization. 5.3.2. Interaction with the system Firstly, Michael creates a new account by filling out a form. He is then asked to self-asses himself by filling out a special form. This form contains questions about what he is interested in, what skills he wants to improve, his expectations as well as his self-assessment of a series of skills. The skills are domain-specific: meaning they are taken from domain ontologies. He then proceeds to the Main GUI where he comes in contact with the list of the multidisciplinary groups on the server. After iterating through the groups’ list, he decides that nothing sparks his interest, so he clicks on the ViewRecommendedGroups tab. At this moment, the Particle Swarm Optimization algorithm does the job of recommending something similar or approximate, in the domain of Mathematics and Algorithm Complexity. He decides that this could be an interesting option so he enrolls himself in the multidisciplinary group. Unfortunately, he arrived late. The group has already started doing the activities coordinated by two trainers. Fortunately, the trainers grant him permission to enroll but he will have extra activities to be done at the end. At this moment, he begins learning about Mathematics and Algorithm Complexity by reading and doing the activities coordinated by the trainers. After each submitted answer, he is granted points. At the end of the process, an overall re-

Please cite this article in press as: Dascalu, M.-I., et al. Improving e-learning communities through optimal composition of multidisciplinary learning groups. Computers in Human Behavior (2013), http://dx.doi.org/10.1016/j.chb.2013.01.022

6

M.-I. Dascalu et al. / Computers in Human Behavior xxx (2013) xxx–xxx

Fig. 2. The classes diagram of the e-learning system for self-built eLCs.

Please cite this article in press as: Dascalu, M.-I., et al. Improving e-learning communities through optimal composition of multidisciplinary learning groups. Computers in Human Behavior (2013), http://dx.doi.org/10.1016/j.chb.2013.01.022

M.-I. Dascalu et al. / Computers in Human Behavior xxx (2013) xxx–xxx

7

Fig. 3. The functionalities of the e-learning system for self-built eLCs.

Please cite this article in press as: Dascalu, M.-I., et al. Improving e-learning communities through optimal composition of multidisciplinary learning groups. Computers in Human Behavior (2013), http://dx.doi.org/10.1016/j.chb.2013.01.022

8

M.-I. Dascalu et al. / Computers in Human Behavior xxx (2013) xxx–xxx

Fig. 4. A scenario for group building in the e-learning system for self-built eLCs.

port of his gained knowledge in the domains is generated and added to his permanent profile. After a while, our user has finally reached a point where he meets the requirements to become a level one trainer. He decides that it is time to share the gained knowledge to others. Michael is now in the second phase. He views a list of other trainees of his level, as well as groups that do not have enough trainers. After contacting the main group coordinator, they agree to create, together with Michael, their own multi-disciplinary group in order to teach Calculus and Biology to low level users. Our user is now a trainer and gains points by spreading the knowledge. But he decides that this is not enough, and wishes to discover other domains. He goes back to become a trainee and again uses the Particle Swarm Optimization algorithm to start another journey. The scenario is available in the activity diagram from Fig. 4.

6. Results and discussion The PSO algorithm was implemented using the Java package Jswarm (http://jswarm-pso.sourceforge.net/), by making several customizations. The performance of the algorithm was tested by changing the following parameters (see Table 1): the number of iterations in the PSO algorithm, the number of particles in the swarm, the number of students with available profiles and the number of groups. Also, for this stage of the experiment we computed the average satisfaction of our grouping, by providing a simple questionnaire, in which the students had to give a grade from 0 to 10, taking into account their automated assignation to a learning group. We noticed an increase in the computational time, as we used more iterations/more particles. Also, this brought a moderate increase in the value of the satisfaction indicator. As a validation

Please cite this article in press as: Dascalu, M.-I., et al. Improving e-learning communities through optimal composition of multidisciplinary learning groups. Computers in Human Behavior (2013), http://dx.doi.org/10.1016/j.chb.2013.01.022

9

M.-I. Dascalu et al. / Computers in Human Behavior xxx (2013) xxx–xxx Table 1 Evaluation results of the PSO algorithm. Students

Groups

10 Iterations/10 particles (s)

Satisfaction metric (avg)

10 Iterations/20 particles (s)

Satisfaction metric (avg)

20 Iterations/10 particles (s)

Satisfaction metric (avg)

10 15 20 25

2 3 4 5

0.079 0.142 0.289 0.321

6.1 6.4 6.3 7.2

0.119 0.164 0.297 0.343

6.4 6.9 6.5 7.6

0.126 0.178 0.311 0.358

6.5 7.1 6.6 7.8

threat for our evaluation method, one could clame that the number of students used for the experiment was not satisfactory. Still, the trend is the following: as long as the number of students increases, the satisfaction metric for our grouping recommendation also increases. The result is sustained by other studies, also (Lin et al., 2010). 7. Conclusions In this paper, we propose a flexible system which can be used by e-learning communities. The system self-evolves, meaning the enrolled trainees can become, in the end, trainers. The learning process is based on the collaborative learning paradigm, meaning we make automated recommendations to our users to join a certain learning group, via a PSO algorithm, taking into account their learning objectives (interests) and their background knowledge. The learning interests and their background knowledge is taken from the trainees’ profiles, initially filled via a formular, then updated using the activity of the trainees within our system (e-tests, participation in e-classes, etc). The list of the interests and skills from the initial formular is taken from a domain ontology. Our system communicates with an ontological repository via a database: each domain which can be learnt in the proposed eLC is reflected in a domain ontology available in the repository. In such way, we ‘‘force’’ our users to respect a baseline of skills necessary to master a certain domain, as a domain ontology is a formal specification of a shared conceptualization of a domain. As future activities, we plan a more detailed evaluation of our system, with more active users. In our evaluation experiment, we ‘‘forced’’ all the users to follow our recommendation and evaluate their satisfaction degree. But the degree of confidence into our system could be also useful. As the system uses the Particle Swarm Optimization algorithm for recommending groups to trainees, we propose as a metric of quality evaluation, the number of users enrolling in a group based on the algorithm’s recommendation. Ultimately, the user decides if he or she is interested in the domains of the recommended group. As a result, we can compare the choices of users with the recommendations provided by the algorithm. In the future, we plan to implement this metric. The PSO algorithm can be further improved: we can consider new factors when creating the fitness function for optimal group building. Such a factor can be the gender of our users, as the gender issue is seen as a type of interest at young students: they are willing to be more communicative in mixed teams, if they want to meet people of opposite gender. In eLCs, there is usually a large number of students involved. If we want to apply several grouping criteria, the problem of dividing the students in learning groups is NP-hard. We consider that applying a PSO-based algorithm, the problem can be solved in an optimal manner. Also, the current paper contributes to developing the field of eLC, in the context of communities of practice. References Alani, H., Dasmahapatra, S., O’Hara, K., & Shadbolt, N. (2003). Identifying communities of practice through ontology network analysis. IEEE Intelligent Systems, 2, 18–25.

Awerbuch, B., & Kleinberg, R. (2008). Competitive collaborative learning. Journal of Computer and System Sciences, 74(8), 1271–1288. Barab, S., Kling, R., & Gray, J. H. (2004). An introduction: designing for virtual communities in the service of learning. In S. Barab & J. Gray (Eds.), Designing for virtual communities in the service of learning (pp. 3–15). Cambridge, MA: Cambridge University Press Cambridge. Baran, B., & Çag˘ıltay, K. (2006). Knowledge management and online communities of practice in teacher education. The Turkish Online Journal of Educational Technology – TOJET, 5(3), 12–19. Barros, B., Verdejo, M., Read, T., & Mizoguchi, R. (2002). Applications of a collaborative learning ontology – MICAI. Mexico: Springer, pp. 301–310. Becks, A., Reichling, T., & Wulf, V. (2003). Supporting collaborative learning by matching human actors. In Proceedings of the Thirty-Sixth Annual Hawaii International Conference on System Sciences (HICSS-36). Big Island, Hawaii: IEEE Computer Society. Bekele, R. (2005). Computer-Assisted Learner Group Formation Based on Personality Traits – PhD Thesis. Hamburg: Hamburg University. Bodea, C.-N., Mogos, R., & Dascalu, M.-I. (2012). ow e-learning experience enhances the social presence in community of practice. An empirical analysis. In C. El Morr, & P. Maret (Eds.), Virtual Community Building and the information Society: Current and Future Directions (pp. 75–120). IGI Global. Bulu, S., & Yildirim, Z. (2008). Communication behaviors and trust in collaborative online teams. Educational Technology & Society, 11(1), 132–147. Chan, T., Chen, C., Wu, Y., Jong, B. S., Hsia, Y. T., & Lin, T. W. (2010). Applying the genetic encoded conceptual graph to grouping learning. Expert Systems with Applications, 37, 4103–4118. Cross, K. P. (1998). Why learning communities? Why now? About Campus. Dale, E. (1946). Audio-visual methods in teaching. New York: The Dryden Press. de Moor, A. (2005). Ontology-guided meaning negotiation in communities of practice. In Proc. of the Workshop on the Design for Large-Scale Digital Communities at the 2nd International Conference on Communities and Technologies. Milano, Italy. Gannon-Leary, P., & Fontainha, E. (2007). Communities of practice and virtual learning communities: Benefits, barriers and success factors. eLearning Papers (5). Graf, S., & Bekele, R. (2006). Forming heterogeneous groups for intelligent collaborative learning systems with ant colony optimization. In Intelligent Tutoring System Conference (pp. 217–226). Jhongli, Taiwan: Springer. Hanewald, R. (2009). Sustainability in an online community of practice. The case study of a group of secondary school educators in Victoria. Australian Journal of Teacher Education Group of Secondary School Educators in Victoria, 34(5), 26–42. Hara, N., & Kling, R. (2002). IT Supports for Communities of Practice. An Empiricallybased Framework. Bloomington, Indiana: Bloomington: Center for Social Informatics, Indiana University. Huang, Y.-M., & Wu, T.-T. (2011). A systematic approach for learner group composition utilizing U-learning portfolio. Educational Technology & Society, 14(3), 102–117. Inaba, A., Tamura, T., Ohkubo, R., Ikeda, M., & Mizoguchi, R. (2001). Design and analysis of learners’ interaction based on collaborative learning ontology. In Proceedings of EuroCSCL01 (pp. 308–315). Isotani, S., Inaba, A., Ikeda, M., & Mizoguchi, R. (2009). An ontology engineering approach to the realization of theory-driven group formation. International Journal of Computer Supported Collaborative Learning, 4(4), 445–478. Knowles, M., Holton, E., & Swanson, R. (2005). The Adult Learner. San Diego, California: Elsevier Inc.. Kumar, V., Gress, C. L., Hadwin, A. F., & Winneb, P. H. (2010). Assessing process in CSCL: An ontological approach. Computers in Human Behavior, 26(5), 825–834. Li, Y., Dong, M., & Huang, R. (2009). Toward a semantic forum for active collaborative learning. Educational Technology & Society, 12(4), 71–86. Lin, Y.-T., Huang, Y.-M., & Cheng, S.-C. (2010). An automatic group composition system for composing collaborative learning groups using enhanced particle swarm optimization. Computers & Education, 55, 1483–1493. Moule, P. (2006). Developing the Communities of Practice. Framework for On-Line Learning Electronic Journal of E-learning, 4(2), 133–140. Mykota, D., & Duncan, R. (2007). Learner characteristics ad predictors of online social presence. Canadian Journal of Education, 30(1), 157–170. Ou, X., & You, D. (2011). A digital collaborative design model based on learner characteristics in art design. American Journal of Engineering and Technology Research, 11(9), 3056–3061. Pedersen, M. E., & Chipper, A. J. (2010). Simplifying Particle Swarm Optimization. Applied Soft Computing, 10(2), 618–628. Ragan, C. L., & Tello, S. (2005). Building the learning community: the power of online learning. In Sloan-C International Conference. Orlando, Florida, USA.

Please cite this article in press as: Dascalu, M.-I., et al. Improving e-learning communities through optimal composition of multidisciplinary learning groups. Computers in Human Behavior (2013), http://dx.doi.org/10.1016/j.chb.2013.01.022

10

M.-I. Dascalu et al. / Computers in Human Behavior xxx (2013) xxx–xxx

Razmerita, L., & Brun, A. (2011). Collaborative learning in heterogeneous classes. In The 3rd International Conference on Computer Supported Education (pp. 189–194). Noordwijkerhout, Netherlands: CSEDU. Riel, M., & Polin, L. (2004). Learning communities: Common ground and critical differences in designing technical environments. In S. A. Barab & J. Gray (Eds.), Designing for Virtual Communities in the Service of Learning (pp. 16–52). Cambridge, MA: Cambridge University Press. Rosu, S., & Dragoi, G. (2011). VPN solutions and network monitoring to support virtual teams work in virtual enterprises, 8(1), 1–26. Slavin, R. (1995). Cooperative learning: Theory, research, and Practice. Boston: Allyn and Bacon. Stahl, G., Koschmann, T., & Suthers, D. (2006). Computer-supported collaborative learning: An historical perspective. In Cambridge handbook of the learning sciences (pp. 409–426). Cambridge, UK: Cambridge University Press. Tifous, A., El Ghali, A., Giboin, A., & Dieng-Kuntz, R. (2007). O’CoP, an Ontology dedicated to Communities of Practice. Retrieved November 30, 2012, http://iknow.tugraz.at/wp-content/uploads/2011/11/Tifous_2007.pdf.

Trausan-Matu, S., Stahl, G., & Sarmiento, J. (2008). Supporting polyphonic collaborative learning. e-Service Journal, 58–74. Van Brakel, P. S. (2010). Mentoring doctoral students within a virtual community of practice – with special reference to their information recording behaviour. In Proceedings of Informing Science & IT Education Conference (pp. 439–448). Vygotsky, L. (1978). Mind in society: The development of higher psychological processes. Cambridge, MA: Harvard University Press. Wenger, E. (1998). Communities of practice. Learning, meaning and identity. Cambridge: Cambridge University Press. Wenger, E. (2006). Communities of practice - a brief introduction. Retrieved November 16, 2012, http://www.ewenger.com/theory/communities_of_ practice_intro.htm. Zimitat, C. (2007). Capturing community of practice knowledge for student learning. Innovations in Education and Teaching International, 44(3), 321–330.

Please cite this article in press as: Dascalu, M.-I., et al. Improving e-learning communities through optimal composition of multidisciplinary learning groups. Computers in Human Behavior (2013), http://dx.doi.org/10.1016/j.chb.2013.01.022