A knowledge extraction and representation system for narrative analysis in the construction industry

ESWA 9257

No. of Pages 13, Model 5G

16 April 2014 Expert Systems with Applications xxx (2014) xxx–xxx 1

Contents lists available at ScienceDirect

Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa 5 6

A knowledge extraction and representation system for narrative analysis in the construction industry

3 4 7 8 9 10 1 2 2 1 13 14 15 16 17 18 19 20

Q1

C.L. Yeung, C.F. Cheung ⇑, W.M. Wang, Eric Tsui Knowledge Management and Innovation Research Centre, Department of Industrial and Systems Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong

a r t i c l e

i n f o

Keywords: Construction industry Knowledge management Knowledge representation Knowledge extraction Narrative analysis

a b s t r a c t Many researchers advocate that the real-world narratives shared by experts or knowledge workers are helpful in teaching and educating novices to learn new knowledge and skills. Narrative analysis is a useful method for experts to understand narratives. However, it does not produce any clear or explicit layouts. This is not easy for a new learner without prior knowledge to glean the right messages from narratives within a short time. In this paper, a narrative knowledge extraction and representation system (NKERS) is presented to extract and represent narrative knowledge in an effective manner. The NKERS is composed of a narrative knowledge element extraction algorithm, a narrative knowledge representation method and a narrative knowledge database. A prototype system has been built and trial implemented in the construction industry. The results show that the domain experts agree that the narrative maps generated by the NKERS can effectively represent narrative elements and flows. Three-quarters of respondents expressed that they will use the produced narrative maps in their training courses to facilitate students’ learning. Ó 2014 Published by Elsevier Ltd.

22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

37 38

1. Introduction

39

Narratives exist in the human world with an infinite diversity of forms (Barthes & Duisit, 1975). Narrative is an international, transhistorical and transcultural medium for human beings. Narrative has a special function in the area of memorization and learning (Taylor, 1989). It helps to retain humans’ memory especially culture memory of things that have happened in the past (Ong, 1982). For example, narratives can record ancient incidents such as Trojan wars and foster people’s memories of incidents and benefit from lessons learnt in the past. People can recall additional information by reading others’ narratives (Gabriel, 2000). By relating critical ideas with previous knowledge and experience, people can construct situations mentally by using narratives, simulating actions and predicting the consequences before they actually perform tasks in the real world (Gee, 2004). Narrative comprises human rationality and narrative structures (Johnson, 1987). It helps people in repairing and restoring meaning when they are in adversity (Bury, 2001). Narratives are useful for people to remember and learn how to solve problems and make decisions (Jonassen & Hernandez-Serrano, 2002). How humans read and understand narrative texts greatly depend on human capability. Hence, people

40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58

⇑ Corresponding author. Tel.: +852 2766 7905; fax: +852 2362 5267.

with different reading capability can have a different understanding or interpretation after reading the same narrative texts. People may misunderstand the information in the narratives. The misunderstandings generate great hindrances to gaining new knowledge (Guzzetti, 1990). This is not easy to rectify in their minds. If people do not have any relevant knowledge or experience, it is not easy for them to understand the correct information in the narratives in a short time. To facilitate humans to understand narratives, it is important to develop a method to systematically analyze narratives and present correct narrative information in a clear and explicit way. The construction industry is well-known for being one of the highest-risk industries in the world due to its high number of fatalities and accident rate (Al-Humaidi & Tan, 2010; Navon & Sacks, 2007). Governments and organizations have stipulated several regulations and guidelines to achieve better safety performance (Government of Alberta, 2011; Queensland Government, 2011; United States Department of Labor, 2011). However, there are still some areas that are not covered by laws and in which lessons have not been learnt from overseas and therefore deserve more attention. Researchers have contributed a lot to developing different intelligent systems to improve the situation, especially in the area of genetic algorithms, neural networks, and knowledge-based and expert systems (Irani & Kamal, 2014). Recent studies focus on process management (Hajdasz, 2014) or company failure detection

E-mail address: [email protected] (C.F. Cheung). http://dx.doi.org/10.1016/j.eswa.2014.03.044 0957-4174/Ó 2014 Published by Elsevier Ltd.

Please cite this article in press as: Yeung, C. L., et al. A knowledge extraction and representation system for narrative analysis in the construction industry. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.044

59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83

ESWA 9257

No. of Pages 13, Model 5G

16 April 2014 2

C.L. Yeung et al. / Expert Systems with Applications xxx (2014) xxx–xxx

106

(Horta & Camanho, 2013). However, limited studies are found to use narratives to support crisis management as well as workers’ learning in the construction industry. Currently, construction investigators document incidents as narratives to disseminate important messages to workers. The traditional approach requires knowledge of experts and knowledge workers to understand the causes, developments and consequences of the incidents. This information is critical for crisis management (Paraskevas, 2006). However, it is immersed in the narrative texts without a clear indication. It is also not easy for a new learner to perceive the right messages in narratives within a short time. Once a new learner digests a concept wrongly or misunderstands the moral of a narrative, extra effort is needed to rectify his thinking. As a result, it is vital to know what can help to analyze narratives, extract and represent narrative knowledge from the narratives and facilitate people to understand and learn the narratives in a clear way. In this paper, a narrative knowledge extraction and representation system (NKERS) is presented to effectively extract narrative knowledge elements and represent narratives in a clear layout. This system incorporates techniques in computational linguistics and rulebased reasoning and provides a semi-automatic method to conduct narrative analysis and generate narrative maps. Encouraging results are found through a case study in the construction industry.

107

2. Literature review

108

2.1. Narrative

109

Narratives are everywhere and enterprises and organizations are no exception. Organizational documents such as handbooks, incident reports, newspaper articles and personal experience are examples of narratives. Most business knowledge and expertise are embedded in narratives in organizations. Instead of learning from incidents happening currently, people can learn from the narratives in organizations or from human experience. Indeed, narratives have different dimensions. As shown in Table 1, there are four different narrative dimensions: story, recount, newspaper report and procedure. According to Labov (1972), a story includes six elements which are abstract, orientation, complication, evaluations, resolutions and coda. An abstract is a brief summary introducing the main idea of the story. Orientation shows the background information (such as personas, time and place) of the story to the readers. Complication means a series of events before showing the climax or highpoint of the story. The information for the readers to know the reason of telling the story is found in the evaluation. Resolution is the attempt to handle the complicated situation. The coda is the consequence of the attempt or the ending of the story. Due to the presence of various stories, it should be noted that not all six elements are included in every story (McCabe & Peterson, 1991). In general, orientation, complication, resolution and coda are usually found in stories. Elements such as abstract and evaluation seem always to be overlooked. However, the evaluation gives the key information regarding why you need to read the story. Recounting can also be found in organizations. The experience of knowledge workers can be regarded as an example of a personal recount while incident reports are regarded as factual reports. It descriptively relates a real incident or an imaginary event to the readers (Schleppegrell, 2003). Newspapers, having different scaffolds from factual recounting, report current issues to the public. They highlight the most critical information in the news by using headlines (Van Dijk, 1985). Instructions or manuals are examples of procedure in organizations. Sometimes procedure is classified as a part of narrative although it contains detailed information regarding the flow and sequences of events (Lewis & Wray,

84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105

110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145

Table 1 A review of different narrative functions and scaffolds. Narrative Dimensions (Examples)

Function(s)

Story (e.g. novels, fables, folktales, legend, etc.)

 To encourage readers to think about the issues  To provide a lesson to be learnt by the readers

Recount

 To descriptively relate a real incident (such as an author’s experience or a particular incident or event) or an imaginary event (like fiction) to the readers

Newspaper report

 To report the current issues to the public

Procedure (e.g. user manual, instruction materials, etc.)

 To indicate how to make or do something

Scaffold

Abstract Orientation (who, where, when) ; Complication ; Evaluation ; Resolution ; Coda Labov (1972) Setting (protagonist, situation, time, etc.) ; Series of events (in time order) ; Concluding statement or ending Schleppegrell (2003) Headline ; By-line ; The Lead (summary of the most important information, i.e. protagonist, situation, time, etc.) ; Next most important point ; Least important point ; Conclusion (consequences, possible future leads) abcteach (2004) ; Introduction of the aim or goal ; Materials required for completing the procedure ; Series of events in the correct order ; Evaluation Lewis and Wray (1996)

1996). The usage of stories is prevalent as it is good at motivating readers to think about the issues and providing a lesson to be learnt by the readers (Labov, 1972; Özyıldırım, 2009). As a result, this study selects stories as the main narrative scaffold for investigation.

146

2.2. Narrative analysis and narrative map

151

Different disciplines have drawn increasing attention to narrative analysis (Bury, 2001). It is popular to use narrative analysis to understand clients’ experience such as dangerous or embarrassing experiences (Stephens, 2011; Özyıldırım, 2009). There is a wide range of narrative analysis methods (Delamont, 2012). Özyıldırım (2009) suggested that one of the most influential narrative models is presented by Labov and Waletzky (1967) and Labov (1972). Labov’s model can be used to analyze narratives in written or oral form

152

Please cite this article in press as: Yeung, C. L., et al. A knowledge extraction and representation system for narrative analysis in the construction industry. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.044

147 148 149 150

153 154 155 156 157 158 159

ESWA 9257

No. of Pages 13, Model 5G

16 April 2014 C.L. Yeung et al. / Expert Systems with Applications xxx (2014) xxx–xxx 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193

(Özyıldırım, 2009). For written narratives, researchers firstly read the collected narratives. Then they match each narrative to the six categories as mentioned in Section 2.1. For oral narratives, researchers define a set of questions to ask respondents to elaborate narratives in terms of the six categories (Delamont, 2012; Stephens, 2011; Özyıldırım, 2009). Narrative analysis provides a framework for experts to understand narratives in different domains (Labov & Waletzky, 1967). However, prior knowledge is required to conduct narrative analysis. It is not easy for a new learner to perceive the right messages in narratives within a short time. Narrative or story maps were developed based on the concept of story mapping. Story mapping facilitates readers to identify critical elements and their interrelationship in stories (Beck & McKeown, 1981; Pearson, 1982). The critical elements include character, problem, event and action (Dimino, Taylor, & Gersten, 1995). Pearson (1982) improved the work of Beck and McKoewn by using a graphic to display the connections between critical information and story maps. Fig. 1 depicts a blank form of the story map. It shows background information, encountered problem, expected goal, and corresponding action and outcome. A story map acts as a visual tool to outline the most critical narrative elements and their linkages within a narrative (Reutzel, 1985). It is generally used in education disciplines to improve the reading comprehension skills of students especially young students with learning disabilities (Burke, 2004; Dimino et al., 1995; Stringfield, Luscre, & Gast, 2011). To identify narrative elements such as characters and events presented in a narrative, students are invited to fill in a story map template independently after reading the narrative (Dimino et al., 1995). It facilitates students to organize the sequences and interconnections of the narrative elements (Pearson 1982, 1985). Students improve their reading comprehension if they study the defined story map before reading the narrative (Davis, 1994). Story maps have been proven to be an effective tool to improve the reading comprehension of skilled readers, less skilled

Fig. 1. Template of a story map (adapted from Idol (1987)).

3

readers and readers with learning disabilities (Idol, 1987). Different narrative maps are reviewed (Burke, 2004; Idol, 1987; Stringfield et al., 2011). Two fundamental characteristics are found. Firstly, critical narrative elements such as background information and event information are identified. Secondly, the events are presented in a sequential order in the narrative. Traditional narrative analysis and narrative maps are constructed manually based on users’ understanding of narratives (Dimino et al., 1995; Stringfield et al., 2011). They require knowledge of experts and knowledge workers. A systematic approach is needed to improve the performance of narrative analysis and narrative maps. Automatic concept mapping tools are useful to identify concepts and relationships in documents (Cañas, Bunch, Novak, & Reiska, 2013; Cline, Brewster, & Fell, 2010; Zubrinic, Kalpic, & Milicevic, 2012). Fuzzy Association Concept Mapping (FACM) (Wang, Cheung, Lee, & Kwok, 2008) is capable of handling multi-word concepts and implicit propositions when they are compared with other existing tools which are restricted to individual word concepts. It is also a domain independent tool which generates concepts and relations based on original texts or documents without initial knowledge. A modified-FACM is built to conduct concept association in this study. From the literature, persona, location and time are critical elements in both narrative analysis and narrative maps (Delamont, 2012; Idol, 1987; Labov, 1972; Stringfield et al., 2011). This information is immersed in narrative texts. A knowledge representation method is introduced to systematically represent extracted narrative elements in order to facilitate narrative analysis and narrative map construction.

194

3. Methodology of narrative knowledge extraction and representation system for narrative analysis

222

The methodology of the proposed framework for the narrative knowledge extraction and representation system (NKERS) is shown in Fig. 2. It is composed of a narrative knowledge element extraction algorithm, narrative knowledge representation method and narrative knowledge database. The narrative texts are firstly preprocessed by a narrative segmentation algorithm and a sentence restructuring algorithm. Narrative knowledge elements are then associated and extracted by a concept association tool. A representation method is introduced to represent narrative knowledge elements to conduct narrative analysis. A narrative map is then produced by NKERS after conducting narrative analysis. A narrative knowledge database is developed for retaining narrative knowledge elements extracted from the narrative to support narrative analysis. The details of each process are depicted in Sections 3.1 and 3.2, respectively.

224

3.1. Narrative knowledge element extraction

239

A modified-Fuzzy Association Concept Mapping (modifiedFACM) algorithm is proposed to conduct narrative knowledge element extraction. The modified-FACM is composed of a narrative segmentation algorithm, a sentence restructuring algorithm and FACM concept association tool (Wang et al., 2008). The FACM tool is capable of recognizing simple text structures and generating linkages between relevant elements from narrative texts. However, its performance is greatly affected if the sentence structure is too complex. In this study, the modified-FACM is introduced to conduct narrative segmentation, sentencing restructuring and concept mapping. It helps to convert complex sentences into simple clauses from narratives and then conduct concept mapping. Narrative texts are firstly preprocessed by a narrative segmentation algorithm. Several prerequisites are required for the narrative texts. The narrative texts should include three main narrative sections

240

Please cite this article in press as: Yeung, C. L., et al. A knowledge extraction and representation system for narrative analysis in the construction industry. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.044

195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221

223

225 226 227 228 229 230 231 232 233 234 235 236 237 238

241 242 243 244 245 246 247 248 249 250 251 252 253 254

ESWA 9257

No. of Pages 13, Model 5G

16 April 2014 4

C.L. Yeung et al. / Expert Systems with Applications xxx (2014) xxx–xxx

Fig. 2. Narrative knowledge extraction and representation system (NKERS).

255 256 257 258 259 260 261 262 263 264

265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295

including a beginning section, a middle section and an ending section in chronological order. The narrative texts which have the sentence patterns identified in Appendix A undergo a sentence restructuring process. The main function of the narrative segmentation algorithm is to identify different sections based on the narrative scaffold and wording. Hence, a sentence restructuring algorithm converts complex sentences into simple clauses to transform the human language into machine-readable language. A concept association tool is then used to extract narrative knowledge elements. 3.1.1. Narrative segmentation algorithm Narrative elements and flows are stored in narratives. People express narrative contents with different wordings or narrative scaffolds. Although different narratives have different scaffolds, they can be generalized into scaffolds that include a beginning section, a middle section and an ending section (Altman, 2008; Cooper, 1947). As different narratives have different functions, their scaffolds have some variations. For instance, the beginning section combines with the middle section. The middle section also includes the ending section. A narrative segmentation algorithm is developed to identify different sections. The pseudo code of the narrative segmentation algorithm is shown in Fig. 3. The algorithm is used to identify three main sections in the narrative by using narrative scaffolds and wordings. Labov (2006) suggested that the consequence is the most reportable event in a narrative. The algorithm starts by recognizing the ending section. Generally, the ending section occurs in the last paragraph. It includes some wordings such as ‘‘as a result’’ and ‘‘finally’’. In this paper, those wordings are categorized into ending words. In order to locate the ending section, the algorithm checks the sentences in the last paragraph with ending words. If the ending words are absent in the last paragraph, the last sentence will be chosen as the consequence of the narrative. After this, the algorithm begins to search the middle section. This section is hard to recognize as it always combines with the beginning section or the ending section. It is found that the middle sections usually start with a sentence in the form of continuous tense with the presence of some keywords such as ‘‘when’’ and ‘‘while’’. Some wordings such as ‘‘on the day of the incident’’ and ‘‘at the material time’’ also indicate the start of middle sections. Therefore, these wordings are grouped into middle words. If the

Fig. 3. Pseudo code of the narrative segmentation algorithm.

middle words are absent from the narrative, the algorithm will select the first sentence of the last paragraph as the beginning of the middle section. The middle section ends with the sentence prior to the start of the ending section. For the wording lists such as middle words and ending words, they are constructed based on finding synonyms through online dictionaries (www.thefreedictionary.com; http://www.synonym.com). Finally, the algorithm identifies the beginning section. The beginning section always includes background information about the narrative and it is in the first paragraph. The algorithm indicates the first paragraph as the start of the beginning section. The beginning section ends with the sentence prior to the start of the middle section.

296

3.1.2. Sentence restructuring algorithm Traditional human wiring involves complex and compound sentences. According to the Oxford Dictionary, a complex sentence is a sentence containing a subordinate clause or clauses while a compound sentence is a sentence with more than one subject or predicate. Because of human intelligence, human beings are capable of

308

Please cite this article in press as: Yeung, C. L., et al. A knowledge extraction and representation system for narrative analysis in the construction industry. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.044

297 298 299 300 301 302 303 304 305 306 307

309 310 311 312 313

ESWA 9257

No. of Pages 13, Model 5G

16 April 2014 C.L. Yeung et al. / Expert Systems with Applications xxx (2014) xxx–xxx 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356

understanding and assimilating the meanings of these sentences. However, natural language is ambiguous and imprecise. It is not easy for computers to interpret natural language. As a result, a sentence restructuring algorithm is developed to facilitate the transformation of natural language into computer language. It firstly undergoes sentence segmentation. The narrative texts are divided into sentences based on the detection of a new line character or terminating punctuation marks such as full stops and question marks. Hence, each word in a sentence is tagged with part-ofspeech (POS) by using a POS parser. This algorithm adopts the POS parser from WordNet (Fellbaum, 1998). If the sentence is a complex or compound sentence, it is then divided into several simple sentences. The simple sentences basically include a subject, a verb and an object. The designed rule sets of the sentence restructuring algorithm are shown in Appendix A. Let’s consider an ‘‘and’’ sentence as an example. If the original sentence has an ‘‘and’’ conjunction, the algorithm checks the sentence pattern of the original sentence. If the sentence pattern is subject1 + verb1 + object1 + ‘‘and’’ + subject2 + verb2 + object2, the sentence is restructured into two simple sentences, i.e. the first sentence is subject1 + verb1 + object1; the second sentence is subject2 + verb2 + object2. After that, the algorithm also correlates the subject in each sentence which is a pronoun with a proper noun or noun phase. After restructuring the complex and compound sentences into simple sentences, the algorithm checks the subject in each sentence. If the subject of a sentence is a pronoun, the algorithm then identifies and extracts the subject of the previous sentence into the proper format to restructure the sentence. After all pronouns are changed, the algorithm then conducts normalization of wordings. The wording found in the narrative, e.g. ‘‘The platform’’ matches ‘‘A wooden platform’’ which appeared in the former part of the narrative. The algorithm then changes the wording to ‘‘The wooden platform’’ in the entire narrative to assure consistency of wording. Fig. 4 shows examples of sentence restructuring algorithm conversion. The schematic diagram of the narrative segmentation algorithm and sentence restructuring algorithm is shown in Fig. 5. The narrative segmentation algorithm identifies the three main sections from the narratives. The sentence restructuring algorithm processes sentence segmentation and wording normalization. The preprocessed texts can then be retained in the narrative knowledge database before concept association. The narrative texts undergo narrative analysis. Narrative analysis is conducted based on Labov’s model as mentioned in Sections

5

2.1 and 2.2. As mentioned by Labov (1972), the abstract is an optional element in narratives. In some cases, the abstract is absent in the narrative. If the abstract is present, it indicates as a separate section with a subtitle namely abstract, summary, recap, outline, etc. The separate section usually appears in front of the main text of narratives. For orientation, the background information such as personas, place and time in the narratives need to be extracted. The background information can be found in the beginning section of the narrative. Complication and resolution are the events that can be extracted in the middle section. Evaluation indicates the reasons for telling the narrative. However, it is not compulsory to include evaluation in the narrative. Evaluation can be found in a middle section or an ending section. The coda is the consequences of the narrative and can be found in the ending section.

357

3.1.3. Fuzzy Association Concept Mapping (FACM) tool In order to conduct narrative analysis, the preprocessed narrative texts undergo a concept association process. The concept association process generates linkages between relevant elements in the narrative. A method named Fuzzy Association Concept Mapping (FACM) is used to conduct concept association (Wang et al., 2008). FACM is capable of recognizing simple text structures and generating linkages between relevant elements from narrative texts. It can handle multi-word concepts and implicit propositions. It is also a domain independent tool which generates concepts and relations based on original texts or documents without the initial acquisition of expertise and domain knowledge (Wang et al., 2008). A schematic diagram of narrative knowledge element extraction is shown in Fig. 6. After narrative segmentation and sentence restructuring, the preprocessed texts undergo concept association by the FACM. FACM makes use of rule-based and case-based reasoning to handle anaphoric resolution. An example of syntactic rules to extract noun phrases and verb phrases is shown in Fig. 7. In case-based reasoning, FACM uses the Eqs. (1)–(3) to measure the similarity between the new case and old case in anaphoric resolution (Wang et al., 2008).

371

Pm

Similarity ¼

0 r j¼1 wj simðv j ; v j Þ Pm j¼1 wj

ð1Þ

where m is the number of inputs, wj is the weighting of the jth POS, are types of the jth POS of the input case and that of the retrieved cases, and simðv 0j ; v rj Þ is the similarity function for the jth POS as follows:

v 0j and v rj

Fig. 4. Example of sentence restructuring algorithm conversion.

Please cite this article in press as: Yeung, C. L., et al. A knowledge extraction and representation system for narrative analysis in the construction industry. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.044

358 359 360 361 362 363 364 365 366 367 368 369 370

372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391

392

394 395 396 397 398

ESWA 9257

No. of Pages 13, Model 5G

16 April 2014 6

C.L. Yeung et al. / Expert Systems with Applications xxx (2014) xxx–xxx

Fig. 5. Schematic diagram of the narrative segmentation algorithm and sentence restructuring algorithm.

Fig. 6. Schematic diagram of narrative knowledge extraction.

399 401

simðv

0 j;

v

r jÞ

simðv

0 j;

v

r jÞ

¼ 1 if

v

0 j

¼ 0 if

v

0 j

¼v

r j

–v

r j

ð2Þ

402 404 405 406

ð3Þ

After using the FACM, the sentences are converted into subjectverb-attribute formats. Elements associated with their attributes

mentioned in the narrative are then extracted. A snapshot of critical elements extracted from a narrative is shown in Table 2. The extracted elements are associated with their verbs and attributes, respectively.

407

3.2. Narrative knowledge representation

411

After extracting the critical elements from a narrative, a systematic narrative representation is required. A method called Subject, Verb, Object, Relevant Information, Location and Time (SVORLT) is presented to organize narrative knowledge in a structured manner. After narrative knowledge extraction, each element and its related verbs and attributes are analyzed and transformed into the SVORLT format. The SVORLT format includes six dimensions which are Subject, Verb, Object, Relevant Information, Location and Time. They are the core information for a narrative. The subject and object dimensions are further divided into sublevels which are persona and non-persona. The structure of SVORLT representation is shown in Fig. 8. After representing all narrative knowledge, the narrative knowledge database retains the narrative knowledge in a structured way. Table 3 shows the representation after conducting SVORLT. A narrative map is designed and constructed to visualize the result of narrative analysis. A visual map with two layers transforms extracted information from the narrative analysis into a visual format. The first layer includes five main parts which are abstract, orientation, complication, resolution and coda. The abstract of the narrative map shows concise or summarized information related to the narrative. A summary of the narrative is a separate section which usually appears in front of the main text of narratives. If the summary is present in the narrative, the algorithm extracts it directly as the abstract of the narrative map. The title of the narrative is selected as abstract if the summary of the narrative is

412

Please cite this article in press as: Yeung, C. L., et al. A knowledge extraction and representation system for narrative analysis in the construction industry. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.044

408 409 410

413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437

ESWA 9257

No. of Pages 13, Model 5G

16 April 2014 7

C.L. Yeung et al. / Expert Systems with Applications xxx (2014) xxx–xxx

Fig. 7. Example of syntactic rules of FACM (adapted from Wang et al. (2008)).

Table 2 A snapshot of critical elements extracted from a narrative. Element

Verb 1

Attribute 1

Verb 2

Attribute 2

Verb 3

Attribute 3

A sub-contractor

Had

The lift shaft

Involved in

A contract with the main contractor to undertake site clearing in a construction site The accident

Was about

5 m long

Was about

3.5 m wide

Fig. 8. Structure of SVORLT representation.

Table 3 A snapshot of SVORLT representation. Subject (S) Persona

Verb (V) Non-persona

A sub-contractor The lift shaft

438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459

Object (O) Persona

Had Involved in Was about Was about

Relevant information (R)

Location (L)

With the main contractor to undertake site clearing The accident 5 m long 3.5 m wide

In a construction site

Time (T)

Non-persona A contract

absent. Orientation includes information of persona, time and place. The information is immersed in narrative texts. The SVORLT format helps to represent the information in a structured format. For personas in the narrative, the algorithm selects the persona under the subject dimension in the SVORLT format. For location and time in the narrative, the algorithm selects narrative elements under the location and time dimension in the SVORLT format respectively. For complication and resolution information, the algorithm selects the sentences in the middle section. Resolution is an attempt by a persona to handle the complicated situation. The algorithm firstly identifies the sentences with the persona and action words such as ‘‘tried’’, ‘‘had to’’ and ‘‘started’’ from the last sentence in the middle section. After identifying the target sentences with the persona and action words, the algorithm then extracts the target sentences to the last sentence in the middle section as resolution. The remaining sentences in the middle section are regarded as complication. For the evaluation, the algorithm selects some reason words such as ‘‘because’’, ‘‘because of’’, ‘‘due to’’ and ‘‘contributory factor’’ in order to extract the reasons for telling the narratives. Finally, the algorithm searches the sentences in the ending section. The consequences in the ending section are selected as the

coda. In the second layer, detail orientation and evaluation are shown. The linkages of critical concepts and relationships are displayed in detailed orientation. The evaluation of the narrative is shown in the second layer. Fig. 9 shows a snapshot of a narrative map. It includes abstract, orientation, complication actions, resolution, coda, detailed orientation and evaluation.

460

4. Trial implementation and case study in the construction industry

466

To realize the capability of the proposed narrative knowledge extraction and representation system (NKERS), a case study is conducted. This study involves four main stages as shown in Fig. 10. Stage 1 mainly focuses on the design of methodology and the development of the algorithms as shown in Section 3. Industry selection and data collection are conducted in stage 2. Rules refinement, trial implementation and evaluation are carried out in stages 3 and 4, respectively. An evaluation questionnaire is designed to collect organization feedback. The construction industry is well-known for being one of the highest-risk industries in the world. Falling of a person from height

468

Please cite this article in press as: Yeung, C. L., et al. A knowledge extraction and representation system for narrative analysis in the construction industry. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.044

461 462 463 464 465

467

469 470 471 472 473 474 475 476 477 478

ESWA 9257

No. of Pages 13, Model 5G

16 April 2014 8

C.L. Yeung et al. / Expert Systems with Applications xxx (2014) xxx–xxx

Fig. 9. Snapshot of a narrative map.

Fig. 10. Workflow of the case study.

479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501

is regarded as a critical issue that should be avoided since it has long been a major problem in the construction industry (Chan et al., 2008; Glasgow Caledonian University, 2005). The situation is not only prevalent in the United Kingdom, but also in Hong Kong. According to Figs. 11a and 11b, nearly 20% of industrial accidents and more than 50% of industrial fatalities take place in the construction industry. On average, more than 20% of industrial accidents and 75% of industrial fatalities occurred in the construction industry during the past decade. The construction industry recorded the highest number of accidents and fatalities among various industry sectors in Hong Kong (Labor Department, 2013). Moreover, the particular environment and culture in Hong Kong also contribute to the surprisingly high figures. Hong Kong is a densely populated city with a large number of high-rise buildings. The expansion of the population and aging of buildings have led to increasing demand for construction work, especially related to working at height. Hong Kong is the only region in the world that uses bamboo truss-out scaffolding in construction work (Chan et al., 2008). The uniqueness of the scaffolding leads to limited lessons being learnt with regard to the improvement of construction safety. The construction industry in Hong Kong is selected as the industry for this case study due to its alarming figures and increasing public concern. Of all the construction-related

incidents, the falling of a person from height, which is the leading category of death, is chosen as the main theme for the investigation. An organization in the construction industry is selected as a reference site. It provides qualified trainings for new-entrants or professional staff. Apart from transferring technical skills, the organization introduces knowledge for construction safety and accident prevention to the learners to reduce the number of industrial accidents and fatalities. The organization possesses a lot of training materials in the form of narratives in unstructured and semi-structured formats. Knowledge workers require lots of time to gather and analyze materials so as to extract key elements and attributes to generate new narratives. The scope of this case study is narrowed down to narratives regarding the falling of a person from height. Eight data sets of around two to three hundred words with three to seven paragraphs in the narratives are used as shown in Table 4. The organization has been trial implementing the narrative knowledge extraction and representation system (NKERS) for narrative analysis. Several rules are refined before carrying out the trial implementation. Detailed information is described in Section 3. User feedback is collected in stage 4 via interviews. The detail of the results and user feedback are shown in Section 5.

502

5. Results and discussion

524

Fifty elements, 117 verbs, 117 attributes, 47 flows and 14 consequences are extracted after the data processing and narrative knowledge extraction processes. The narrative knowledge is transformed into the SVORLT format. A snapshot of extracted critical elements and SVORLT representation are shown in Tables 2 and 3, respectively. After the representation of the narrative knowledge, 50 subjects (15 personas; 35 non-personas), 117

525

Please cite this article in press as: Yeung, C. L., et al. A knowledge extraction and representation system for narrative analysis in the construction industry. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.044

503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523

526 527 528 529 530 531

ESWA 9257

No. of Pages 13, Model 5G

16 April 2014 9

C.L. Yeung et al. / Expert Systems with Applications xxx (2014) xxx–xxx

Fig. 11a. Percentage of total number of industrial accidents in the construction industry and other industries from 2003 to 2012.

Fig. 11b. Percentage of total number of industrial fatalities in the construction industry and other industries from 2003 to 2012.

Table 4 Information of the collected data sets. Data set

Title

Number of words (number of paragraphs)

1 2 3 4 5 6 7 8

Fatal accident due to the collapse of a wooden platform erected inside a lift shaft Death caused by falling into an unprotected caisson Unsuitable and insufficient safe means of access and egress can cause serious injury Fatality resulting from falling through an unprotected lift shaft opening Substandard working platform can cause fatal accidents A worker fell from an unfenced working platform resulting in a head injury Wooden platform collapsed during dismantling and resulted in a fatality Unsuitable and insufficient safe means of access and egress can kill

309 287 255 232 324 205 373 237

(5) (4) (3) (5) (6) (5) (7) (4)

Please cite this article in press as: Yeung, C. L., et al. A knowledge extraction and representation system for narrative analysis in the construction industry. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.044

ESWA 9257

No. of Pages 13, Model 5G

16 April 2014 10

C.L. Yeung et al. / Expert Systems with Applications xxx (2014) xxx–xxx

Table 5a Feedback score of Q1–Q10 of narrative maps 1–4. Feedback Score of Narrative Map 1

Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8 Q9 Q10

2

3

4

Mean

Standard deviation

Mean

Standard deviation

Mean

Standard deviation

Mean

Standard deviation

4.25 4.50 4.25 4.25 4.50 4.50 4.25 4.00 3.75 3.25

0.50 0.58 0.50 0.50 0.58 0.58 0.96 1.41 1.26 1.50

4.25 4.25 4.00 4.25 4.25 4.25 4.00 4.00 4.00 4.25

0.50 0.50 0.00 0.50 0.50 0.50 0.82 0.82 0.82 0.96

4.25 4.25 4.00 4.25 4.25 4.25 3.75 3.75 4.00 4.25

0.50 0.50 0.00 0.50 0.50 0.50 1.26 1.26 1.41 1.50

4.25 4.50 4.25 4.25 4.25 4.50 4.00 4.00 4.25 4.50

0.50 0.58 0.50 0.50 0.50 0.58 0.82 0.82 0.96 1.00

Table 5b Feedback score of Q1–Q10 of narrative maps 5–8. Feedback score of narrative map 5

Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8 Q9 Q10

532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561

6

7

8

Mean

Standard deviation

Mean

Standard deviation

Mean

Standard deviation

Mean

Standard deviation

4.25 4.25 4.00 4.25 4.25 4.25 4.25 4.25 4.00 4.50

0.50 0.50 0.00 0.50 0.50 0.96 0.96 0.96 0.82 1.00

4.25 4.25 4.00 4.25 4.25 4.00 4.00 4.00 4.00 4.25

0.50 0.50 0.00 0.50 0.50 0.82 0.82 0.82 0.82 0.96

4.25 4.25 4.00 4.25 4.25 4.00 4.00 4.25 4.25 4.50

0.50 0.50 0.00 0.50 0.50 0.82 0.82 0.96 0.96 1.00

4.25 4.25 4.00 4.25 4.25 4.50 4.50 4.50 4.50 4.75

0.50 0.50 0.00 0.50 0.50 0.58 0.58 0.58 0.58 0.50

verbs, 76 objects (14 personas; 62 non-personas), 57 related pieces of information, 17 locations and six times were found. All of them are retained in the narrative knowledge database. Eight narrative maps are produced by using the proposed narrative knowledge extraction and representation system (NKERS) for narrative analysis. An evaluation questionnaire is designed (see Appendix B) to evaluate the performance of the proposed NKERS. The questionnaire includes personal information and evaluation questions of the narrative map. Six questions related to respondents’ profile are included in part one while 11 questions regarding the quality and performance of narrative analysis and narrative map are included in part two. The questionnaire adopts the 5-point Likert scale (5 = strongly agree; 4 = agree; 3 = neutral; 2 = disagree; 1 = strongly agree) for Q1–Q10 of the evaluation process. Q11 is a closed-ended question. Each respondent is required to fill in a questionnaire for each narrative map. Half of the construction safety experts in the organization participate in the evaluation process. They come from Hong Kong with prior knowledge of the construction industry. They have a master’s degree with advanced English level. Thirty-two questionnaires are collected. Table 5 shows the mean feedback score and standard deviation of their score. The mean feedback scores range from 3.50 to 4.75. The standard deviations are between 0 and 1.50. Table 5 shows the mean scores of Q1 to Q5 of narrative map 1 is more than 4. All respondents agree that narrative map 1 is useful for extracting and identifying critical elements and events in narrative 1. It is interesting to note that similar mean scores (>4) are obtained in the other seven narratives. The performance of narrative maps in extracting and identifying the narrative

elements is consistent in the collected narratives. For narrative map 1, the mean scores of Q6–Q7 are more than 4. The mean score of Q8 is equal to 4. The respondents generally agree that the narrative map assists learners to understand a narrative, to remember the important things in the narrative, and to learn the critical elements and events leading to the consequences of the narratives. The performance of narrative map 3 is slightly lower. The mean score of Q6 is more than 4. The mean scores of Q7 and Q8 are less than 4. For the remaining narrative maps, the mean scores of Q6–Q8 are equal to or larger than 4. Although the performance of narrative maps 1 and 3 are lower when compared with the other narrative maps, the respondents agree that narrative maps can help learners to better understand, remember and learn narratives. For Q9–Q11, the narrative maps obtain equal to or more than 3.25 mean scores. The performance of narrative maps 1 and 3 are lower than the other narrative maps. However, the respondents agree that narrative maps are helpful for retaining narrative knowledge in relation to training and learning. Of the respondents, 75% agree that they will use narrative maps in their training courses to facilitate trainees’ learning.

562

6. Conclusion

583

This section concludes the research work of this paper. The assumptions, strengths and weaknesses of the proposed research method are presented followed by the impact and insights of the research work. In this paper, there are two assumptions regarding the proposed method which include (1) The narrative texts should

584

Please cite this article in press as: Yeung, C. L., et al. A knowledge extraction and representation system for narrative analysis in the construction industry. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.044

563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582

585 586 587 588

ESWA 9257

No. of Pages 13, Model 5G

16 April 2014 C.L. Yeung et al. / Expert Systems with Applications xxx (2014) xxx–xxx 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652

include three main sections which are a beginning section, a middle section and an ending section in chronological order. (2) The sentence restructuring algorithms should focus on restructuring and converting the sentences which are in the identified sentence patterns in this study. This research is a new approach to systematically convert paragraph-based narrative texts into a clear narrative map layout. It can indicate and extract narrative elements from narrative texts. A SVORLT narrative representation method is proposed in order to semi-automate the narrative analysis process. An explicit narrative map is also generated. It not only presents the important narrative elements from the narratives, but also shows the linkages of critical concepts and relationships within the narrative texts. The explicit narrative map provides a holistic way for readers to understand narratives clearly based on Labov’s narrative model. The evaluation in this study is performed by domain experts. Domain experts agree that the narrative maps generated by the NKERS can identify and extract narrative elements and narrative flows from narrative texts. The respondents generally agree that the narrative map assists learners to understand narratives, to remember the important things in the narrative and to learn the critical elements and events leading to the consequences of the narratives. Therefore, the proposed research method is capable of conducting narrative analysis in a systematic way and generating visual narrative maps to clearly represent narratives. There are some limitations of the proposed research method. This research method is developed based on several narrative theories on narrative structure and narrative modeling. This method is only capable of analyzing narratives which include three main sections which are a beginning section, a middle section and an ending section in chronological order. Narratives with different types of narrative order are not taken into account in the present study. This method can conduct sentence restructuring based on the sentence patterns identified in this study. Other sentence patterns are not considered. The impact and insight of this research work include research contributions to theory, and practical applications and concrete future directions are also indicated. Traditionally, narrative analysis is manually conducted by experts or knowledge workers. Due to the diversity of human practice and capability, there are different ways to carry out narrative analysis. The outputs of narrative analysis also vary and are inconsistent. This study is a new attempt to investigate how to systematically conduct narrative analysis. Apart from semi-automating the narrative analysis process, this study firstly proposes the use of Labov’s narrative model and narrative map to visualize the output of narrative analysis. It helps to provide a systematic method to carry out narrative analysis. A clear and explicit layout is produced to present narrative information. This paper presents a narrative knowledge extraction and representation system (NKERS) for narrative analysis. Modified-Fuzzy Association Concept Mapping (modified-FACM) is developed based on a Fuzzy Association Concept Mapping (FACM) tool. Two algorithms for narrative segmentation and sentence restructuring respectively are constructed to convert complex sentences into simple clauses first so as to improve the performance of concept mapping. A Subject, Verb, Object, Relevant Information, Location and Time (SVORLT) narrative representation method is used to represent narrative knowledge. By using the NKERS, narrative analysis is conducted in a semiautomatic way. A narrative map is developed as a visual tool to represent a narrative. It uses Labov’s narrative model to show important information regarding incidents such as background, development, consequences and causes. It helps to provide a clear and explicit layout to present the content of narratives and educate novices to understand critical crisis information. It is useful

11

to facilitate crisis management and training. Hence, a prototype of NKERS is established and trial implemented at a selected reference site in the construction industry. The results show that the proposed NKERS can be successfully implemented in a reference site in the construction industry. Through the trial implementation in the reference site, the NKERS is realized to be capable of conducting narrative segmentation, extraction and represent narrative knowledge. It also facilitates narrative analysis and narrative map construction. Domain experts agree that the proposed NKERS can effectively extract and represent narrative knowledge. The use of visual narrative maps can retain narrative knowledge and support novices to understand and remember the correct messages from the narratives. Three-quarters of the respondents express that they will use the narrative maps in their future training courses to facilitate trainees’ learning for construction safety. There are several practical implications for knowledge management practices that can be drawn from this study. Knowledge management has attracted increasing concern from researchers and the public. One of the reasons is that executives have started to face challenges induced by the retirement tsunami of the baby boomers since 2012 (Toossi, 2004). The retirement tsunami has triggered a lot of highly skilled and experienced employees to leave their workplaces. Hence, researchers have started to investigate different methods to facilitate knowledge retention and transfer. Narrative is known as one important means for storing (Ong, 1982), transferring and sharing experience and lessons learnt (Gabriel, 2000). Human knowledge is immersed in the narratives. Traditionally, researchers advocated the use of narrative databases to retain narrative content. Users search proper indexes to retrieve the narrative content and assimilate the content by themselves (Snowden, 2002). It is not easy for novices to understand the cor- Q2 rect messages from the narratives. The narrative maps produced by NKERS can generate a clear and explicit layout to present narrative information. Systematically analyzing narratives and extracting narrative information can help to capture and retain critical human knowledge from predecessors from narrative texts. Advanced computational approaches in knowledge engineering can help to analyze or discover the causal relationships between narrative information from narratives in order to support narrative evaluation and generation in the future. For further research, performance evaluation with existing tools can be conducted. The developed system and algorithms can be improved and their performance compared with that of other natural language processing tools like ANNIE (A Nearly-New Information Extraction System). A triangular evaluation approach can be used to further verify the system’s performance. In this case, crowdsourcing evaluation can be conducted to cross-check the results of expert evaluation. Apart from this study, narrative-related research such as narrative knowledge extraction or narrative evaluation can be one of the future areas of investigation in knowledge management and knowledge engineering.

653

Acknowledgements

705

The work described in this paper was fully supported by a grant from the Research Grants Council of the Hong Kong Special Administrative Region, China (Project No. PolyU 5145/09E). The authors would also like to express their sincere thanks to the Research Committee of The Hong Kong Polytechnic University for its financial support of the research work. Many thanks are also due to the Construction Industry Council (CIC) of Hong Kong for its technical support of the work (Ref. No. RPKX).

706

Please cite this article in press as: Yeung, C. L., et al. A knowledge extraction and representation system for narrative analysis in the construction industry. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.044

654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704

707 708 709 710 711 712 713

ESWA 9257

No. of Pages 13, Model 5G

16 April 2014 12 714 716 715 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778

C.L. Yeung et al. / Expert Systems with Applications xxx (2014) xxx–xxx

Appendix A. Rule set of sentence restructuring algorithm

Appendix B. Narrative map evaluation questionnaires

(1) If a sentence contains a word ‘‘and’’ AND the sentence pattern is hs1i + hv1i + ho1i + ‘‘and’’ + hs2i + hv2i + ho2i Then the sentence is restructured to: hs1i + hv1i + ho1i; hs2i + hv2i + ho2i. (2) If a sentence contains a word ‘‘and’’ AND the sentence pattern is hs1i + hv1i + ho1i + ‘‘and’’ + hv2i + ho2i Then the sentence is restructured to: hs1i + hv1i + ho1i; hs1i + hv2i + ho2i. (3) If a sentence contains a word ‘‘and’’ AND the sentence pattern is hs1i + hv1i + ho1i + ‘‘and’’ + ho2i Then the sentence is restructured to: hs1i + hv1i + ho1i; hs1i + hv1i + ho2i. (4) If a sentence contains a word ‘‘and’’ AND the sentence pattern is hv1i + ho1i + ‘‘and’’ + hs2i + hv2i + ho2i Then the sentence is restructured to: hs2i + hv1i + ho1i; hs2i + hv2i + ho2i. (5) If a sentence contains a word ‘‘and’’ AND the sentence pattern is hs1i + ‘‘and’’ + hs2i + hv2i + ho2i Then the sentence is restructured to: hs1i + hv2i + ho2i; hs2i + hv2i + ho2i. (6) If a sentence contains a comma ‘‘,’’ AND the sentence pattern is hs1i + hv1i + ho1i + ‘‘,’’ + hs2i + hv2i + ho2i Then the sentence is restructured to: hs1i + hv1i + ho1i; hs2i + hv2i + ho2i. (7) If a sentence contains a comma ‘‘,’’ AND the sentence pattern is hs1i + hv1i + ho1i + ‘‘,’’ + hv2i + ho2i Then the sentence is restructured to: hs1i + hv1i + ho1i; hs1i + hv2i + ho2i. (8) If a sentence contains a comma ‘‘,’’ AND the sentence pattern is hs1i + hv1i + ho1i + ‘‘,’’ + ho2i Then the sentence is restructured to: hs1i + hv1i + ho1i; hs1i + hv1i + ho2i. (9) If a sentence contains a comma ‘‘,’’ AND the sentence pattern is hv1i + ho1i + ‘‘,’’ + hs2i + hv2i + ho2i Then the sentence is restructured to: hs2i + hv1i + ho1i; hs2i + hv2i + ho2i. (10) If a sentence contains a comma ‘‘,’’ AND the sentence pattern is hs1i + ‘‘,’’ + hs2i + hv2i + ho2i Then the sentence is restructured to: hs1i + hv2i + ho2i; hs2i + hv2i + ho2i. (11) If a sentence contains a word ‘‘when’’ OR ‘‘while’’ OR ‘‘whilst’’ AND a comma ‘‘,’’ AND the sentence pattern is ‘‘when’’ OR ‘‘while’’ OR ‘‘whilst’’ + hs1i + hv1i + ho1i + ‘‘,’’ + hs2i + hv2i + ho2i Then the sentence is restructured to: hs2i + hv2i + ho2i; hs1i + hv1i + ho1i. (12) If a sentence contains a word ‘‘when’’ OR ‘‘while’’ OR ‘‘whilst’’ AND a comma ‘‘,’’ AND the sentence pattern is ‘‘when’’ OR ‘‘while’’ OR ‘‘whilst’’ + hv1i + ho1i + ‘‘,’’ + hs2i + hv2i + ho2iThen the sentence is restructured to: hs2i + hv2i + ho2i; hs2i + hv1i + ho1i. (13) If a sentence contains a word ‘‘when’’ OR ‘‘while’’ OR ‘‘whilst’’ AND the sentence pattern is hs1i + hv1i + ho1i + ‘‘when’’ OR ‘‘while’’ OR ‘‘whilst’’ + hs2i + hv2i + ho2i Then the sentence is restructured to: hs1i + hv1i + ho1i; hs2i + hv2i + ho2i. (14) If a sentence contains a word ‘‘who’’ AND the sentence pattern is hs1i + ‘‘who’’ + hv1i + ho1i + hvb2i + ho2i Then the sentence is restructured to: hs1i + hv1i + ho1i; hs1i + hv2i + ho2i. (15) If a sentence contains a word ‘‘which’’ AND the sentence pattern is hs1i + ‘‘which’’ + hs2i + hv2i + ho2i + hv1i + ho1i Then the sentence is restructured to: hs1i + hv1i + ho1i; hs2i + hv2i + hs1i. (16) If a sentence contains a word ‘‘which’’ AND the sentence pattern is hs1i + hv1i + ho1i‘‘which’’ + hv2i + ho2i + Then the sentence is restructured to: hs1i + hv1i + ho1i; ho1i + hv2i + ho2i. (17) If a sentence contains a word ‘‘where’’ AND the sentence pattern is hs1i + hv1i + ho1i‘‘where’’ + hs1i + hv2i + ho2i Then the sentence is restructured to: hs1i + hv1i + ho1i; hs2i + hv2i + ho2i + ‘‘at’’ + ho1i. Please cite this article in press as: Yeung, C. L., et al. A knowledge extraction and representation system for narrative analysis in the construction industry. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.044

779 782 780 781 783

786

ESWA 9257

No. of Pages 13, Model 5G

16 April 2014 C.L. Yeung et al. / Expert Systems with Applications xxx (2014) xxx–xxx 787 785 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842

References abcteach. (2004). Writing a news article. Retrieved 31 May, 2012, from . Al-Humaidi, H., & Tan, F. (2010). Construction safety in kuwait. Journal of Performance of Constructed Facilities, 24(1), 70–77. Altman, R. (2008). A theory of narrative. New York, Columbia: University Press. Barthes, R., & Duisit, L. (1975). An introduction to the structural analysis of narratives. New Literary History, 6(2), 237–272. Beck, I. L., & Mckeown, M. G. (1981). Developing questions that promote comprehension: The story map. Language Arts, 58(8), 913–918. Burke, M. D. (2004). Use of story-mapping to increase the story-grammar text comprehension of elementary students with learning disabilities. Learning Disability Quarterly, 27(2), 105–121. Bury, M. (2001). Illness narratives: Fact or fiction? Sociology of Health & Illness, 23(3), 263–285. Cañas, A. J., Bunch, L., Novak, J. D., & Reiska, P. (2013). Cmapanalysis: An extensible concept map analysis tool. Journal for Educators, Teachers and Trainers, 4(1), 36–46. Chan, A., Wong, F., Chan, D., Yam, M., Kwok, A., & Lam, E. (2008). Work at height fatalities in the repair, maintenance, alteration, and addition works. Journal of Construction Engineering and Management, 134(7), 527–535. Cline, B. E., Brewster, C. C., & Fell, R. D. (2010). A rule-based system for automatically evaluating student concept maps. Expert Systems with Applications, 37(3), 2282–2291. Cooper, L. (1947). Aristotle, on the art of poetry (revised ed.). Ithaca and London: Cornell University. Davis, Z. T. (1994). Effects of prereading story mapping on elementary readers’ comprehension. The Journal of Educational Research, 87(6), 353–360. Delamont, S. (2012). Handbook of qualitative research in education. Edward Elgar Publishing. Dimino, J. A., Taylor, R. M., & Gersten, R. M. (1995). Synthesis of the research on story grammar as a means to increase comprehension. Reading & Writing Quarterly, 11(1), 53–72. Fellbaum, C. (1998). WordNet: An electronic lexical database. Cambridge: MIT Press. Gabriel, Y. (2000). Storytelling in organizations: Facts, fictions, and fantasies. London: Oxford University Press. Gee, J. P. (2004). An introduction to discourse analysis. New York, London: Routledge. Glasgow Caledonian University (2005). A technical guide to the selection and use of fall prevention and arrest equipment. Retrieved 30.12.11. Government of Alberta (2011). Legislation and enforcement. Retrieved 1.12.11. Guzzetti, B. J. (1990). Effects of textual and instructional manipulations on concept acquisition. Reading Psychology, 11(1), 49–62. Hajdasz, M. (2014). Flexible management of repetitive construction processes by an intelligent support system. Expert Systems with Applications, 41(4, Part 1), 962–973. Horta, I. M., & Camanho, A. S. (2013). Company failure prediction in the construction industry. Expert Systems with Applications, 40(16), 6253–6257. Idol, L. (1987). Group story mapping: A comprehension strategy for both skilled and unskilled readers. Journal of Learning Disabilities, 20(4), 196–205. Irani, Z., & Kamal, M. M. (2014). Intelligent systems research in the construction industry. Expert Systems with Applications, 41(4, part 1), 934–950. Johnson, M. (1987). The body in the mind: The bodily basis of the meaning, imagination, and reason. Chicago: University of Chicago Press.

13

Jonassen, D. H., & Hernandez-Serrano, J. (2002). Case-based reasoning and instructional design: Using stories to support problem solving. Educational Technology Research and Development, 50(2), 65–77. Labor Department (2013). Occupational safety and health statistics bulletin. Retrieved 30.06.13. Labov, W. (1972). Language in the inner city. Studies in the Black English vernacular. Philadelphia: University Pennsylvania Press. Labov, W. (2006). Narrative pre-construction. Narrative Inquiry, 16(1), 37–45. Labov, W., & Waletzky, J. (1967). Narrative analysis: Oral versions of personal experience. In J. Helm (Ed.), Essays on the verbal and visual arts. Proceedings of the 1966 annual spring meeting of the American ethnological society (pp. 12–44). University of Washington Press: Seattle. Lewis, M., & Wray, D. (1996). Writing frames scaffolding children’s non-fiction writing in a range of genres. United Kingdom: Reading and Language Information Center. McCabe, A., & Peterson, C. (1991). Developing narrative structure. Hillsdale, New Jersey: Lawrence Erlbaum Associates. Navon, R., & Sacks, R. (2007). Assessing research issues in automated project performance control (APPC). Automation in Construction, 16(4), 474–484. Ong, W. J. (1982). Orality and literacy: The technologizing of the word. New York: Methuen. Özyıldırım, I. (2009). Narrative analysis: An analysis of oral and written strategies in personal experience narratives. Journal of Pragmatics, 41(6), 1209–1222. Paraskevas, A. (2006). Crisis management or crisis response system? Management Decision, 44(7), 892–907. Pearson, P. D. (1982). Asking questions about stories (15 ed.). Columbus, OH: Ginn and Company. Ginn occasional papers. Pearson, P. D. (1985). Changing the face of reading comprehension instruction. The Reading Teacher, 38(8), 724–738. Queensland Government (2011). Law and prosecutions. Retrieved1.12.11. Reutzel, D. R. (1985). Story maps improve comprehension. The Reading Teacher, 38(4), 400–404. Schleppegrell, M. J. (2003). Grammar for writing: Academic language and the ELD standards. Retrieved 31.05.12. Stephens, C. (2011). Narrative analysis in health psychology research: Personal, dialogical and social stories of health. Health Psychology Review, 5(1), 62–78. Stringfield, S. G., Luscre, D., & Gast, D. L. (2011). Effects of a story map on accelerated reader postreading test scores in students with high-functioning autism. Focus on Autism and Other Developmental Disabilities, 26(4), 218–229. Taylor, C. (1989). Sources of the self: The making of the modern identity. Cambridge: Cambridge University Press. Toossi, M. (2004). Labor force projections to 2012: The graying of the U.S. workforce. Retrieved 20.11.10. United States Department of Labor (2011). Regulations. Retrieved 1.12.11. Van Dijk, T. A. (1985). Structures of news in the press. Discourse and Communication: New Approaches to the Analysis of Mass Media Discourse and Communication, 10, 69–93. Wang, W. M., Cheung, C. F., Lee, W. B., & Kwok, S. K. (2008). Mining knowledge from natural language texts using fuzzy associated concept mapping. Information Processing & Management, 44(5), 1707–1719. Zubrinic, K., Kalpic, D., & Milicevic, M. (2012). The automatic creation of concept maps from documents written using morphologically rich languages. Expert Systems with Applications, 39(16), 12709–12718.

843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900

Please cite this article in press as: Yeung, C. L., et al. A knowledge extraction and representation system for narrative analysis in the construction industry. Expert Systems with Applications (2014), http://dx.doi.org/10.1016/j.eswa.2014.03.044