An integrated proactive knowledge management model for enhancing engineering services

An integrated proactive knowledge management model for enhancing

engineering services

Ji-Wei Wu

, Judy C.R. Tseng

, Wen-Der Yu

⁎

, Jyh-Bin Yang

, Shun-Min Lee

, Wen-Nung Tsai

Dept. of Computer Science, National Chiao Tung University, Hsinchu, Taiwan, ROC

b_{Dept. of Computer Science and Information Eng., Chung Hua University, Hsinchu, Taiwan, ROC} c

Dept. of Const. Management, Chung Hua University, Hsinchu, Taiwan, ROC d

Grad. Inst. of Const. Engg. and Management, National Central University, Jhongli City, Taoyuan, Taiwan, ROC e

CECI Engineering Consultants, Inc., Taiwan, ROC f

Dept. of Multimedia and Game Science, Yu-Da University, Miaoli, Taiwan, ROC

a b s t r a c t

a r t i c l e i n f o

Article history:

Accepted 16 February 2012 Available online 15 March 2012 Keywords:

Engineering consulting Knowledge management Proactive problem-solving Business intelligence

More and more construction organizations have adopted Knowledge Management (KM) to enhance their engi-neering services. However, most of the traditional KM methods suffer from their“reactive mode” of problem solving. To cope with this problem, a newly developed model, the Integrated Proactive Knowledge Management Model (IPKMM), is proposed in this paper. A leading engineering consultingfirm in Taiwan was selected as a case study to implement the proposed model. The system implementation of IPKMM, the Integrated Proactive Knowl-edge Management System (IPKMS), is verified with real world cases. A novel Business Intelligence Index (BII) is also proposed in this paper to evaluate the relative competitiveness of different KM models. It is confirmed from the case study that IPKMM can significantly improve the efficiency of problem-solving and the competitiveness of an engineering consultingfirm in the service market. This study demonstrates that IPKMM has great potential in enhancing emergent problem-solving for engineering consultants.

© 2012 Elsevier B.V. All rights reserved.

1. Introduction

Construction is a knowledge-intensive industry and an experience-based discipline. It relies heavily on the knowledge and experience accumulated from previous projects. In Taiwan and many other countries, engineering consultingfirms have developed their own Knowledge Management Systems (KMSs) to enhance the firms' Knowledge Management (KM) initiatives in the past decade. The traditional KMSs adopt Communities of Practice (COPs) as an im-portant means of knowledge generation, sharing, exchanging, storing, and retrieving. A COP is defined as a group of people informally bound together by shared expertise and passion for a joint enterprise[1]. Emergent problem solving is an important application of the COP in engineering services. Problem solving is related to almost all kinds of engineering services including proposal preparation, feasibility studies, architectural and engineering designs, procurement, con-struction supervision, and project management.

A problem-solving process in a KMS is usually divided into two stages: (1) a problem is posed by a COP member (the questioner); then (2) other members (the responders) read the description of

the problem and provide their solutions voluntarily. The essential problem for such a problem-solving approach is that the raised prob-lem should wait for the responder (the so-called“domain expert”) who has the expertise or knowledge to solve the problem, to provide solutions. However, the domain experts are not always available to answer problems immediately. They may not see the raised problem or they may be too busy to answer the problem in time. Although pre-vious research has improved this drawback by developing an emer-gent problem-solving system (called the SOS system)[2], it was found that the average waiting time for the raised problems to be solved by domain experts is 2.68 days[3]. Such a Reactive Problem-Solving (RPS) approach is inef_{ficient in timeliness and cost} effective-ness. It is unacceptable for many emergent problems that require prompt responses, such as construction disasters or crises happening on site. Even for other problems, instant answering will undoubtedly improve the satisfaction of the clients. Hence, to shorten the problem-solving duration is very desirable in engineering consulting services. Yu et al.[4]have identified the key barrier that causes the delay of problem solving as the“reactive mode” of KM. That is, the problem raised by the questioner has to wait (passively) for responses from members of the COP in a KMS. Such a passive/reactive mode is related to the original design of the_{“knowledge creation organization” in} Nonaka's theory[5]. Such a barrier needs to be broken in order to im-prove the timeliness of problem solving and thus enhance the engi-neering services of a consulting_firm.

⁎ Corresponding author. Tel.: +886 3 5186748; fax: +886 3 5370517. E-mail address:[email protected](J.W. Yu).

0926-5805/$– see front matter © 2012 Elsevier B.V. All rights reserved.

doi:10.1016/j.autcon.2012.02.006

Contents lists available atSciVerse ScienceDirect

Automation in Construction

The presented research aims at enhancing engineering services through the Integrated Proactive Knowledge Management Model (IPKMM). The proposed IPKMM consists of an Enhanced Proactive Problem Solver (EPPS), which is based on the previously developed Model of Proactive Problem Solver (MPPS), a Knowledge Value Adding System (KVAS), and an Intellectualization System (IS). The integration of the abovementioned subsystems provides required functions for a proactive problem solver. A prototype system called the Integrated Pro-active Knowledge Management System (IPKMS) that implements the proposed IPKMM has been developed and integrated with the existing KMS of a leading engineering consultingfirm in Taiwan. A quantitative measure, namely the Business Intelligence Index (BII), is defined to evaluate the performance of the proposed IPKMM in enhancing the en-gineering services of an enen-gineering consultingfirm. Besides, several quantitative experiments for evaluating the performance of the pro-posed method are also conducted.

The rest of the paper is organized as follows. InSection 2, previous research by the research team and other teams related to IPKMM is revisited following the introduction in order to provide the required backgrounds. InSection 3, the proposed IPKMM is described in detail.

InSection 4, the IS system is demonstrated. InSection 5, a set of

com-prehensive quantitative experiments are conducted to evaluate the performance of the proposed IS. InSection 6, the BII is defined and employed to evaluate the proposed IPKMM. Finally, conclusions and recommendations are addressed inSection 7.

2. Review of related works

In recent years, Knowledge Management (KM) has been recognized as a core business concern and many KM related approaches have been proposed to enhance the performance of engineering services. Since construction is an experience-based discipline, KM in the construction industry plays a very important role. New works can be solved ef ficient-ly by reusing the knowledge and experience accumulated from previ-ous projects[6]. Hence, the business performance can be improved and the competitive advantage can be gained. In the construction indus-try, more and more companies have developed their own Knowledge Management Systems (KMSs)[6–8]. It has been proved that thefirms have received benefits from the KMS implementations.

On the other hand, due to the contracting nature, fragmented orga-nization, uncertain environment, and the changeable construction site, construction managers and engineers are faced with emergent prob-lems and crises in their daily business operations[2]. Problem-solving has become the fundamental activities in construction management [9_–12]. Previous researchers have pointed that the speci_fic characteris-tics of construction problems need to be tackled in order to solve them quickly, correctly, and cost-effectively. Such characteristics include

[9,12]: the ill-structure nature, inadequate vocabulary, little

generaliza-tion and conceptualizageneraliza-tion, and temporary multi-organizageneraliza-tion. Apart from the nature of construction problems, Dave and Koskela [13] noted that industry fragmentation and the ad-hoc nature of construc-tion projects further complicate the capture and reuse of valuable knowledge. Such a nature results in“dynamic knowledge” that needs constant updating to create new practices to achieve enhanced solu-tions[14]. Such a complicated nature has made the adoption of KMS in a construction organization unique compared with that in the other industries. In the following, we review some related works and experi-ences of the research team gained from the applications of KMS in an engineering consulting organization. Such works provide required backgrounds of the proposed Integrated Proactive Knowledge Manage-ment Model (IPKMM) in this paper.

2.1. Emergency problem solving system

In 2002, China Engineering Consultants, Inc. (CECI) developed an integrated system that combined the Knowledge Management

System (KMS) with an emergency problem-solving system (called the SOS system), namely the Knowledge Management integrated Problem-Solver (KMiPS)[2]. The SOS is a special subsystem of the KMS, which provides a tentative forum for solving emergency prob-lems encountered by engineers/managers. Once the problem is posed as an SOS-problem, it is posted on the SOS board on the portal page of the KMS for emergent discussion. Such an arrangement forces every member of the KMS to take a look at the posed problem when-ever he/she logs on. As a result, it generally receives attentions and usually has a better chance of being solved by responders. Problems posted on the SOS board which receive no response within one work-ing day (24 h) will be automatically removed and transferred to the relevant Communities of Practice (COPs). After this, it becomes a reg-ular topic for discussion in the relevant COPs.

The KMiPS demonstrated its capability in the improvement of the effectiveness of KMS for problem solving. It was measured by Yu et al. that the average time benefit of KMiPS is 63%; the average man-hour benefit is 73.8%; and the average cost benefit is 86.6%[15]. In another study, it was found that the KMiPS has shortened the average problem-solving time for emergency problems from 5.54 days to 2.68 days[3,6]. The underlying reason for this improvement in time effectiveness is the Medici's Effect[16]that brings together experts of all relevant disciplines and provides a forum for intersections of the domain experts with different contexts.

2.2. The Performance Improvement Strategy Planning (PISP) Model Yu et al.[11]proposed a model called Performance Improvement Strategy Planning (PISP) that combines a Knowledge-management-activity Surveying Module (KSM), a Benefit Quantification Module (BQM), a Performance Data-mining Module (PDM), and a Strategy Planning Module (SPM) to form an integrated system that suggests performance improvement strategies systematically. With the mass performance data measured by BQM, the Data Mining (DM) tech-niques can be applied tofind out effective strategies such as: “Devel-opment of a Pre-selection Mechanism” to screen out the insignificant problems,“Setup of a Warning System” to avoid wasting members' time on unsolvable problems[17].

The study of PISP also suggests building an“Expert Map” to direct the emergency problem to the most relevant domain experts, and de-veloping“Lesson-Learned Files” for questioners to find the solution directly. The above strategies suggest the need for a “proactive mode” of problem solving[17]which: (1) searches for a solution to the problem from previous lessons learnedfirst; or (2) if the solution is not found, directs the problem to the most appropriate domain ex-pert who knows the solution best. The abovementioned ideas resulted in the development of the Model of Proactive Problem-Solving (MPPS)[4].

2.3. Model of Proactive Problem-Solving (MPPS)

The MPPS was proposed to improve the drawbacks of the traditional reactive mode of KM[17]. The MPPS consists of four major components: (1) a Knowledge/Expert Map (K/EMap)—providing a classification scheme for the knowledge and expertise of the domain experts; (2) an Automatic Problem Answering (APA) module—solving the posed prob-lem automatically based on the historic probprob-lem-solving cases and lessons-learned; (3) an Automatic Problem Dispatching (APD) module— dispatching the posed problem to the most appropriate domain expert when the problem was not solved by APA; (4) a Lesson-Learned Wizard (LLW)—accumulating historic lessons-learned based on the classification scheme of K/EMap.

The major ideas behind MPPS are: (1) accumulating historic lessons-learned and classifying them according to some coding cri-teria; (2) identifying the domain experts related to a specific domain based on the classification of previous lessons; (3) searching for

historic solutions before posing problems in the COP; and (4) if the solution is not found, dispatching the problem to the domain experts who are most relevant to the problem.

The major contribution of MPPS is to establish a new framework for KM; however, it was soon discovered that the MPPS cannot work effectively if the historic lessons-learned are not suf_{ficient for} the diverse emergent engineering problems encountered by the firm. It is desirable to establish an automated method for collection of historic lessons learned.

3. Proposed Integrated Proactive Knowledge Management Model (IPKMM)

The Integrated Proactive Knowledge Management Model (IPKMM) is composed of three major components: an Enhanced Pro-active Problem Solver (EPPS), a Knowledge Value Adding System (KVAS), and an Intellectualization System (IS). The framework of the proposed IPKMM is shown inFig. 1.

Based on the framework of Model of Proactive Problem Solver (MPPS) [4], an Enhanced Proactive Problem Solver (EPPS) is pro-posed. The EPPS enhances the Lesson-Learned File (LLF) of the previ-ous MPPS with more comprehensive and general knowledge sources called the Intellectual Asset Repository (IAR) by introducing two new components—the Knowledge Value Adding System (KVAS) and the Intellectualization System (IS)—to automatically collect historical lessons-learned and Knowledge Cases (KCs) recorded in the Commu-nities of Practice (COPs), and to generate the intellectual assets of the IAR. The components of IPKMM are described in the following. 3.1. Enhanced Proactive Problem Solver (EPPS)

The EPPS is the kernel of the IPKMM; it performs the main func-tions of proactive problem solving. The major components of EPPS are described in the following:

1. Knowledge/Expert Map (K/EMap)

The EPPS is focused on the knowledge and the experts holding the knowledge. In EPPS, the domain knowledge is represented by

Knowledge Map (KMap), while the domain experts are character-ized by Expert Map (EMap)[4]. The Knowledge Map and Expert Map (K/EMap) provide the ontology for modeling the knowledge repository of the engineering consultingfirm. In this paper, the K/EMap refers, but not limited, to that of the target engineering consulting_firm[4]. The K/EMap of the target engineering consult-ingfirm is constructed by a Multi-dimensional Knowledge Ontolo-gy (MKO) consisting of three dimensions: lifecycle code, product code, and technical code. 270 engineering areas are contained in the K/EMap, including civil, structural, geotechnic, environmental, transportation, and construction engineering.

2. Automatic Problem Answering (APA) module

The APA[4]is an Automatic Problem-solving System (APS). Once a problem is issued by the questioner, APA will automatically search the IAR tofind out the most relevant solutions, and then these so-lutions are given to the questioner.

3. Automatic Problem Dispatching (APD) module

In the APD module[4], the unsolved problem is automatically dis-patched to the most relevant experts. The problem characteristics analyzed in the APA module are used tofind the most appropriate domain experts based on the EMap described previously. Then, the problem is dispatched to the most relevant domain experts for a possible solution. Finally, the experts respond to the problem in a special COP called SOS.

4 Intellectual Asset Repository (IAR)

The IAR is an extension of the LLF[4]in the original MPPS. The IAR consists of three major sources of knowledge: (1) LLFs—obtained from previous experiences of problem solving in Knowledge Man-agement Integrated Problem-Solver (KMiPS) and COPs; (2) Knowledge Cases (KCs)—a compilation of the reusable knowledge sharing cases by the KVAS from COPs; (3) Corpuses—the knowl-edge corpuses generated automatically by the IS.

3.2. Knowledge Value Adding System (KVAS)

The KVAS is comprised of three components[18](1) the Knowledge Value Adding Survey Module (KVASM)—a web-based questionnaire sys-tem that surveys the COP members participating in discussions of

problem-solving or knowledge-sharing processes; (2) the Knowledge Value Adding Quantification Module (KVAQM)—a module that gener-ates Knowledge Value Adding (KVA) values by quanti_{fication of KM} per-formance with the questionnaire survey results; and (3) the Lesson-Learned Recorder (LLR)—a computer aided information system that helps the COP members to compile and record previous learned knowl-edge, named Lesson-Learned Files (LLFs), that is useful to solve future problems. The outputs of KVAS are: (1) the Lesson-Learned Files (LLFs) and Knowledge Cases (KCs); and (2) the calculated Knowledge Value Adding (KVA) values associated with all surveyed cases and the partici-pants (the domain experts). As the KVA values are calculated, the rele-vance of a specific LLF/KC and its associated domains (classified codes) is established. The relevance is used in searching for the historic LLF/KC for problem solving andfinding out the most appropriate domain ex-perts when the solution is not found.

3.3. Intellectualization System (IS)

The IS is a text-mining based corpus generation system that pro-duces knowledge corpuses automatically from existing documents (the explicit knowledge) of thefirm, such as final reports, plans, pro-posals, notes, minutes, etc. The current version of IS can tackle corpus generation tasks for three types of corpuses: (1) a text corpus—the segment of text that contains specialized meaning related to a specific domain; (2) a table corpus—a table that contains special meaning re-lated to a specific domain; and (3) a figure that contains special meaning related to a specific domain. Details of the functionality of the IS are described inSection 4.

3.4. Integrated operations of IPKMM

The integrated operation of the IPKMM is shown inFig. 1, where the solid arrows show the problem-solving processes and the dashed arrows show the intellectual asset generation processes. The problem-solving process starts with the questioner posing a question in EPPS. The APA of the EPPS searches the KMap tofind candidate so-lutions from the most relevant Intellectual Assets for the questioner. If the questioner is not satisfied with all candidate solutions, the ques-tion is diverted to the relevant COP. Simultaneously, the EPPS searches the EMap tofind out the most appropriate domain experts, and a message is automatically sent to those experts. The selected ex-perts are requested to respond to the question with their solutions in the COP. Should the problem be solved, the participants in the problem-solving process are surveyed via KVAS, and a new LLF is gen-erated in the IAR.

4. Knowledge extraction through the Intellectualization System As addressed at the end ofSection 2, the critical bottleneck of problem-solving process in the previously developed Model of Proactive Problem Solver (MPPS) was due to the insufficient Lesson-Learned Files (LLFs) to provide acceptable solutions for the ques-tioners. This is resolved in the Integrated Proactive Knowledge Management Model (IPKMM) with the Intellectualization System (IS), which can extract explicit knowledge (namely knowledge cor-puses), represented as text, table orfigure corpuses, from engineering documents such as the project'sfinal reports, plans, proposals, and other knowledge documents. Three types of knowledge corpuses are generated by IS, i.e., text, table andfigure corpuses.

The text corpuses are generated via the following steps: (1) ana-lyzing the structure of the document with the Structure Analysis Module (SAM); (2) the structured document is segmented into puses with the Semantic Segmentation Module (SSM); (3) the cor-puses are classified with the Knowledge Categorization Module (KCM); (4) the classified corpuses are stored in the Knowledge Cor-pus Base (KCB); and (5) finally, the corpuses stored in KCB are

integrated with other existing corpuses by the Knowledge Integration Module (KIM).

The table and_{figure corpuses are generated via a similar process,} as follows: (1) analyzing the structure of the document with SAM; (2) thefigures/tables are extracted with the Figure and Table Extrac-tion Module (FTEM); (3) the_{figures/tables are classified with the} Fig-ure and Table Categorization Module (FTCM); (4) the classified figure/table corpuses are stored in KCB; and (5) finally, the figure/ table corpuses stored in KCB are integrated with existing corpuses by the Knowledge Integration Module (KIM).

4.1. Structure Analysis Module (SAM)

In the Structure Analysis Module (SAM), headings of chapters and sections (such as“Chapter 1 Introduction”) included in a knowledge document are extracted and organized as a structured list of content with multiple levels. The structured list of headings is useful for users to grasp the overall structure of the document.

To implement SAM, existing Microsoft Office® Word functions are employed to extract predefined headings, which are defined by the authors through Microsoft Office® Word built-in functions, such as Bullet and Style. However, according to the examples collected for the case study, most of the documents in the real world contain no predefined headings. Authors usually key in the headings manually, instead of using the Microsoft Office® Word functions. In such cases, headings are not able to be extracted by Microsoft Office® Word functions.

To cope with this problem, a heading extraction method is devel-oped. Firstly, the documents are analyzed. A heading usually consists of“Chapter” or “Section” followed by an ordinal number and a brief description, such as“Chapter 1 Introduction”. In other cases, a head-ing may only consist of an ordinal number and a brief description, such as“1. Introduction”. The commonly used forms of headings are collected from the documents beforehand. Each of the collected forms of headings is then compiled as a regular expression[19]. This provides a basis for automatically constructing the structured list of a document.

Secondly, each paragraph in a knowledge document is evaluated to decide whether it is a heading or not. If a paragraph is formed as a heading and the paragraph is too short (i.e., the length of the para-graph is less than n words, n is empirically designated), the parapara-graph will be considered as a heading. To identify whether a paragraph is a heading, the regular expressions are used to match with the para-graph. The paragraph is regarded as a heading if one of the regular ex-pressions matches the form of the paragraph.

Finally, the headings are used to construct the structured list of a document. Each of the headings is assigned a number representing its level in the structured list. The level of a heading is assigned according to its extracted order and the regular expression it matches. For example, thefirst extracted heading is assigned as level one. If the second extracted heading matches the same regular expression as that of the first heading, the second heading will be assigned as level one as well. Otherwise, the second heading will be assigned as level two, and so forth.

4.2. Semantic Segmentation Module (SSM)

The purpose of semantic segmentation is to divide a document into several shorter segments, with the sentences within a segment sharing a subtopic. Semantic segmentation enables the Automatic Problem Answering (APA) of Enhanced Proactive Problem Solver (EPPS) to provide only the fragment(s) of text that the user is inter-ested in, rather than the whole document. It will save the questioner great effort and time as it is not necessary to read through the whole document.

To segment a document into several thematic segments, topic boundaries between text paragraphs need to be identified. At first, each chapter of a document is divided into the sections and subsec-tions indicated by the author. The regular expressions applied in the Structure Analysis Module (SAM) are used as delimiters to split a doc-ument into sections and subsections.

A chapter/section usually consists of several subsections, each of which consists of paragraphs which share the same subtopic. Based on such an observation, the boundaries of subsections are regarded as the topic boundaries, and hence each subsection is regarded as a semantic segment. Finally, all the semantic segments are regarded as knowledge corpuses and saved into Intellectual Asset Repository (IAR).

4.3. Figure and Table Extraction Module (FTEM)

In the Figure and Table Extraction Module (FTEM), bothfigures and tables included in knowledge documents are also extracted as knowledge corpuses. Extracted _{figures and tables are correlated} with the most relevant semantic segments to provide integrated knowledge corpuses. Hence, knowledge corpuses included in a docu-ment could be completely utilized while solving emergent problems. The process of FTEM can be divided into three stages. Firstly, fig-ures and tables are extracted from knowledge documents. The built-in object model provided by Microsoft Office® Word is employed to extractfigures and tables from the knowledge documents. Secondly, figure captions and table captions are extracted. Generally speaking, figure captions are located below figures while table captions are lo-cated above tables. For this reason, FTEM uses regular expression to search for paragraphs which include the keyword“Figure” and are short (e.g., the length of the paragraph is less than ten words), located below a_{figure, as the figure caption. A similar process is conducted} for extracting table captions. Finally, the extractedfigures, tables, fig-ure captions, and table captions are saved in IAR.

4.4. Knowledge Categorization Module (KCM)

The Knowledge Categorization Module (KCM) is used to classify semantic segments created by SSM. KCM applies Case Based Reason-ing (CBR)[20,21]to deduce the knowledge category of a semantic segment. CBR is an artificial intelligence method that finds solution for a new problem by using the previous experience. In KCM, accord-ing to the categories of previous Knowledge Cases (KCs) and Lesson-Learned Files (LLFs), which are served as training cases, the knowledge category of each semantic segment can be deduced automatically.

The process of knowledge categorization can be divided into three stages: the similarity between a semantic segment and each training case is firstly determined by the similar method described in

Section 4.2. Then the average similarity of each knowledge category

(represented with a classification code [6]) is computed. Finally, knowledge categories with the top 3 average similarities are regarded as the knowledge categories of the semantic segment. For example, suppose there is a semantic segment Seg1created in SSM, the

similar-ities between Seg1and each knowledge category are computed based

on a previously developed algorithm[4]and shown inTable 1. The summarized similarity of each knowledge category is shown in

Table 2. The knowledge categories with the top 3 average similarities,

“510D15”, “416D80” and “071X10”, are regarded as the knowledge categories of Seg1.

4.5. Figure and Table Categorization Module (FTCM)

The Figure and Table Categorization Module (FTCM) is used to classify figures and tables according to the knowledge categories of the semantic segments referring to them. For example, suppose

“Fig. 1” is referred to by semantic segments Seg1and Seg2which

be-long to the knowledge categories shown inTable 3, the knowledge categories of“Fig. 1” are the union of the knowledge categories of Seg1and Seg2(i.e., 510D15, 416D80, 071X10 and 542D15).

4.6. Knowledge Integration Module (KIM)

The Knowledge Integration Module (KIM) is used to integrate KCB, KCs, and LLFs into structured knowledge assets containing se-mantic index. The Vector Space Model (VSM)[22–24]has been the most popular model to represent a document in thefield of informa-tion retrieval. Due to its popularity, the integrainforma-tion of the unstruc-tured knowledge assets is based on VSM. At first, the original knowledge assets (i.e., KCB, KCs, and LLFs), which are in the form of unstructured natural language discourses, are processed by domain keywords extraction and importance weighting identification. After the processing, each of the unstructured knowledge assets is repre-sented as a Characteristic Vector:

CV KAð iÞ ¼ ðk1; wi1Þ; kð2; wi2Þ;…; kj; wij



; kð n; winÞ

n o

ð1Þ where CV(KAi) represents the Characteristic Vector of the knowledge

asset i; kjrepresents keyword j; and wijrepresents the importance

weighting value of kjin knowledge asset KAi. The importance

weight-ings of the keywords are calculated by using the Importance Factor (IMF) method[25]. Finally, the characteristic vectors are stored in the IAR as the structured knowledge assets. The knowledge assets are used as the sources for searching solutions in solving future problems.

5. System testing and performance evaluation

A computer system that implements the proposed IPKMM has been developed with the similar name, Integrated Proactive Knowl-edge Management System (IPKMS). In order to evaluate the problem-solving performance of IPKMS, 200 testing sets were ran-domly selected from the testing data of our previous work on the Model of Proactive Problem Solver (MPPS) (as mentioned in

Section 2.3)[6], where the testing problems were comprised of two

different sources: (1) original problem set—consists of the 908 LLFs; and (2) similar problem set—consists of 1304 derived problems, each of which is a similar problem description modified from one of the original LLFs. Totally 1304 similar problems were generated by 63 domain experts who were managers/senior engineers of the case engineering consultingfirm. Each problem description of the 908

Table 1

Similarities between an example semantic segment Seg1and each of the knowledge categories.

Case ID Knowledge category Similarity

1 416D80 0.4 2 542D15 0.2 3 071X10 0.5 4 542D15 0.2 5 510D15 0.7 6 542D15 0.2 Table 2

Summarized similarities of the knowledge categories inTable 1.

Knowledge category Summation of similarity Average similarity

510D15 0.7 0.7

071X10 0.5 0.5

416D80 0.4 0.4

LLFs was presented to the 63 domain experts to provide 1 to 3 similar but differently articulated problem descriptions for testing.

In this study, the 100 original problems were randomly sampled from the original problem set and denoted as Test Set 1; while the other 100 similar problems were selected from similar problem set and denoted as Test Set 2, which are associated with the selected original problems.

In order to show the benefits of the proposed Intellectualization System (IS) for problem-solving, this study focuses on evaluating the capability of IPKMS in providing the most relevant solutions for each posed problem. In other words, the aim was to validate whether or not IS could nourish the Intellectual Asset Repository (IAR), so that IPKMS could provide the more relevant solutions for each problem. For each of the test sets, two experiments were performed: (1) IPKMS search solutions (through Automatic Problem Answering (APA)) from LLFs only; (2) IPKMS search solutions (through APA) from the IAR which consists of LLFs, Knowledge Cases (KCs), and knowledge corpuses extracted from the engineering documents. Av-erage similarity was adopted as the measure for evaluating the per-formance of IPKMS: AvgSim¼₁₀₀1 X 100 i¼1 Xn j¼1 Simij n ð2Þ

where Simijrepresents the similarity between problem i and retrieved

solution j; n represents the number of all retrieved solutions. The evaluation results of IPKMS for the two test sets based on Eq.(2)are shown inFigs. 2 and 3. It is found that the average similar-ity decreases as n increases. This is expected because the retrieved so-lutions are sorted by the similarity between problem i and each of the retrieved solutions. The average similarity is highest when only the most relevant solution is retrieved (i.e. n = 1) and the average simi-larity decreases gradually when n > 1. More retrieved solutions may lead to lower average similarity, but will improve the performance of problem solving because the user can alwaysfind the most rele-vant solutions from the top n retrieved solutions and can alsofind more relevant solutions while he/she explores the other solutions.

As can be seen fromFigs. 2 and 3, under different n settings, the average similarity is higher when the LLFs, KCs, and knowledge cor-puses extracted from the knowledge documents are incorporated in IAR than merely using LLFs. This means that using IPKMS to search

for solutions (through APA) from the IAR (comprising LLFs, KCs, and knowledge corpuses) will improve the relevance of the obtained so-lutions for the posed problems. As a result, the proposed IPKMS is verified to be able to enhance IAR and hence improve construction problem solving.

6. Measurement of organizational business intelligence

In order to measure the business intelligence capacity of an engi-neering consulting organization, an operational indicator, the Busi-ness Intelligence Index (BII) for engineering consulting firms, is defined. The definition of BII and analysis of the problem-solving per-formance in three different Knowledge Management (KM) imple-mentation phases are described in the following.

6.1. Defining the Business Intelligence Index (BII) for an engineering con-sultingfirm

The proposed Integrated Proactive Knowledge Management Model (IPKMM) and its prototype system, Integrated Proactive Knowledge Management System (IPKMS), have been implemented for the case engineering consultingfirm, China Engineering Consul-tants, Inc. (CECI). In order to evaluate the performance of the pro-posed IPKMM, a measurement, namely Business Intelligence Index (BII) is defined in Eq.(3)to measure the relative competitiveness of different KM implementations.

BII¼½IA pð ; i; cÞ

S

T : ð3Þ

In Eq.(3), the IA(p, i, c) is the collection of all Intellectual Assets (IAs) that are functions of: (1) the staff of the organization (p); (2) the internal process (i); and (3) the customers (c); S is the“times of knowledge-sharing” (or share number) of the IA, which can be measured by the number of COP members participating in the problem-solving process; T is the responding time required tofind the solution.

The numerator of Eq.(3)represents the capacity of intelligence of the organization. According to Nonaka[5]and Johansson[16], more intersections of the people with different contexts shall result in more valuable knowledge creations for problem solving. Due to the leverage of knowledge in Knowledge Management System (KMS), the share number of IA should be exponential in measuring BII. Therefore, more IA will result in a higher BII; a higher share number will increase the BII exponentially. In the proposed IPKMM, there are three types of IAs, as depicted in Fig. 1 and explained in

Section 3.1: (1) the Lesson-Learned Files (LLFs) compiled from

emer-gency problem solving system (SOS);(2) the Knowledge Cases (KCs) recorded from the voluntary discussions in COPs; and (3) the

Table 3

Test sets used for the performance evaluation of IPKMS. Type of sample Description

Test set 1 100 existing problems The same problems exist in LLFs Test set 2 100 similar problems Similar problems as those in test set 1

1 2 3 4 5 6 7 8 9 10 LLF 1.000 0.652 0.512 0.434 0.384 0.347 0.320 0.298 0.281 0.266 IAR 1.000 0.704 0.579 0.507 0.459 0.423 0.396 0.373 0.355 0.338 0.000 0.100 0.200 0.300 0.400 0.500 0.600 0.700 0.800 0.900 1.000 Average Similarity LLF IAR n

Fig. 2. Comparison of average similarity in test set 1.

1 2 3 4 5 6 7 8 9 10 LLF 0.670 0.480 0.394 0.342 0.309 0.284 0.265 0.249 0.237 0.226 IAR 0.697 0.550 0.478 0.434 0.402 0.379 0.359 0.342 0.328 0.316 0.000 0.100 0.200 0.300 0.400 0.500 0.600 0.700 0.800 0.900 1.000 A v erage Similarity LLF IAR n

knowledge corpuses that are automatically generated from engineer-ing documents with Intellectualization System (IS).

The denominator of Eq.(3)implies that the shorter responding time will increase the value of BII since problem-solving is always time-constrained. This is especially true for engineering consulting services, since the clients of engineering projects usually require the consultants to respond to their problems promptly. Delays in problem solving will not only dissatisfy the client but also reduce the value of the solution.

The value of BII can be obtained by substituting the estimated values of IA(p, i, c), S, and T into Eq.(3). An illustrative example of BII calculation for the case engineering consultingfirm is demonstrat-ed in the following subsection.

6.2. Measuring the BII of the case engineering consultingfirm

The BII defined previously is applied to the case engineering con-sulting firm to measure the problem-solving performance in the three different KM implementation phases: (1) phase with no KM im-plementation; (2) reactive KM phase; and (3) IPKMM implementa-tion phase. Assume that the IA of the organizaimplementa-tion is the same for the three periods. Before KM implementation, the problem solving is conducted by group meetings with an average of 4.26 staff mem-bers; the number of problem-solving participants increased to 7.14 after KMS was established[3]. To be conservative, here we assume that the implementation of IPKMM involves the same number of staff as the reactive KM period in the problem solving. The responding times for the three periods are on average[3]: 5.54 days before KM implementation and 2.68 days for reactive KM, but about 5 s after IPKMM implementation. Although the reactive KM and IPKMM need additional time to set up a knowledge management system, the in-stallation is conducted only once. In the long term, the set up time can be ignored. Therefore, the set up time required to construct a knowledge management system is not taken into account in the responding times for the three periods. Substituting the above pa-rameters into Eq.(3), the ratios of BII for the traditional approach: BII of reactive KM:BII of IPKMM are shown in Eq.(4):

BIItrad:: BIIreact:KM: BIIIPKMM

¼½IA p₄₇₈ð ; i; c_{; 656 s}Þ4:26:½IA p₂₃₃ð ; i; c_{; 280 s}Þ7:14:½IA pð ; i; c_{5 s}Þ7:14: ð4Þ To measure the BII of the case engineering consultingfirm more specifically, totally 908 LLFs, 20,045 KCs (from 36 COPs), and 522 knowledge documents were collected from 50 real world projects of the case A/E consultingfirm. The LLFs and KCs were collected from Jan. 2005 to Aug. 2009. Finally, 3665 semantic paragraphs (consisting of 70,519 paragraphs), 11,959figures, and 3223 tables were extracted from 522 knowledge documents by IS.

Substitute“39,800” (the total number of LLFs, KCs, and knowledge corpuses extracted) for the base, IA, then substitute the responding time with 5.54 days, 2.68 days, and 5 s for the three methods, respec-tively. Considering the traditional approach (no KM implementation) as the basis, the ratios of Eq.(4)become Eq.(5):

BIItrad:: BIIreact:KM: BIIIPKMM

¼½39; 800 4:26 478; 656 s : 39_{; 800} ½ 7:14 233; 280 s : 39_{; 800} ½ 7:14 5 s ≈1013 : 1027 : 1032 : ð5Þ

Using the order of the components in Eq.(5)as indexes, the BII measure for the three periods of KM implementation are 13, 27, and 32 respectively. The BII index provides a simple and meaningful index of competitiveness for the engineering consulting industry to measure their problem-solving capabilities of thefirms in providing engineering services.

7. Conclusion and future work 7.1. Conclusions

Engineering consulting is a knowledge intensive industry. Problem solving has been the hard core to almost all kinds of engineering ser-vices including proposal preparation, feasibility studies, architectural and engineering designs, procurement, construction supervision, and project management. More and morefirms have adopted knowledge management systems as initiatives to enhance their professional ser-vices, especially for emergency problem solving. However, the under-lying “reactive mode” of the traditional Knowledge Management System (KMS) has hindered further shortening of problem-solving time and resulted in ineffective utilization of the organization's intel-lectual assets. This research proposes a novel model, namely the Inte-grated Proactive Knowledge Management Model (IPKMM) for proactive problem solving. The proposed IPKMM consists of an En-hanced Proactive Problem Solver (EPPS), a Knowledge Value Adding System (KVAS), and an Intellectualization System (IS). The integration of the three subsystems supports the required functionalities of a “proactive mode” of Knowledge Management (KM), which dramati-cally changes the paradigm of the traditional KM approach to problem-solving in engineering consulting.

The proposed IPKMM does not only provide a useful framework for implementation of proactive knowledge management, but also estab-lishes a mechanism to accumulate the business intelligence in the Intel-lectual Asset Repository (IAR) that is essential for competing in the market. The knowledge corpuses embedded in reports and documents, which are collected over the years in an engineering consultingfirm, are extracted by IS automatically. The extracted knowledge corpuses can nourish IAR and be utilized for solving future emergent problems. Thus, the capability of problem solving in traditional KMS is enhanced. A Business Intelligence Index (BII) is proposed and defined as an operational index to measure the organization's competitiveness level in providing their engineering services. With BII, the case engi-neering consultingfirm, China Engineering Consultants, Inc. (CECI), has an index value of 13 before KM implementation. The index creases to 27 after adopting KMS. It is found that the BII further in-creases to 32 as the IPKMM is implemented. It is concluded that the proposed IPKMM can significantly improve the performance of the professional services for engineering consultingfirms, and thus en-hance their competitiveness in the service market.

One major limitation of the proposed IPKMM is the requirement of user involvement. In KVAS, COP members have to participate in a questionnaire survey for quanti_{fication of KM performance. Also, in} Knowledge Categorization Module (KCM) of IS, COP members have to provide the categories of previous Knowledge Cases (KCs) and Lesson-Learned Files (LLFs) for collecting training cases to deduce the knowledge categories of each semantic segment. Another limita-tion found from the case study is the misclassification problem. Some of the training cases (i.e., LLFs and KCs) were not classified correctly, which can lead to incorrect knowledge categorization in KCM. 7.2. Future work

In the current stage, the Integrated Proactive Knowledge Manage-ment System (IPKMS) is installed in China Engineering Consultants, Inc. (CECI). After the setup, valuable information, such as user feed-backs, can be collected for further validating and verifying the pro-posed method.

As mentioned inSection 4.1, we have observed that most real world knowledge documents do not adopt predefined headings. Au-thors of the documents usually key in the headings manually, instead of using the Microsoft Office® Word built-in functions. In such cases, the extraction performance deteriorates. It requires a more complex implementation of the Structure Analysis Module (SAM). As the

performance of the IS could be improved when the knowledge docu-ments are well prepared, a document preparation guide for authors has been developed by the research team and provided to the case engineering consultingfirm as a facilitating tool for the document preparers. Besides, it is important to review whether documents are well written before submitting them to the system. To this end and to reduce the reviewer's burden, an automatic program should be de-veloped for automatic document checks.

The powerful functionality of proactive knowledge management pro-vided by the proposed IPKMM cannot only be applied to emergency problem solving; it also offers potential benefits to other construction op-erations as long as knowledge and experiences are incorporated. Future applications of the proposed IPKMM for design, procurement (or bid-ding), and construction processes will be explored by the research team. Acknowledgments

The valuable case study data presented in this paper was provided by CECI, Taipei. The authors would like to express sincere apprecia-tion to the involved staffs of the Department of Business and Research and the Department of Information Systems, CECI Engineering Con-sultants, Inc., Taipei, Taiwan.

Appendix A List of acronyms

APA Automatic Problem Answering APD Automatic Problem Dispatching APS Automatic Problem-solving System BII Business Intelligence Index BQM Benefit Quantification Module CBR Case Based Reasoning

CECI China Engineering Consultants, Inc. COP Communities of Practice

DDS Decision Support Systems DM Data Mining

EMap Expert Map

EPPS Enhanced Proactive Problem Solver FTCM Figure and Table Categorization Module FTEM Figure and Table Extraction Module IA Intellectual Asset

IAR Intellectual Asset Repository

IPKMM Integrated Proactive Knowledge Management Model IPKMS Integrated Proactive Knowledge Management System IS Intellectualization System

KC Knowledge Case KCB Knowledge Corpus Base

KCM Knowledge Categorization Module K/EMap Knowledge/Expert Map

KIM Knowledge Integration Module KM Knowledge Management KMap Knowledge Map

KMiPS Knowledge Management integrated Problem-Solver KMS Knowledge Management System

KSM Knowledge-management-activity Surveying Module KVA Knowledge Value Adding

KVAQM Knowledge Value Adding Quantification Module KVAS Knowledge Value Adding System

KVASM Knowledge Value Adding Survey Module LLF Lesson-Learned File

LLR Lesson-Learned Recorder LLW Lesson-Learned Wizard

MKO Multi-dimensional Knowledge Ontology MPPS Model of Proactive Problem Solver PDM Performance Data-mining Module

PISP Performance Improvement Strategy Planning

RPS Reactive Problem-Solving SAM Structure Analysis Module SPM Strategy Planning Module SSM Semantic Segmentation Module VSM Vector Space Model

References

[1] E.C. Wenger, W.M. Snyder, Communities of practice: the organizational frontier, Harvard Business Review (January–February 2000) 139–145 (Product Number: R00110).

[2] W.D. Yu, C.T. Lin, C.T. Yu, S.J. Liu, H.C. Luo, P.L. Chang, Integrating emergent problem-solving with construction knowledge management system, Proceedings of the CME 25 Conference, July 16–18, 2007, University of Reading, Reading, UK, 2007, (10 pp.). [3] W.D. Yu, J.B. Yang, J.C.R. Tseng, Introduction of PISP model to CECI and the pilot study of PPS, Internal Final Report to CECI, Project No.: 96940, Taipei, Taiwan, 2007, (in Chinese).

[4] W.D. Yu, J.B. Yang, J.C.R. Tseng, C.T. Yu, Model of proactive problem-solving for construction knowledge management, Proceedings of 2007 International Sympo-sium on Automation and Robotics in Construction, Session 1-C—Management and Production Aspects (1), Sept. 19–21, 2007, Kochin, India, 2007, pp. 493–500. [5] I. Nonaka, A dynamic theory of organizational knowledge creation, Organization

Science 5 (1) (1994) 14–37.

[6] W.D. Yu, J.B. Yang, J.C.R. Tseng, S.J. Liu, J.W. Wu, Proactive problem-solver for con-struction, Automation in Construction 19 (6) (2010) 808–816.

[7] T. Mezher, M.A. Abdul-Malak, L. Ghosn, M. Ajam, Knowledge management in me-chanical and industrial engineering consulting: a case study, Journal of Manage-ment in Engineering, ASCE 21 (3) (2005) 138–147.

[8] T. Mezher, M.A. Abdul-Malak, M. Khaled, L. El-Khatib, Building a knowledge man-agement system in a designfirm: the case of XYZ structural department, Journal of Cases on Information Technology 11 (3) (2009) 1–17.

[9] H. Li, P.E.D. Love, Developing a theory of construction problem solving, Construc-tion Management and Economics 16 (6) (1998) 721–727.

[10] G. Runeson, The future of building research, Australian Institute of Building Pa-pers 5 (1994) 3–8.

[11] W.D. Yu, P.L. Chang, S.J. Liu, A model of performance improvement strategy planning for a construction knowledge management system, in: B.H.V. Topping, G. Montero, Montenegro (Eds.), Proceedings of The Fifth International Conference on Engineer-ing Computational Technology (ECT 2006), Sept. 12–15, Las Palmas de Gran Canaria, Spain, Session VII—Modeling of Decision Making and Risk Analysis in Construction Management, Civil-Comp Press, Stirlingshire, Scotland, 2006, pp. 70-1–70-18. [12] G.F. Smith, Towards a theory of managerial problem—solving, Decision Support

Systems 8 (1) (1992) 29–40.

[13] B. Dave, L. Koskela, Collaborative knowledge management—a construction case study, Automation in Construction 18 (7) (2009) 894–902.

[14] P. Chinowsky, K. Molenaar, A. Realph, Learning organizations in construction, Journal of Management in Engineering, ASCE 23 (1) (2007) 27–34.

[15] W.D. Yu, P.L. Chang, S.J. Liu, Quantifying benefits of knowledge management sys-tem: a case study of an engineering consultingfirm, in: N. Kano, T. Arai (Eds.), Proceedings of International Symposium on Automation and Robotics in Con-struction 2006 (ISARC 2006), Session A4—Planning and Management (1), Oct. 3–5, 2006, Tokyo, Japan, 2006, pp. 124–129.

[16] F. Johansson, The Medici Effect: Breakthrough Insights at the Intersection of Ideas, Concepts, and Cultures, Harvard Business School Press, Cambridge, MA, USA, 2002. [17] W.D. Yu, C.T. Lin, C.T. Yu, S.J. Liu, H.C. Luo, P.L. Chang, Mining knowledge in KMS

performance data—a case study of an A/E consulting firm, in: S.M. Ahmed, S. Azhar, S.T. Mohamed (Eds.), Proceedings of the Fourth International Conference on Construction in the 21st Century (CITC-IV), Session of Knowledge Manage-ment in Construction, July 11–13, 2007, Gold Coast, Australia, 2007, pp. 574–582. [18] W.D. Yu, P.L. Chang, H.H. Yao, S.J. Liu, KVAM: model for measuring knowledge management performance of engineering community of practice, Construction Management and Economics 27 (8) (2009) 733–747.

[19] J.E. Hopcroft, R. Motwani, J.D. Ullman, Introduction to Automata Theory, Lan-guages, and Computation, third ed. Addison Wesley, Boston, MA, 2006. [20] Q.F. Sun, L. Tan, L. Wang, C.C. Yu, Application of case-based reasoning technology

in decision-making system of rural new energy utilization, Proceedings of the 3rd International Conference on Information Management, Innovation Management and Industrial Engineering, Kunming, China, 2010, pp. 235–239.

[21] A. Aamodt, E. Plaza, Case-based reasoning: foundational issues, methodological variations, and system approaches, AI Communications 7 (1) (1994) 39–59. [22] G. Salton, M.J. McGill, Introduction to Modern Information Retrieval,

McGraw-Hill, New York, 1983.

[23] G. Salton, A. Wong, C.S. Yang, A vector space model for automatic indexing, Com-munications of the ACM 18 (11) (1975) 613–620.

[24] J.C.R. Tseng, G.J. Hwang, Development of an automatic customer service system on the internet, Electronic Commerce Research and Applications 6 (1) (2007) 19–28. [25] J.W. Wu, C.C. Wei, J.C.R. Tseng, Development of an enhanced web-based

automat-ic customer servautomat-ice system, Proceedings of the 2007 International Conference on Enterprise Systems and Application (ICESA), Taipei, Taiwan, 2007.