Know-how Sharing Using a Knowledge Sharing System KIDS A Knowledge Management Practice at a Research Laboratory Yasuko Nakayama Kazuo Sumita Kouichi Sasaki Corporate R&D Center Corporate R&D Center Corporate R&D Center Toshiba Corp., 212-8582 Japan Toshiba Corp., 212-8582 Japan Toshiba Corp., 212-8582 Japan yasko.nakayama@toshiba.co.jp kazuo.sumita@toshiba.co.jp kouichi.sasaki@toshiba.co.jp Toshihiko Manabe Masaru Suzuki Corporate R&D Center Corporate R&D Center Toshiba Corp., 212-8582 Japan Toshiba Corp., 212-8582 Japan toshihiko.manabe@toshiba.co.jp masaru1.suzuki@toshiba.co.jp enterprise. Consequently, how best to facilitate knowledge sharing within organizations has emerged as an Abstract important issue for management [Apo98]. The capability of individuals and organizations to gather, accumulate and We developed a knowledge and information on dispatch information has improved due to information demand system, KIDS, which structures free-form technology such as groupware, document management text as knowledge to provide the user with systems, and intra- or internet-based search. It has become knowledge fit for his natural language query. For clear that two factors, namely system architecture and a period of four years, we carried out knowledge richness of content, have a particular important bearing on sharing experiments using the system at a the ability of a knowledge sharing system to function research center that includes around 1,000 effectively. researchers engaged in various fields such as computer science, display devices, advanced LSIs Nonaka and Takeuchi modeled the knowledge creation in and mechanical systems. In the experiment we a company as a knowledge conversion cycle between focused on sharing the knowledge processed by individuals; technical papers, patents and the like explicit knowledge and tacit knowledge [Non95]. The were excluded from the scope of the experiments explicit knowledge is knowledge that could be transmitted because a document management system has using formal and logical language such as a technical already been implemented. An individual’s document or a manual. On the other hand, the tacit knowledge includes a knack or tips useful in the knowledge is difficult to formalize and could not be performance of research. The result shows the transmitted using formal and logical language, because feasibility of integrating organized domain such knowledge is contextual. An individual’s tacit knowledge and personal free-form know-how. knowledge is difficult to transmit without his experiences We discuss key factors for success in knowledge and actions. By accumulating individuals’ knowledge with sharing such as text structurization, building a the context and by making a network of the knowledge, knowledge sharing culture, knowledge quality the creativity of individuals is raised and the creativity of maintenance. the whole organization increases. 1 Introduction In order to develop a method of accumulation, retrieval 1.1 Rich stock of knowledge is key for success and practical use of useful knowledge for an organization and its individuals, it is first necessary to collect the There is growing recognition that intellectual capital, organization’s intellectual capital and promote knowledge including individuals’ know-how, technical skills and externalization. cognitive skills, constitutes a valuable resource for any 1.2 Background knowledge for comprehension The copyright of this paper belongs to the paper’s authors. Permission to copy without fee all or part of this material is granted provided that the copies are not Among researchers there is keen interest in advanced made or distributed for direct commercial advantage. information retrieval, classification, extraction, and Proc. of the Third Int. Conf. on Practical Aspects of automatic organization of information for detecting Knowledge Management (PAKM2000) knowledge contained in extensive information, as Basel, Switzerland, 30-31 Oct. 2000, (U. Reimer, ed.) evidenced by various studies [Fuj99], [Iwa96], [Mil95], http://sunsite.informatik.rwth-aachen.de/Publications/CEUR-WS/Vol-34/ [Gar98]. K.Sumita, Y.Nakayama, K.Sasaki, T.Manabe, M.Suzuki 21-1 In order to obtain knowledge useful for an individual from Various consulting firms have focused on knowledge the stock of knowledge and information, the system has to management, but the systems they have developed, e.g. perform not superficial retrieval but the smart retrieval KnowledgeSpace [Kno], share only typical explicit- according to the individual’s intention situation. For the knowledge such as documents on best practice, purpose, the system has to comprehend the stock of presentation manuscripts, and educational materials. They knowledge and information. Since the internet includes exchange consulting know-how through teamwork or various types of information, it is very difficult to build face-to-face communication. Explicit and tacit knowledge general-purpose background knowledge or ontology to are complementary. A knowledge management system for manage all of the information. However, a company sharing know-how or tacit knowledge, not a document engages in business with the aim of attaining certain management system, has to be developed. specific goals, and consequently, it is easier to describe the specific background knowledge in order to manage the A document management system has already been knowledge and information. implemented at the Corporate Research & Development Center, Toshiba Corp., for sharing technical knowledge; it First, we developed a knowledge base, used for stores patents and technical reports. The patents and understanding the stock of information, which involves technical reports written by researchers based on their lexical knowledge and concept hierarchy. We also research are stored in the document management system. developed a knowledge and information sharing system Presentation at a conference and publish of academic (KIDS--Knowledge and Information on Demand System) papers are an incentive for the researcher to write such which utilizes the knowledge base to comprehend both of documents. However, he has no motivation to impart his the stock of text and users’ queries [Nak97]. In order to know-how to others. To facilitate the imparting of the verify the validity of the knowledge-sharing framework know-how to others, a knowledge management system for using the system, we have been accumulating content and know-how sharing is needed. performing an evaluation of the practice of knowledge sharing at the Corporate Research & Development Center, The main subjects of know-how sharing in this paper are Toshiba Corp. [Rdc] (1)acquisition of the knowledge possessed by individuals, (2)knowledge retrieval according to the user's intention, 1.3 Know-how sharing practice (3)cultivation of a knowledge-sharing culture, and (4)reinforcement of the accumulated knowledge. The purpose of the knowledge sharing practice here is not to manage explicit knowledge such as technical papers, The following sections describe how we tackled these patents, and manuals, but to share know-how such as a subjects in order to promote know-how sharing and what knack or tips useful in the performance of research. In this we learned as a result of the knowledge sharing. paper, we focus on know-how sharing for research scientists. Figure 1: System configuration K.Sumita, Y.Nakayama, K.Sasaki, T.Manabe, M.Suzuki 21-2 2 Knowledge information sharing system (native)", and "rokuon (record)" as terms to be stored. These structure data are stored in the know-how base. 2.1 System configuration Currently, the number of structurization rules is about 200, Basically a knowledge information sharing system needs and the number of context labels is 12 for the know-how three functions: (1)acquisition of knowledge, structurization. (2)management of knowledge, and (3)retrieval of At the time of retrieval, the system carries out the same knowledge. In order to manage the accumulated analysis for the user's query, which is a natural language knowledge and information at the semantic level or sentence or a phrase, and it matches the extracted conceptual level, the system requires "knowledge" for structure with that of each nugget of know-how in the comprehending them. know-how base. Thus, since it retrieves a nugget of know- how based on a consideration of the natural language Figure 1 shows the configuration of KIDS. The system context, unlike a conventional keyword search system, consists of a knowledge database, text structurization, and text appropriate for a user's intention can be retrieved. dialog interface. It stores know-how and enables users to access nuggets of know-how by using natural language as On the basis of the above structurization the similarity Sqd shown in Figure 1. A knowledge database consists of two between document d, a nugget of know-how, and query q types of databases: the knowledge base representing is defined as follows: domain knowledge and the know-how base storing individuals’ know-how. Sqd = Σt in d δtq Wtd Ctqd . The knowledge in the knowledge base includes concept hierarchy and synonym dictionary structurization rules. Here, each variable is defined as follows: The concept hierarchy represents relations between words, e.g., library is a kind of facility. The concept hierarchy Wtd : ((1 – p1) + p1 (TFtd / TFd)) log2 (D / Dt) , and the synonym dictionary are used to expand words Ctqd : ((1 – p2) + p2 (TFtq / TFq)) max u in d δtu TCqu (1 + p3 extracted from a user’s query. Currently, the number of (Kqu – 1)) , words in the concept hierarchy and the synonym p1 , p2 , p3 : weighting factor (constant) , dictionary is about 9,200. TFtd : frequency of term t in document d, TFd : max t in d TFtd, 2.2 Text structurization and retrieval TFtq : frequency of term t in query q, The know-how base accumulates knowledge offered by TFq : max t in q TFtq, the users. The know-how, which is stored in the form of Dt : number of documents including term t, text, includes a shallow semantic structure so that the D : number of all documents in DB, know-how can be treated as a kind of knowledge with its δtq : 1 if t in term q, otherwise 0, context. We defined a descriptive structure required for δtu : 1 if t in a field labeled by u, otherwise 0, the user to search and reuse know-how easily. TCqu : correlation factor between query q and field u, Kqu : frequency of terms in query q occurring in a field u. Figure 2 shows the steps of the text structurization process. There is no explicit word boundary in Japanese sentences because Japanese is an agglutinative language. The system 2.3 Know-how sharing functions executes morphological analysis to extract words in the Every user is able to register his know-how and also text. The context presumption/structurization step, a accesses all know-how registered by other users. The context label is assigned to each phrase according to the system also has two functions to promote registration of structurization rules. In order to deal with ungrammatical know-how by users: a question registration function and a sentences and phrases, a word and/or phrase spotting content evaluation function. If the user is unable to find an method is applied to extract the structure of the input appropriate answer, he can register his question using the know-how. For every phrase of a sentence, the context question registration function. The user can enter a label is determined by referring to the structurization rules, comment for each nugget of know-how, with five rank and the terms are extracted by referring to the lexical evaluations: very useful, useful, old, incorrect, and other. knowledge. Figure 3 shows a snapshot of the system and the English In the example, the system assigns the context label translation. The figure shows the retrieval result acquired “situation” to the first phrases, and extracts "kokusai-kaigi when a user made an inquiry in connection with an oral (international conference)", "eigo (English)", and presentation to be made at the time of making an oral "happyou (presentation)" as terms to be stored. Also the system assigns the label "advice" to the later phrases, and presentation at an international conference. The titles of extracts "happyou-genkou (oral manuscript)", "neitibu the nuggets of know-how and the abstract are listed in K.Sumita, Y.Nakayama, K.Sasaki, T.Manabe, M.Suzuki 21-3 Figure 2: Structurization process order of similarity to the user's query. Also the author's The knowledge sharing system entered service in name, the date of registration, and the reference count are November 1996 and some 1,000 researchers and displayed under the title. The system is a web system, administrative staffs use it to share know-how useful in and so the server software works on a PC server whose their work. The knowledge stored consists of OS is Windows NT and the user accesses it through a explanations of expertise, glossaries of technical terms, WWW browser. miscellaneous tips respecting computers and peripheral equipment, hints respecting how best to tackle various 3 Practice and evaluation research activities, links to useful web pages, and so on. It also includes information how to do perform various 3.1 Practice history company procedures and how to compose document We have been practicing knowledge sharing using the related to those procedures, as well as know-how system described above for approximately four years at concerning welfare and everyday life. Figure 4 shows the the Corporate Research and Development Center, trend of the number of items of know-how registered. Toshiba Corp. The extent to which rich, know-how content is accumulated has a decisive impact on the The histogram in Figure 4 represents the number of items usefulness of the system. of know-how registered, and the polygonal line graph represents the cumulative number. Three knowledge administrators were assigned in May 1996. To K.Sumita, Y.Nakayama, K.Sasaki, T.Manabe, M.Suzuki 21-4 accumulate know-how and to construct a know-how base, ex. Procedure for contracting a NDA they tried various methods: they questioned experts by e- Who could take a child-care holiday? mail, and interviewed them face to face. Approximately (2)Tips relating to research activities … 21% 3,500 nuggets of know-how were registered in the know- ex. What is a `motion processor?’ how base before entering the service. According to the Web links on wearable PC result of an experiment we conducted using 2,500 queries (3)Tips on computers and software … 20% collected from 36 people, the coverage rate of this know- ex. How to make LINUX recognize a network card how base was only 50% at the initial stage. Then, we How to print a PDF file in multi-page format collected 7,200 queries limited to company procedures (4)Other useful tips … 28% from the same 36 people, and registered over 90% of the ex. Be careful when you take a taxi in Greece know-how corresponding to the gathered queries, so that How to use a library of Tokyo University the system could reply to any question about company procedures in the first phrase after its introduction. Figure 5 shows a typical know-how example registered in the system, which is translated into English. We added a maintenance function so that the knowledge administrators could reply to a non-registered question Title: How to insert a figure created using tgif in MS and register the reply in the know-how base to surely WORD? reply to any question as soon as possible. After Content: When you would like to insert a figure practicing for four months in a department consisting of created using tgif, you should do the following. about 300 people, the practice was expanded in March First, in tgif, save a figure by an eps file format. If 1997 to include the entire Corporate Research & you try to insert the eps file in MS WORD, an error Development Center, Toshiba Corp., consisting of about may occur. So, second, edit the file to rewrite the 1,000 people. Also, we enacted guidelines for the first line “%!” into “%!PS-Adobe-2.0.” Finally, copyright of know-how in July 1998, reviewed all the insert the file in MS WORD document. contents again, and deleted know-how that was violating the guidelines, useless or old. A commendation system, in which prizes are awarded to those users who Figure 5: A typical know-how registered in the system contribute most know-how, was introduced in October 1998. With the aim of enriching the content of the system, If they wish to, users can enter an evaluation for any knowledge managers were assigned to each technical nugget of know-how. The number of nuggets of know- field in August 1999. Since then, the amount of content how evaluated as very useful or useful was 970 (8.3%). has continued to increase and 11,600 nuggets of know- how were registered as of September 2000. To evaluate contents registered in the system, we selected 90 queries at random from the system’s query 3000 14000 log. Then by investigating the know-how base, we found correct answers for 60 of those queries: the answers 2000 12000 corresponding to 66% queries exist in the system. For the 60 queries, we calculated 11pt average precision, whose 1000 10000 value was 0.73. 8000 0 4 Problems in knowledge sharing /5 8 11 96/2 5 8 11 97/2 5 8 11 98/2 5 8 11 99/2 5 8 11 2 5 8 95 6000 In this section, we discuss problems in knowledge -1000 4000 sharing that have come to light in the course of our use of the system. -2000 2000 (1)Acquisition of an individual's knowledge -3000 0 Regarding the externalization of an individual’s knowledge, it is necessary to motivate the individual to impart his knowledge and an easy method of Figure 4: The number of items of know-how registered disseminating the knowledge is also required. We acquired knowledge by means of questions or interviews, 3.2 Evaluation and these approaches proved to be effective. Also, since it is impossible to externalize all tacit knowledge, face- Contents registered in the know-how base are classified to-face communication is important, too. "Know-who", into the following four categories. which is a kind of knowledge about a person, e.g., what (1)Tips on company procedures … 31% is his domain of expertise, is also important knowledge, K.Sumita, Y.Nakayama, K.Sasaki, T.Manabe, M.Suzuki 21-5 and so we are developing a know-who sharing system knowledge management to be one element in his overall based on the profiles of people. work, and therefore, we are investigating a method that would enable knowledge managers to acquire knowledge (2)Content retrieval of knowledge and maintain it in the course of their work. In order to reuse knowledge efficiently, sophisticated retrieval according to the intention of the user is required. 5 Conclusion Natural language retrieval, especially the content retrieval using lexical knowledge and concept hierarchy The practice of knowledge sharing using a knowledge of an object field improved retrieval efficiency. The and information on demand system, KIDS, was described. knowledge base, which represents background A thousand nuggets of know-how were gathered and knowledge, is the key to understanding the user’s made available for use. Individual's know-how was intention, and we intend to strengthen it. stored and managed by means of know-how base whose contents have a shallow structure and the knowledge base (3)Organization culture conductive to knowledge sharing to enable the know-how to be shared and utilized For a member of an organization to offer his know-how effectively. Thereby, acquisition of individuals’ proactively, it became clear that not only system knowledge became easy and the quantity and the quality functionality but also organizational culture and the of explicit knowledge improved. In order to achieve individual’s attitudes and values are decisive factors. In successful implementation and utilization of knowledge particular, it is important to provide motivation and sharing, it is essential to give full consideration to both implement guidelines regarding the offering of know- technology and organizational culture. In addition to how. Ideally, everyone would be conscious of the providing the system, it is important to provide desirability of offering knowledge to others, and act motivation and implement guidelines regarding the accordingly. In practice, it is necessary to raise people’s offering of know-how. consciousness and inculcate a proactive attitude throughout the organization toward the offering of know- In addition to the study at the Corporate Research & how. The system introduced in August 1998, whereby Development Center, using KIDS we performed prizes are awarded to those who offer the most know- knowledge sharing at a customer service center for a how, motivated some, but by no means all members of period over two years. The system is used for sharing the organization. This insufficient motivation is question-answer text, which the staffs store according to attributable, in part, to the lack of clarity regarding what the questions received from the customers. Also, using types of knowledge should be stored in the system. Our KIDS we have started knowledge sharing at a systems experience indicates the desirability of clarifying the integration department. In future, through the promotion purpose of knowledge sharing. of knowledge sharing at various departments, we intend to investigate effective methods of knowledge sharing (4)Reinforcement of the accumulated knowledge that correspond to the objectives and characteristics of The success of a knowledge sharing system is to a great any given department. extent determined by the quality of the content stored in the system. Regarding management of the quality of the We have enhanced KIDS to accommodate various types unnecessary content, and avoidance of duplication of of file format and strengthen the file security. We have data. Initially, we concentrated on increasing the amount developed the commercial version, of know-how, and, reflecting that approach, some of the eFilingMeister/KnowledgeServerTM. nuggets of know-how proved to contain useless or incorrect information. The evaluation feedback was References effective for content management. [Apo98]Knowledge Management: Lessons from the Leading Edge, Proceedings from APQC's Third Also, with the aim of enriching the content of the system, Knowledge Management Symposium, 1998. knowledge managers were assigned to each technical [Non95]I. Nonaka, et al.: The Knowledge-Creating domain from August 1999 onward. To be effective, a Company, Oxford University Press, 1995. knowledge manager needs to have a sure grasp of the [Fuj99]S. Fujita: An Approach for Information Retrieval technical field, and be able to select the knowledge that and Classification using Natural Language Processing, should be registered, collect the knowledge, examine the Information Processing Society of Japan, Vol.40, No.4, content of the knowledge, etc. Currently, knowledge pp.352-357, 1999 (in Japanese). managers are not exclusively assigned to knowledge [Iwa96]M. Iwazume, et al.: Ontology-Based Information management, but do this work in addition to their regular Capturing from the Internet, Proc.4th Int. Conf. on the work. To facilitate the knowledge manager’s Society of Knowledge Organization, pp.261-272, 1996. maintenance of his expertise, it is preferable for K.Sumita, Y.Nakayama, K.Sasaki, T.Manabe, M.Suzuki 21-6 [Mil95]G.A. Miller: WordNet: A Lexical Database for “Advice/Help on Demand”, Interaction'97, Information English, Comm. ACM, Vol.38 No.11, pp.39-41, 1995. Processing Society of Japan Symposium, pp.103-110, [Gar98]J. Gareth, et al.: Experiments in Japanese Text 1997 (in Japanese). Retrieval and Routing Using the NEAT System, Proc. [Rdc]http://www.toshiba.co.jp/rdc/ 21st ACM SIGIR Conf., pp.197-205, 1998. [Kno]KnowledgeSpace: [Nak97]Y. Nakayama, et al.: Development of a http://www.knowledge.space.com/ Knowledge /Information Sharing System original Figure 3: A snapshot of the system and the translation to English. question title translation abstract Figure 3: A snapshot of the system and the translation to English K.Sumita, Y.Nakayama, K.Sasaki, T.Manabe, M.Suzuki 21-7