Скачать презентацию Knowledge Management for Learning Organizations or Скачать презентацию Knowledge Management for Learning Organizations or

1d85e95c1a9872c8f3d4dceb6a34699a.ppt

  • Количество слайдов: 64

( Knowledge Management for Learning Organizations ) or Context-Aware, Proactive Delivery of Task-Specific Knowledge ( Knowledge Management for Learning Organizations ) or Context-Aware, Proactive Delivery of Task-Specific Knowledge Andreas Abecker, Ansgar Bernardi, Knut Hinkelmann Otto Kühn, Tino Sarodnik, Michael Sintek German Research Center for Artificial Intelligence (DFKI) Gmb. H, Kaiserslautern Knowledge Management Group Knowledge Management Research Group 1 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H The Know. More Project

– The Know. More project was done in the DFKI Knowledge Management Group as – The Know. More project was done in the DFKI Knowledge Management Group as application-oriented basic research in order to promote a better understanding and tool support for intelligent software solutions in Knowledge Management and Organizational Learning. – The project was funded by the German National Ministry for Research and Education (bmb+f). – Know. More ran from April 1997 til March 1999, with an extension til October 1999. Knowledge Management Research Group 2 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Administrativa

Knowledge Management Research Group 3 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. Knowledge Management Research Group 3 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Introduction

Knowledge Management and Organizational Learning are emerging paradigms in industry – – – shorter Knowledge Management and Organizational Learning are emerging paradigms in industry – – – shorter product life cycles, lean organizational structures, concurrent engineering efforts, globally dispersed virtual enterprises, enterprise reengineering, . . . make knowledge management an urgent need for enterprises managers are biased towards non-technological issues, like human resource management, cultural aspects, organizational changes etc. . . . which are crucial for KM, anyway several IT communities recently „discovered“ the area: workflow systems, CSCW, expert systems, case-based reasoning, intranets, data mining, document management systems, . . . are considered to be useful for KM However, a commonly agreed-upon approach and methodology is still lacking. Knowledge Management Research Group 4 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H CURRENT SITUATION

Knowledge Management can be supported by exchange of information ORGANIZATIONAL LEARNING knowledge – knowledge Knowledge Management can be supported by exchange of information ORGANIZATIONAL LEARNING knowledge – knowledge socialization management externalization information communication storage internalization – – continued training & experience Learning by sharing experiences – cooperation & observation Learning through communication – supply-driven learning information retrieval repository IT support Individual Learning – – demand-driven learning Learning through development of a knowledge repository – storing and monitoring lessons learned IT people tend to concentrate on either the communication / collaboration, or the repository aspects (Organizational Memory). Knowledge Management Research Group 5 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H – individual learning

Basically, research on Organizational Memory can concentrate on knowledge explication, or on knowledge capitalization Basically, research on Organizational Memory can concentrate on knowledge explication, or on knowledge capitalization – explication of tacit knowledge: – – – the typical expert system approach [Kühn. Abecker 97]: cost-benefit problems [Rittel 72], [Buckingham Shum 97]: feasibility for “wicked problems”? [Davenport. Jarvenpaa+96]: construction and maintenance problems capitalization on implicit and existing explicit knowledge: – existing documents and knowledge sources often severely underutilized – ease finding, access, and exploitation – increase utilization potential The Know. More approach concentrates on the second goal. Knowledge Management Research Group 6 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H TWO COMPLEMENTARY APPROACHES

Practical solutions require different degrees of formalization OBJECTIVES OF AN ORGANIZATIONAL MEMORY Ensure the Practical solutions require different degrees of formalization OBJECTIVES OF AN ORGANIZATIONAL MEMORY Ensure the utilization of “formal” organizational knowledge: business rules, design guidelines, standard procedures, . . . can be formalized to allow automatic processing – Enable sharing and reuse of experiences: lessons learned, best practice reports, case bases, . . . can be stored as semi-structured electronic documents – Ease the exploitation of implicit knowledge, personal knowledge, and knowledge contained in documents and databases technical documentation, hypertexts, personal notes, minutes of meetings, graphics, images, product data sheets, business letters, . . . must be effectively accessible How can several kinds of knowledge synergetically interact? Knowledge Management Research Group 7 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H –

Knowledge Management Research Group 8 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. Knowledge Management Research Group 8 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H The Know. More approach

One solution approach: Knowledge Management oriented on Business Process Management BPMS-METHODOLOGY: PRODUCT/PROCESS-PHILOSOPHY [Karagiannis, 1994] One solution approach: Knowledge Management oriented on Business Process Management BPMS-METHODOLOGY: PRODUCT/PROCESS-PHILOSOPHY [Karagiannis, 1994] based on Which products do we offer? Strategic Decisions Products created by How are the products made? Processes Modeling done by How are the processes realized? Information Technology Employees Implementation of the processes What kind of improvement potential exists? Execution Finished business processes Evaluation The BPMS Methodology is supported by the ADONIS system Knowledge Management Research Group 9 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Company

Knowledge Management adds a new quality to Business Process Management – Strategic Decision – Knowledge Management adds a new quality to Business Process Management – Strategic Decision – – Modeling – Implementation Evaluation Execution Organizational Memory Workflow-Management Systems Document-Management Systems Knowledge Management Research Group Internet Groupware. Products Intranet etc. Conventional business process models represent procedural knowledge Business Process Management optimizes efficiency of the whole process Knowledge Management improves the result of the process It focuses on additional aspects: – Modeling • identification of required knowledge • analysis of existing knowledge – Implementation • structuring and recording of knowledge • strategies for the elimination of knowledge deficits • determining the access points – Execution • context-based information retrieval • active assistance 10 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H TOWARDS A KNOWLEDGE-MANAGEMENT METHODOLOGY

Business Process Models represent control flow of business activities THE DFKI PURCHASING PROCESS Check Business Process Models represent control flow of business activities THE DFKI PURCHASING PROCESS Check Budget no Support Demand Reject supp. ? yes Hardware or Software? yes no Specify HW/SW Details price > 800, - ? no Specify Details Install HW / SW Deliver Goods yes appr. ? Update Purchasing Database Receive Goods Send Order yes price > 800, - ? Receive Delivery Note no Sign Invoice Approve Demand no yes Hardware or Software? yes Sign Invoice Receive Invoice Update Database Pay Invoice Allocate Inv. no. Attach Inv. no. The main complexity of this process is hidden in few knowledge-intensive activities. Knowledge Management Research Group 11 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Specify Demand

An ideal Knowledge Management system would answer manifold questions related to a given knowledge-intensive An ideal Knowledge Management system would answer manifold questions related to a given knowledge-intensive activity Check Budget Specify Demand no Support Demand Reject supp. ? Hardware or Software? no no Specify Details price > 800, - ? yes Install HW / SW Specify HW/SW Details yes no Sign Invoice Approve Demand yes appr. ? Send Order no Receive Goods Hardware or Software? Deliver Goods yes – Update Purchasing Database yes price > 800, - ? Receive Delivery Note Are there general guidelines for buying computer devices? Sign Receive Invoice – Who bought a graphics card Update Database recently? Pay – Are there any experiences with Invoice card Matrox Mystique? Allocate Inv. no. Attach Inv. no. – Can anyone recommend a good graphics card? Which requirements can be derived for an Organizational Memory Information System which is able to answer such questions? Knowledge Management Research Group 12 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H yes

OM technology has to face a number of demanding challenges – – predominance of OM technology has to face a number of demanding challenges – – predominance of non-formal knowledge representation (text, drawings, . . . ) heterogeneity on all levels (domain conceptualizations, kinds of knowledge, computer systems, storage formats, . . . ) – – – active knowledge supply instead of passive information retrieval self-adaptivity task-independence OM technology grows out of an application-driven integration of enabling technologies. Knowledge Management Research Group 13 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H CHARACTERISTICS OF ORGANIZATIONAL MEMORY SETTINGS

Enabling technologies cover the whole cycle of capturing, storage, and utilization of corporate knowledge Enabling technologies cover the whole cycle of capturing, storage, and utilization of corporate knowledge How do the pieces fit together? Knowledge Management Research Group 14 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H SOME OM CONTRIBUTING FIELDS

Know. More supports knowledge-intensive tasks (KITs) by active delivery of context-specific information OVERVIEW process Know. More supports knowledge-intensive tasks (KITs) by active delivery of context-specific information OVERVIEW process support information processing and retrieval domain ontology enterprise ontology information ontology – – Know. More concentrates on knowledge -intensive tasks the supply of relevant information must exploit domain and context knowledge in Know. More the context is given by the business process support is based on formal knowledge descriptions based on ontologies the task at hand is performed by the user The cooperative model of active support has been realized in a first demonstration prototype. Knowledge Management Research Group 15 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H –

 • • To purchase a graphics card details need to be specified: – • • To purchase a graphics card details need to be specified: – name – price – supplier Support is provided by – computation of suggestions – presentation of relevant information • business rules • supporting documents • references to experienced persons • Knowledge Management Research Group Process state is taken into account 16 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Context-specific information supports the user in knowledge-intensive activities

When the user decides to buy the Matrox Mystique, the system automatically constrains the When the user decides to buy the Matrox Mystique, the system automatically constrains the set of supporting documents to the subset dealing with this specific product. Knowledge Management Research Group 17 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Changes in the process state result in refined support

– – The Know. More system is not a hard-wired approach for exactly the – – The Know. More system is not a hard-wired approach for exactly the support described before, but merely provides the representation facilities for easily building such business-process information assistants. To this end, the following topics must be addressed: – Question (I): How are task-specific information needs expressed in order to enable the system for a context-specific, automatic information delivery? – Question (II): How do BPM enactment (i. e. , the workflow engine) and information assistant cooperate? – Question (III): How does the information assistant process information needs, knowledge descriptions, context information, and background knowledge for precise information retrieval? – Question (IV): How are knowledge items in the OM described with respect to the conceptual structures (ontologies) underlying the domain of application? – Question (V): What about system design and implementation? – Question (VI): What about modeling support for knowledge item descriptions and underlying ontologies? These issues are discussed in the following parts of the talk. Knowledge Management Research Group 18 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H How to realize such functionalities?

Knowledge Management Research Group 19 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. Knowledge Management Research Group 19 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Technical realization (I)

A detailed description of Information Needs extends the process model Task participant activity input/output A detailed description of Information Needs extends the process model Task participant activity input/output variables precondition: parameters: } product_type isa Hardware product_type, specification, price info agent: retrieval-agent-1 contributes-to: product_detail {post-processing rules} Knowledge Management Research Group – what to do, resources. . . – data flow extensions { Info-Need-1, Info-Need-2, . . . Info-Need-n The conventional process model provides context information – referring to • process status • input / output variables – provides the information – variables influenced by the result – govern presentation/computation 20 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H • • •

Modeling of knowledge-intensive tasks is integrated with business process modeling (BPM) • • • Modeling of knowledge-intensive tasks is integrated with business process modeling (BPM) • • • ADONIS tool used for process modeling additional variables for data flow Info Needs are modeled – currently in the ‘comment’ field – ADONIS could be adapted by BOC • Knowledge Management Research Group a parser translates into the representation of some workflow engine 21 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H CREATION OF EXTENDED BPMs AT PROCESS DEFINITION TIME

Preconditions and postprocessing rules of information needs allow to formulate information-seeking strategies 1. Present Preconditions and postprocessing rules of information needs allow to formulate information-seeking strategies 1. Present formally derived recommendations result produced 1 a. Offer supporting information 2. Present relevant business rules 3. Provide info about products important purchase or unexperienced user Decisions are based on • process context • KIT variables • results of previous steps 4. Suggest contact to competent employees Knowledge Management Research Group 22 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H STRATEGY FOR PURCHASING SUPPORT

Knowledge Management Research Group 23 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. Knowledge Management Research Group 23 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Technical realization (II)

The prototype system illustrates the key players in a workflow scenario (which is an The prototype system illustrates the key players in a workflow scenario (which is an extension to the Workflow Mgt Coalition‘s scenario) WF Control Data invokes Workflow Engine Applications Worklist Handler support WF Relevant Data + Extensions Information Agent(s) Information Sources Knowledge Management Research Group 24 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Pars pro toto: Attribute editor

The KIT model is processed by the extended worklist handler workflow activity Worklist Handler The KIT model is processed by the extended worklist handler workflow activity Worklist Handler knowledge agent invoke application present variables to the user suggest check preconditions suggest evaluation initiate perform Information Retrieval user fills variables support by display / calculation changes in parameters return result process results observe changes We realized a communication model for the assistance by the information agent Knowledge Management Research Group 25 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Knowledge-Intensive Task

Knowledge Management Research Group 26 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. Knowledge Management Research Group 26 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Technical realization (III)

The information agent uses formal knowledge to retrieve the information relevant for the task The information agent uses formal knowledge to retrieve the information relevant for the task at hand Parameters: • instantiated WF variables Ontology Information Agent • specification of the search heuristic Result: Description Frames of relevant information items Postprocessing Information Sources Knowledge Management Research Group 27 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H From the extended modeling of Knowledge. Intensive Tasks (KIT):

 • formal calculate values of variables from the retrieved information message type • • formal calculate values of variables from the retrieved information message type • informal create a suitable presentation by • grouping • sorting • html generation The presentation in a WWW Browser ensures flexibility Knowledge Management Research Group 28 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Post-processing governs the presentation of supporting information

From the extended modeling of Knowledge. Intensive Tasks (KIT): info agent formal inferences computed From the extended modeling of Knowledge. Intensive Tasks (KIT): info agent formal inferences computed values Parameters: • instantiated WF variables • specification of the search heuristic thesaurus maps NL to concepts Domain Knowledge / Thesaurus Knowledge Management Research Group find relevant concepts retrieve relevant information Enterprise Model Information Model 29 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Play-together of formal inferences and background knowledge in the information agent

Purchasing a graphics card example: Find competent employees • determine parameters of heuristic search Purchasing a graphics card example: Find competent employees • determine parameters of heuristic search consider direct subconcepts retrieve personal or group expertise with: • determine retrieval constraints demand urgent? • if demand urgent: expertise owner immediately available • relevant concept Î expertise expand concept to related concepts map value of variables to domain concepts „Grafik-Karte“ >> graphics card + matrox mystique + matrox millenium +. . . Domain Knowl. / Thesaurus results: Enterprise Model Information Model Ontologies with domain-specific relations are traversed using task-specific search heuristics to retrieve relevant information items Knowledge Management Research Group 30 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H • test for rule applicability unexperienced user?

- -1 1 1. ( has. Compete nce ) - -1 1 1. ( has. Compete nce ) "Search people for directly linked a search to concept. " - - + 2. ( has. Compete nce 1 )1 o (is. Sub. Field Of 1 ) "Look people for competent some in subfield. " Knowledge Management Research Group 1. (has. Competence 1 )1 2. ( works. In 1 )1 o (uses. Technology 1 )1 "Look people for working a project in applying the technologyquest. " in - 3. (has. Competence 1 )1 o (is. Sub. Field. Of )1 "Look people for experience thedirect din superconce pt of thetopic quest. " in 31 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Intelligent conceptual information retrieval: search heuristics describe how to navigate in the ontologies

Knowledge Management Research Group 33 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. Knowledge Management Research Group 33 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Technical realization (IV)

Information retrieval maps information need descriptions to knowledge item descriptions – [van. Rijsbergen 89] Information retrieval maps information need descriptions to knowledge item descriptions – [van. Rijsbergen 89] identified the three basic constituents of intelligent information retrieval: – a semantic representation of documents – a semantic representation of queries – an inference procedure mapping the latter one onto the former ones – logic-based formalisms have a clear semantics and powerful processing mechanisms which can incorporate background knowledge – in Know. More, stable structures of conceptual domain models (ontologies) are the reference models, all other representations refer to Knowledge Management Research Group 34 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H LOGIC-BASED INFORMATION RETRIEVAL

Ontologies organize information models and background knowledge – Information Ontology context Enterprise Ontology content Ontologies organize information models and background knowledge – Information Ontology context Enterprise Ontology content Domain Ontology – information ontology: – kinds of information sources – logical structure -> relevance propagation – meta properties (reliability, message type, availability, creation context, intended usage context) – link to information content enterprise ontology: – provides usage and creation context – basis for BPMs – enterprise organizational structure – domain ontology / thesaurus: – description of information content – natural language expressions linked to formal concepts – usually incomplete We extended standard modeling approaches by the context dimension. Knowledge Management Research Group 35 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H DIMENSIONS OF INFORMATION MODELING

Simplified example of the three information modeling ontologies information ontology department group expertise employee Simplified example of the three information modeling ontologies information ontology department group expertise employee personal expertise content information rule document article title enterprise ontology book keyword s-p-o message type section thing isa instance of uses object link part of domain ontology / thesaurus Knowledge Management Research Group 36 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H company

Knowledge representation requirements in the area of OM conceptual information retrieval INTERESTING RESEARCH TOPICS Knowledge representation requirements in the area of OM conceptual information retrieval INTERESTING RESEARCH TOPICS – – – Distinguish between document models and indexing ontologies – for document models additional requirements like order of chapters, concepts as attribute fillers, specific links in hypertexts, „higher-order“ expressiveness for complex content descriptions For indexing ontologies: some basic object-centered modeling formalism: concepts, attributes, instances Some link to thesaurus information Possibilities to freely define other relationships equipped with special inference mechanisms (domain and task specific) Maybe later: special procedures for vague relationships: „has-to-do-with“ Maybe later: a clearer understanding of indexes and the „is-about“ relation Maybe later: uncertainty and vagueness In Know. More, we dealt with the upper four requirements. Knowledge Management Research Group 37 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H –

The Know. More knowledge representation language: OCRA (object-centered relational algebra) • classes from OOP The Know. More knowledge representation language: OCRA (object-centered relational algebra) • classes from OOP • relations from logic programming languages and RDBs • types from functional and imperative programming languages • classes, inheritance, objects, methods • set orientation • rules inherited from OOP from RDBs from logic languages – Designed for the Intelligent Fault Recording project – Designed for efficient processing on an RDBMS – Allows special inferences for specific link types Knowledge Management Research Group 38 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H The Know. More representation formalism integrates the object-oriented and the relational paradigms by unifying

The OCRA is strictly typed human( name age father mother ) : : string, The OCRA is strictly typed human( name age father mother ) : : string, number, human // string and number are // built-in classes man : human() woman : human(maiden. Name : string) Alternative definition of human with set type: human( name : string, age : number, children : {human} ) Knowledge Management Research Group 39 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Class declarations:

human( name : string, . . . , competences : {competence} / strength ) human( name : string, . . . , competences : {competence} / strength ) competence(name : string, . . . ) ann() // the top class of all annotations strength : ann(value : string) // e. g. "good", "medium", "bad" human name competences John Mary Knowledge Management Research Group strength name good bad medium competence name classes English objects French 40 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H OCRA: Annotations allow complex semantic nets to be modeled

woman( name = woman( name = "Mary", competences = {competence(name = "French") / strength(value = "medium")} ) User-defined object identifiers: man( name = "John", competences = {english / good, french / bad} ) english : competence(name = "English") french : competence(name = "French") good : strength(value = "good") bad : strength(value = "bad") Note: predicate symbols and function symbols are not distinguished-they are both class names. Knowledge Management Research Group 41 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H OCRA: Objects have a textual representation

Knowledge Management Research Group 42 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. Knowledge Management Research Group 42 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Technical realization (V)

Know. More fits in the Wf. MC general workflow system architecture EXTENSIONS TO THE Know. More fits in the Wf. MC general workflow system architecture EXTENSIONS TO THE WORKFLOW MANAGEMENT COALITION’S SYSTEM ARCHITECTURE WF Control Data invokes Business Process Model + Extensions for knowledgeintensive Tasks interpreted by Workflow Engine WF Relevant Data + Extensions Applications Worklist Handler support Information Agent Information Sources Knowledge Management Research Group 43 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H WF Application Data

Business Process Definition Tool generates Enterprise WF Application Data WF Control Data interpreted by Business Process Definition Tool generates Enterprise WF Application Data WF Control Data interpreted by • units • employees • roles • business processes Workflow Engine Worklist Handler invokes Applications WF Relevant Data DBs & KBs Knowledge Management Research Group Server Clients 44 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H The client / server architecture builds upon Wf. MC standards

Extended BPMs and background knowledge allow access to various supporting information sources Adonis generates Extended BPMs and background knowledge allow access to various supporting information sources Adonis generates Enterprise interpreted by • units • employees • roles • business processes + KITs Competences Product Data Sheets Test Reports Rules WF Application Data WF Control Data Workflow Engine Worklist Handler invokes Applications WF Relevant Data provide terminology Enterprise O. Information O. Domain O. Knowledge Item Descriptions DBs & KBs Knowledge Management Research Group Server Clients 45 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Business Process Definition Tool

The extended variable blackboard bridges between workflow applications and Know. More support Adonis generates The extended variable blackboard bridges between workflow applications and Know. More support Adonis generates Enterprise interpreted by • units • employees • roles • business processes + KITs Competences Product Data Sheets Test Reports Rules WF Application Data WF Control Data Workflow Engine Worklist Handler invokes Applications WF Relevant Data provide terminology Enterprise O. Information O. Domain O. Information Agent request infos provide infos Knowledge Item Descriptions DBs & KBs Knowledge Management Research Group Excel Netscape Java Var. Editor Inference Engine Server Clients 46 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Business Process Definition Tool

Knowledge Management Research Group 47 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. Knowledge Management Research Group 47 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Technical Realization (VI)

Know. More envisions a comprehensive toolbox to create the Organizational Memory ADONIS © — Know. More envisions a comprehensive toolbox to create the Organizational Memory ADONIS © — Business Process Modeling Tool KIT Modeling Facilities – Modeling of the business process using the ADONIS – process support information processing and retrieval Creating the KIT descriptions – information ontology – specify relevant WF variables – enterprise ontology select from a library of info agents – domain ontology specify search heuristics (a) Arbitrary knowledge items can be integrated and manually annotated – Knowledge Item Editor TCW — Text Classification Workbench Knowledge Management Research Group Ontology Editor (b) A learning text classification tool automatically creates meta information for text documents – (c) The Ontology Editor employs thesaurus information to support the construction of ontologies TRex — Similarity Thesaurus Generator 48 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H BPM tool (seen earlier)

Two examples: annotating a personal homepage and a conference paper with ontology concepts Knowledge Two examples: annotating a personal homepage and a conference paper with ontology concepts Knowledge Management Research Group 49 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H (a) The knowledge item description relates OM content to concepts from the domain ontology

– – – The Text Classification Workbench TCW is trained with manually categorized example – – – The Text Classification Workbench TCW is trained with manually categorized example documents The system learns characteristic complex text patterns After training, new documents can be categorized automatically This can be applied to all text documents added to the OM TCW originated from the READ and Virtual Office document analysis projects Automatic creation of more detailed formal descriptions will be investigated using information extraction techniques. Knowledge Management Research Group 50 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H (b) The TCW Tool for learning text classification has been integrated to automatically create meta information for text documents in the OM

Knowledge Management Research Group 51 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. Knowledge Management Research Group 51 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H The TCW automatically suggests to the user ontological categories as potential indices for text documents

(c) Ontology development: Our approach for acquisition + maintenance is based on automatic thesaurus (c) Ontology development: Our approach for acquisition + maintenance is based on automatic thesaurus generation – – routinely created during work processes – contain relevant terms in task contexts thesaurus generator TRex – ontology thesaurus + knowledge base Knowledge Management Research Group automatic thesaurus generation – efficient processing of large document corpora – extract important terms and relations based on frequency and co-occurrence similarity thesaurus interactive knowledge acquisition and update tool documents – interactive knowledge acquisition – perform a semantic classification of identified similarity relations – update knowledge base and ontology thesaurus 52 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H documents

Documents are a plentiful source of information available in any application domain Example from Documents are a plentiful source of information available in any application domain Example from FAKT project: similar terms to ‘backup’ tape mount device not ready restore data safety Example from ‘Die WELT’ Articles on the German spelling reform similar terms to ‘Rechtschreibung’ Reform Kultusministerkonferenz Duden Regeln (112 Regeln) Orthographie Knowledge Management Research Group – – – In an Organizational Memory it is important to handle large amounts of knowledge First results confirm our expectations that thesaurus generation methods may be profitably exploited for knowledge acquisition Even a rough analysis of word frequencies and correlations. . . identifies core topics in a new domain. . . offers guidance for subsequent knowledge acquisition An analysis of term similarities points out interesting relationships and dependencies More sophisticated analyses based on additional knowledge are needed to separate meaningful from spurious results 53 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H EXPLOITING THESAURUS GENERATION FOR KNOWLEDGE ACQUISITION

documents - schemata - stopwords - non-text important terms (from ontology) – document parsing documents - schemata - stopwords - non-text important terms (from ontology) – document parsing term and phrase generation parameters – construction of term-context-matrix analysis of term-context-matrix computation of term-similarities similarity thesaurus Knowledge Management Research Group term-context matrix to be updated – TRex can be easily adapted to domainspecific document collections – specification of document schemata, stopwords and non-text – adjustment of term and phrase generation parameters TRex offers a variety of techniques for computing term-similarities – different term contexts (document, window) – various weighting schemes – singular value decomposition of termcontext matrix – numerous similarity scores TRex can exploit lists of important terms (extracted from an ontology) – for focussing term and phrase generation – for weighting of term similarities 54 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H The thesaurus generation tool TRex was extended and enhanced

An integrated Ontology/Thesaurus is constructed or updated semi-automatically from the term associations generated by An integrated Ontology/Thesaurus is constructed or updated semi-automatically from the term associations generated by TRex newly found associations for term t explanation by known relations for t from the Ontology/Thesaurus t t t t t isa a t haspart p, p haspart b t antonym c t 0. 25 d ? t isa u, f isa u ? t haspart v, v synonym g 0. 65 a 0. 47 b 0. 33 c 0. 31 d 0. 26 e 0. 19 f 0. 17 g 0. 16 h performed updates t 0. 28 d t haspart e (ignore) Besides building/updating an integrated Ontology/Thesaurus, the identified term associations can also be used for updating the knowledge base Knowledge Management Research Group 55 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H EXAMPLE

A first evaluation shows the utility of automatic thesaurus generation techniques for building and A first evaluation shows the utility of automatic thesaurus generation techniques for building and maintaining Organizational Memories TEST 2: Computer technology – – 67 documents created by DFKI knowledge management group rudimentary ontology of names and competencies text-window contexts of size 100 – 2685 documents from a computer magazine on CD no previous ontology – document contexts Terms associated with Unternehmensgedächtnis Terms associated with Prozessor wissen unternehmen informationsueberflutung wissensverarbeitung mitarbeiter informationstechnische arbeitsablauf arbeitsprozess ug wissen unternehmen intel mhz pentium cpu amd cyrix mmx k 6 kbyte sockel Knowledge Management Research Group 56 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H TEST 1: DFKI competencies

The Know. More ontology editor supports the cumbersome task of ontology construction & maintenance The Know. More ontology editor supports the cumbersome task of ontology construction & maintenance An automatically created similarity thesaurus provides correlations between terms – Terms indicate possible concepts – Term correlations indicate possible links – The editor visualizes these relations – Identification of concepts and creation of links is done manually Knowledge Management Research Group 57 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H –

Knowledge Management Research Group 58 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. Knowledge Management Research Group 58 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H The user is responsible for the final decision about concepts and link types

Knowledge Management Research Group 59 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Know. More Knowledge Management Research Group 59 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Know. More 3/98

– IT support for Knowledge Management and Organizational Learning is an emerging, still open – IT support for Knowledge Management and Organizational Learning is an emerging, still open research topic. – We propose intelligent assistance for knowledge-intensive tasks which is based on context-sensitive, active information supply. – Information supply essentially amounts to a demanding multimedia and hypermedia retrieval task. – We propose ontology-based information modeling with special focus on the context dimension and information meta properties. – Though the framework still provides many basic research questions, pragmatically designed prototypes already yield promising results. Knowledge Management Research Group 60 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H OVERALL SUMMARY

The Know. More research mainly investigated three research areas workflow-oriented knowledge management – a The Know. More research mainly investigated three research areas workflow-oriented knowledge management – a model for knowledge-intensive tasks has been defined – an activation method has been specified • processes – a knowledge item description schema has been developed – a representation formalism has been defined and implemented – ontology-based retrieval with search heuristics was investigated Know. More inferencing & retrieval Knowledge Management Research Group ontology/ thesaurus integrated processing and retrieval of knowledge • integrated ontology and thesaurus – a prototypical approach for the conjoint construction of a domain ontology and a thesaurus has been developed – a system for learning a similarity thesaurus has been implemented 61 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H •

The Know. More system prototype illustrates key ideas of a three-layered OM approach Organizational The Know. More system prototype illustrates key ideas of a three-layered OM approach Organizational Memory idea: process – workflow integration for active support information processing and retrieval domain ontology enterprise ontology information supply – handling of informal knowledge (“documents”) by reasoning on formal knowledge descriptions – interplay of several ontologies and specific search heuristics – support tools for knowledge input Knowledge Management Research Group 62 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H Know. More realized key concepts of the

Know. More results spawned several application projects active information delivery, integrated modeling of processes Know. More results spawned several application projects active information delivery, integrated modeling of processes & information need, information transfer between different contexts CLOCKWORK (EU) Help Me Find Person (USU AG) un-intrusive ontology acquisition, text categorization, retrieval heuristics Process representation, OM realization ENRICH (EU) KONARC Configuration support (TELEKOM) Thesaurus-based ontology construction, text categorization support Ontologies, search heuristics, information agent Know. Net (EU) ESB Electronic Fault Recording (Saarbergwerke) Representation formalisms, search heuristics Knowledge Management Research Group Know. More Knowledge description formalism, information ontology, info agent for push mechanism 63 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H DECOR (EU)

An OM technology requires further research on all three levels of the Know. More An OM technology requires further research on all three levels of the Know. More conceptual architecture SOME FUTURE WORK Towards business-process oriented knowledge management – a methodology is missing for integrated business process and knowledge modeling – workflow notion: „a smooth transition from ad-hoc cooperative work of humans and standardized, automated interaction between autonomous information systems“ [De. Michelis et al. – 97], i. e. more flexible workflow concepts are required to describe knowledge work Precise-content retrieval for OM knowledge items – investigate XML and RDF / Schema for representation of ontologies – discuss uncertainty processing in conceptual information retrieval – consider metadata & retrieval constraints for – – ontology mapping for distributed knowledge sources Continous, self-adaptive knowledge capture and organization – exploit document analysis and understanding techniques to automatically extract content and meta-level descriptions from text documents – context-enriched document storage Agent technology could provide the software basis for an OM middleware. Knowledge Management Research Group 64 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H –

Knowledge Management Research Group 65 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. Knowledge Management Research Group 65 Know. More 3/98 Deutsches Forschungszentrum für Künstliche Intelligenz Gmb. H The show goes on. . .