1. Fabien.Gandon@sophia.inria.fr Fabien GANDON - INRIA - ACACIA Team - KMSS 2002 CoMMA in a Nutshell

2. Fabien.Gandon@sophia.inria.fr 2 Introduction: issues and motivations  Knowledge and information management:  Needs: improve reaction time & address turnover -Persistent memory: store and/or index knowledge -Nervous system: capture and diffuse knowledge  O.M.: an explicit and persistent representation and/or indexing of knowledge in an organization, in order to facilitate its access and reuse by members of the organization, for their individual and collective tasks.  Current trend: reuse internet and web technologies to build intranets and intrawebs -Same advantages: standardised technology, browser unique access means, distributed architecture, etc. -Same drawbacks: human-understandable but only machine readable; problem of retrieval, automation,...  Corporate Memory Management through Agents:  Assist new employee integration  Support technology monitoring activities

3. Fabien.Gandon@sophia.inria.fr 3 Positioning and pointers  Dynamically integrating heterogeneous sources of information: Manifold [Kirk et al.,1995] ; InfoSleuth [Nodine et al., 1999] ; InfoMaster [Genesereth et al., 1997] ; Carnot [Collet et al., 1991] ; RETSINA [Decker and Sycara, 1997] ; SIMS [Arens et al., 1996] ; OBSERVER [Mena et al., 1996]  Digital libraries: SAIRE [Odubiyi et al., 1997] UMDL [Weinstein et al., 1999]  Knowledge management:  Collaborative gathering, filtering and profiling: CASMIR [Berney and Ferneley, 1999]; Ricochet [Bothorel and Thomas, 1999]  Mobile access to memory and domain model for classification: KnowWeb [Dzbor et al., 2000]  Taxonomy, profiling and push: RICA [Aguirre et al., 2000]  Ontology and corporate memory: FRODO [Van Elst and Abecker, 2001]

4. Fabien.Gandon@sophia.inria.fr 4 How ?  In CoMMA:  Corporate memories are heterogeneous and distributed information landscapes  Stakeholders are an heterogeneous and distributed population  Exploitation of CM involves heterogeneous and distributed tasks  Choices: Materialization CM Exploitation CM XML: Standard, Structure, Extensible, Validate, Transform RDF: Annotation, Schemas Multi-Agent System: Modularity, Distributed, Collaboration Machine Learning: Adaptation, Emergence Corporate Memory Management through Agents

5. Fabien.Gandon@sophia.inria.fr 5 Overall Schema Corporate Memory Multi-Agents System Learning User Agent Learning Profile Agent Ontology and Models Agent User Agent Learning Interconnection Agent Knowledge Engineer Ontology Models - Enterprise Model - User's Profiles

6. Fabien.Gandon@sophia.inria.fr 6 Overall Schema Corporate Memory Multi-Agents System Learning User Agent Learning Profile Agent Ontology and Models Agent User Agent Learning Interconnection Agent Author and/or annotator of documents Annotation Document

7. Fabien.Gandon@sophia.inria.fr 7 Overall Schema Corporate Memory Multi-Agents System Learning User Agent Learning Profile Agent Ontology and Models Agent User Agent Learning Interconnection Agent End User Annotation Document Annotation Document Annotation Document Query

8. Fabien.Gandon@sophia.inria.fr 8 Functionalities studied in CoMMA: Annotate, Pull and Push Corporate Memory Multi-Agents System Learning User Agent Learning Profile Agent Ontology and Models Agent User Agent Learning Interconnection Agent Knowledge Engineer Author and/or annotator of documents End User Annotation Document Annotation Document Annotation Document Annotation Document Ontology Models - Enterprise Model - User's Profiles Query

9. Fabien.Gandon@sophia.inria.fr 9 Overall Schema Corporate Memory Multi-Agents System Learning User Agent Learning Profile Agent Ontology and Models Agent User Agent Learning Interconnection Agent Knowledge Engineer Author and/or annotator of documents End User Annotation Document Annotation Document Annotation Document Annotation Document Ontology Models - Enterprise Model - User's Profiles Query

10. Fabien.Gandon@sophia.inria.fr 10 Illustration of the cycle a - Reality  Ontology: explicit partial account of concepts used in the corporate memory management scenarios and their relations Organizational Entity (X) : The entity X is or is a sub-part of an organization. Person (X): The entity X is living being pertaining to the human race. Include (Organizational Entity: X, Organizational Entity / Person Y) : the organizational entity X includes Y as one of its members. Manage (Person: X, Organizational Entity: Y) : The person X watches and directs the organizational entity Y b - Ontology Person(Rose) Person(Fabien) Person(Olivier) Person(Alain) Organizational Entity(INRIA) Organizational Entity(Acacia) Include(INRIA, Acacia) Manage(Rose, Acacia) Include(Acacia, Rose) Include(Acacia, Fabien) Include(Acacia, Olivier) Include(Acacia, Alain) c - Situation & Annotations

11. Fabien.Gandon@sophia.inria.fr 11 Interesting aspects of XML & RDF(S)  XML leitmotiv: Bring structure to the memory to improve search and manipulation of documents using an emerging standard in industry.  RDF leitmotiv: If the corporate memory becomes an annotated world, software can use the semantics of these annotations an through inferences help the users exploit the corporate memory. OS AD + + + + Memory  Corporate semantic web: (annotated info world)  Ontology in RDFS (O'CoMMA)  Description the Situation in RDF: -User Profiles (annotate person) -Organization model (annotate groups)  Annotations in RDF describing Documents (manipulation at semantic level)  Annotated persons & organizational entities  context awareness

12. Fabien.Gandon@sophia.inria.fr 12 Use & Users (1) Scenarios and Data collection Building the O'CoMMA ontology Observations & Internal Interviews Documents Reuse External Expertise (Meta-) Dictionaries Scenarios (2) From semi-informal to semi-formal (3) RDF(S) Conceptual Vocabulary Relations - signature & sub. ex: person  (author)  document Terms & natural language definitions ex: 'bike', 'cycle', bicycle' - (bicycle) Concepts - subsumption ex: document  report (4) Navigation and Use

13. Fabien.Gandon@sophia.inria.fr 13 Scenarios : Summary table  Define some guidelines (influences ex: KADS)

14. Fabien.Gandon@sophia.inria.fr 14 Scenario Report : From industrial partners  Scenario Report : Rich story-telling document: Document type Event type Role type Very rich document... Function

15. Fabien.Gandon@sophia.inria.fr 15 Semi-structured Interview  Semi-structured (individual / group) guide for end-users: Definition of role (  nt tasks) Position : Personal definition  Official definition

16. Fabien.Gandon@sophia.inria.fr 16 Example of Observation

17. Fabien.Gandon@sophia.inria.fr 17 New Employee Route Card  What to do  Where to go  Who to contact  How to contact  In what order  Multi-lingual  NLP and Graphics

18. Fabien.Gandon@sophia.inria.fr 18 Quick reminder of the steps  State of the art & Reuse:  Enterprise Ontology,  TOVE Ontology,  Cyc Ontology,  PME Ontology,  CGKAT & WebKB top ontology  Other sources e.g.:  “Using Language” Herbert H. Clark,  MIME, Dublin Core,  Meta-dictionary, etc.  Terminological study: term to notions  Continuum Informal  Formal: Informal (textual)  Lexicons (semi-informal)  Structured tables (semi-formal)  RDF(S) (formal)  Structuring: Bottom-Up // Top-Down // Middle-Out

19. Fabien.Gandon@sophia.inria.fr 19 O'CoMMA  Content:  470 concepts (taxonomy depth = 13 subsumption links).  79 relations (taxonomy depth = 2 subsumption links).  715 terms in English and 699 in French.  547 definitions in French and 550 in English. Top : abstract Middle : common Extension: Specific EnterpriseDocumentUserDomain

20. Fabien.Gandon@sophia.inria.fr 20 Browsing the ontology

21. Fabien.Gandon@sophia.inria.fr 21 Other example: hyperbolic interface of OntoBroker

22. Fabien.Gandon@sophia.inria.fr 22 Other example: Protégé 2000

23. Fabien.Gandon@sophia.inria.fr 23 Functionalities studied in CoMMA: Annotate, Pull and Push Corporate Memory Multi-Agents System Learning User Agent Learning Profile Agent Ontology and Models Agent User Agent Learning Interconnection Agent Knowledge Engineer Author and/or annotator of documents End User Annotation Document Annotation Document Annotation Document Annotation Document Ontology Models - Enterprise Model - User's Profiles Query

24. Fabien.Gandon@sophia.inria.fr 24 Interfacing Users  User Interfaces  Annotating documents  Querying the memory  Present the results  Hide complexity (ontology, agents,...)  Machine learning leitmotiv:Represent, learn and compare current use profiles to improve future use.  Learning during a login session  Ranking results  Push technology  Improve information flowing  Proactive diffusion of annotations  Communities of interest

25. Fabien.Gandon@sophia.inria.fr 25 Interface Agent: Ontology-guided query on the corporate semantic web

26. Fabien.Gandon@sophia.inria.fr 26 Presenting and documenting results with the ontology

27. Fabien.Gandon@sophia.inria.fr 27

28. Fabien.Gandon@sophia.inria.fr 28

29. Fabien.Gandon@sophia.inria.fr 29

30. Fabien.Gandon@sophia.inria.fr 30

31. Fabien.Gandon@sophia.inria.fr 31 Profiles for ontology browsing

32. Fabien.Gandon@sophia.inria.fr 32 Shopping metaphor : Using a 'notion cart' for query building [CORESE specified by A. Giboin and implemented by J. Guillaume]

33. Fabien.Gandon@sophia.inria.fr 33 Complex interface [CORESE specified by implemented by Phuc Nguyen]

34. Fabien.Gandon@sophia.inria.fr 34 MAS Architecture Mémoire d'entreprise Système Multi-Agents Apprentissage Agent Utilisateur Apprentissage Agent groupe d'intérêts Agent Ontologie et Modèles Agent Utilisateur Apprentissage Agent d'inter- connexion Ingénieur de la connaissance Auteur et/ou Annotateur de documents Utilisateur final Annotation Document Annotation Document Annotation Document Annotation Document Ontologie Modèles - Modèle d'entreprise - Profils d'utilisateurs Requête

35. Fabien.Gandon@sophia.inria.fr 35 Principal interest of MAS in CoMMA  Leitmotiv: One functional architecture leading to several possible configurations in order to adapt to the broad range of environments that can be found in a company  Architecture: Agent kinds and their relationships Fixed at design time  Configuration: Exact topography of a given MAS Fixed at deployment time  One architecture  Several configurations Adapt to context  Agent paradigm adequacy:  Agent collaboration  Global capitalization  Agent autonomy & individuality  Local adaptation  Integration of different technologies  interesting for a domain that requires multidisciplinary solutions

36. Fabien.Gandon@sophia.inria.fr 36 Multi-agents information system for the CM  CoMMA is an heterogeneous multi-agents information system  From Macroscopic to Microscopic  Functional analysis for high level functions: societies  Four aspects in our scenarios: ontology and model availability, annotation management, user management and yellow pages services. Users' society Annotations Society Ontology and Model Society Interconnection Society

37. Fabien.Gandon@sophia.inria.fr 37 CoMMA Society Sub-societies and Roles Users' society Annotations Society Ontology and Model Society Interconnection Society Ontologist Agents MediatorsArchivists Profile Managers Profiles Archivists InterfaceControllers FederatedMatchmakers

38. Fabien.Gandon@sophia.inria.fr 38 Role cards AMLocal:AM*:AM *:AA 1:cfp 2:cfp 3:propose :protocol fipa contract net :content :language CoMMA-RDF :ontology CoMMA Ontology 5:accept/ reject :protocol fipa contract net :content :language CoMMA-RDF :ontology CoMMA Ontology 4:propose 6:accept/ reject 6:accept/ reject 7:inform 8:inform

39. Fabien.Gandon@sophia.inria.fr 39 Annotation allocation: contract-net based on pseudo-semantic distance bids

40. Fabien.Gandon@sophia.inria.fr 40 Conclusion  What you did not hear me say:  "ontology, DAI, etc. are a silver bullets for KM"  "an information system is the solution to KM problems"  "an ontology is easy to build, use, etc."  What you did hear me say:  "Knowledge-based system are not the old expert systems"  "Ontology is a new concept of knowledge representation that can be used in the supporting infrastructure of a complete solution"  "Distributed A.I. offers paradigms that can be used to build software architectures adapted to KM distributed" Just an example of the fact that the use of formal knowledge and (distributed) artificial intelligence can go a long way in K.M. support

41. Fabien.Gandon@sophia.inria.fr 41 That's all folks