Dynamic Generation of Agent Communities from Distributed Production and Content-Driven Delivery of Knowledge AAAI Spring Symposium on Agent-Mediated Knowledge Management (AMKM-03) Sinuhé Arroyo Juan Manuel Dodero Richard Benjamins Institut für Informatik IFI Intelligent Software Components Computer Science Department Next Generation Research Group Universidad Carlos III de Madrid (iSOCO), S.A. Madrid, Spain University of Innsbruck, Austria 1
Dynamic Generation of 1. Introduction Agent Communities 2. Multi-agent collaborative from Distributed production Production and Content-Driven Delivery Features and structure � of Knowledge Interaction within marts � Consolidation protocol � 3. Case study Course of the protocol � Results � 4. Dynamics of markets 5. Conclusions 2
1. Introduction � Collaborative knowledge management Intro Intro � KM processes acquisition production � Distributed system Multi-agent system � Collaborative creation � Task coordination needed delivery � Creation or production � Different interaction policies: Case study compete, cooperate, negotiate � Structured interaction � Delivery � Content-driven Dynamic of � Communities of interest markets Conclusions –3– Dodero, Arroyo. — AMKM 2003
2. Multi-agent collaborative production � Producers’ collaboration (e.g. instructional Intro designers) � Asynchrony Multi-agent Multi-agent system system • Development, exchange and evaluation of proposals are asynchronous. • Different pace of creation � Different levels of knowledge (Domain-level knowledge) Case � Decision privileges (e.g. lecturers vs. assistants) study � Conflicts � Multi-agent architecture motivation � Facilitates coordination when collaborating (e.g., compose a new educational resource) Dynamic of markets � Allows different interaction styles (e.g., compete, cooperate, or negotiate) Conclusions � Organizes interaction in distributed, but interconnected domains of interaction –4– Dodero, Arroyo. — AMKM 2003
System features � From a functional perspective… Intro � Consolidation of knowledge that is produced Multi-agent Multi-agent system system � From a structural perspective… � Multi-tiered structure Interaction group M � Agents operate in tightly-coupled Agent hierarchical knowledge marts Case Proxy Agent study � Progressive consolidation of knowledge Interaction group S1 Interaction gr Agent Agent Agent Proxy � From a behavioural Agent Agent Agent Agent perspective… � Affiliation of agents into marts Dynamic of markets � Evolution of marts Conclusions –5– Dodero, Arroyo. — AMKM 2003
Interaction within marts � Principles Intro � Agent rationality modeled as preference relationships k1 > k2 or relevance functions u ( k ) Multi-agent Multi-agent system system � Relevant aspects modeled as RDF triples (object, attribute, value): • Submitter’s hierarchical level • Fulfilment of goals Case study • Time-stamp � Message exchange � Message types • proposal ( knowledge, interaction ) Dynamic of • consolidate ( knowledge, interaction ) markets � Multicast, reliable transport facility Conclusions –6– Dodero, Arroyo. — AMKM 2003
Consolidation protocol receive any Intro worse-evaluated Multi-agent Multi-agent system system start receive consolidation (send proposal) Distribution Distribution better-evaluated any message Idle t 0 expires receive proposal Failure Case better-evaluated study receive any receive consolidation Consolidation Consolidation worse-evaluated better-evaluated Success t 1 expires Dynamic of markets Conclusions –7– Dodero, Arroyo. — AMKM 2003
3. Case study � Learning Object Intro � Course titled “Introduction to XML” � Roles Multi-agent system � 3 instructional designers, represented by agents A 1.. A 3 � A 1 is a docent coordinator � Task Case Case study study � Development of the TOC � A 1 submits p, A 2 submits q, A 3 does nothing � Proposals � p = Proposed manifest file with 6 chapters Dynamic of markets � q = Modified manifest file, divides up chapter 5 in two � Evaluation criteria Conclusions � Fulfillment of objectives � Actor’s rank –8– Dodero, Arroyo. — AMKM 2003
Course of the protocol Intro Proposal p Start timeout Start timeout Proposal q t 0 t 0 Multi-agent A 1 A 2 Proposal q A 1 A 2 system Proposal p Proposal q u ( p ) < u ( q ) u ( p ) < u ( q ) A 3 A 3 Start timeout t 1 Reply with OK Case Case study study t 0 expires Initial exchange of proposals After receiving proposals Termination: Finish: unsuccessful successful Start timeout t 1 A 1 A 2 Consolidate q A 1 A 2 Dynamic of markets Consolidate q t 1 expires A 3 A 3 Conclusions OK Consolidation after t 0 expiration After t 1 expiration –9– Dodero, Arroyo. — AMKM 2003
Results: quality (grade of fulfilment) Intro Issued in two-mart scenario Issued in one-mart scenario 80 75 75 75 Multi-agent system 70 60 55 Grade of fulfillment (%) 55 54 54 50 43 43 Case Case 43 study 40 study 30 20 10 Dynamic of markets 0 Conclusions 0 2 4 6 8 10 12 14 16 18 Proposals ordered by submission time –10– Dodero, Arroyo. — AMKM 2003
Results: consolidation lifetime Intro 183027 Multi-agent 190288 200000 system Consolidation lifetime (time units) 180000 160000 140000 120000 Case Case study 100000 study 80000 60000 40000 13439 20000 7291 Dynamic of 7401 3976 2183 1622 0 markets 1 2 3 4 Conclusions Consolidated proposals ordered by instant of consolidation –11– Dodero, Arroyo. — AMKM 2003
Results: number of conflicts No. of conflicts (two-mart) No. of conflicts (one-mart) Intro 70,00 Multi-agent system 60,00 No. of conflicts/time unit 50,00 40,00 Case Case study study 30,00 20,00 10,00 Dynamic of markets 0,00 I1 A1 A3 Conclusions Agents –12– Dodero, Arroyo. — AMKM 2003
4. Dynamics of markets � Dynamics of collaborative groups Intro � Agents affiliate to marts depending on the kind of knowledge that they produce Multi-agent system � Marts evolve (merge or divide) depending on the kind of knowledge consolidated within them � Agents arrangement � Cognitive distance d k between agents and marts Case study � Defined from dissimilarity between issued proposals’ attributes � Agents operate in the nearest mart � Agents relocate based on Knowledge production � Evolution of groups Dynamic of Dynamic of markets markets � Mart fusion/division Conclusions � MajorClust algorithm –13– Dodero, Arroyo. — AMKM 2003
Dynamic of markets Intro � Information brokering services Multi-agent � Content-driven delivery system � Filters to deliver contents of interest � Publish/subscribe pattern Case � Communities of users study � User agents subscribe to items of interest � User agents produce (publish) items � Brokers’ routing tables are built Dynamic of Dynamic of markets markets � Routing tables contain (hide) users’ layout into communities of interest Conclusions –14– Dodero, Arroyo. — AMKM 2003
Goal Intro � Effective communications Multi-agent � Reduce amount of info shared by brokers system � Reduce distance among agents and their interested marts � Evaluate Case � Mart’s optimal size study � Cost of agent’s relocation related to brokers communication efforts � Impact of mart’s evolution in the service � Find best clustering algorithm Dynamic of Dynamic of markets � K -means, COBWEB, MajorClust,… etc markets Conclusions –15– Dodero, Arroyo. — AMKM 2003
5. Conclusions � Features Intro � Bottom-up, multi-agent approach to collaborative knowledge production systems Multi-agent system � Dynamic building of user communities � Applicable to other collaborative KM production tasks • e-Book & learning objects composition • Calendar organization Case study • Software development (analysis & design) � Improvements � Further validation in multi-tiered scenarios � Test of mixed interaction styles (retract, substitute, Dynamic of reject) markets � Evaluation of dynamic evolution of marts Conclusions Conclusions –16– Dodero, Arroyo. — AMKM 2003
Recommend
More recommend
