1 Towards Decentralized Communities and Social Awareness Pierre Maret Université de Lyon (St Etienne) Laboratoire Hubert Curien CNRS UMR 5516
2 Who I am? Pierre Maret PhD in CS (1995) Ass. Prof. at INSA Lyon ( ) Prof. at Univ of St Etienne (Univ. of Lyon) since 2008 Research background : DB, IS, electronic documents, knowledge management, knowledge modeling
3 Talk on: Towards Decentralized Communities and social Awareness
4 A Community ? What is it? A set of participants? A topic? A protocol for the exchange of messages? A data base for storing some information? Actually, what is/are the objectives?
5 Improve information exchanges Increase efficiency Create new opportunities for relevant exchanges Enable exchange of new types of information Deliver the right information, at the right moment, and to the right person
6 Domains addressed Knowledge modeling Information diffusion, sharing, retrieval Recommendation systems
7 Social Networks Sites Great success 4 types: Content Sharing (i.e. U-Tube) Social Notification (i.e. Facebook) Expertise Promotion (i.e. Wikipedia) Virtual life, games (i.e. Second life) Great tools for building communities
8 Social Networks Sites Regarding Content sharing and Social notification: People trust people they know Social network ↔ Decision making Decision making = to follow recommendations to imitate behavior to support in real-life activities
9 Social Networks Sites Social networks can be useful but SNS have some drawbacks
10 Some drawbacks of SNS Multiple registration Close world (no interoperability) Privacy issues No control on data deletion Towards a unique governmental secure SNS ? No Then what?
11 Need for an open approach An open approach for community- related information exchanges include interoperability avoid personal data dispersion Proposal: A community abstraction Decentralized + bottom-up approach
12 Towards a decentralized approach 1 st step : Actors 2 nd step : Communities 3 rd step : Context
13 Towards a decentralized approach 1 st step : Actors Actors : an abstraction to model any participant Person Personnel assistant (artifact) Autonomous system (artifact) An actor has Knowledge Behavior (decision abilities, actions)
14 Actors as SW agents 2 types of agents: Context agent Dedicated to sensors From raw data to information Personal agent Personal assistant. Pro-active (internal goal) Contains some user's knowledge Knowledge is "delivered to" and "gathered from" the environment Mobility scenario or in-office scenario
15 Personnel agent Role of a user assistant Piece of software Autonomous software with communication abilities Knowledge = abstraction of the owner's knowledge Decision abilities = actions (managed by the owner), related to the present knowledge
16 Actor abstraction Expressed using web semantic techniques : OWL { k i } knowledge { b i } behavior { k i } knowledge Tulip is_a Flower Red is_a Color Tulip has_property Red T1 instance_of Tulip { b i } behavior Send message Receive message Extract Instances Set Value { k i } knowledge { b i } behavior Actor
17 Making behavior exchangeable Knowledge (RDF/OWL ontologies) can be exchanged Behavior is generally hardcoded : not exchangeable A model for expressing agent's behavior in SWRL (expression of rules on OWL) Work of Julien Subercaze (PhD candidate)
18 Making behavior exchangeable Behavior as a finite state machine If (transition from State A to State B) then (execute list of actions)
19 Describing information Using Tags to describe agents information/knowledge Tag = Annotations, Meta-data Concerns any information/knowledge/document picture signal , etc.
20 Tagging activity on personal agents Tagging activity Automated Semi-automated Manual Useful regarding information retrieval Several dimensions/processes for tags Location, environmental information, body information, thoughts, …
21 Tagging activity on personal agents Work of PhD candidate Johann Stan Main idea : the meaning of tag changes dynamically according to the user and circumstances. Circumstance : communities the user belongs to context
22 2 nd step : Communities 1 st Step : Actors Community : A set of actors with compatible communication abilities and shared values (common domain of interest) VKC = Virtual Knowledge Communities An abstraction for the exchange of information in- between actors
23 Features for communities Community-related knowledge of the agents List of (some) communities List of (some) agents Community-related domain knowledge (about the community topic) Community-related primitives Protocol: create, inform, request… Knowledge selection (extract from its knowledge) Knowledge evaluation and insertion (received through exchanges)
24 Features for communities Communities Knowledge Mappings
25 Agent communities Community protocol Create community (with a topic) Join, Leave Inform, request Specific role (any agents) Yellow page Knowledge = existing communities and topics
26 Example { ki } //joint communities C1 (on Car) C2 (on Flower)(Owner) { ki } Tulip is_a Flower C1 is a Community C2 is a Community //joint communities C2 (on Flower) { ki } Tokyo is_a City //joint communities C1 (on Car) A1 A2 A3 A3 has previously joined A1's community on Flowers. A3 wants to send some info to this community A2 needs more info about Japan. A2 is about to create a community on Japan
27 Communities and social network Memory of interactions builds my social network With who? The topic? The context? The environment? Carried out with tags Used to propose interaction facilities (prediction)
28 Communities and social network Example of annotations of interactions (manual) Automatic annotations: context, content analysis More about the context…
29 Step 3 : Context Context data: gathered from the environment Location Internal state Environment Activity (…) Situation = f(context data) SAUPO model: situation ↔ communication preferences
30 SAUPO model S ituation ↔ Communication preferences
31 Agent's context User's current activity as context data Identifying the user's current activity to promote exchanges Event + Content analysis and filtering Target : more accurate solicitations Contextual Notification Framework
32 Agent's context Contextual Notification Framework (Work of Adrien Joly, PhD Candidate) Filtered ambient awareness Main idea : maintain cooperation in-between people while reducing overload Context model Context sniffer (with user acceptance) Matchmaking process (context + social network) and notification
33 Contextual Notification Framework
34 Conclusion Improving knowledge exchanges Used techniques Semantics modeling: ontologies, owl Context awareness Social networks Leveraged into several scenarios or projects Leading idea : bottom-up approach
35 Thank you for your attention