Hide Notes 

Slide 1: Master Thesis Final Presentation RWTH Aachen University Multidimensional Patterns of Disturbance in Digital Social Networks Dimitar Denev Lehrstuhl für Informatik V Information Systems Prof. Dr. Matthias Jarke Advisors: Ralf Klamma Marc Spaniol Lehr- und Forschungsgebiet Knowledge-based Systems Prof. Gerhard Lakemeyer Ph.D.
Slide 2: Agenda  Motivation  Problem Analysis Approach  State of the Art  Model of Digital Social Networks  Pattern Language  PALADIN  Conclusions and Outlook 2
Slide 3: Motivation  Trolls – persons who post only in threads, started by themselves  Context   Yahoo! Mailing list „Greek Mythology Link“ Discussion about the movie „Troy“  Message of a troll  Troy is a MOVIE – message containing deliberate error  Movies are current mythology – message posted as a generally accepted fact without a proof or analysis  Is Christianity and all that other stuff myth, history, religion or what – inflammatory message including a contemptuous comment on religious thematic. 3
Slide 4: Problem Statement  Disturbance as a new source of information and a starting point for learning processes  Hinders the communication in the network  Compels individuals to leave the network  Difficulties for the disturbances to be discovered or predicted    Multidimensional context of the digital social networks Large size of the networks Knowledge about the disturbances is mostly from experience and observation 4
Slide 5: Solution Approach  A pattern language overcomes the difficulties for discovering and describing disturbances  Pattern – a general repeatable solution to a commonly recurring problem [Alexander 1978]  Machine-readable description of the patterns - XML- based Pattern Language for Multidimensional Disturbances  Automatic Analysis of digital social networks for disturbances with the pattern language 5
Slide 6: Solution Approach  The model of the digital social networks is a based on  Actor-Network Theory (ANT)  Graph Representation  Social Network Analysis (SNA)  I* Framework  Multidimensionality of the digital social networks reflected in the model      Sociology Computer Science Media Theory Graph Theory Social Capital Theory 6
Slide 7: State of the Art Digital Social Networks Projects Media Actors Relations Analysis Approach Social Network Analysis, Statistics COMB [Boudourides et al. 2002] Mailing List Individuals, Mails, Threads, Genres Developers, Software Components Individuals Posting in the same thread Ariadne [de Souza et al. 2004] Flink [Mika 2005] Eclipse IDE, CVS Repository Friend-Of-AFriend network, Google results Dependencies derived from the technical dependencies Relations built on the information from Google, Friend-Of-A-Friend network, Bibliography Temporal Analysis Social Network Analysis, Semantic Web 7
Slide 8: Model of Digital Social Networks Actor-Network Theory [Latour 1997]  Actor - the basic unit of the model, no difference between technical and social actors.  Semantics, given to the actors from the interpretation in the context of digital social networks:    Member – any person or group, part of the digital social network Medium – an actor which enables the members to exchange information Artefact – objects created by the members using some medium  Relation – a relation between two actors  Network – set of actors along with their relations 8
Slide 9: Model of Digital Social Networks Digital Social Network Digital Media Artefacts I* Dependencies Member Network Members 9
Slide 10: Model of Digital Social Networks Members  Member types defined according to patterns of behavior      Answering Person Questioner Troll Spammer Conversationalist  Member properties, defined with the help of SNA  Centrality types: degree centrality, closeness centrality, betweenness centrality - determined by the position of the member in the network  Efficiency – describes the existence of structural holes 10
Slide 11: Model of Digital Social Networks Media  Medium – an actor which enables the members to exchange information   Every network supports a set of media A medium affords the creation of a certain set of artefacts  Media types  Email  Discussion group  Chat room  Blog    Wiki Transaction-based web sites URL 11
Slide 12: Model of Digital Social Networks Artefacts  Artefact – objects created by the members using some medium  Artefact types        Message Burst Thread Blog entry Comment Conversation Feedback (Rating)  Artefact properties – author, date of creation, reply to 12
Slide 13: Model of Digital Social Networks I* Framework [Yu et al. 1997]  I* Dependency types  Goal  Resource  Task  Soft goal  Dependencies in digital social networks  Structural dependencies  Communication dependency  Cross-media dependencies  Coordination dependency  Artefact dependency 13
Slide 14: Model of Digital Social Networks I* Dependencies Example Coordinator Coordinatio n Iterant Broker isA Member Artefact isA Gatekeeper isA isA Hub URL Communication Network 14
Slide 15: State of the Art Pattern Languages Projects Domain Public Sphere Project [Schuler 2002] PoInter [Viller et al. 2000] PLML [Fincher 2004] E-LEN [Steeples et al. 2004] Social Studies Pattern Structure Problem, Context, Discussion, Solution Pattern Examples „Citizen access to simulations“, „Online Community Service Engine“ „Working in small groups“, „Overlapping responsibilities“ no patterns available Formal Definition Not available ComputerSupported Collaborative Work HumanComputer Interface e-Learning Essence, Context, Discussion, Implication, Pattern Relations Synopsis, Problem, Context, Forces, Rationale, Pattern Link Problem, Analysis, Solution, Context Not available XML Schema „Asynchronous collaborative learning“, „Student group management“ Not available 15
Slide 16: Pattern Language Pattern Structure  Pattern – a general repeatable solution to a commonly recurring problem [Alexander 1978]  Pattern structure       Disturbance Forces and force relations Solution Rationale Dependencies Pattern relations 16
Slide 17: Pattern Language FELP  Variables – simple variables (troll, thread), properties (thread.author) and set variables (v1,…,vn).  Operations  Arithmetic (+, -, *, / )  Aggregate (SUM, COUNT, AVERAGE)  Logical (&, |, ~, FORALL and EXISTS)  Comparison (=, !=, >, <).  Rules for variable binding    Simple variables – pattern parameters, actors or set variables Properties – actor properties or relations Set variables – actors 17  Interpreted by a finite state automaton
Slide 18: Pattern Language Sample Pattern Troll Pattern: This pattern tries to discover the cases when a troll exists in a digital social network. A troll in the network is considered a disturbance. Disturbance: (EXISTS [medium | medium.affordance = threadArtefact]) & (EXISTS [troll |(EXISTS [thread | (thread.author = troll) & (COUNT [message | (message.author = troll) & (message.posted = thread)]) > minPosts]) & (~EXISTS[ thread1, message1| (thread1.author1 != troll) & (message1.author = troll & message1.posted = thread1 ]))])]) Forces: medium; troll; network; member; thread; message; url Force Relations: neighbour(troll, member); own thread(troll, thread) Solution: No attention must be paid to the discussions started by the troll. Rationale: The troll needs attention to continue its activities. If no attention is paid, he/she will stop participating in the discussions. Pattern Relations: Associates Spammer pattern. 18
Slide 19: Pattern Language Algorithm for Pattern Application Pattern Disturbance Variables Pattern Template 1. Set pattern Disturbance parameters Variables Pattern Parameters  v1,...,vn – variables bound to actors a1,...,an  p1,…, pm – pattern parameters  d – disturbance with d=(v1,...,vn, p1,…, pm).  μ1,…, μm – substitutions for the pattern parameters  Set Pattern Parameters: d = d(v1,...,vn, p1/μ1,…, pm/μm) 19
Slide 20: Pattern Language Algorithm for Pattern Application Pattern Disturbance Variables Pattern Template 1. Set pattern Disturbance parameters Variables Pattern Parameters  α1,..., αk – actor instances in Digital Social Network 2. Instantiate disturbances the social network  I(ai)=(αi1,…,αir) – instances of Pattern Template Instance the actor ai  S = (s1,…,st)= I(a1)×…×I(an) Disturbance Instances Variables Pattern Parameters  Instantiate disturbances: D = (d(s1),…, d(sp)), where d(si) = d(v1/α i1,...,vn/αin,p1/μ1,…,pm/μm) 20
Slide 21: Pattern Language Algorithm for Pattern Application Pattern Disturbance Variables Pattern Instance Disturbance Variables Pattern Parameters Pattern Template 1. Set pattern Disturbance parameters Variables Pattern Parameters 4a. Change Pattern Parameters Digital Social Network 4b. Apply Pattern Solution Pattern Template Instance 2. Instantiate disturbances Forces Force Relations Solution Disturbance Instances Variables Pattern Parameters Description Rationale Dependencies 3. Evaluate disturbances Pattern Relations 21
Slide 22: PALADIN Architecture Implementation  PALADIN – PAttern LAnguage for DIsturbances in digital social Networks   ANT Subsystem Web Interface  XML Repository  Pattern Subsystem  Formal Expression Module  XML Pattern Repository  Web Interface  Social Network Subsystem  Base Social Network Module  JUNG Interface  IBM DB2 Database  Pattern Application Module  Formal Expression Evaluation  Pattern Instance Repository 22
Slide 23: PALADIN Web Interface  Step 1:  define disturbance expression  enter pattern properties  Step 2:  bind variables to actors  store pattern in the pattern repository 23
Slide 24: PALADIN JUNG Interface Extension  Troll  Spammers  Members  Size reflects centrality of the member  Members who participate in other disturbances, such as bursts or structural holes can be displayed as well 24
Slide 25: PALADIN Results  Case study - 10 patterns of disturbance over 119 social network instances, 17359 individuals, 215 345 mails Pattern Burst No Conversationalist No Questioner No Answering Person Troll Spammer Leader No Leader Structural Hole Independent Discussions Occurrences 22 76 67 61 2 86 37 40 67 13 Remarks The pattern finds out topics which were very important for certain period of time. Scalability is necessary. The existence implies little communication in the network. The existence implies that the network is not popular. Occurs in small networks. The effects of the lack of an answering person must be further checked with content analysis. Troll occurs very rarerly in cultural communities. True negatives exist. Spammers can be found often in discussion groups. False positives exist. The pattern occurs in the network centered around a member. Occurs in big networks where the members are distributed in different clusters. Occurs for members having neighbours with only one contact. Occurs in large networks where disconnected subnetworks exist. 25 Scalability is necessary.
Slide 26: Conclusion Media COMB [Boudourides et al. 2002] Ariadne [de Souza et al. 2004] Flink [Mika 2005] Mailing List Actors Individuals, Mails, Threads, Genres Developers, Software Components Individuals Relations Posting in the same thread. Analysis Approach Social Network Analysis, Statistics Eclipse IDE, CVS Repository Friend-Of-AFriend network, Google results Dependencies derived from the technical dependencies. Relations built on the information from Google, FOAF, Mails, Bibliography Depends on the used media in the network Temporal Analysis Social Network Analysis, Semantic Web PALADIN Any Type of Digital Social Network Media, Members, Artefacts Disturbance-oriented, Pattern Repository, Social Network Analysis, Temporal Analysis, Statistics 26
Slide 27: Outlook  Interoperability with applications based on Semantic Web, such as Flink  Methodology for visualization of multidimensional disturbances, must reflect   Media Artefacts   SWAP-it [Seeling et al. 2004] InfoSky [Tochterman 2002]  Dependencies  Integration with simulation environment for social networks – can predict disturbances earlier 27
Slide 28: THANK YOU FOR YOUR ATTENTION! 28