Ifip wg-galway-


Published on

summary of current work. presented to the IFIP Working Group on Social Semantics

Published in: Technology, Business
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide
  • Slide1: who you areFuture Internet – here KMi is playing a major role in shaping the future Internet in a large EU initiative, which envisages a global network encompassing both a variety of devices and a variety of services; Knowledge Management – developing new methods for capturing, interpreting, organising and sharing knowledge in a variety of learning and knowledge management contexts – e.g., we work with OU students, school children, the corporate world, etc.; Multimedia & Information Systems (MMIS) – developing new solutions for indexing, searching, organising and interacting with different types of media content; Narrative Hypermedia – developing new infrastructures to support collaborative discourse and sensemaking in fields such as open, participatory learning, e-democracy, scholarly research and knowledge management; New Media Systems – developing and applying new media solutions, such as desktop video conferencing, video blogs, podcasting, etc, in a variety of learning and commercial contexts; Semantic Web & Knowledge Services – researching the emerging Semantic Web (or Web 3.0), to develop new methods for locating, organising and making sense of web content; Social Software – this can be thought of as "software which extends, or derives added value from, human social behaviour - message boards, musical taste-sharing, photo-sharing, instant messaging, mailing lists, social networking”. Here we investigate various contexts of work, learning and play to better understand the trade-offs involved in designing effective large-scale social software application, which can be effective in a variety of contexts.
  • Semantics to facilitate integrating all this info and adding meaning to some of the SNS data. Semantics makes it easier to integrate and analyse such multidimensional networks.
  • Risk and opportunity management tools and methods for online communitiesCloud based data management and processing to support real-time analytics Representations, measures, and monitors for user, subgroup behaviour and community evolution in online communitiesLarge scale simulation for predicting impact of user behaviour and policies on community evolution and the risks and opportunities for online business.Scalable real time tools and algorithms for community analysis including dynamics and interactions
  • http://www.viralheat.com/homeInfluence is based on number of followers! Viral analysis – analyses content that is going viralDetecting sales leads from intent analysis to identify what users are interested in
  • Content features play a greater role than user featuresThe combination of all features provides the best resultsBoxplots help to visualise the distribution of the data, by splitting them into quartiles, with top max and bottom min, and outliers and median all shown on the plot. Boxplots: http://flowingdata.com/2008/02/15/how-to-read-and-use-a-box-and-whisker-plot/
  • Receiver Operator Characteristic CurveTrue Positive RateFalse Positive Rate
  • Ifip wg-galway-

    1. 1. Behaviour and Health Analysis ofOnline CommunitiesHarith AlaniKnowledge Media institutetwitter.com/halanidelicious.com/halanilinkedin.com/pub/harith-alani/9/739/534facebook.com/harith.alaniIFIP WG 12.7 – Galway, October 12, 2012
    2. 2. Milton Keynes
    3. 3. Knowledge Media institute (KMi)• Set up in 1995 to bring the OU to the forefront ofresearch and development• Different from the rest of the OU– 100% focus on research and development• has around 60 researchers, lead by 8 senior staff• Over 100 projects, and 1000 publications• Core research areas:– Future Internet, Knowledge Management, Multimedia &Information Systems, Narrative Hypermedia, New MediaSystems, Semantic Web & Knowledge Services, Social Software
    4. 4. 5 9 13 17 21 25 29 33 37 41 45H-Index F2F Degree F2F Strengthhealthy scien fic & socialprofiles. freq chairs/OCsin LSS teamgood scien fic, andsocial signalsshy scien st?outsider,high profileStudents, PG, developers.whos the next star researcher?First encounter with ‘Behaviour analysis’• Integration of physicalpresence and onlineinformation• Semantic user profilegeneration• Logging of face-to-face contact• Social network browsing• Analysis of online vs offlinesocial networks
    5. 5. eParticipation is about reconnecting ordinary people with politics andpolicy-making [….] Governments and the EU institutions working with citizensto identify and test ways of giving them more of a stake in the policy-shapingprocess, such as through public consultations on new legislation• Problem is that people don’t use government portals, minister blogs, opinion collecting web sites• Instead, they use social media• Targeted at developing methods to understand and manage the business, social and economicobjectives of the users, providers and hosts and to meet the challenges of scale and growth inlarge communities• Management and risk analysis in business online communities• Scalable, real time analysis of behaviour, value, and health of communitieshttp://robust-project.eu/http://wegov-project.eu/
    6. 6. “specifically designed forpoliticians, enabling them to monitor debate,filter out the background "noise" and zoom inon what people are saying about them andtheir policies in a particular geographical area”http://www.wegov-project.eu/
    7. 7. Management of Online Communities Health– Which are strong and healthy?– Which are aging and withering?– What health signs should we lookfor?– How these signs differ betweendifferent communities?• Evolution– Can we predict their futureevolution?– How can their evolution beinfluenced?• Behaviour– How can behaviour be detected?– How are their member behaving?– Which behaviour is good/bad inwhich community type?– What’s the lifecycle of behaviourroles?• Goals and Values– What are the goals of thesecommunities?– Are they fulfilling the goals oftheir owners?– Are they fulfilling the goals oftheir members?– Which members are valuable?
    8. 8. 8Tools for monitoring social networks
    9. 9. http://www.ubervu.com/9• Analytics:– Mention volume– Sentiment– Discussion clouds– Activity graphs andmetrics– Language andgeolocation filtering– Filter by socialplatform– Comparisons
    10. 10. http://www.viralheat.com/home• Analytics:– Influencing users– Sentiment and opinion analysis– Viral content analysis– Detecting sales leads– Filter by geo-location
    11. 11. Tweet recipe for generating more attention• Identifying seed postsTop features: Time inDay, Readability, Out-Degree, Polarity, InformativenessAccuracy of the classification (J48)F1: 0.841 (User + Content)Top features: Referral Count, TopicLikelihood, Informativeness, Readability,User AgeAccuracy of the classification (J48)F1: 0.792 (User + Content + Focus)For both datasets:• Content features play a greater rolethan user features• The combination of all featuresprovides the best results• Predicting discussion activity Top features: Referral Count(-),Complexity(-)User features harm the performanceTop features: Referral Count(-), Polarity(-),Topic Likelihood(+), Complexity (+)Best with Content +FocusFor both, a decrease in Referral Count isassociated with heightened activity.Language and terminology are moresignificant for Boards.ie.
    12. 12. Semantic engine for behaviour analysis• Bottom Up analysis– Every community member isclassified into a “role”– Unknown roles might beidentified– Copes with role changes overtimeinitiatorslurkersfollowersleadersStructural, social network,reciprocity, persistence, participationFeature levels change with thedynamics of the communityAssociations of roles with a collection offeature-to-level mappingse.g. in-degree -> high, out-degree -> highRun rules over each user’s featuresand derive the community role composition
    13. 13. Correlation of behaviour with community activityForum 246 – Commutingand TransportForum 388 – Rugby Forum 411 – Mobile Phones and PDAs
    14. 14. Online Community HealthAnalytics0.0 0.2 0.4 0.6 0.8 RateFPRTPR0.0 0.2 0.4 0.6 0.8 CountFPRTPR0.0 0.2 0.4 0.6 0.8 / Non−seeds PropFPRTPR0.0 0.2 0.4 0.6 0.8 CoefficientFPRTPR• Machine learning models topredict community health basedon compositions and evolutionof user behaviourHealthcategories0.0 0.2 0.4 0.6 0.8 / Non−seeds PropFPRTPR0.0 0.2 0.4 0.6 0.8 CoefficientFPRTPRFalse Positive RateFalse Positive RateFalse Positive RateFalse Positive RateTruePositiveRateTruePositiveRateTruePositiveRateTruePositiveRate
    15. 15. Behaviour evolution patterns• Can we predict futurebehaviour role?• Who’s on the path tobecome a leader? anexpert? a churner?• Which users we want toencourage staying/leaving?experts to-beabout to churnon right pathto leadership
    16. 16. OU Communities• Many FB groups existfor students of OUcourses• Created and used bystudents to discuss andshare opinions oncourses and get supportBehaviourAnalysisSentimentAnalysisTopicAnalysisCourse tutorsReal timemonitoring• How do students likethis course?• What main topics arethey busy discussing?• Do students get theanswers and supportthey need?• Which students arelikely to drop out?
    17. 17. What’s next!• Community-type analysis• Stability of results over time and events• Health metrics (what’s good/bad?)• Influence/change in behaviour
    18. 18. Relevant Publications• Rowe, W. and H. Alani. What makes Communities Tick? Community Health Analysis using Role Compositions. Proceedings ofthe Fourth IEEE International Conference on Social Computing. Amsterdam, The Netherlands (2012)• Rowe, M., M Fernandez, S Angeletou and H Alani. Community Analysis through Semantic Rules and Role CompositionDerivation. In the Journal of Web Semantics (2012)• Burel, G.; He, Y. and Alani, H. Automatic identification of best answers in online enquiry communities. In: 9th ExtendedSemantic Web Conference, Crete, (2012)• Rowe, Matthew; Fernandez, Miriam; Alani, Harith; Ronen, Inbal ; Hayes, Conor and Karnstedt, Marcel (2012). Behaviouranalysis across different types of Enterprise Online Communities. In: ACM web Science Conference 2012 (WebSci12),Evanston, U.S.A, (2012)• Rowe, M., Stankovic, M., and Alani, H. Who will follow whom? Exploiting semantics for link prediction in attention-information networks. In: 11th International Semantic Web Conference (ISWC 2012), Boston, USA, (2012)• Wagner, C., Rowe, M., Strohmaier, M. and Alani, H. Ignorance isnt bliss: an empirical analysis of attention patterns in onlinecommunities. In: 4th IEEE International Conference on Social Computing, Amsterdam, The Netherlands, (2012)• Angeletou, S., Rowe, M. and Alani, H. Modelling and Analysis of User Behaviour in Online Communities. InternationalSemantic Web Conference. Bonn, Germany (2011)• Karnstedt, M., Rowe, M., Chan, J., Alani, H., and Hayes, C. The Effect of User Features on Churn in Social Networks. In: ACMWeb Science Conference 2011 (WebSci2011), Koblenz, Germany, (2011)• Rowe, M., Angeletou, S., and Alani, H. Predicting discussions on the social semantic web. In: 8th Extended Semantic WebConference (ESWC 2011), Heraklion, Greece, (2011)http://oro.open.ac.uk/view/person/ha2294.html