<is web> Information Systems & Semantic Web University of Koblenz Landau, Germany

Transcription

1 Information Systems University of Koblenz Landau, Germany On Understanding the Sharing of Conceptualizations, Klaas Dellschaft

2 What is an ontology? Gruber 93 (slightly adapted by Borst): An Ontology is a formal specification of a shared conceptualization of a domain of interest Executable, Discussable Group of persons About concepts Between application and unique truth Where do we get the ontologies from? 2of 38

3 Collaboration and Knowledge Degree of Collaboration ( shared ) high Folksonomies, Text memes formalize Ontologies share none to low Keywords, Text Knowledge Base, Logical Schemata, Most Ontologies none to low high Degree of Formalization 3of 38

4 Language Games Features Ball Duck For playing Round shape Certain sizes Not for eating Not liquid 4of 38

5 Narrow Folksonomy One resource tagged by at most one user 5of 38

6 Broad Folksonomy 6of 38

7 Broad Folksonomy Add your own tag-annotated bookmark In delicious 7of 38

8 Collaboration and Knowledge Degree of Collaboration ( shared ) high Folksonomies, Text memes formalize Ontologies share Conceptualization: An epiphenomenon of sharing and explication/ formalization! none to low Keywords, Text Knowledge Base, Logical Schemata, Most Ontologies none to low high Degree of Formalization 8of 38

9 Influence Factors on Conceptualization User interface Something else? Conceptualization Tagging Behavior Own Knowledge Shared terminology Observation Folksonomies serve as our fruit flys! Not, the real thing, but part of the real world for which we have data! 9of 38

10 Dynamics of Knowledge Sharing in Folksonomies 10 of 38

11 Better understanding of the tagging process Cooperative classification of resources Which factors influence the tagging process? Background knowledge of the user? Tag assignments of other users? Hypothesis: Tagging involves imitation of other users AND selection of tags from background knowledge of users. 11 of 38

12 Overall Scheme User interface Something else? Conceptualization Tagging Behavior Own Knowledge Model of User Interface Influence Joint Stochastic Model Shared terminology Comparison of Statistics Simulated Tagging Behavior Model of Own Knowledge Model of Sharing 12 of 38

13 Components of Analysis Properties of Tag Streams Stream view of Folksonomies Co-occurrence streams Resource streams Dynamic model for Tagging Systems Simulating background knowledge Simulating tag imitation Simulation Results Co-occurrence streams Resource streams Observations in the real world Stochastic models of influence Which models best fit the reality? 13 of 38

14 Properties of Tag Streams 14 of 38

15 Stream Views of a Folksonomy Folksonomies: Vertices: Users, tags, resources Edges: Tag assignments Postings: Tag assignments of a user to a single resource Can be ordered according to their time-stamp 15 of 38

16 Co-occurrence Streams Co-occurrence Streams: All tags co-occurring with a given tag in a posting Ordered by posting time Co-occurrence stream for 'apple': {mackz, r1, {apple, tree}, 13:25} {klaasd, r2, {apple, mac, ibook}, 13:26} {mackz, r2, {apple, macintosh, stevejobs}, 13:27} tree, mac, ibook, macintosh, stevejobs Tag Y U T R ajax blog xml of 38

17 Co-occurrence Streams Tag Growth linear growth 17 of 38

18 Co-occurrence Streams Tag Frequencies power law 18 of 38

19 Resource Streams Resource Streams: All tags assigned to a resource Ordered by posting time Resource stream for 'r2': {mackz, r1, {apple, tree}, 13:25} {klaasd, r2, {apple, mac, ibook}, 13:26} {mackz, r2, {apple, macintosh, stevejobs}, 13:27} apple, mac, ibook, apple, macintosh, stevejobs 19 of 38

20 Resource Streams Tag Frequencies 20 of 38

21 Resource Streams Tag Frequencies 21 of 38

22 Simulating the Evolution of Tag Streams 22 of 38

23 Simulating tag streams Which of my concepts represent this web page? How do I tag this web page? Inspiration for conceptualization from: 1. Most popular tags 2. Most recently used tags 3. Tags used for this resource 4. Tags co-occuring with similar text documents 5. Creating completely new tags 6. Which combination of inspirations develop the same statistics as the one observed for delicious? 23 of 38

24 The Delicious User Interface Imitating previous tag assignments: Recommended tags: Intersection of tags of a user and tags already assigned to the resource. Your tags: Tags of the user. Popular tags: 7 most popular tags assigned to the resource. 24 of 38

25 Simulating a Tag Stream Start with empty tag stream Each simulation step appends a new tag assignment Simulation of a single tag assignment: P BK p(w t): Probability of selecting word w for topic t. Modeled by word distributions in a topic centered text corpus. 1-P BK n: Number of visible previous tags. h: Maximal number of previous tag assignments used for determining ranking of the n distinct tags. 25 of 38

26 Modeling Background Knowledge Text Corpora Text Corpora Del.icio.us P BK : Probability of selecting from background knowledge p(w t): Probability of selecting word w for topic t. Modeled by word distributions in a topic centered text corpus. p(w r): Probability of selecting word w for resource r. 26 of 38

27 Modeling Tag Imitation P BK t t-1 t-2 t-3 t-4 t-5 t-h 1-P BK n P I = 1 P BK : Probability of imitating a previous tag assignment n: Number of visible top-ranked tags h: Maximal number of previous tag assignments used for determining ranking of the n distinct tags 27 of 38

28 Simulation Results 28 of 38

29 Overall Scheme User interface Something else? Conceptualization Tagging Behavior Own Knowledge Model of User Interface Influence Joint Stochastic Model Shared terminology Comparison of Statistics Simulated Tagging Behavior Model of Own Knowledge Model of Sharing 29 of 38

30 Simulating Co-occurrence Streams Tag growth: Influenced by P BK and p(w t) Tag Frequencies: Influenced by P BK, p(w t), n, h n: Semantic breadth of a topic (blog: 100 tags, ajax: 50 tags, xml: 50 tags; Cattuto et al. 2007) h: No hint for realistic values. Good guesses may be 500 and of 38

31 Co-occ. Streams Simulated Tag Growth 31 of 38

32 Co-occ. Streams Simulated Tag Frequencies 32 of 38

33 Co-occ. Stream Simulated Tag Frequencies 33 of 38

34 Simulating Resource Streams P I and P BK : Values comparable to co-occurrence streams p(w r): Approximated by p(w t) n: 7 tags are visible (cf. Delicious user interface) h: Smaller value than for co-occurrence streams 34 of 38

35 Res. Streams Simulated Tag Frequencies 35 of 38

36 Res. Streams Simulated Tag Frequencies 36 of 38

37 Conclusions Imitation AND background knowledge are needed for explaining properties of tag streams Probability of imitating previous tag assignments: ~70-90% Frequency Rank Co-occur. Streams Resource Streams Tag Growth Polya Urn Model o o fixed size Simon Model o o linear YS Model w/ Memory + o linear Halpin et al. Model o o linear Our Model + + power-law 37 of 38

38 Overall Scheme User interface Something else? Conceptualization Tagging Behavior Own Knowledge Model of User Interface Influence Joint Stochastic Model Shared terminology Comparison of Statistics Simulated Tagging Behavior Model of Own Knowledge Model of Sharing 38 of 38

39 Thank You! 39 of 38