Analysing Online Social Network Data with Biclustering and Triclustering
Download as PPTX, PDF5 likes1,604 views
The document discusses methods for analyzing online social network data through biclustering and triclustering techniques, addressing the limitations of traditional methods that produce a vast number of patterns. It introduces the concept-based bicluster approach for efficient community representation and presents a pseudo-triclustering algorithm for analyzing three-way network data like folksonomies. The findings suggest potential applications for community detection, recommendations, and the need for further refinement of the methods to enhance scalability and analysis quality.
![Formal Concept Analysis
[Wille, 1982, Ganter & Wille, 1999]
Definition 1. Formal Context is a triple (G, M, I ), where G is a
set of (formal) objects, M is a set of (formal) attributes, and I
G M is the incidence relation which shows that object g
G posseses an attribute m M.
Example
Car House Laptop Bicycle
Kate x x
Mike x x
Alex x x
David x x x](https://image.slidesharecdn.com/semenovslides-120320081828-phpapp02/85/Analysing-Online-Social-Network-Data-with-Biclustering-and-Triclustering-5-320.jpg)
![Triclustering
[Ignatov et al., 2011]
One dense tricluster VS 33 =27 formal triconcepts](https://image.slidesharecdn.com/semenovslides-120320081828-phpapp02/85/Analysing-Online-Social-Network-Data-with-Biclustering-and-Triclustering-16-320.jpg)
Analysing Online Social Network Data with Biclustering and Triclustering
- 1. Analysing Online Social Network Data with Biclustering and Triclustering Alexander Semenov1 Dmitry Ignatov1 Dmitry Gnatyshak1 Jonas Poelmans2,1 1NRU Higher School of Economics, Russia 2KU Leuven, Belgium
- 2. Motivation I • There are large amount of network data that can be represented as bipartite and tripartite graphs • Standard techniques like maximal bicliques search result in huge number of patterns (in the worst case exponential w.r.t. of input size)… • Therefore we need some relaxation of this notion and good measures of interestingness of biclique communities
- 3. Motivation II • Applied lattice theory provide us with a notion of formal concept which is the same thing as biclique • L. C. Freeman, D. R. White. Using Galois Lattices to Represent Network Data Sociological Methodology 1993 (23). • Social Networks 18(3), 1996 – L. C. Freeman, Cliques, Galois Lattices, and the Structure of Human Social Groups. – V. Duquenne, Lattice analysis and the representation of handicap associations. – D. R. White. Statistical entailments and the Galois lattice. • J.W. Mohr, Vincent D. The duality of culture and practice: Poverty relief in New York City, 1888—1917 Theory and Society, 1997 • Camille Roth et al., Towards Concise Representation for Taxonomies of Epistemic Communities, CLA 4th Intl Conf on Concept Lattices and their Applications, 2006 • And many other papers on application to social network analysis with FCA
- 4. Motivation III • Concept-based bicluster (Ignatov et al., 2010) is a scalable approximation of a formal concept (biclique) – Less number of patterns to analyze – Less computational time (polynomial vs exp.) – Manual tuning of bicluster (community) density threshold – Tolerance to missing (object, attribute) pairs • For analyzing three-way network data like folksonomies we proposed triclustering (Ignatov et al., 2011)
- 5. Formal Concept Analysis [Wille, 1982, Ganter & Wille, 1999] Definition 1. Formal Context is a triple (G, M, I ), where G is a set of (formal) objects, M is a set of (formal) attributes, and I G M is the incidence relation which shows that object g G posseses an attribute m M. Example Car House Laptop Bicycle Kate x x Mike x x Alex x x David x x x
- 6. Formal Concept Analysis Definition 2. Derivation operators (defining Galois connection) AI := { m M | gIm for all g A } is the set of attributes common to all objects in A BI := { g G | gIm for all m B } is the set of objects that have all attributes from B Example Car House Laptop Bicycle {Kate, Mike}I = {Car} Kate x x {Laptop} I = {Mike, Alex, David} Mike x x Alex x x {Car, House} I = {}G David x x x {} IG =M
- 7. Formal Concept Analysis Definition 3. (A, B) is a formal concept of (G, M, I) iff A G, B M, AI = B, and BI = A . A is the extent and B is the intent of the concept (A, B). B (G, M , I ) is a set of all concepts of the context (G, M, I) Example • A pair ({Kate, Mike},{Car}) is a formal concept Car House Laptop Bicycle Kate x x • ({Alex, David} ,{Laptop}) doesn‘t form a formal concept, because Mike x x {Laptop} I {Alex, David} Alex x x • ({Alex, David} {House, Laptop}) David x x x is a formal concept
- 8. FCA and Graphs a b c d Kate x x Mike x x Alex x x David x x x Formal Context Bipartite graph Formal Cocept Biclique (maximal rectangle)
- 9. Formal Concept Analysis Definition 4. A formal concept (A,B) is said to be more general than (C,B), that is (A,B) (C,D) iff A C (equivalently D B) The set of all concepts of the context (G, M, I) ordered by relation forms a complete lattice B (G, M , I ) called concept lattice (Galois lattice). Example Car House Laptop Bicycle ({Alex, David, Mike} ,{Laptop}) Kate x x is more general than concept Mike x x ({Alex, David} {House, Laptop}) Alex x x David x x x
- 10. Concept Lattice Diagram a b c d a - Car Kate x x b - House Mike x x c - Laptop Alex x x David x x x d - Bicycle
- 12. Biclustering Example Car House Laptop Bicycle Kate x x Mike x x Alex x x David x x x Since (House, David) is in the context (HouseI, DavidI)= ({Alex, David}, {House,Laptop, Bicycle}) (HouseI, DavidI)=5/6
- 13. Biclustering properties • Number of all biclusters for a context (G,M,I) not greater than |I| vs 2min{|G|,|M|} formal concepts. Usually |I| « 2min{|G|,|M|}, especially for sparse contexts. • Probably dense biclusters ( (bicluster) min) are good representation of communities, because all users inside the extent of every dense bicluster have almost all interests from its intent.
- 14. Triadic FCA and Folksonomies Definition 1. Triadic Formal Context is a quadruple (G, M, B, Y ), where G is a set of (formal) objects, M is a set of (formal) attributes, B is a set of conditions, and Y G M B is the incidence relation which shows that object g G posseses an attribute m M under condition. Example. Folksonomy as triadic context (U, T, R, Y), where U is a set of users T is a set of tags R is a set of resources
- 15. Concept forming operators in triadic case To define triclusters we propose box operators
- 16. Triclustering [Ignatov et al., 2011] One dense tricluster VS 33 =27 formal triconcepts
- 17. Pseudo Triclustering for Social Networks
- 18. Algorithm
- 19. Algorithm
- 20. Data Pseudo-triclustering algorithm was tested on the data of Vkontakte, Russian social networking site. Student of two major technical and two universities for humanities and sociology were considered: Bauman MIPT RSUH RSSU # users 18542 4786 10266 12281 # interests 8118 2593 5892 3733 # groups 153985 46312 95619 102046
- 21. Biclustering results Bauman MIPT RSUH RSSU UI UG UI UG UI UG UI UG Time # Time # Time # Time # Time # Time # Time # Time # 0.0 9188 8863 1874458 248077 863 2492 109012 46873 3958 5293 519772 116882 2588 4014 693658 145086 0.1 8882 8331 1296056 173786 827 2401 91187 38226 3763 4925 419145 93219 2450 3785 527135 110964 0.2 8497 6960 966000 120075 780 2015 74498 28391 3656 4003 330371 68709 2369 3220 402159 79802 0.3 8006 5513 788008 85227 761 1600 63888 21152 3361 3123 275394 50650 2284 2612 332523 58321 0.4 7700 4308 676733 59179 705 1270 56365 15306 3252 2399 232154 35434 2184 2037 281164 40657 0.5 7536 3777 654047 53877 668 1091 54868 13828 3189 2087 224808 32578 2179 1782 270605 37244 0.6 7324 2718 522110 18586 670 775 44850 5279 3075 1367 174657 10877 2159 1264 211897 12908 0.7 7250 2409 511711 15577 743 676 43854 4399 3007 1224 171554 9171 2084 1109 208632 10957 0.8 7217 2326 508368 14855 663 654 43526 4215 3032 1188 170984 8742 2121 1081 209084 10503 0.9 7246 2314 507983 14691 669 647 43216 4157 2985 1180 174781 8649 2096 1072 206902 10422 1.0 7236 2309 511466 14654 669 647 43434 4148 3057 1177 173240 8635 2086 1068 207198 10408
- 22. Pseudo triclustering results Bauman MIPT RSUH RSSU Time, ms Count Time, ms Count Time, ms Count Time, ms Count 0.0 3353426 230161 77562 24852 256801 35275 183595 55338 0.1 76758 10928 35137 5969 62736 5679 18725 5582 0.2 80647 8539 31231 4908 58695 5089 16466 3641 0.3 77956 6107 27859 3770 53789 3865 17448 2772 0.4 60929 31 2060 12 9890 14 13585 12 0.5 66709 24 2327 10 9353 14 12776 10 0.6 57803 22 2147 8 11352 14 12268 10 0.7 68361 18 2333 8 10778 12 13819 4 0.8 70948 18 2256 8 9489 12 13725 4 0.9 65527 18 1942 8 10769 12 11705 4 1.0 65991 18 1971 8 10763 12 13263 4
- 23. Pseudo triclustering results Bauman MIPT 1000000 100000 100000 10000 10000 1000 1000 100 100 10 10 1 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 RSUH RSSU 100000 100000 10000 10000 1000 1000 100 100 10 10 1 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
- 24. Examples. Biclusters • =83,33% Gen. pair: {3609, home} G: {3609, 4566} M: {family, work, home} • =83,33% Gen. pair: {30568, orthodox church} G: {25092, 30568} M: {music, monastery, orthodox church} • =100% Gen. pair: {4220, beauty} G: {1269, 4220, 5337, 20787} M: {love, beauty}
- 26. Conclusion • It is possible to use pseudo-triclustering method for tagging groups by interests in social networking sites and finding tricommunities. E.g., if we have found a dense pseudo-trciluster (Users, Groups, Interests) we can mark Groups by user intersts from Interests. • It also make sense to use biclusters and tricluster for making recommendations. Missing pairs and triples seem to be good candidates to recommend potentionaly interesting users, groups and interests.
- 27. Conclusion • The approach needs some improvements and fine tune in order to increase the scalability and quality of communities – Strategies for approximate density calculation – Choosing a good thresholds for n-clusters density and communities similarity – More sophisticated quality measures like recall and precision in Information Retrieval • It needs comparison with other approaches like iceberg lattices (Stumme), stable concepts (Kuznetsov), fault- tolerant concepts (Boulicaut) and different n-clustering techniques from bioinformatics (Zaki, Mirkin, etc.) • Current version also requires expert’s feedback on the output data analysis and interpretation
- 28. Questions?