1. Interactive Discovery of Influential Friends from Social Networks By: Behzad Rezaie In the Name of God Professor: Dr. Mashayekhi May 11, 2014 brezaie@shahroodut.ac.ir

2. Experimental Results Proposed Method Problem Description Introduction State of the Art Conclusion

3. Cameron JJ, Leung CKS, Tanbeer SK (2011) Finding strong groups of friends among friends in social networks. In: SCA 2011, pp. 824–831. Jiang F, Leung CKS, Tanbeer SK (2012) Finding popular friends in social networks. In: SCA 2012, pp. 501–508. 5 % Completed

4. Experimental Results Proposed Method Problem Description Introduction State of the Art Conclusion

5. Social networks have become popular to facilitate collaboration and knowledge sharing among users Interactions or interdependencies among users are deeply important in social networks Such interactions or interdependencies can be dependent on or influenced by user characteristics such as connectivity, centrality, weight, importance, and activity in the networks 10 % Completed

6. Experimental Results Proposed Method Problem Description Introduction State of the Art Conclusion

7. A Facebook user may want to identify those prominent friends who have high impact (e.g., in terms of knowledge or expertise about a subject matter) in the social network. A LinkedIn user may want to get introduced to those second- degree connections who have rich experience in some profession. 15 % Completed

8. Finding influential friends from social networks may also help corporations and business organizations in making important business decisions. A Twitter use may also be interested in following (and subscribing to a Twitter feed from) those who are highly influential in the whole network. 20 % Completed

9. Experimental Results Proposed Method Problem Description Introduction State of the Art Conclusion

10. G = {Ana, Carlos} LC = {L1, L2, L5, L7} Freq(G, LC) = 4 25 % Completed

11. The prominence, which is represented by a non-negative number, indicates the status (such as importance, weight, value, reputation, belief, position, or significance) of a friend in a social network. 30 % Completed

12. Inf({Ana, Carlos}, LC) = Prom({Ana, Carlos}) * Freq({Ana, Carlos}, LC) = 0.5 * 4 = 2.0 35 % Completed

13. When mining frequent patterns, the frequency measure satisfies the downward closure property: if a pattern is frequent, then all its subsets are also frequent. Equivalently, if a pattern is infrequent, then all its supersets are also infrequent. Influence does not satisfy the downward closure property. minInf = 2.0 Inf({Carlos}) = 4 * 0.4 = 1.6 Inf({Ana, Carlos}) = 4 * 0.5 = 2.0 40 % Completed

14. Example minInf = 2.0 According to prominence value, we have: 45 % Completed L1 = {Carlos, Eva, Beto, Ana} L2 = {Carlos, Beto, Ana} L3 = {Eva, Beto, Fabio} L4 = {Beto, Ana, Davi} L5 = {Carlos, Eva, Beto, Ana} L6 = {Eva, Beto, Fabio} L7 = {Carlos, Eva, Beto, Ana}

15. L1 = {C, E, B, A} L2 = {C, B A} L3 = {E, B, F} L4 = {B, A, D} L5 = {C, E, B, A} L6 = {E, B, F} L7 = {C, E, B, A} IF-tree construction 50 % Completed

16. DIFSoN Mining Routine Using PromGMax 55 % Completed

17. Enhanced DIFSoN Mining Routine Using PromLMax 60 % Completed

18. 65 % Completed

19. Experimental Results Proposed Method Problem Description Introduction State of the Art Conclusion

20. WFIM vs. DIFSoN WFIM is an FP-tree based weighted frequent pattern mining algorithm that requires two database scans. Differences: WFIM uses a secondary support threshold to calculate weighted frequent patterns. 70 % Completed

21. Datasets IBM synthetic datasets  T10I4D100K (http://www.almaden.ibm.com/cs/quest or http://www.cs.loyola.edu/*cgiannel/assoc_gen.html) Real datasets  Mushroom (http://fimi.ua.ac.be/data)  Pumsb (http://fimi.ua.ac.be/data)  Kosarak (http://fimi.ua.ac.be/data) 75 % Completed

22. Runtime 80 % Completed

23. Compactness of the IF-tree 85 % Completed

24. Scalability of the DIFSoN 90 % Completed

25. Experimental Results Proposed Method Problem Description Introduction State of the Art Conclusion

26. DIFSoN comprises the IF-tree and a mining routine. Although the notion of influential friends does not satisfy the downward closure property, we addressed this issue using the global maximum prominence values of users. To enhance the model, we proposed to use the local maximum prominence values. 95 % Completed

27. 100 % Completed!!! Results show that: the IF-tree is compact and space efficient the tree-based mining routine within the DIFSoN model is fast and scalable for both sparse and dense data

28. Any Questions?

29. Thank You So much

30. Cameron JJ, Leung CKS, Tanbeer SK (2011) Finding strong groups of friends among friends in social networks. In: SCA 2011, pp 824–831 Jiang F, Leung CKS, Tanbeer SK (2012) Finding popular friends in social networks. In: SCA 2012, pp 501–508 Leung CKS, Medina IJM, Tanbeer SK (2013) Analyzing social networks to mine important friends. In: Social media mining and social network analysis: emerging research, pp 90–104