sides could be made by the nature of the groups, and classified the but the amount of over than the overlap between random pairs, We also considered that two kinds of users on the both extreme was significantly la groups into three categories- one having large portion of perfect similarity between the group and the matched users, another one having large portion of non-matched We saw that the little overlap between group members compared sers, and the last one having relatively equal portion of these two with the large overlap between the groups and the members may xtremes. We compared the number of members and the number be caused by the interface problem of Citeulike. According to the of items in group collection for these three categories of groups result about the group and group members infomation sharin and failed to find any significant results. patterms, there were two kinds of users -one kind was the people who take advantage of the group's information to the large extent sItem and another kind was the people who just neglect the group information and try the information seeking strategy of their own meta with very specific information needs. 1000 As the future direction, it is necessary to examine the timely change of information similarity and dynamics of memberships Using different kinds of social networks such as friendships or groups or the similarity of a group to the members. In order to reinforce our findings in this study, we plan to add different data sets as well 7. REFERENCES Group Fractions [1] Backstrom, L, D. Huttenlocher, et al.(2006) Group formation in large social networks: membership, growth, and Figure 10. Group Fractions of Shared Items and Metadata volution. Proceedings of the 12th ACM SIGKDD from Groups'Point of view) Knowledge discovery and data mining, Philadelphia, PA, USA. [2] Guy, L, N. Zwerdling, et al. (2009) Personalized recommendation of social software items based on social ≌1400 ■ macro relations. Proceedings of the third ACM conference on 1200 Recommender systems. New York. New York, USA. [3 Hotho, A, R. Jaschke, et al. (2006). Information Retrieval in Folksonomies. Search an 4 Hung, C.-C, Y.-C. Huang, et al.(2008). Tag-based User Profiling for Social Media Recommendation. Workshop on Intelligent Techniques for We Recommender Systems, AAAl 2008. Tools(D). D-Lib Magazine 11(4): 1-1 [61 OHara, K. H. Alani, et al. (2002) Identifying Communities of Practice: Analysing Ontologies as Networks to Support e ll. Group Fractions of Shared Items and Metadata Groups’ Point of view) Challenge. Montreal Canada last analysis, the view point of groups'side was taken into [7 Zhou, D, E. Manavoglu, et al. (2006)Probabilistic models account.16.35% of items and 18.98% of metadata in group's for discovering e-communities. Proceedings of the 15th collection are overlapped with the members'personal collection international conference on World Wide Web, Edinburgh, on average. In addition, the number of group members whose Scotland. personal collection contains all the items of the group(100% group fraction) is just 64. Put differently, this 100% overlap of group collection means, for example, if a group has 50 items in the group collection and one of the members, user'A' has the all 50 items in his collection. 469. 22 members have more than 50% of the group collection. aforementioned. the information space is larger than the members hence the portion of overlapped information in groups much smaller than the portion of the overlap in members'spaces 6. CONCLUSION AND DISCUSSION In this paper, we explored how much the information of my group d the information of my group members are similar with mine We found that the information overlap between group membersWe also considered that two kinds of users on the both extreme sides could be made by the nature of the groups, and classified the groups into three categories – one having large portion of perfect matched users, another one having large portion of non-matched users, and the last one having relatively equal portion of these two extremes. We compared the number of members and the number of items in group collection for these three categories of groups and failed to find any significant results. Figure 10. Group Fractions of Shared Items and Metadata (from Groups’ Point of View) Figure 11. Group Fractions of Shared Items and Metadata (from Groups’ Point of View) As the last analysis, the view point of groups’ side was taken into account. 16.35% of items and 18.98% of metadata in group’s collection are overlapped with the members’ personal collection on average. In addition, the number of group members whose personal collection contains all the items of the group (100% group fraction) is just 64. Put differently, this 100% overlap of group collection means, for example, if a group has 50 items in the group collection and one of the members, user ‘A’ has the all 50 items in his collection. 469.22 members have more than 50% of the group collection. As aforementioned, the groups’ information space is larger than the members’ personal spaces; hence the portion of overlapped information in groups’ spaces is much smaller than the portion of the overlap in members’ spaces. 6. CONCLUSION AND DISCUSSION In this paper, we explored how much the information of my group and the information of my group members are similar with mine. We found that the information overlap between group members was significantly larger than the overlap between random pairs, but the amount of overlap was small. However, the information similarity between the group and the members were quite large. We saw that the little overlap between group members compared with the large overlap between the groups and the members may be caused by the interface problem of Citeulike. According to the result about the group and group member’s information sharing patterns, there were two kinds of users – one kind was the people who take advantage of the group’s information to the large extent and another kind was the people who just neglect the group information and try the information seeking strategy of their own with very specific information needs. As the future direction, it is necessary to examine the timely change of information similarity and dynamics of memberships. Using different kinds of social networks such as friendships or unilateral relationships (as ‘following’ in twitter), it may be possible to see how personal similarities flows to the formation of groups or the similarity of a group to the members. In order to reinforce our findings in this study, we plan to add different data sets as well. 7. REFERENCES [1] Backstrom, L., D. Huttenlocher, et al. (2006) Group formation in large social networks: membership, growth, and evolution. Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining, Philadelphia, PA, USA. [2] Guy, I., N. Zwerdling, et al. (2009) Personalized recommendation of social software items based on social relations. Proceedings of the third ACM conference on Recommender systems, New York, New York, USA. [3] Hotho, A., R. Jäschke, et al. (2006). Information Retrieval in Folksonomies: Search and Ranking: 411-426. [4] Hung, C.-C., Y.-C. Huang, et al. (2008). Tag-based User Profiling for Social Media Recommendation. Workshop on Intelligent Techniques for Web Personalization and Recommender Systems, AAAI 2008. [5] Lund, B., T. Hammond, et al. (2005). Social Bookmarking Tools (II). D-Lib Magazine 11(4): 1-1. [6] O'Hara, K., H. Alani, et al. (2002) Identifying Communities of Practice: Analysing Ontologies as Networks to Support Community Recognition. In Proceedings IFIP World Computer Congress. Information Systems: The E-Business Challenge., Montreal, Canada. [7] Zhou, D., E. Manavoglu, et al. (2006) Probabilistic models for discovering e-communities. Proceedings of the 15th international conference on World Wide Web, Edinburgh, Scotland. 0 200 400 600 800 1000 1200 1400 1600 1800 No. of Members Group Fractions item meta 0 200 400 600 800 1000 1200 1400 1600 1800 No. of Members Group Fractions micro macro