Finding Social Network for Trust Calculation Yutaka Matsuo, Hironori Tomobe, Koiti Hasida and Mitsuru Ishizuka National Institute of Advance Industrial Science and Technology (AIST) J University of Nagoya, Japan AIST, Japan University of Tokyo, Japan ECAI 2004

2 Outline Abstract Introduction Social Network Extraction Invention of Nodes and Edges Extraction of Edge Label Example and Evaluation Trust Calculation Social Trust Individual Trust Related Works and Conclusion

3 Abstract Trust is a necessary concept to realize the Semantic Web. But how can we build a “Web of Trust”? Small “Web of Trust” => A huge “Web of Trust.” Focus on an academic community : as a “microcosm” of a “Web of Trust” to generate a social network automatically. Each edge is given a label Coauthor, Lab, Proj, Conf.

4 Introduction Based on the trust network, the computer can decide how trustworthy persons, resources, and pieces of information are. At the beginning : A person or an organization will trust some acquaintances. A trust network appears locally and grows gradually by adding new nodes and edges. According to social scientists : A person can name 200 to 5000 people Relations are dynamic New relations appear every day and old relations weaken gradually.

5 Introduction Aspects of Knowledge Transfer Current Study Structural strong vs. weak ties Relational trust Knowledge tacit vs. explicit Hansen, 1999 Tsai & Ghoshal, 1998 Mayer et al., 1995 Zand, 1972 Zaheer et al., 1998 Nonaka, 1994 Polanyi, 1966 Zander & Kogut, 1995 Szulanski, 1996 Krackhardt, 1992 Ghoshal et al., 1994 Granovetter, 1973

6 Introduction Berners-Lee : Layer Cake metadata, ontologies, rules, proofs,

7 Social Network Extraction An academic society retains member profiles name, affiliation, qualification,contact address … Rregular annual conference: JSAI99, JSAI2000, JSAI2001, and JSAI people Choose 150 members to illustrate network Edge label : Coauthor: Coauthors of a technical paper Lab: Members of the same laboratory or research institute Proj: Members of the same project or committee Conf: Participants of the same conference or workshop

8 Social Network Extraction

9 For example ‘Yutaka Matsuo” (denoted X) “Hironori Tomobe” (denoted Y) query “X and Y” to get a documents query “X or Y” to get b documents “X and (A or B or...)”.. “Y and (A or B or...)”

10 Social Network Extraction Edge Label: Retrieved by the query “X and Y” and get 3 pages. First checked 275 pages manually and assigned labels to each page. manually-selected word groups to characterize pages

11 Social Network Extraction C4.5

12 Social Network Extraction

13 Trust Calculation PageRank-like model to measure authoritativeness of each member. v : member number v = 1509 n : iterations number set n=1000 Neighbor(v) : set of nodes each of which is connected to node v c : constant for normalization E(v) : uniform over all nodes

14 Trust Calculation

15 Trust Calculation Individual Trust n=300, Vtarget = Yutaka Matsuo

16 Relate Works and Conclusion First extract a list of members in the community, and try to determine their social network. Used the contents of the retrieved documents to classify the relation into four categories. Dan Brickley and Libby Miller invented an RDF vocabulary called FOAF (Friend-of-a-Friend) to create a social network. In this paper, we argue how local trust networks will finally constitute a huge “Web of Trust.”