Social networks are transforming the way people interact with each other. Not only do general-purpose social networks like Facebook, Linkedln, and Twitter shape day-to-day communications among people, but enterprises and organizations are also beginning to model internal communication among their members according to social paradigms. In this context, this paper explains how to build a recommendation system, or a system that assists its users in selecting people with whom to interact. The authors start from the now-familiar concept of "friendship," or relation between two people, and build their system in order to exploit two of its properties: triadic closure, or the probability that two people who are both friends with a third person will become friends, and homophily, or the probability that the more friends in common a group of people have, the more new friends this group will collectively attract. In the paper, the authors develop a three-step algorithm to test the degree of existence of these properties in a given network. The algorithm works roughly as follows: first, for each person in the network, it finds the most trusted people in his or her neighborhood based on network reputation; then, it checks the similarity of each couple of people using the Tversky index; and, finally, it combines the previous two findings to rank other people in the neighborhood of each person.
展开▼
