Social Network Based Friend Recommender System

Abstract

Earlier, we make friendship with our neighbors, colleagues based on geographical area. This is the traditional method of making friends. With the evolution in Internet, a social network comes in existence for connecting with distant people and friends for communicating with them. Existing social network uses social graph and pre-existing relationship between users for recommending friends to user. Such as Facebook uses mutual friends that is friends of friend for recommending friends. This may not be most appropriate method for recommending friends and selection of those by user in real life. We are presenting Friendbook, a social network based friend recommender system, which is based on semantic-based friend recommendation for friend recommendation. Friendbook recommends friends based on users life-style not on social graph. Friendbook discovers the life-style of user, using the user centric data, by taking the advantage sensor rich smart-phones. We model user’s daily life as a life document and extract his/her daily activities inspired by text mining through life document by using Latent Dirichlet Allocation Algorithm. We proposed similarity metric to measure the similarity of life styles between users. Friend matching graph is constructed based on impact ranking which is calculated in terms of users’ life style. Friendbook returns a list of people with highest recommendation scores to query user. We also integrate feedback mechanism with Friendbook to improve the accuracy of recommendation. The result reflects recommendations preferences of users in choosing friends accurately.