WIR-A Graph-Based Algorithm for Friend Recommendation

Abstract

Content-based [6] and collaborative-based methods [7] [8] [9] are two main underlying techniques used in recommendation system. Some researchers proposed hybrid recommendation method to compromise the advantages and disadvantages of content-based and collaborative recommendation approaches [4] [10]. These three approaches share the common goal of assisting in the user’s search for items of interest not recommending themselves to other users. For friend recommendation, we proposed a graph-based method, in this research, named Weighted Information Ratio (WIR) borrowed idea from information theory [1] [2]. We compare the method WIR with our prior algorithm in friend recommendation named Minimum-message Ratio (WMR) [3]. The precision and recall of WMR method are 15% and 8% for a target member with 15 recommendations in the testing prediction, respectively. This result is acceptable compared to a hybrid recommender system for digital library [4], where the testing precision and recall are 3% and 14% for 100 customers. Both recommendation algorithms generate a limited, ordered and personalized friend lists by the real communication number among web users. Communication number is more representative than most of the population variables because they are lack of diversity [5].