Research And Application Of Group Recommendation System Based On Rank Aggregation

Recommendation system attempts to recommend items that are likely to be of interest to the user according to some reference characteristics of the user. These characteristics may be from the information item or the user's social environment. Recommendation system has become an important information filtering system to assist user to make a choice. There are many movies and restaurants recommendation systems. Recommendation systems have traditionally recommended items to individual users, but there has recently been a proliferation of recommendation that address their recommendations to groups of users. Recommendation system for individual is built to recommend items that meet individual user's interest. However, group recommendation system is a system for measuring individual interests as an aggregate towards collective decision. Group recommendation is a process of social choice like voting.This paper describes the framework and the key technology of group recommendation system. The group recommendation system needs to deal with aggregation rules and the relationship between members. This paper bring recommendation aggregation framework into our study. Recommendation aggregation firstly produces recommendation for each member and then aggregates all of the recommendations. This paper discusses the recommendation system for individual and focuses on Tangent algorithm. We build an experiment to evaluate the effectiveness of Tangent. Next, after the analysis of traditional aggregation rules, we bring rank aggregation (Medrank) into group recommendation system. We propose Weighted-Medrank to meet the need of social relationship between members. This paper gives a series of experiments on Medrank and Weighted-Medrank. Based on the theory above, aiming at restaurant applications, this paper designs and implements the GroupR canteens recommendation system. GroupR will collect ratings and comments. GroupR generates a recommendation with Tangent according the ratings one user. Users can form groups. GroupR provides three options: Medrank, Least Misery and Multiplicative Utilitarian to assist user to make the final decision.