A Group Recommendation Method Based On Optimized Collaborative Filtering And The Weighted Average

Nowadays, recommendation systems have been a powerful tool to address information overload issues. In recent few decades, recommendation systems mainly offer personalized services for individuals and work successfully. However, with the rapid development of information technology and social networks, consumers tend to form groups to participate in activities. Users in the same group often have similar preference, they participate in group activities and be affected by others. Personalized recommendation systems for individuals have been unable to recommend items to groups. As the sizes of groups in social networks and e-commerce sites increase, building recommendation systems and offering personalized services to group users, so as to satisfy the needs of improving information searching efficiency as well as saving searching time and effort, have been significantly important.In this paper, a novel group recommendation technique based on optimized collaborative filtering algorithm and the weighted average method was proposed. The method includes hybrid collaborative filtering algorithm and user rating aggregation algorithm. The hybrid collaborative filtering algorithm integrated user-based and item-based rates predicting method, as well as taking items’rating similarities and type similarities into consideration to calculate similarities for items, and then the unrated items were imputed with values. When integrating rates of one certain group users, the aggregation algorithm first uses the median score of all group users to calculate the credibility of users, and then uses the credibility of the users to fix their rating frequency weight. It considered the influence of the whole group users so as to eliminate the effects caused by some individuals whose prediction scores were largely in differ with others.In this paper, Movielens film score data set was used, the impact of item type similarity weight and user rating similarity weight was explored. In the end, the hybrid collaborative filtering algorithm was compared with traditional algorithms; the aggregation strategy based on weighted average was compared with five classic aggregation strategies. The results show that two algorithms are better than classical ones in terms of accuracy, which indicates that our strategy is valid.