A Group Recommendation Algorithm Based on Group Members' Weights and Project Popularity
Min Li, Chunming Wu, Ye Li · 2017
With the rapid development of many types of social networks, the number of people involved in group activities is increasing, and thus group recommendation systems have been widely studied.This paper proposes a collaborative filtering algorithm to mine groups' interests.In view of the different roles played by different users in the group, the algorithm combines the members' weights factor.And because of the differences in the degree of concern among groups and members for the various categories, the weighted mean similarity method is used.In order to solve the problem of a groups' narrow field of view of the existing algorithms and to more effectively mine unpopular items, the algorithm considers project popularity to improve the novelty of the recommendations.Experiments on the MovieLens 1M GroupLens data set show that the algorithm can effectively improve the accuracy of the recommendations.And to a certain extent, it improves the novelty of the ecommendations. Related workCollaborative filtering.The collaborative filtering recommendation algorithm was the first proposed algorithm that provided the most thorough research in the recommendation system.According to whether to join the model, the collaborative filtering recommendation algorithm can be divided into two categories: heuristic (also known as memory-based) and model-based.According to the purpose and function, it can be divided into two categories: score prediction and TopN recommendation.Among them, the most classic memorybased collaborative filtering algorithm mainly includes project-based [11] and user-based collaborative filtering.Group user recommendation.The group recommendation algorithm integrates the preferences of all the members of the group and provides the recommended service for the group.Over the past few years, some well-known group recommendation systems have been developed and applied in practice [5,6,7] .However, more accurate personalized recommendation service can't be provided.More recently, many scholars have conducted in-depth research on how to improve the effect of group recommendation, and put forward many related algorithms.Chao et al. [8] found the nearest neighbors of the group by computing groups and users of a similarity matrix, so as to predict the behavior of the group according to 2nd Joint