Social Recommendations Based On Multi-relations Discovery And Analysis

Recommendation is the critical technology of social network analysis,data mining,information retrieval,etc,and has been widely applied in online emotion consultation,spam filtering,webpage ranking,etc.In addition,with the development of recommendation technologies and sophisticated,there are an increasing number of scenarios where recommendations are required for groups of people rather than individuals.Therefore,group recommendation has become increasingly important.The goal of this paper is to investigate the existing recommendation approaches,analyze the main problem,find out the deficiencies,and design the effective solutions.Meanwhile,in order to meet the new needs of society,we will try to describe user and group preference in a complete manner,and propose the novel recommendation model to compensate for the deficiencies of the existing recommender systems.Nowadays,many approaches have demonstrated that combining ratings with text information allows us to predict ratings more accurately than approaches that consider either of the data sources in isolation.Therefore,we propose a new approach which combines rating and text information of items,it can comprehensively detect the latent features of users and items,and describe the item features and user behaviours in a right way.First,we simultaneously learn items' latent features and users' interested topics by combining the items' ratings with their content information based on the probabilistic topic model;Then,on the basis of the topic model,we globally discover the latent relationships between users and items through random walk method to improve the prediction accuracy.Group members are society persons,their preference is closely related to the social relationships.In other words,taking group member relationships into recommendation model can help us better describe the group preference,and improve the performance of recommender systems.In this paper,we propose a new approach based on random walk method,which can globally discover the associations among groups,users,and items.In addition,on the basis of combining the two approaches described above together,we propose a new approach for group recommendations based on the combination of an integrated probabilistic topic model and the random walk with restart method.The topic model provides a latent framework of users,groups,and items by exploiting both the users' preference profiles and the items' content information,which together can describe group interests and item features in a more complete manner.This latent framework is then combined with random walk with restart to predict the preference degrees of groups to unrated items by detecting comprehensive latent relationships.In particular,we devised various group-based recommendation algorithms on the basis of different aggregation policies and recommendation strategies.Finally,we test all the approaches described above,and compare them with the state-of-the-art methods.The experiments indicate that our approaches can better resolve the problem and improve the performance of recommender systems.