Research On Group Recommendation Based On Potential Relationship

With the development of Information Technology,emergence of a large number of web services leads to the information overload problems.Recommendation systems have become one of the important mitigation methods,and effective recommendation schemes can help users to quickly select the services they are interested in.In real life,there are various activities in the form of groups.Traditional personalization recommendation systems can not meet a group of people's needs,so group recommendation has attracted more and more attention.In group recommendation systems,the interaction information between users and services contains various potential relationships,such as the association relationship between different services,the similar relationship between group members and the historical interaction relationship between groups and services.Due to the influence of these potential relationships,group members' preferences are difficult to obtain accurately,resulting in poor recommendation performance.Therefore,how to study the potential relationship between groups and services and carry out group recommendation system are the main research work of this thesis.The main contributions are as follows:Firstly,this thesis proposes a group discovery method based on knowledge graphs' association relationship.In this method,attention neural network is used to train a knowledge graph for extraction of the potential association relationship between different service entities.The collaborative filtering method and Word2 vec technology are used to obtain users' preferences,and a two-stage clustering method is used to discover potential groups.Secondly,this thesis proposes a group recommendation method based on group members' similar relationship.This method uses self attention mechanism to study the similarity relationship among all members in fixed groups for the weighted integration of these members' preferences into the overall group's preferences,and then the common vector space of group's preferences and service characteristics are trained by the Collaborative Metric Learning method for group recommendation.Finally,according to the above group discovery and group recommendation methods,this thesis constructs a prototype system of group recommendation based on potential relationships.The prototype system is a movie recommendation system for a group based on Movie Lens datasets.Its construction process includes requirement analysis,overall design,detailed design and implementation display.The functions of personal recommendation,group discovery and group recommendation are realized in the system.Meanwhile,it verifies the feasibility of the proposed methods and shows recommendation performances of these methods for potential relationships in practical applications.