Research And Design Of Group Recommendation System Based On Attention And Latent Relationship

With the rapid development of the information age,the exponential growth of massive information data leads to the problem of information overload.As one of the important ways to alleviate the problem of information overload,recommendation system can provide users with the services they are interested in according to their preferences.However,at present most of the recommendation system research is aimed at a single user personalized recommendation,with the continuous development of social network,in group activities,such as company team building,team travel,etc.,but the personalized recommendation is difficult to meet the needs of in group activities,therefore group recommendation system arises at the historic moment,can be of interest to the service for the group,It has become one of the hot spots in current research.In the group recommendation system,most of the existing group recommendation systems are based on the inherent group recommendation,it is difficult to mine the user’s potential preference,capture the user’s potential similarity relationship to discover the group.In addition,the dynamic complexity of the group needs to be considered during group recommendation.User preferences will affect each other and ultimately affect the result of group recommendation.Therefore,this thesis mainly describes the technology and method of mining users’ latent preferences to discover latent groups and how to alleviate the dynamic complexity of groups,and designs a group recommendation system based on the relationship between attention and latent.The main work is as follows:(1)From the perspective of group discovery,aiming at the potential relationship between users,a group discovery method combining attention mechanism and knowledge graph is proposed.The method by user interaction with the history of the project data to build knowledge map and knowledge map and attention mechanism is used to mining of high order potential relationships between entities in the information obtained through polymerization potential user preferences,and then using the method of clustering to divide to form potential group of user preferences,Finally,the relevant methods are analyzed experimentally.(2)From the perspective of group recommendation,aiming at the problem of dynamic complexity in group recommendation,a group recommendation algorithm integrating contextual information and self-attention mechanism is proposed.The method uses the context information of the group members to learn the interaction between the members through the self-attention mechanism,so as to alleviate the dynamic complexity of the group,and the group view as a user to make personalized recommendation.Finally,the proposed group recommendation algorithm is compared with the benchmark algorithm.The results show that the proposed method improves the two evaluation indexes of recommendation hit ratio HR and NDCG compared with the benchmark algorithm,effectively alleviates the dynamic complexity problem in the group,and can obtain better group recommendation results.(3)From the perspective of group system,this thesis designs a group system based on the relationship between attention and potential based on the above methods.This system is a movie recommendation system based on Movielens dataset,and use Python to complete the establishment of system interaction.