Research On Group Recommendation Algorithm Based On Review Analysis

Personalized recommendation system can provide accurate recommendation for individual users by analyzing users' behavior data such as rating and review in the network,learning users' preference characteristics.However,in real life,users often perform daily activities in different groups.How to recommend group has become a research hotspot in recent years.Most existing group recommendation algorithms recommend a group by fusing the preferences of members in the group.However,in most cases,we don't know the group information of users,and because the result of group detection directly affects the performance of group recommendation,so group detection is particularly important in the group recommendation algorithm.This paper analyzes the existing problems of the group recommendation algorithm from two aspects: group detection and group recommendation,and proposes the density peak clustering group detection algorithm based on GRU-CNN and the group recommendation algorithm based on the leader mechanism.(1)In the group detection stage,most of the existing group detection algorithms have the following problems: first,only using the static preference features of users to find the group structure in the network,ignoring the change of users' interest with time;second,using the users' subject features extracted from the reviews to divide the group,and it is difficult to mine the depth features of user review text.In view of the above problems,this paper proposes a density peak clustering group detection algorithm based on CNN-GRU.Firstly,the topic model is used to analyze the topic of user reviews,and then the representative topic words are extracted.Then,the time sequence information of review topics is modeled based on the gated recurrent neural network,and the dynamic topic features of users are extracted.At the same time,it fuses Convolutional neural network which is used to mine the deep features of review topics.Finally,based on the deep mixed features,the density peak clustering algorithm is used to group detection.The experimental results on real dataset show that the features mined by the fusion depth neural network model can effectively capture users' dynamic interest preferences,and the effect of group detection is better than the existing algorithms.(2)In the group recommendation stage,the existing group recommendation algorithm problems are as follows: firstly,the influence of the group structure on the calculation scale of the recommendation system is ignored;secondly,the influence of the authoritative users in the group on the members of the group is ignored.In view of the above problems,this paper proposes a group recommendation algorithm based on leader mechanism.Firstly,the influence of members in the group is modeled.Considering the interaction and activeness of members,the authoritative users in the group are selected as leaders,and then the group recommendation results are generated by leader decision-making.Because most leaders are highly professional,the recommendation results can satisfy the preferences of most members.Experiments on real dataset show that the accuracy of the proposed algorithm is significantly higher than that of the mainstream group recommendation algorithm.There are 33 figures,9 tables,and 83 references in this paper.