The Research Of Group Recommendation Model Based On User Implicit Feature

The recommendation system can match items for users by satisfying users' preferences intelligently,quickly and conveniently,and it has important application value.Online applications make it easy for people to build various user groups,and with the enrichment of data,how to provide accurate recommendation services to user groups has become a challenge.Group recommendation needs to take the differences of users' preferences into account.The way of predicting single user preferences and modeling user group preferences has an important impact on the quality of recommendation services.Among them,user preference aggregating is the main method for modeling group preferences,for example,using a predefined preference aggregating strategy to obtain group preferences,assisting modeling group preferences based on user explicit features,and so on.In the above method,the member preferences are merged in a static manner,the influence of the group members' personal characteristics on the group decision-making is not fully considered,which has a certain influence on the group recommendation effect;or due to the facts that group members have different preferences and the user data does not fully express the user information or characteristics,which affect the recommendation performance.In summary,how to use the available data to establish a suitable preference model for the user group is the key to the group recommendation service.This paper proposes two novel preference aggregating strategies for group preference modeling.Based on the user's historical behavior,the weights of the group members which represent the users' influence can be calculated,and the preferences model can be reconstructed based on the users' weights to make recommendations for the group.The feature of the above method is to use only the interaction data between the users and the items to mine the implicit features of the users and to assist in building group preferences accurately.In order to further improve the satisfaction of the group members with the recommendation results,this paper designs an attention network-based user preference prediction model,which incorporates category factors and simultaneously models users' preferences for items and users' preferences for categories.This model is characterized by improving the prediction model of the preference of a single user,which improves the accuracy of the users' predicted ratings and then promotes the quality of group recommendations.This paper evaluates the proposed method from the two aspects including the accuracy of the recommendation results and the relevance of the ranking,and compares it with the benchmark method on the real data set.The experimental results confirm that the proposed method has better recommendation performance.