Group Recommendation Based On User And Group Feature

In the information age,the development of e-commerce is very rapid and has been widely popularized in daily life.With the rapid growth of network information,the problem of information overloads is becoming increasingly serious,and the recommendation system can effectively alleviate the problem.Therefore,the recommendation system has gradually become the focus of attention and research.Group recommendation,as an important branch of recommendation system,mainly deals with the recommendation processing when the target user changes from single user to multi-user.At present,most of the existing group recommendation methods are mostly limited to the aggregate users' preference itself,but ignore the particularity of group recommendation is different from single-user recommendation,ie,the characteristics of users and groups.Based on the full analysis of the existing group recommendation techniques,combined with real-life application scenarios and movie recommendation applications,this paper studies the group recommendation based on users and group characteristics.The main contributions are as follows:(1)As the low accuracy of the discrete attribute labels of items in traditional recommendation systems,we propose a method of continuousizing discrete attribute labels to make the recommendation system more accurate.The method first gets the expert in each attribute label domain,and then uses the domain expert's rating information to measure the attribute label of the item,thereby realizing the process of continuous items' discrete attributes.In addition,based on this method,we propose a recommendation algorithm DASC and a hybrid recommendation algorithm M-DASC to verify the validity of the continuous method.(2)Aiming at the lack of user interaction in the current group recommendation method,we propose an algorithm model of adding preference interaction in the rating prediction process of group recommendation.The model divides the predicted rating into two parts: self-prediction and preference interaction.Firstly,we analyze and design the preference interaction part according to the user interaction characteristics.Then,we propose a user rating prediction method based on the preference interaction framework.Finally,aiming at each user in the group,we propose a method of generating personalized interaction parameters through historical group activity and post-recommendation rating feedback mechanism to improve the real satisfaction of the group recommendation.(3)In order to realize a group recommendation system based on user and group characteristics and combining the above two methods and models,we extend the preference interactive group recommendation model and propose a comprehensive IMDG group recommendation algorithm.Firstly,we extend the model from the perspective of user classification.Then,we propose a preference interactive group recommendation framework according to the characteristics of the model.Finally,apply this framework and combine above methods,we propose and implement a comprehensive group recommendation algorithm IMDG,and the experimental results verify its effectiveness.