Research On Group Recommender System Incorporating Social Trust Relationships

In today's society,people are in the fast-changing information age,and they have more and more ways to access information.The resulting huge amount of information makes it increasingly difficult for people to find what they are really interested in,and this kind of "information overload" problem has begun to plague people.As a result,the recommender system was born.It helps people process and analyze the huge amount of relevant information to get the most suitable information for their needs and provide personalized recommendation results.However,with the development of the Internet and social media,there is an increasing demand for group recommendations for multiple users,such as playing the right music for a client in a gym or recommending the right attractions for a tour group.This requires expanding the traditional recommender system from a single user to a multi-user group.A group recommender system generally consists of two parts: fitting the preferences of a single user and fusing the preferences of users in a group to obtain group preferences.In the traditional recommender system,the probability matrix decomposition algorithm adds the a priori information of user and item feature vectors on the basis of matrix decomposition,assuming that the user potential hidden vector and item potential hidden vector obey Gaussian distribution,and obtains more accurate potential hidden vectors by maximizing their posterior probabilities.In recent years,deep learning has also started to enter the field of recommender systems and has quickly gained a place in the field with its excellent expressive power and feature pattern mining ability.Group preference fusion is usually divided into preference fusion and recommendation result fusion.Preference fusion refers to the fusion of member preferences into group preferences after obtaining the preferences of all members in a group,and using the group preferences to make recommendations.Recommendation result fusion is to obtain the recommendation results of all members in a group and merge them into the group recommendation results according to a certain strategy.The previous group recommender algorithms do not take into account the interaction between users,which makes the utilization of social relationship information low,but the interaction between group members often has a great influence on the recommender results.To address this problem,the following research is conducted in this paper:(1)A probability matrix decomposition group recommender system that incorporates user trustworthiness is proposed.The method first uses the probability matrix decomposition algorithm to calculate users' trustworthiness,then improves the posterior probability derivation of the probability matrix decomposition,adds the trustworthiness information among users,and maximizes the posterior probability to obtain the prediction score.Finally,a weighting strategy based on user trustworthiness is used in the preference fusion process of group members.Experiments based on the Epinions dataset and the Film Trust dataset show that this method outperforms other comparative methods such as Neu MF and Ripple Net on two evaluation metrics,root mean square error and hit rate.(2)A deep learning method based TFM model is proposed,which uses a multilayer neural network to add a Multi-Hot vector representing the trust data between users in the model input.The user implicit feature vector is spliced with the item implicit feature vector and trust vector and sent into the deep network,so that the user features are fully crossed with the item features and trust features,and the prediction score of the item is obtained in the final model output layer.In the group fusion part,a weighting strategy based on user trust is used.It has a better recommender effect compared to the comparison methods such as Neu MF and MLP.(3)This paper designs and implements a group recommender system platform based on Java Web.On the basis of deep learning model,the user social system in the platform is implemented to obtain the trust between users,and good recommendation effect is obtained.