Research On Recommender Algorithm Based On Multi Information Fusion And Social Trust Degree

The recommender system is able to mine and capture user's preferences and recommend relevant content for users.It is a bridge between users and projects,and plays a very important role in people's lives.At present,the recommendation system has been rapidly developed in e-commerce,movies,social networking,etc.In the field of social networking,in recent years due to the gradual increase in the size of users,the rate of update of information has become more frequent,and the explosion of network data has caused information overload problems.For a long time,academia and industry have made many efforts to accurately capture the needs or preferences of users and filter out useless or uninteresting content for users.Collaborative filtering is a mainstream method in the field of personalized recommendation.By finding the similarities between different individuals,the most similar individuals are selected to meet the individual needs of different users.Although collaborative filtering can effectively alleviate information overload and is widely used,most existing collaborative filtering-based recommendation methods are not satisfactory in the face of sparse data and cold-start problems.In addition,the characteristics of the social network make it inherently complex.The traditional recommendation method is not ideal for the personalized recommendation of the user in the social network environment.On the basis of thorough analysis of common recommendation algorithms,this paper attempts to combine the idea of multi-source information fusion on personalized collaborative filtering algorithms to help better understand the user's interests;it also uses common user-item rating data in collaborative filtering algorithms.Combining the idea of matrix decomposition and using the social network trust model to improve data sparseness and improve the prediction accuracy of the recommendation system.The main research work of this article is as follows:First of all,this paper introduces the basic ideas,classification and comparison of social network recommender system,and expounds the common recommendation evaluation indicators in detail.For two types of collaborative filtering algorithms: a collaborative filtering algorithm based on neighborhoods and a collaborative filtering algorithm based on models are studied in depth.Analyze and compare the implementation principles of different recommender algorithms,and at the same time point out its defects or deficiencies.Secondly,for the situation that the classic user-based collaborative filtering algorithm makes use of rating information to result in poor recommendation results,a personalized recommendation algorithm based on multi-source information fusion is proposed,which can reflect user needs or preferences in an integrated social network environment.The multi-source information improves the prediction accuracy of the recommendation results and helps to alleviate the cold start problems common to recommendation systems.The algorithm is evaluated using the real data sets Movielens and Last.fm,and the prediction accuracy of the Top N recommendation results is measured and compared with some conventional collaborative filtering algorithms.At the same time tested the performance of the cold start problem.Finally,considering the low performance of the traditional collaborative filtering algorithm in the environment containing social network information,the idea of matrix decomposition and the integration of social network trust information are proposed to propose an SVD collaborative filtering algorithm based on social trust.The algorithm constructs the user's trust matrix first,adds the bias of the trustee and the trusted person,improves the trust model,and reduces the data dimension.Combined with the user's rating information on the item,the accuracy of the algorithm's prediction is improved.The experiment is based on the real data set Film Trust,and the proposed new algorithm is evaluated.The prediction accuracy of the score prediction results is measured,and the experimental results are compared and analyzed to verify the effectiveness of the new algorithm.