| With the rapid development of Internet services and the rapid rise of online social networks,it has brought great convenience to Internet users,but the resulting network services are growing exponentially.Information overload has caused Internet users to be unable to quickly obtain the desired content,which has brought great impact on our online access.Therefore,how to quickly and efficiently find the content users want in massive network data has become one of the important research topics for scholars today.Recommendation systems,as an important means to alleviate the current information explosion problem,can provide effective information to users.Currently,traditional recommendation algorithms mainly rely on collaborative filtering methods.Collaborative filtering recommendation methods are based on user historical behavior and project attribute information to achieve personalized recommendations for users,with good applicability and independence.However,the sparsity of raw data and the existence of user cold start problems exist,"This has a serious impact on the accuracy of the recommendation system.When there is very little scoring data for users in the system,it is difficult to calculate the similarity between users,and collaborative filtering cannot accurately provide personalized recommendations for users.In addition,for cold start users,due to the small amount of their own scoring data,collaborative filtering is also difficult to provide accurate recommendations.".Existing research has shown that fusing trust information can alleviate the problems of data sparsity and cold start to some extent.Combining trust information with collaborative filtering recommendation algorithms can achieve good recommendation results.In existing collaborative filtering recommendation algorithms that fuse trust relationships,explicit trust and positive trust are often used to filter the set of trusted friends,but the user trust matrix is also sparse,Considering only explicit trust and positive trust relationships is still difficult to alleviate the problem of sparse trust matrices,and the user coverage rate of trust relationships is low.Therefore,this article considers fusing implicit trust information and untrusted information to alleviate the impact of sparse raw data.In addition,for the problem of user cold start,we introduce expert trusted users as recommended neighbors to alleviate the problem of user cold start.Based on the above problem analysis,This article has conducted the following two aspects of research:(1)Aiming at the problem of data sparsity faced by traditional collaborative filtering recommendation algorithms,this paper fuses trust relationships between users to alleviate the problem of raw data sparsity;For the problem of sparse trust matrix,the implicit trust relationship between users is mined through the data information of users’ project ratings.The implicit trust relationship between users is used to replace the similarity relationship in traditional collaborative filtering recommendation algorithms,effectively alleviating the problem of sparse original trust information;In addition to considering the positive trust relationship between users,it also integrates the distrust relationship between users,and adds the impact of untrusted users on the predicted score,making the score prediction more objective and accurate.Finally,the matrix decomposition algorithm based on explicit trust and mistrust relationships is improved to integrate implicit trust and mistrust relationships,eliminate the negative impact of untrusted users on target user rating prediction,and provide personalized recommendations for users.Experiments are conducted on the dataset Epinions to verify the effectiveness of the algorithm in this paper.(2)Considering the trust factors of ordinary users towards expert users,expert user recommendations tend to be more convincing to users.Based on this,expert user trust factors are introduced.Define the selection principles of expert users from the perspective of user rating attitude and user trust;For the accuracy of prediction scoring,in addition to considering the user’s own preferences,we also consider the impact of trusting users on the target user’s potential feature vector,adding trust among users to adjust the target user’s potential feature vector.In addition,in combination with the impact of expert users on scoring,we adjust the proportion of expert users’ and user’s own preferences on the prediction scoring to improve the accuracy of the final scoring prediction.Improved on the Bias SVD algorithm,integrating the impact of experts’ trusted users on ratings and the impact of trusted users on the potential feature vectors of target users to perform rating prediction.The algorithm performance evaluation and cold start user experiment evaluation were conducted on the dataset Flim Trust and Ciao,verifying that the algorithm in this paper has good effects on alleviating cold start user problems. |
