An Improved Collaborative Filtering Recommendation Algorithm Based On Expert Trust

With the increase of network resources,the value density of information is getting lower and lower.Personalized recommendation algorithms can actively recommend information that may be of interest to users,changing the situation where people are at a loss before massive amounts of information.Collaborative filtering recommendation algorithm is one of the most researched personalized recommendation algorithms.Due to sparse scoring data,cold start,and low scalability,the algorithm's recommendation accuracy has not been greatly improved.Introducing expert trust can well reduce the effects of cold start and data sparseness and combining clustering algorithms can improve scalability.In this paper,starting from improving the recommendation accuracy,we mainly do the following two points.(1)Introducing the trust of hetero-community cluster experts in the score prediction formula,and put forward a recommendation algorithm,which combing the trust of hetero-community cluster experts,of collaborative filtering.Existing collaborative filtering recommendation algorithms that introduce expert trust only consider the influence of the trust of experts in the same community cluster and the similarity of similar users when scoring predictions,and ignore the influence of the trust of experts in different community clusters,making the recommendation accuracy insufficient.In the improved algorithm,the comprehensive similarity is first calculated by integrating the average scoring factor,popularity,and complementary Jaccard and Pearson correlation coefficient.Users are clustered into various community clusters according to the comprehensive similarity value,and the target users.The community cluster with the highest similarity in the community cluster is defined as a hetero-community cluster.Then calculate the trust of all users in each community cluster,and define a part of users with high trust as experts in the community cluster.Finally,a score prediction formula that combines the trust of experts in the same community cluster,the similarity of similar users in the same community cluster,and the trust of experts in the different community cluster is used to make score prediction and recommendation.Experiments indicate that the accuracy of the recommendation algorithm,which improved,is enhanced.(2)Introduce user similarity and implicit preference similarity in the similarity calculation formula,and put forward an expert trust's recommendation algorithm,which improved the similarity,of collaborative filtering.Existing collaborative filtering recommendation algorithms that introduce expert trust do not fully consider user characteristics and user preferences implicit in the same score when calculating similarity,making the recommendation accuracy not high enough.In the improved algorithm,first,the user's feature similarity,implicit preference similarity,and complementary Jaccard and Pearson correlation coefficient are calculated,and the users are clustered into each community cluster according to the comprehensive similarity value.Then,the trust degree of all users in each community cluster is calculated,and a part of users with high trust degree are defined as experts in the community cluster.Finally,a score prediction formula that combines the trust of experts in the same community cluster and the similarity of similar users in the same community cluster is used to make score predictions and recommendations.Experiments indicate that the coverage and accuracy of the recommendation algorithm,which improved,is enhanced.