Research On Recommendation Algorithm Based On Collaborative Filtering

Today is the era of web2.0,not only the amount of data and data types are still exploding,but the number of netizens is also increasing substantially every year.The problem of information asymmetry is still a current hot issue,and research on this issue is also continuing to deepen.As one of the important means to solve this problem,the recommendation system is a hot field in both industrial and commercial applications and theoretical research.The recommendation algorithm is the core content of the recommendation system.At present,the collaborative filtering recommendation algorithm is the most widely used among the recommendation algorithms.However,this algorithm has serious data sparsity problems that affect the accuracy of score prediction.This paper improves on the two core steps of the collaborative filtering algorithm,using tag data to establish the connection between the user and the item,and reducing the data sparsity.The impact of improving the accuracy of scoring predictions.The research content of this paper includes the following two aspects:1.In order to solve a special data sparsity problem,that is,the user's cold start problem,the first step in the collaborative filtering algorithm—calculating the set of neighbor users has been improved.Based on the concept of model-based collaborative filtering algorithm,tag data is used as the attribute characteristics of users,and an algorithm combining tag data and naive Bayes classification algorithm is proposed.This paper uses the ideas of statistics and probability theory to establish the association between users and tags,thereby more accurately expressing user preference information.And the association between the user and the tag is combined with the Naive Bayes classification algorithm to realize the calculation of the neighbor users of the target user and the category matching of the new user.In addition,consider the two factors of tag scalability and temporal context information to further reduce the impact of data sparsity.Finally,an experiment is designed to verify that the proposed algorithm can improve the quality of the target user's neighbor user set by comparing the RMSE values of different algorithms.2.In order to further improve the accuracy of scoring prediction,for the second step in the collaborative filtering algorithm-scoring prediction,through the user-tag,tag-item relationship between the prediction method Make improvements.The tag data can be used to more accurately express the reasons for the user to score the project.The user-tag association and the tag-project association are used as two independent influencing factors,and after weighted sum,they are used as a weight factor,and then weighted into the traditional scoring prediction method.Finally,a control experiment is designed.RMSE,precision,recall and coverage are used as the evaluation standard,and the algorithm proposed in this paper is compared with other algorithms.Experiments verify that the improved scoring prediction method can effectively improve the accuracy of scoring prediction,and combine the above two research methods before conducting a comparative experiment.The results show that the combined scoring prediction value is more accurate.