Research On User Similarity Measure And Recommendation Service Algorithm Based On Data Enhancement

With the rapid development of Internet and the explosive growth of data,personalized recommender system has gradually become the main tool to solve the problem of information overload.The recommendation algorithm based on collaborative filtering is undoubtedly one of the most popular and mature technologies in personalized recommendation system,but there are still many shortcomings in the current collaborative filtering recommendation.On the one hand,the rapid growth of data makes the data set become highly sparse,which hinders the determination of similar neighbors and the accuracy of user similarity measurement,and affects the effect of recommendation.On the other hand,the imperfection of similarity measure algorithm itself also leads to the inaccuracy of recommendation results.Therefore,in this paper,the above two problems are deeply studied and the corresponding improvements are proposed.Firstly,aiming at the problem of high data sparsity in traditional collaborative filtering algorithm,this paper proposes an improved Bias SVD matrix filling algorithm with time decay function.Bias SVD matrix decomposition model can effectively alleviate the sparsity of data by associating user items and item items with hidden features,but this model does not pay attention to the importance of scoring time,that is,the value of user's interest or behavior will decline with the change of time.This algorithm integrates the time decline function into the Bias SVD matrix decomposition model when filling,so as to achieve better filling effect.Secondly,considering that the traditional collaborative filtering recommendation algorithm based on matrix filling ignores the credibility difference between the real score and filling score,this paper proposes an improved similarity measure algorithm based on filling confidence.In the calculation of user similarity,the concepts of score filling confidence and item filling confidence are introduced to fully distinguish the reference value of real score and filling score.Thirdly,considering that the traditional user similarity measurement algorithm only depends on the common score value among users,and ignores the influence of the number of common score items and the relationship between similar user sequences on the similarity calculation,this paper proposes a similarity measurement algorithm based on the number of common score items for similar user sequences.The similarity between users can be represented by calculating the similarity of similar user sequences arranged in reverse order according to the number of common scoring items.Fourthly,in view of the problem that the traditional similarity computing model can not accurately calculate the user similarity without common scoring items,and ignore the importance of similar items scoring,this paper proposes similarity algorithm based on similar item scoring.The algorithm calculates the similarity of users' scoring items from two aspects of content similarity and scoring similarity,and sets a threshold to screen similar item groups when calculating user similarity,so as to fully consider the value of similar project scoring.Experimental results show that the improved algorithm proposed in this paper can not only effectively alleviate the problem of data sparsity,but also effectively improve the accuracy of user similarity measurement algorithm,so as to improve the recommendation accuracy.