Research On Hybrid Collaborative Filtering Algorithm Based On Weight Adjustment And User Preference

The development of Internet has made it convenient for people to access information,But the subsequent explosive amount of information makes how to let users extract useful information from massive information and improve the effective utilization of information become urgent problems to be solved.The emergence of recommendation system is a quite promising direction for solving these problems.The recommendation system is a bridge between users and information,and it has been applied to create value for users and information in many fields.The success of recommendation system is inseparable from the use of recommendation algorithm.Collaborative Filtering,which is applied relatively frequently now,has become the most commonly used algorithm in Internet recommendation since its inception.Scholars have done a lot of research on collaborative filtering algorithms,but the problems of data sparseness and scalability have not been well solved yet.These problems will become more and more prominent as the amount of data increases.In order to alleviate the problems of data sparseness and inaccurate similarity calculation,this paper has made the following improvements for collaborative filtering algorithm.The specific work is as follows.First,The weighted Slope One algorithm is added to the scoring prediction process to obtain a hybrid collaborative filtering algorithm based on weighted Slope One.The weighted Slope One algorithm is used to obtain project-based scoring predictions which fill projects without scores for target projects in the user's neighbor set,in order to alleviate the sparsity of the user's scoring matrix.This paper verifies the effectiveness of the algorithm based on experiments.Second,The traditional similarity calculation only considers the original user's project scoring information,making the calculation of similarity inaccurate.So ahybrid similarity based on weight adjustment and user preference is proposed.First,the minimum weight of anti-user frequency and scoring behavior are integrated to reduce the overall impact of popular items and the accuracy deviation owing to little common scores.Then the project preference word matrix is introduced to mine users with the same preferences on project characteristic.Finally,the similarity calculations of the two items are combined through parameters to calculate the similarity between users,improving the similarity calculation accuracy.Weighted Slope One prediction filing and the similarity improvement algorithm are applied in the MovieLens dataset,Through experimental comparison and analysis,the improved algorithm has better MAE and RMSE values,and has higher accuracy and can improve recommendation effect.