Research On Collaborative Filtering Based On Merged Ratings And Multidimensional Neighbors

The rapid development of the Internet has brought great convenience to people,but also led to the rapid expansion of network information and caused the problem of information overload.The recommender system is widely researched and applied because it can solve the information overload problem and provide personalized service for users.Collaborative filtering(CF)is one of the most well-known and commonly used techniques in recommender systems.However,its performance has been seriously affected by the issue of data sparsity.Therefore,it has great significance on studying how to alleviate this problem to improve the performance of CF.Through a series of statistical experiments,we analyzed the influence of data sparsity on the performance of CF in detail.Aimed at the sparse data set,a corrected trust-aware rating filling algorithm(CTRF)is proposed which is based on the idea of trust-aware collaborative filtering.It uses the rating similarity of trusted-peer to correct the one-sided trust relationship,and then fills missing ratings by the ratings of trusted neighbors to obtain a merged rating matrix.In order to overcome the shortcomings of the existing neighbors selection algorithm,a dynamic multidimensional neighbors selection algorithm(DMNS)is proposed based on the idea that the ratings of dissimilar users also have the value of aggregation.The similarity relationship among users is divided into four dimensions.According to the distribution of user similarity and the number of neighbors,the multidimensional neighbors are constructed.Based on CTRF and combining with DMNS,we propose a collaborative filtering algorithm based on merged ratings and multidimensional neighbors(MMCF).Firstly,the data set is preprocessed by CTRF to get the merged rating matrix.Then the similarities between users are calculated based on the merged ratings.Next,the multidimensional neighbors are constructed by DMNS.Finally,the prediction is generated by aggregating the ratings of the multidimensional neighbors.Experimental results based on real-world data sets demonstrate that the MMCF algorithm outperforms other related recommendation algorithms in terms of coverage and accuracy,which means that applying the CTRF and DMNS algorithms to the collaborative filtering can alleviate the issue of data sparsity effectively.