Research On The Latent Factor Model Recommendation Algorithm Integrated Users’ Social Relationships And Items’ Associated Relationships

By machine learning,the latent factor model recommendation algorithm maps the sparse users’ rating information into two dense attribute feature matrices,improving the recommendation performance in the case of sparse data,but the algorithm ignores the potential relationships of users,the potential connections of items and the interpretation of recommendation results.Based on the latent factor model algorithm,the thesis’ s research topic is devoted to improving the accuracy of the estimated ratings and giving the explanation of recommendation results by using users’ social relationships and items’ associated relationships expressed by the rating dataset.The main works of the thesis are as follows:(1)Based on the latent factor model algorithm,the latent factor model recommendation algorithm integrated users’ social relationships and items’ associated relationships is proposed to improve the accuracy and the interpretation of the algorithm,as well as the applicability under extreme conditions.(2)For the problems of numerical sensitivity and subjective measurement standard of measurement methods,the adjusted cosine similarity measurement method is proposed to research partial users’ social relationships and intergal-partial items’ associated relationships.(3)Considering the scale of rating data,the minimum hash method is introduced to achieve the association strength measurement of the users’ social relationships and items’ associated relationships.Finally,the adjusted cosine similarity-minimum hash method is proposed to obtain the final users’ social relationships and items’ associated relationships,which enrich the data information of the single user and the single item.To verify the effectiveness of the latent factor model recommendation algorithm integrated users’ social relationships and items’ associated relationships,the proposed algorithm and ten mainstream recommendation algorithms experiment on five public rating datasets.Through the analysis of the accurate rating prediction indicators RMSE and MAE values,compared with the SVD++ algorithm,the proposed algorithm improves the recommendation accuracy by 0.0094～0.0775.To further researches the applicability of the proposed algorithm with extremely sparse data conditions,experiments are carried out on the selected user sets and item sets with different degrees of sparseness.The experimental results show that the proposed algorithm improves the recommendation accuracy by 0.0142～0.3503 compared with the SVD++ algorithm.The above research works show in the public datasets of commodities and movies application scenarios,the proposed algorithm of the latent factor model integrated users’ social relationships and items’ associated relationships improves the accuracy,alleviates the performance in extremely sparse data information conditions.Users’ social relationships described by the specific rating information take into account the interests and preferences with high association strength of users and items’ associated relationships described by the specific rating information consider the attribute characteristics with high association strength of items,which improve the interpretation of recommendation results.