Research On User And Item Attributes Promotion Based Collaborative Filtering

Collaborative filtering recommendation is an important recommendation algorithm in the recommendation system,but it still faces problems such as data sparsity,scalability and cold start.The main improvement focuses on model-based collaborative filtering.There are still many vacancies at the improvement in the improvement of algorithm similarity and score prediction based on user and item attributes.Aiming at the above shortcomings,this paper proposes a collaborative filtering recommendation algorithm based on item and user attributes.The algorithm is divided into the following two parts.This paper introduces two important concepts,attribute importance and attribute credibility,for collaborative filtering based on item attribute scoring.The attribute importance is used to describe the user's preference for the attribute.The attribute credibility is used to describe the credibility of the attribute importance.The user's rating of the item's attributes is calculated using attribute importance,attribute credibility and user ratings of the item.The user-item attribute rating matrix which is used to calculate the user's similarity is constructed using the user's scoring of the item attributes.The user's rating is predicted on the original user-item scoring matrix using the new similarity.Finally,the effectiveness of attribute importance,attribute credibility and multi-attribute scoring is proved by experiments.The collaborative filtering based on item attribute scoring is compared with the recommendation effect of user-based collaborative filtering.Experiments show that the algorithm can improve the MAE evaluation standard by 0.02 on mean-based collaborative filtering.For collaborative filtering based on clustering of user attributes,this paper uses k-means clustering algorithm to cluster users after quantifying user features.The similarity of users which is used to predict user's rating in different clusters is imporved.Finally,it is proved by experiments that users with similar attributes are similar in interest.Collaborative filtering based on user attribute clustering is better than collaborative filtering based on common clustering.When the number of neighbors is small,the recommendation effect of the algorithm based on the user attribute clustering is better than the user-based collaborative filtering.And when the number of neighbors is large,the recommendation effect is not improved.The effectiveness of the similarity improvement method generated by combining the two similarity improvement methods proposed in this paper is verified by compared with other improved similarity algorithms.The algorithm combined the two improvement methods use other similarity calculation methods such as pearson baseline,mean square error similarity and pearson similarity to observe the improvement degree of the experimental results.After merging three kinds of similarities,the algorithm proposed in this paper has a better improvement of effect which proves that the algorithm proposed in this paper has Universal applicability.