Personalized Recommendation Research Fusing Comments Tags

Personalized recommendation is a popular topic in the field.of natural language processing research,especially in the field of electronic business.The personalized recommendation system has been widely used,which aims at minimizing the user' s interest and recommending the products or information that the users might like to reduce the overloaded information.Currently,it has raised a lot of attention from scholars at home and abroad.Some achievements have been made as well.Most of the existing approaches are based on the ratings or reviews of users.However,with the sparsity and poor regularity of comments,it is hard to acquire all of the effective information,which influence the recommendation performance.Thus,in this paper we focus on this problem and aim to extract the valid information from comments tags for recommendation,the details of the research are proposed as followed:(1)As for the drawbacks of poor regularity,non-restrictive form of users'comments in the current fields of electronic commerce.This paper proposes a recommendation algorithm that integrates comments tags into Collaborative Filtering recommendation algorithm based on matrix factorization.It first extracts the users'opinions from the features of products,then makes use of them to construct user interests and product features,finally recommends the products with well-reviews on specific features which users are interested in.The experimental results show that users tend to use comments tags,and more effective information is obtained from comments tags than comments.Meanwhile,the results demonstrate that the proposed algorithm can effectively improve the user coverage and recommendation accuracy and effectiveness.(2)The recommendation algorithm of Collaborative Filtering recommendation algorithm based on matrix factorization only considers users' information to recommend,ignoring other factors,and the Collaborative Filtering recommendation algorithm based on neighborhood model to recommend only using ratings.Thus,we propose a recommendation algorithm that integrates comments tags into Collaborative Filtering recommendation algorithm based on neighborhood model.Firstly,we map the user rating of high-dimensional to low-dimensional space and construct user and item latent factor models by using the LDA.Secondly,the users' opinions for the features of products are extracted to construct user and product factors models,comments tags are decomposed to construct multi-relational of user-feature-opinion-item.And calculates the similarity of opinion between users.Thirdly the whole similarity matrix are calculated by the latent factor models,the product factors models and the similarity of users' opinion.Finally we employ the neighborhood methods to predict unobserved ratings.Experimental results indicate that the proposed algorithm are better than the benchmark algorithm.(3)The recommendation algorithm of Collaborative Filtering recommendation algorithm based on matrix factorization only considers the factor of user,and the recommendation algorithm of Collaborative Filtering recommendation algorithm based on neighborhood model is sensitive to data sparsity,both of them have the limitations.Thus,this paper presents two kinds of recommended strategy algorithm:namely weighted hybrid model and algorithm hybrid model.Weighted hybrid model uses weight factor to mix various predicted scores of algorithm.Algorithm hybrid model is based on neighborhood model in the framework of the recommendation algorithm,and fuses predicted scores of matrix factorization in this framework,then calculates the new prediction ratings.Experimental results demonstrate that the hybrid model of comments tags not only can improve the performance of algorithm,but also improve the stability of the recommendation algorithm.