Topic Model-based Recommender System For Personalized Long-tailed Products

The recommendation system is a type of information filtering system for predicting user preferences or scoring.In the business field,many business platforms use the recommendation system as a way of marketing.In the traditional marketing,the long tailed products are a kind of commodities at the end of the trading curve.Recommendation system is helpful to present long tailed products to target users and develop the potential sales ability of long tailed products.Because of the sparsity of the transaction data of long tailed commodities,the traditional recommendation system is prone to popularity bias and shilling attack problems.In view of the popularity bias problem,this paper jointly models user history behavior,product feature text and user social network information based on the topic model.The product feature text and user social network information can make up the missing behavioral data in the long tail products and increase the connections between users and the long tailed products.The experimental results show that the mixed recommendation system based on the topic model performs well in the recall rate,accuracy rate,product recommendation coverage rate and gini coefficient,and effectively alleviates the popularity deviation of long-tailed products.In view of the shilling attack problem,this paper jointly models the user's credibility and user's historical behavior based on the topic model.The nearest neighbor similarity of the target user is taken as the index to distinguish the true and false users and the model learning is conducted on the basis of the existing index discrimination which can modify the index discrimination result and reduce the dependence of existing related researches on data characteristics.The experimental results show that the recommender system performs well in the data set with attack behaviors and effectively solves the shilling attack problem of long tailed products.