Research And Implementation On Recommendation Algorithm Based On Context And Item Properties Of Food Stores

This paper mainly researches recommendation problems of food stores based on the data set from www.dianping.com. Firstly, we will introduce the related technologies and algorithms about their present trend and development. Then as to the research directions of this paper, we will put forward the key research points and analyze them one by one, such as item properties modeling, sentiment analysis, time utility, and so on.The main research points of this paper areas follows.(1)An improved recommendation algorithm based on the context of text sentiment analysis. In this part we analyze users’sentiment from their comments on food stores, and our purpose is to fill up the original two-dimensional matrix model through sentiment analysis which especially considers negative adverbs. Experimental comparisons verify the reasonableness of our improved algorithm and better accuracy in context recommendation algorithm than in collaborative filtering algorithm.(2) An improved context recommendation algorithm based on time utility and user emotion. Recommendation system always analyze historical data, however, in this part we will combine the context factors of time recession and other characteristics together, such as emotional scores and item properties. Experimental verification has shown that the proposed algorithm has lower MAE and RMSE value, that is to say, our recommendation system has a higher accuracy.(3) An improved recommendation algorithm with the fusion of context and item properties. Firstly, we describe users’interest model with a frequency method which comes from the domain classification model. Secondly, we add emotional analysis in mining users’behaviors which means to improve the modeling accuracy. Finally, we consider the aging recession when calculate user similarity. Verified by experiments, the accuracy of our algorithm in this paper is higher.