Research On Recommendation Algorithm Based On User Context And Comment

Recommendation system as a kind of important means of information filter, which can help users by means of personalized in the choice of many contents they are interested in.Recommendation system mainly includes six kinds of methods, including content-based,collaborative filtering, demographic-based, knowledge- based, community-based and mixed recommendation, among them, collaborative filtering recommendation mainly by the score of the user to the product and other activities to provide users with the recommended, which is the earliest and most widely used algorithm.This paper improves the problem of data sparseness and accuracy in the traditional collaborative, then use the improved algorithm to food recommendation field. Based on the characteristics of data used in this article, the main part improved algorithm is as follows:1. Improved null values fill method, after scoring matrix was established, combining with the comments information mining analysis, using the estimate of the score to fill null value, to reduce the score data sparse influence on the recommendation accuracy.2. Ratings standardization process,this paper used data from the public comments on the real users and restaurant information in the net, because of the five-star evaluation system of the grading range span is larger, and each of us has our own habits of scores, these reasons will result in the user’s ratings appear error. So this article in view of the user to the mean score data centralized processing, improve the accuracy of grading, reduce the error.3. Introducing time utility and trust between users of similarity algorithm of improvement. On the one hand, the user’s score on behalf of his personal preference, and people’s preferences can change over time, the score of time utility can also affect the accuracy of recommendation; On the other hand, trust relationship between users can also affect users’ satisfaction with the outcome of the recommended and adopted, that is similar to the user’s authority and the degree of intimacy between the target user and will affect the result of the recommendation. This paper based on the above two aspects to improve traditional recommendation, and then improve the accuracy of recommendation.On the basis of the algorithm is improved, the article also according to the similarity of the calculation results predict target user preferences,get to the target user’s restaurant recommended list, and carries on the contrast experiment by using real data set, validation of the improved algorithm has higher accuracy than the traditional algorithm. Finally,according to the design ideas in this paper, the algorithm realizes the fusion of context and comment on the food recommendation system of information, and through the system test to prove the feasibility of the system.