Research On Context-aware Recommendation System Based On Factorization Machines Model

Recommender system is an interdisciplinary research which covers the content of various disciplines..it derived from the research on collaborative filtering algorithm in mid-1990 s, and then began as an independent field of research. Currently, the traditional recommender systems have a lot of research literature.They divided the recommender system into three categories based on the different generated method of recommender system, including content-based recommendation, collaborative filtering recommendation, and mixed recommendation. Consensus has been gained by more and more researchers, that, different recommended method should be chosen for different projects,(For different projects, should chose different recommend methods,)or chose mixed recommender,). Researchers also found some problems of the recommender system, such as cold start, data Sparseness and the interpretation of the recommendation. they also raised a number of research methods according to these issues, such as similar labeling, social networks, matrix decomposition and Bayesian inference method etc.However, with the growing demand for personalized, how to give a reasonable, fit and appropriate recommenders to each user is becoming a more important task of the recommender system. Accordingly, context-aware recommender take time, place, peers and other contextual information into account, which persist the advantages of the traditional recommendation system. in order to better improve the recommender accuracy, and better meet the individual needs of users. It improved the recommender credibility of the system. In recent years, researchers have paid more attention to this issue. After analyzing and summarizing the review lectures of context-aware recommender systems as well as other relevant documents, this article have done the following:(1) This paper presents a hierarchical context factorization machine model. The model improved the original factorization machine by considering and hierarchical processing the context, it enhanced the overall relationship between(among) the scoring data, and improve the score predicted more effectively.(2) This paper presents a hybrid model logistic function and factorization machine Bayesian approach. Based on Bayesian processing on the factoring machine model, the model further did a logistic function processing on the shape parameter of gamma function, which able to distinguish different characteristic parameters better, thereby improving the accuracy of prediction score.(3) Through a plurality of experiments on multiple data which sets for validation, first the characteristics of the data set will be briefly described, to describe the process data sets in a certain degree. Besides verify the proposed hierarchical context factoring machine features and benefits through several experiments.