Research Of Recommendation System Based On Hybrid Model

With the advent of the era of big data, people are faced with massive amounts of digital information every day, the problem of information overload is increasingly serious. The traditional way to get information is to traditional portals and search engines through active search, which has been unable to meet the people’s access to timely and effective information. The emergence and development of recommender systems has brought changes in the way to get information. The initial recommendation technology is used in e-commerce sites such as Amazon, Taobao shopping recommendation system. With Web development, recommendation technology is used in news, video, music, quizzes and other Web applications that helps people get timely information from the mass.With the use of recommendation technology to a variety of applications, the research on recommendation algorithm is also gradually deepening. The foucus of academic research is on collaborative filtering algorithm, including memory-based methods, such as user-based KNN algorithm, item-based KNN algorithm,and model-based methods, such as matrix factorization algorithm MF, SVD++ algorithm, the social factors regular based matrix decomposition (SocailRegular-FM) algorithm. Single model and its many variants are mature. To improve single model is still the focus of future research. However, a single model algorithm can only catch part of the characteristic function hypothesis space.Mixing several models attracts more research attention.We foucus on both single model and mixture model.The main work of this paper is as follows, we make research and implementation on the basis model.Then we develop a recommendation engine library, called JRSLib library. We improve the similarity formula using auxiliary information. Social factors is increasingly being important to recommended system,we make research on MF subject to the social factors constraint, then propose the interest groups regular MF model,we use hte probabilistic latent semantic analysis model (PLSA) to learn interest groups,let the internal interest groups member vectors are more similar. We make research on linear regression and gradient boost decision trees (GBDT) hybrid technology model, analysis the advantages of the hybrid model, compare the model with a single model and mixed model. Finally,we design and implement a personalized mobile news recommendation prototype system.We use the recommendation system algorithm in the recommended text fields.