Research And Application On User Context-Based Recommendation Technology

With the development of information technology, Internet and mobile communication technology, traditional Internet and mobile Internet have entered the age of "information overload". On the ALLThingsDD11conference in2013, Mary Meeker, a famous American Internet analyst, pointed out that the users of mobile Internet have increased0.4billion and reached to1.5billion, the information created by mobile Internet has got an unprecedented growth, but mobile Internet was still in the early development. With the rapid growth of mobile Internet users and information, mobile Internet users have got a serious problem of information overload which has been gradually out of mobile users’affordable level. How to ease the problem of information overload has been a challenge of Internet development. The research content of recommender system is to build model about the users’interest through historical behavior records on users and to recommend items to the potential users who may be interested in these items. Recommender system has eased the problem of information overload in some way and got lots of researchers’attention. Traditional recommender system build model about the users’interest by using the "user-item" binary relation without user context. With the development of mobile communication technology and sensor technology, it is easy to get the user context by using mobile terminal such as cellphone. In the recent years, lots of researchers paid attention to the influence of user context to the user preference, but lots of researches only considered a kind of user context factor, ignored the influence of various kinds of user context factors to user preference. It will get high recommender accuracy by considering various kinds of user context factors. In this paper, I gave a user context-based recommendation method to provide users with more accurate personalized recommendation service.In this paper, I give a deep research on historical behavior records by considering the influence of various kinds of user context factors to user preference. The main work of this paper is shown as follows:First, briefly summarize the basic concepts of recommender system and several basic recommendation methods:social recommendation, content-based filtering recommendation, collaborative filtering recommendation and hybrid recommendation. And then, introduce several evaluation indices of Recommender System, the current state of user context-based recommender system research and the challenges of recommender system.Second, provide a detailed description of the core algorithm of this paper-tensor factorization recommendation algorithm based on Impact Factor of user context (IFUC_TF). In this part, briefly describe the definition of user context, the ways acquired user context and the modeling of user context. To the core algorithm, I give a method to compute the impact factor of each user context factor, and describe the tensor factorization algorithm include how we propose the loss function, optimize the parameters in our model and also give the pseudo-code of the algorithm to solve the given model. At last, generate the recommender result based on Multidimensional user context by using the predicted value of user preference based on Single-dimensional user context and the impact factor of each user context factor. Finally, use MovieLens data set to operate the simulation experiment on tensor factorization model based on Single-dimensional user context. In the experiment, we also use three other recommendation algorithms in which we compare the result to our model and then analyze the result. The experimental result shows that tensor factorization model based on Single-dimensional user context get higher recommended accuracy than traditional recommender algorithm, and also get higher recommended accuracy than contextual pre-filtering algorithm. It also shows that tensor factorization model which considered user context will bring high recommended performance.Finally, design a mobile e-commerce recommender system by using mobile app to login the e-commerce platform and receive the recommended result. In this system, the mobile recommendation module uses IFUC_TF algorithm to recommend items to the potential user. And present the overall architecture of the mobile e-commerce recommender system. In the system implementation, I use Java programming language and Android app development technology. During the evaluation of this system, find that user context-based recommended result get a higher user satisfaction than traditional one.