Research On Collaborative Filtering Algorithm Based On Context Modeling

With the wide application of Internet, the information that the user can contact is in the explosive growth through big date analysis of user behavior pattern and forecast trend begin to pop up. By this time, The recommendation system, through the initiative to recommend users of goods and services which may be of interest, greatly reducing the user to find the content they are interested in information from the mass of difficulty.The traditional recommendation systems are generally based on analysis of user behavior, preferences, personal characteristics, calculation of the similarity between users, to generate recommendations. Usually assume that static user preferences, without considering the impact of contextual factors on user selection. Context aware recommendation system expansion in the traditional recommendation system, no longer just concerned about the link between users and items that fully considers the influence of context (such as season, regional, mood etc.) on the user's decision, which will be the user's current context information applied to the recommendation process. This makes the results not only recommended to meet the personalized requirements and in accordance with situational requirements. On this basis, the application of Bayesian methods to determine a context modeling, context, context influence on user decision, improved collaborative filtering algorithm, better for the user to make personalized recommendation. In this paper, the following research work has been done:1. Analysis of the context of meaning and effect, compared several paradigm of context, the choice of the most computationally complex but the best paradigm of context modeling, combined with Bayesian method to construct a context user interest model and calculation user on the probability score out of context, probability as a weight into the model.2. Solving a single user to a single project for all ratings data, taking into account the small sample size, it is difficult to obtain the probability of each context. The user and project clustering process, the experiment proved that this classification helps to solve the problem of data sparseness and computational complexity.3.Improve user-based collaborative filtering algorithm, contextual factors into the algorithm, improve the accuracy of the recommendation.4. In this paper, the two data sets for off-line experiments, this algorithm and the traditional collaborative filtering algorithm, Differential Context Relaxation algorithm comparative experiments, combined with the two indicators of evaluation indexes MAE and RMSE, comprehensive comparison of the data shows that the algorithm can be to a certain extent recommend to improve the system accuracy.