A Hybrid Personalized Recommendation Method Based on Dynamic Bayesian Network
چکیده
With the rapid growth and wide application of electronic commerce, lots of information comes forth to people. However, our experiences and knowledge often do not enough to process the vast amount of information. The problem of obtaining useful information becomes more and more serious. To deal with the problem, the personalized service and recommender system play a more important role in many fields and collaborative filtering is one of the most successful technologies in recommender systems. However, with the tremendous growth in the amount of items and users, most collaborative filtering algorithms suffer from data sparsity which leads to inaccuracy of recommendation. Aiming at the problem of data sparsity, a hybrid personalized recommendation method based on dynamic Bayesian networks is presented. This method uses the dynamic Bayesian network technique to fill the vacant ratings at first, and then employs the user-based collaborative filtering to produce recommendations. Introduction With the Internet and e-commerce development, e-commerce recommendation system is widely used in e-commerce site, to provide product information and advice to help users decide what products to buy. It has ability to improve cross-selling e-commerce sites and customer loyalty for e-commerce sites. Collaborative filtering recommendation system is the current widespread using methods of e-commerce recommendation algorithm, and it is based on the user's information neighbor recommended by the target user [1,2,3]. Collaborative filtering techniques can be divided into user-based collaborative filtering recommendation and item-based collaborative filtering. Collaborative filtering algorithm is generally based on the user item rating matrix. When the matrix is very sparse, the algorithm is difficult to find similar users or items, so that the recommendation quality has been greatly limited. On the other hand, collaborative filtering does not care about product and user information, which information can effectively improve the accuracy of collaborative filtering recommendations [4,5]. The collaborative filtering algorithm assumes that a good way to find a particular user is to find interesting content like-minded people who work with him. This method, the user ratings in order to predict the user is not on matters of interest observed in the observation program [6]. In this paper, a hybrid personalized recommendation method based on dynamic Bayesian networks is presented. This method uses the dynamic Bayesian network technique to fill the vacant ratings at first, and then employs the user-based collaborative filtering to produce recommendations. Prediction using the dynamic bayesian network[7,8,9] Markov hypothesis. S(Y[t+1], { Y [1],..., Y [t1]}| Y [t]) is the set of state variables at each moment only a moment before the value of the relevant state variables. Then p(Y [1],..., Y [t])=p(Y [1])p(Y [2]| Y [1])...p(Y [t]| Y [t1]) Programs Supported by Ningbo Natural Science Foundation (Grant No. 2010A610118). A Project Supported by Ningbo Training Base of Textile and Fashion (Grant No. JD090312). For limited time periods 0, ..., T of the dynamic Bayesian network can be expanded into a set of variables containing Y [0], ..., Y [T] the Bayesian network, and on the Y [0], ..., Y [T] The joint probability distribution can be expressed as: 0 (Y[0], ,Y[ ]) (Y[0]) (Y[ 1] | Y[ ]) T