Application Research On On-line Guiding Shopping Of Bayesian Network

With the developmemt of electronic commerce ,shopping on-line was accepted by more and more people. The information provided by shopping website is more and more plentiful. A good navigation system for the shopping web site is very important, therefore, how to provide better guiding shopping service is must been solved problem. Application on on-line guiding shopping of Bayesian network can better solve this problem.The belief network(BN),proposed by Pearl in1982,also called Bayesian network, probability network, is a new mechanism for knowledge representation based on probability theory and graph theory. Its distinct performance in representing and reasoning about uncertainty makes it a hot research Goods in artificial intelligence. At present the research on belief network include three area: inference over belief network, learning belief network from data, and application of belief network. The research and application for the belief network have gain great progress these years, In this paper, some summary and discussion on the three area were given. Moreover, a mode of on-line guiding shopping of Belief net—works was given.The inference was one of main research about belief network. The junction tree algorithm and bucket elimination algorithm was two important inference algorithm on belief network. The junction tree algorithm has two defect.: the one is that it need too much store space, the other is that it can't utilized the conditional independence exhibited in the structure of belief network. In this paper, The junction tree algorithm was optimized and then built a new algorithm--the junction tree algorithm based on the bucket elimination. The new algorithm need less store space, and can utilize the conditional independent in a belief network to reduce the computation. The complexity of structure is a main factor in the bucket elimination algorithm. We build a algorithm which can translate a belief network into a new one based on the causal independence assumption. we also prove that the inference on the two network was equivalent. the structure of new one was simpler than the old one, so the inference on the new one was faster than on the old belief network.The method of learning belief network from data was required for application with belief network in practice. At present the main method of learning belief network was based on the scoring metric. The BDe metric was used widely. But this metric was build on many strong assumptions, and the specification of the prior probability of structure and parameter was difficult. Based on the information theory, we proposed a new scoring metrics for learning belief network--Sum of Mutual Information. Comparing to the BDe metric, the new metric have no strong assumption, and can automatically learned the belief network ,which need not to specify the prior of the structure and parameter.The target of research on belief network is to apply it. Now the belief network has already been used in many area, such as data mining, pattern recognition etc. In this paper we find a new application for the belief network, it can be used to build the navigation system for the web site. By analyzed the browser data with the learning and inference algorithm, the navigation system can help the browser to find the Goodss that he was interesting to quickly. It also can be used to optimized the structure of the web site, and find a good place to insert the ad bar in the web.