Research Of Bayesian Learning And Its Application On Personalized Search And Recommendation

Bayesian learning is an important research direction of machine learning theory. It is based on Bayes law, by understanding prior knowledge of data distribution and combining with sample training data to estimate the mathematical model of the overall data. Bayesian learning is to get the result of the joint probability distribution of a set of variables. Bayesian learning is in the form of probability to represent the uncertainty knowledge, so to the uncertainty problem it has a unique description form and an advantage of calculation. Through the Bayesian learning model to calculate a set of input, the model will give you the probability value of each solution results that may occur, rather than a determine one. Personalized search and recommendation is a hotspot in the research of the Internet, it has important significance to the search engine, e-commerce, etc. Because each person’s choice tendency is affected by various factors, therefore, personalized customization is also a kind of uncertain problems. For this reason, while studying the related theory of Bayesian learning, at the same time, the author focus exploration on the application of Bayesian learning in personalized search and recommended applications.The main contents and innovations of the thesis are as follows:1. Research the theory of Bayesian classifier and its application in personalized search, we design a Bayesian classifier applied to goods classification. The classifier dives the commodity attributes into two categories:one is text attribute, which is used to describe the commodity name, brief introduction and evaluation, etc. The other attributes are discrete quantity, such as price, sales, etc. For text description attribute, classifier based on classic naive Bayesian text method to compulate every classifications’probabilities, and then multiplied by other discrete quantity attributes’ probabilities. By using the classifier to classify goods, it can realize the goal to add more targeted tag of goods descriptions, which can not only provide user more terms of choice to locate commodity, but also is advantageous to search engine to parse the user’s description semantics so that return personalized search results.2. Research Bayesian belief network and its application on personalized recommendation. According to the entire network customers’purchase records, we construct a Bayesian network to descript the users’purchases. For recommended behavior, the essence of the algorithm idea is found users who have similar purchases records and recommend other goods in the purchases chain to users. The network uses goods bought by users as network’s nodes, constructed by dependency analysis algorithms which based on information theory, then trains the parameters of the network node table. User recommendation is the process of bayesian networks inference. As to recommendation algorithm design,we chose a kind of approximate reasoning of random sample algorithm after Comprehensive considerations of the scale and precision of calculation, etc.