User Interest Oriented Behavior Analysis Method And Its Application

With the rapid expansion of the resources on the Web, the current search engines have been difficult to meet uses’needs. Personalized search results based on the users’actual interests can improve user satisfaction. Numerous studies have shown that user’s actual interest is closely related to their browsing behaviors. By analyzing browsing behaviors, user interest information could be obtained which could be used to make search results closer to user expectation. However, the current implicit relevance feedback (IRF) techniques do not consider the change in browsing behavior for different search patterns. Moreover, the current user interest obtaining methods usually consider some specific sorts of interaction pattern to predict page interest rate, thus returning not very satisfied results.In response to these problems, we explore the user browsing behaviors in different types of search tasks and propose a method to analyze multiple types of user browsing behavior. By using this information, we build an appropriate model to make the search result closer to user expectation.This thesis divides search tasks into three types:navigational, informational and transactional tasks, and considers four kinds of action which are the dwell time, number of mouse clicks, visit times and number of scrolls, as the objects of study. Search task type is identified using Bernard’s algorithm, and the four behavioral events are analyzed in the different search task types. In the users’behavior analysis phase, user behaviors are analyzed usingM5model tree, in which the tree pruning and the correlation coefficient smoothing must be considered in the calculation of user interest. In model evaluation phase, we use the model accuracy to evaluate the models. In order to clearly and effectively express the user interest information, we propose a user interest model on the basis of classification, which leverages the document feature extraction, Sogou corpus based on SVM classifier and other technologies. We make use of accuracy and sorting accuracy rate to evaluate the baidu search engine, the VSM based model and the model based on the classification. Experimental results show that the proposed user-oriented interest in user behavior analysis model can effectively improve users’satisfaction for the search results.