The Research Of Personalized Search Engine Technology Based On User Interest

With the maturity of general search engine technology, personalization and intelligentialize gradually become the main direction of search engine technology. The key point of personalized search engine is the creation and synchronization of user interest model, because the quality of personalized service is decided by the quality of user interest model, and most personalized processing procedures are based on user interest model. Therefore, how to establish high quality user interest model, and use it to optimize the query results is a research subject with practical application value.This thesis aims at the problems that personalized search engine is confronted with at the present stage, points out the insufficiency of modeling based on traditional TF-IDF algorithm, and combines content anlysis with behavior analysis to improve TF-IDF algorithom. The thesis designs a method to create user interest model based on the improved TF-IDF algorithm, which uses user manual custom-made modeling technology and automatic user modeling technology. This thesis presents an interest model update algorithm, which uses the interest information from user group as complementary interest, and takes user feedback information for reference to resolve the problem that the update algorithm which is base on user behaviour or user subject vector can not obtain user interest information comprehensively and accurately. The experiment results prove that the algorithm can efficiently improve the comprehensiveness and accuracy of user interest model. Based on the interest model, the thesis studies the methods of optimizing the query results. The thesis improves the query expansion method which is based on user interest degree by further calculating the weight of the candidate keywords to resolve the problem that the original query expansion method only takes the clicks of the search results into consideration, and pays no attention to the quality of the search results. The thesis also improves the personalized sorting method which is based on historical querys by referencing user feedback information to resolve the problem that the attitudes of users towards the search results are uncertain. The experiment results show that the above improved methods make the search results sequence closer to the user click sequence. Finally, the research results are used for the design and implementation of the personalized search engine system, and the above-mentioned methods are proved to be effective through the system test.