The Study Of Latent Semantic-Based Personalized Search Key Technology

With the rapid development of network technology, the information explosion resulting from the information fatigue and information pressure makes the search engine become more and more important. The search engine has become a veritable information hub and information portal, which has become the preferred tool for users to get web information. However, in the huge results list, only a small part of information is in line with the user’s preferences, even in the top K results, there is no information in line with user preferences. Faced with this dilemma, people had to re-examine how to provide users with their preferred personalized information.This dissertation analyzes the main reason for the fact that there is no real understanding of the latent semantic motives behind user queries. Don’t know what users need, we are unable to provide high-quality personalized service for them.Search engines directly face to the different users who have different background knowledge and search intention. Therefore, it is impossible that there exists a universal query way that can make clear what potential motives maybe when they enter the same keyword between different users. For example, when the query keyword is "Northeastern University", the user may want to know the current dissemination of information, or want to know take a look at this year’s admissions policies, or want to know about the outside evaluation of Northeastern University. Therefore, the user’s potential semantic motivation analysis is the basic of personalized search, for the following personalized work are all carried out upon this work. Because input ’keyword’ is the first step of search, this keyword on behalf of the users’ latent search needs TAG expression. The Internet "search holographic theory" that the founder of the wind believed that we must give deep understanding to the relationship between ’keyword’ generate process in users heart and output process in his search. The dissertation tries to find out the holographic relationship among user search motivation, the thinking volumn before search, and search feedback. In fact, the user enter the "keyword" is actually a process in which user introduce this calculated TAG in his heart into search behavior. And such, the TAG should be the best quality TAG; because it combines the users first reflected unconscious wisdom of the heart.This dissertation just focuses on this, from the user latent semantic motivation to begin the study on multiple personalized service key technologies, mainly include the following:(1) In the field of computer research, we give detailed analysis of user search behavior from philosophy, psychology aspects, and give a user behavior model based on probabilitic latent semantic motivation from cognitive view. We highly summarized a variety of specific search behavior, from an abstract point of view to understand the user’s search behavior. The model we put forward provides a basis for study search behavior further.(2) In document latent semantic space, we apply combination zipf distribution with the probabilistic latent semantic analysis algorithm to extract the documents topics, improve the efficiency of document topic extraction.(3) On Dirichlet priori limited mixture model theory basis, a highly effective unsupervised web page clustering algorithm is suggested, which can effectively overcome the traditional clustering algorithm ineffective in dealing with high dimensional, sparse text, definition similarity function difficulties between text data, which can improve clustering effect and fill up a deficiency of low clustering quality and low efficiency, which improves the ability to capture the user preference latent topic.(4) A novel query expansion technology based on use latent semantic analysis is suggested. That is to say, it combines the general search query expansion technology with user motivation mining. Develop a novel query expansion technology. Solve the user-oriented personalized information processing capacity insufficiency because search engine general character. Give a fundamental solution to polysemy and multi-word synonyms problem in query, from understanding user semantic search motivation and understanding interaction between cognitive and mind. Effective semantic disambiguation has been made in personalized search process.(5) A user-oriented re-ranking algorithm is suggested against the query-oriented ranking algorithm deficiency. That is, on a basis of original web page ranking algorithm, a kind of local optimization algorithm is proposed according to user preferences. In line with user personalized preferences and without affecting the search results recall. To achieve ranking results coincides with user semantic motivation as soon as possible.In short, the starting point of the dissertation is user latent semantic motivation. Focus on personalized search every aspect of key technologies, such as query expansion technology, web page local re-ranking technology, clustering technology and so on, we make our study. Try to achieve the high-performance match between user’s query and the search engine returns results.