Research On Recommendation Algorithm Of Personalized Search Engine

With the rapid development of the Internet and the networks information technology, network resources increase exponentially, the results of the traditional search engine only depend on the query keywords, in fact, even if the same query keywords, the purpose of different user may be different. In view of this situation, people need a special search engine to provide personalized search engine services which can return precise results for their own features, and a user-centric personalized search engine was proposed.This thesis introduced the basic theory and state-of-art of personalized search engine, and made performance evaluation and comparison on the exsiting technology for personalized recommendation. These theories provided foundation stone of researches behind.Then, the thesis researched on the most important collaborative filtering algorithms in the field of recommendation,user-based collaborative filtering can recommend new interested potential resources for users, but it has some shortages, such as sparse; project-based collaborative filtering can solve the problem of sparsity, and it is simple and effective, but only the similar information can be found. To solve these problems, an improved algorithm for collaborative filtering algorithms was proposed, improved the quality of collaborative filtering recommendation system by single value decomposition and increaseing the impact collection, and improved the performance of the system.However, in the recommendation system based on the algorithm of improved collaborative filtering, the problems of cold-to-start, expansibility were still bad to the personalized recommendation system, when the system began to run, the recommended services was difficult to implement. In this section, a personalized fusion algorithm was proposed, based on the excellent user-besd collaborative filtering, it combined with the content-based recommendation and the project-based collaborative filtering to solve the problems of sparse,scalability,cold-to-start and the difficulty to mine potential interest with matrix technology and extending influence sets, improved the quality of the recommendation system. And on this basis, a strategy was proposed to predict the user's score to solve the problem of the great difference of score caused by different strictly levels of users .Finally, the thesis analysisd the open-source full-text search tools Lucene, added the personalized search modules to the platform, and made algorithm simulation on the improved collaborative filtering algorithms and the fusion of personalized recommendation algorithms. The experimental results show that: the quality of improved collaborative filtering algorithm is better than the traditional collaborative filtering recommendation algorithms, and on the conditions of the cold-to-start, the quality of personalized recommendations fusion algorithm is better than the improved collaborative filtering algorithm, the prediction meet the user's actual score more,search results were more in line with user needs, improved the quality of personalized search engine service.