In this paper, a supervised learning framework with strong expansibility is first established for search engine joint ranking problem. It can transform existing algorithms into corresponding learning algorithms, and design new algorithms under this framework. Second, with Markov chain model as the core algorithm, this paper combines the ranking results of three main factors, including content relevance, hyperlink prediction, and query click behavior, and transforms the joint problem of ranking results into a positive semi-definite programming problem, and deduces the detailed process of solving the problem. Finally, this paper analyzes the rationality and efficiency of the joint ranking recommendation model based on Markov chain by setting the weight coefficient through experimental data. KEYWORDS joint ranking, Markov chain, search engine
INTRODUCTIONThe joint problem of search engine ranking results is to combine the sequence formed by various ranking factors to form a new result sequence which is more in line with the actual situation and user needs. The combined ranking is called basic ranking, and the newly generated ranking is called joint ranking. There are still many technical difficulties to be solved in the practical application of search engines. On the one hand, it is difficult for users to find appropriate phrases to accurately express their retrieval content. On the other hand, users have different personal interests and preferences, and even when using the same query words, they will represent different query needs. In recent years, many scholars have made corresponding research in these fields. The CORI fusion algorithm proposed by Khalil et al 1 assumed the number of overlapping documents in each repository. It combined the information resources score and documents score linearly. First, it evaluated the relevance of each repository to the query, and then combined the retrieval scores of each repository to get the final ranking score. Liu and Hou 2 proposed a microblog hot topic mining method based on heat joint ranking. The authoritativeness of microblog text and the related characteristics of hot topic words within a certain time period were fused by building a corresponding model framework, which greatly enhanced the interdependence of each other and enabled us to find hot topics more comprehensively and effectively. Zhao 3 proposed a joint learning framework of feature group information and ranking structure, which was a DSSVM based on deep feature learning. By introducing induced variables, the structural support vector machine and feature learning tasks were well combined to obtain better ranking results.In order to overcome the problems of unreasonable ranking basis and inaccurate ranking results caused by single ranking factor in traditional retrieval recommendation model, this paper proposes a recommendation model that combines content relevance of pages, hyperlink analysis, and query click behavior to more accurately judge and recommend the web pages related to user needs information.