Research And Implementation Of Virtual Region Re-Rnking In Information Retrieval

With the development of Web technology, the information quantity is growing and growing.Users want the channels and ways of obtaining information to become more convenient and efficient.Especially when searching information, users expect what they need at the front page, this is the question of Search Engine Optimization (SEO).Re-ranking of search results is a main branch of SEO, whose research methods are becoming increasingly diverse.In recent years, a novel and effective concept of virtual region is proposed for query expansion.There are a variety of methods to create virtual region.However, due to the huge workload of calculating virtual region, there’s a limitation that only one method can be used for query expansion at one time. We proposed a new re-ranking model based on virtual region, which is an integrate of many virtual region methods and only process relevant documents in the initial retrieval. Experimental results show that the re-ranking model based on virtual region can improve the efficiency of re-ranking.Re-ranking model based on virtual region is focus on achieving multiple virtual region from the operation among a variety of query terms,and we calculate the new relevance between the document to sort(the first k documents) and the query by using the method of linear interpolation to combine multiple virtual region,and then give a new sorting according to the relevance.According to the query operator,for example: or,and,gram,phrasal,we design and achieve query mapping algorithm and establishing virtual region algorithm.The calculation of correlation is based on the traditional BM25,and on the basis of it we achieve the re-ranking model of virtual region. We propose five methods of calculating the virtual region with the combination of or,or-and,or-gram,or-and-gram,or-and-gram-phrasal and so on,and we research and analysis how it influence on the result of re-ranking. We optimize all of the parameters in the model by a lot of experiments coherently the previous experience.The innovation of this article is that we calculate the virtual region by combining multiple operators,which breaks the limitations of the original single operator,and then our relevance between a document and the query will be more accurate.We carry out the experiment in several TREC sets,and the experiment results show that the model this article proposed can improve the performance of re-ranking.