Research On Relevance Ranking

For relevance ranking in search engine technology, this paper mainly researches on three technologies: link analysis, paragraph retrieval, relevance feedback. They all can improve retrieval with different method. The main innovation contributions of this paper are listed below:First, this paper proposes a storing method of PageRank algorithm based on link analysis. As all known, PageRank algorithm needs to compute the score of every web page, which makes the problem of storing these web nodes critical. This paper uses formula deduction and characteristics of sparse array to solve this problem, and at the result, reduced the space complexity from square to linear, and improved the computation efficiency meanwhile.Second, this paper proposes a method of combining paragraph ranking and full text ranking to improve retrieval result. The method of ranking based on paragraphs can improve precision but will make recall reduce, to avoid this, this paper combined the score of paragraph ranking with full text ranking, that will make sure the recall of retrieval.Third, this paper analysis the advantages and disadvantages of Rocchio relevance feedback algorithm, which is based on vector space model.