Query Expansion Research

Search engine often returns a large number of results, but users do not have enough time to read so many documents, they need the most related ones. The exploding increasing of information and the short input query make it more difficult to find related information.Query Expansion is often used to enhance performance of search system. Query Expansion expands the original query and get a new query which is more clear. There are many query expansion methods, local analysis, and global analysis and so on. This paper presents a novel query expansion method, used in scientific papers. This method combines phrase dictionary and local analysis, and expand the query both in semantics and syntax. In details, this paper conducts the following researches:1) Query Expansion based on phrase dictionary. Phrase dictionary is given. This paper use mutual information to compute relationship between words in phrase dictionary and make a new phrase dictionary. When user inputs a query, search the query in phrase dictionary and get the most related phrases. These phrases are candidates of expanding words.2) Local analysis technique. Compute the weight of every word in local documents by frequency, and choose the most important ones as candidates. Meanwhile, design a word segment method based on phrase dictionary. This method has better performance in phrase segment.3) The retrieval model of query expansion. Retrieval model in this paper contains two parts. First, in the first retrieval, use Boolean model which returns all documents with original query. Second, in the second search, computer the correlation between local documents and the expanded query and rank the documents. 4) Retrieval evaluation method. Two methods are used. One is R@10, which compute average relationship of first 10 results. Another is read the first 10 documents by human and judge whether they are related with query.