Research On The Relevant Feedback Algorithm In Information Retrieval

Information retrieval is about the structure of the information, the field of analysis, organization,storage, searching and retrieval. Broadly speaking, information retrieval is from unstructured informationcollection, to identify information related to the user needs. Information retrieval is a core issue focus onusers and their information needs, because the evaluation of search based on user-centered. This philosophyhas led to a lot of research on how people interact with search engines, especially developed to help usersto express their information needs of technology.In the retrieval process that users participate, users submit a short query, the system returns to theinitial query results, the user on the partial results Note indicating information to a related or unrelated tocalculate a better query, the system based on user feedback needs and back a group are more likely to meetuser needs new search results, this process is called relevance feedback. Relevance feedback in informationretrieval process can optimize the query results and improve query efficiency.Start from the status of the introduction of relevance feedback, relevance feedback algorithm,including the relevance feedback in the vector space model, the probability model and Boolean model.Among them, Rocchio relevance feedback algorithm based on vector space model, described in detail theidea of the algorithm and implementation process and its query ineffective in certain circumstances, such asa collection of answers to a query itself requires different types of document composition and usually amore specific concept or relationship to the words of these two aspects, to improve Rocchio relevancefeedback algorithm, the algorithm in these two exceptional circumstances can get a good return results.This document made the following contribution:1. In the condition of query contains more than one word, based on the idea of Rocchio relevancefeedback algorithm: the new modified queries move from the initial query, go toward the centric vectors ofrelated documents. These documents which contain the information of the two query word will be regardedas a new class. Firstly, using Rocchio classification method to calculate the boundary of the class, and thenuse the mean vector of the vector space model document vector calculation method to identify the twocentric of the cross class. Finally, move from the center of which the initial query vector near by, toward the other centric, and found the documents that users are looking for.2. In the condition of query contains polysemy, using the concept of multi-label classificationalgorithm to cluster the results that returned, while return the query word contains all semantic tags forusers.