Research And Implementation Of Negative Feedback In Information Retrieval Based On Language Model

The 21 st century is the times of network economy, with the rapid development of internet, the amount of information on the internet is increasing, however, how to get the information they need from the vast ocean of information become more meaningful.In information retrieval, search engine use ranking algorithm to sort the retrieved documents according the relevance between the query and documents, the researchers presented a mathematical retrieval model of relevance. At present, language model has better performance. For the difficult query, the search documents sorted front of the results are poor, and little documents are associated with the user's query needs. In this case, how to use the no-relevant information to improve the retrieval accuracy? Researchers proposed a negative feedback which is a special case of relevance feedback.This paper proposed a new method of combining negative feedback and relevant feedback altogether in information retrieval based on language model. Under the language model framework, I applied the approach of feedback algorithm similar to Rocchio feedback in the vector space model to expend query and change the probability of query words. In this paper, we considered the first ten documents of the initial search results, as the local situation of query expansion; we traditionally view these ten documents as relevant documents, used for pseudo-relevance feedback. In this paper, the first ten documents will be considered separately, according to the comparison with standard judgment of relevance, separating collection of documents into set of documents relevant to a query and set of documents no-related to the query, generating positive model and negative model with the original query separately, the words which appear in the negative model and positive model be added to the query to expand query, we appropriately increase probability of query terms appeared in the documents of relevant, and reduce probability of query terms appeared in the documents of no-relevant, combining relevant feedback and negative feedback, to improve the accuracy of the expanded query, making the search results performance of the new query model have greatly improved than the query after pseudo-relevance feedback and original query.