Modified term weighting with relevant terms for retrieval model

摘要

This paper introduces a method that improves web search accuracy by using modified term weighting with relevant terms that are extracted from various information resources by the application of various techniques. The existing query expansion method, which is typical of the techniques that use relevant terms, offers improved search accuracy, but its weakness is that it can use only a few extension relevant terms because of its high computational cost. Our method can improve the web search accuracy without increasing the cost. Another advance is that we store weighted term frequency calculated with relevant terms in the search index, and use the index to conduct BM25 base searches. We can get highly accurate search ranking results without increasing the response time. Experiments show that the relevant terms are extracted from Wikipedia, query logs or click logs, and accuracy is improved by 17%.