Key Concepts Identification and Weighting in Search Engine Queries

摘要

It has been widely observed that queries of search engine are becoming longer and closer to natural language. Actually, current search engines do not perform well with natural language queries. Accurately discovering the key concepts of these queries can dramatically improve the effectiveness of search engines. It has been shown that queries seem to be composed in a way that how users summarize documents, which is so much similar to anchor texts. In this paper, we present a technique for automatic extraction of key concepts from queries with anchor texts analysis. Compared with using web counts of documents, we proposed a supervised machine learning model to classify the concepts of queries into 3 sets according to their importance and types. In the end of this paper, we also demonstrate that our method has remarkable improvement over the baseline.