Mining named entities from search engine query logs

Abstract

We present a seed expansion based approach to classify named entities in web search queries. Previous approaches to this classification problem relied on contextual clues in the form of keywords surrounding a named entity in the query. Here we propose an alternative approach in the form of a Bag-ofContext-Words (BoCW) that is used to represent the context words as they appear in the snippets of the top search results for the query. This is particularly useful in the case where the query consists of only the named entity without any context words, since in the previous approaches no context is discovered. In order to construct the BoCW, we employ a novel algorithm, which iteratively expands a Class Vector that is created through expansion by gradually aggregating the BoCWs of similar named entities appearing in other queries. We provide comprehensive experimental evidence using a commercial query log showing that our approach is competitive with existing approaches.