Ryen W. White, Matthew Richardson, and Wen-tau Yih
Search systems have traditionally required searchers to formulate information needs as a set of keywords rather than in a more natural form, such as questions. Recent studies have found that search engines are observing an increase in the fraction of Web search queries that take the form of natural language. As part of building better search engines, it is important to understand the nature and prevalence of these intentions, and the impact of this increase on search engine performance and searcher efficiency. In this work, we study the behaviors of search engines when handling keyword-based queries and natural language questions, as well as the costs incurred by searchers in creating query statements of each form. We show that although informational search intentions are often expressed as keyword queries, when given the same search intent expressed as a query and as a natural language question, search engines in fact perform equally well in terms of relevance. Since creating queries has been assumed to be challenging, this equality should support an increase in question-querying. However, question formulation has an associated cost, e.g., we show that generating natural language questions for search engines takes much longer than keyword queries for the same intent. Our findings suggest that searchers should stick with keyword queries and that the increase in question prevalence is related to factors beyond search engine performance.
© Microsoft Research