Temporal Analysis of Web Search Query-Click Data

Understanding the intent of a query is paramount to build- ing an eectiv e web search engine. This is challenging, due to the sparsity in the number of words provided by the user (typically less than three), and also because the nature of queries evolves over time. In particular, distinguishing re- quests specically for recently published content such as blog post requests or news queries is a very dicult task. This is due primarily to the transient nature of the topics for these queries. To improve query classication, we propose to understand and extract specic temporal signatures for dieren t types of queries. In this regard, we construct a time series of query-clicks as a sequence of bipartite graphs, and develop a click rank score measure to capture the dierence in terms of rank entropy across the time series. We demon- strate the benets of oine temporal analysis in studying and isolating dieren t types of queries such as navigational, adult and news queries.