StarrySky

Abstract

Query intent mining is a critical problem in various real-world search applications. In the past few years we have witnessed dramatic advances in the field of query intent mining area. In this paper, we present a practical system---StarrySky for identifying and inferring millions of query intents in daily sponsored search with high precision and acceptable coverage. We have already achieved great advantages by deploying this system in Sogou sponsored search engine\footnote {http://www.sogou.com}. The general architecture of StarrySky consists of three stages. First, we detect millions of fine-grained query clusters from two years of click logs which can represent different query intents. Second, we refine the qualities of query clusters with a series of well-designed operations, and call the final refined clusters as concepts. Third and foremost, we build a flexible real-time inference algorithm for assigning query intents to the detected concepts with high precision. Beyond the description of the system, we employ several experiments to evaluate its performance and flexibility. Our inference algorithm achieves up to 96% precision and 68% coverage on daily search requests. We believe StarrySky is a practical and valuable system for tracking query intents.