Query Concept Interaction over Time

Abstract

With the ever-increasing amount of data available through various kinds of search engines, the need for better ways of identifying what users are looking for increases as well. In recent years, there have been many attempts at improving search results by trying to identify the true nature of the users' intent. An aspect of this that is often overlooked is that the intent of users is dynamic. One example of this comes from health care. Queries such as flu or fever occur at a regular frequency under normal circumstances. How- ever, a search for bird flu may indicate that an outbreak of the disease is at hand. In this paper we present a novel method for discovering hidden, time-varying interaction patterns in search query relationships. The method revolves around a probabilistic graphical model, capable of inferring interactions between groups of queries. Given sequences of query expressions and a base graph, the model produces sequences of interaction strengths. We perform synthetic experiments confirming the effectiveness of our model in recovering latent interaction dynamics. Furthermore, we compare the performance of our model to existing methods for network dynamics. In an application to search query data, we use our model to perform keyword suggestion and evaluate the results. The results of evaluating our model, shows that it has the potential to benefit a wide variety of applications within web-search including keyword suggestion and suggestion for related queries.