On Effective Dynamic Search in Specialized Domains
Felipe Moraes, Rodrygo L. T. Santos, Nívio Ziviani · 2017
Dynamic search in specialized domains is a challenging task, in which systems must learn about the user's need from his or her interactive exploration. Despite recent initiatives to advance the state-of-the-art for this task, limited progress has been achieved, with the best performing dynamic search systems only marginally improving upon vanilla ad-hoc search systems. In this paper, we perform a comprehensive analysis of the impact of several components of a prototypical dynamic search system on the effectiveness of the entire system. Through a series of simulations, we discuss the impact of: producing an initial ranking of candidate documents, modeling the possible aspects underlying the user's query given his or her feedback, leveraging the modeled aspects to improve the initial ranking, and deciding when to stop the interactive process. Our results using data from the TREC 2015-2016 Dynamic Domain track shed light on these components and provide directions for the design of effective dynamic search systems for specialized domains.