Overview of the TREC 2006 ciQA task

Abstract

Growing interest in interactive systems for answering complex questions lead to the development of the complex, interactive QA (ciQA) task, introduced for the first time at TREC 2006. This paper describes the rationale and design of the ciQA task and the evaluation results. Thirty complex relationship questions based on five question templates were investigated using the AQUAINT collection of newswire text. Interaction forms were the primary vehicle for defining and capturing user-system interactions. In total, six groups participated in the ciQA task and contributed ten different sets of interaction forms. There were two main findings: baseline IR techniques are competitive for complex QA and interaction, at least as defined and implemented in this evaluation, did not appear to improve performance by much.