TREC 2016 Dynamic Domain Track Overview

The main goal of TREC dynamic domain track is to explore and evaluate systems for interactive information retrieval, which reflects real-world search scenarios. Due to the importance of learning from user interactions, this track has been held for the second year. The name of the track contains two parts. “Dynamic” means that the search system dynamically adapts the provided ranking to the user through interactions. “Domain” stems from the fact that the search task in the track is on the domains of special interests, which tend to bring information needs that would not be met within a single interaction. The task is inspired by interested groups in government, including the DARPA Memex program. Each search task in the DD track involves some interactions between a user and a search system. In the first iteration, the user submits a query and the target domain of interest to the search system. The system provides the user with an initial ranking of documents, and receives feedback from the user on the provided ranking. This interaction, providing the user with a ranking and receiving the user’s feedback, continues until the search system stops the search task. The DD track introduces a new challenging search problem with the following important assumptions.