Re-ranking via User Feedback: Georgetown University at TREC 2015 DD Track

There are two principal components involved in a search process, the user and the search engine. In TREC DD, the user is modeled by a simulator, called “jig”. The jig and the search engine exchange many messages among themselves, including the relevant passages returned by the jig, user cost spent on examining the documents, etc. In this work, we don’t apply any dynamic search algorithms to model these interactions. Instead, we produce a basic re-ranking baseline. Our algorithm starts at taking in an initial query from the simulator. During the search, we collect the relevance feedback from the simulator and use them to re-rank the initial retrieval results. Our algorithm terminates itself automatically when it senses that the user has gained enough information about the search topic or that no further relevant documents can be retrieved for the user.