A User-in-the-Loop Process for Investigational Search: Foreseer in TREC 2013 Microblog Track
Cheng Li, Yue Wang, Qiaozhu Mei · Text REtrieval Conference · 2013
Traditionally, ad hoc retrieval aims at satisfying an information need with a few highly relevant documents. This highprecision approach works well for simple and clear queries. When the information need becomes complex, a few topranked documents may not provide satisfactory answer. As a result, the user adapts to reformulate the query to explore more relevant information; the search engine evolves to diversify results to cover the user’s real information need. In cases where the user puts emphasis on high recall of the results, the search process becomes increasingly laborious. We propose a retrieval system that interacts with the user to obtain high precision and high recall search results, while minimizing the user eort. It iteratively explores the collection by a series of queries to optimize the recall, and renes an active learning classier to maintain the precision. We