Human aided computer assessment for exhaustive search

Human-computer interaction has been proposed as a means for more effectively dealing with the challenges posed by information retrieval. However, within the hierarchy of information needs, certain kinds of needs are only poorly handled by existing techniques. Information needs requiring thoroughness (exhaustive research, information discovery) are not well served by existing models of human-computer information retrieval (HCIR). Information needs such as these are commonplace in legal, patent, medical and intelligence searches. In such applications, high recall with high precision is the primary consideration, and the standard model, which treats human interaction as a kind of post-processing filter, does not yield a system with desired characteristics. In this paper, we propose an alternative model of HCIR which yields systems whose properties are more closely aligned with the needs of the exhaustive research task, and describe an implementation of such a system, demonstrating its effectiveness over the standard model in a task that models real-world conditions.