How Does Exploration Impact IR Performance in Large Document Collections

A significant problem for IR researchers is how to efficiently handle large collections of electronic documents. Manual review is time consuming and expensive. Automated methods can be imprecise and fail to yield relevant documents in the retrieval set. This problem is receiving significant attention in two major domains: the legal community, with the increase in search of electronic documents in litigation (eDiscovery), and in the medical community, with the increase of mandates for ehealth systems such as electronic patient records (EMR and EHR) and health informatics. This paper examines how the construct of exploration may be implemented as a methodology to improve user performance when searching and sorting through large electronic document collections by facilitating context and content understanding through multiple iterations. The study reported in this paper examines the research questions of: How does exploration impact IR performance, and how can exploration be implemented to achieve improvement in IR results? The study examines the correlation between an individual user’s exploration of iterated sample selections from a large corpus of electronic documents and the individual’s IR performance. Our findings support that: (1) IR performance can be manipulated by using an exploration method, (2) Time spent exploring a collection is correlated with performance in both recall and precision, and (3) Number of documents viewed in a collection is correlated with performance in precision.