Clairvoyance Corporation Experiments in the TREC 2003 High Accuracy Retrieval from Douments (HARD) Track.

The Clairvoyance team participated in the HARD Track, submitting fifteen runs. Our experiments focused primarily on exploiting user feedback through clarification forms for query expansion. We made limited use of the genre and related text metadata. Within the clarification form feedback framework, we explored the cluster hypothesis in the context of relevance feedback. The cluster hypothesis states that closely associated documents tend to be relevant to the same requests [Van Rijsbergen, 1979]. With this in mind we investigated the impact on performance of exploiting user feedback on groups of documents (i.e., organizing the top retrieved documents for a query into intuitive groups through agglomerative clustering or document-centric clustering), as an alternative to a ranked list of titles. This forms the basis for a new blind feedback mechanism (used to expand queries) based upon clusters of documents, as an alternative to blind feedback based upon taking the top N ranked documents, an approach that is commonly used.