Examining assessor attributes at HARD 2005

The TREC HARD (High accuracy Retrieval from Documents) track was motivated to investigate techniques for personalised retrieval of documents. Through the use of a limited dialogue with the TREC assessors, the track facilitated the gathering and exploitation of information about the assessors' personal search context (e.g. knowledge of search topic) which could be used to improve document retrieval. In this paper we describe experiments, run within the context of the 2005 HARD track, which indicate that assessor attributes such as familiarity, interest and confidence when searching a topic can help determine when the utilisation of automatic query expansion improves retrieval over the original document ranking.