Patent retrieval – A question of access

The development of models and systems in Information Retrieval (IR) has been driven by the empirical measurement of effectiveness. However, in recall-oriented domains such as patent search where there is a significant cost of missing a relevant document, standard IR effectiveness measurement only reveals part of the truth. Since credible estimates of recall are not available, it is difficult to evaluate or design systems for this domain. Here, we propose a measure of corpus access, retrievability, and show using four large patent corpora that it can be used both to evaluate models for patent retrieval and also the corpora themselves for the ease with which a document can be retrieved.

► Information retrieval and the measurement of effectiveness.
► Patent search as a recall-oriented domain.
► A measure of corpus access and retrievability in the patent domain is proposed.
► Evaluation of models of patent retrieval and for the corpora themselves is described.