So many topics, so little time

Abstract

In the context of creating large scale test collections, the present paper discusses methods of constructing a patent test collection for evaluation of prior art search. In particular, it addresses criteria for topic selection and identification of recall bases. These issues arose while organizing the CLEF-IP evaluation track and were the subject of an online discussion among the track's organizers and its steering committee. Most literature on building test collections is concerned with minimizing the costs of obtaining relevance assessments. CLEF-IP can afford to have large topics sets since relevance assessments are generated by exploiting existing manually created information. In a cost-benefit analysis, the only issue seems to be the computing time required by participants to run (tens or hundreds of) thousands of queries. This document describes the data sets and decisions leading to the creation of the CLEF-IP collection.

CLEF-IP 2009: Retrieval Experiments in the Intellectual Property Domain

“…For an in-depth discussion of topic selection for Clef Ip see [13]. This section describes the two types of relevance assessments used in Clef Ip2009: (1) assessments automatically extracted from patent citations as well as (2) manual assessments by patent experts.…”