Evaluation of Computational Linguistic Techniques for Identifying Significant Topics for Browsing Applications
Proceedings of the Second International Conference on Language Resources and Evaluation (LREC 2000)
Abstract
Evaluation of natural language processing tools and systems must focus on two complementary aspects: first, evaluation of the accuracy of the output, and second, evaluation of the functionality of the output as embedded in an application. This paper presents evaluations of two aspects of LinkIT, a tool for noun phrase identification linking, sorting and filtering. LinkIT [Evans 1998] uses a head sorting method [Wacholder 1998] to organize and rank simplex noun phrases (SNPs). LinkIT is to identify significant topics in domain-independent documents. The first evaluation, reported in D.K.Evans et al. 2000 compares the output of the Noun Phrase finder in LinkIT to two other systems. Issues of establishing a gold standard and criteria for matching are discussed. The second evaluation directly concerns the construction of the browsing application. We present results from Wacholder et al. 2000 on a qualitative evaluation which compares three shallow processing methods for extracting index terms, i.e., terms that can be used to model the content of documents. We analyze both quality and coverage. We discuss how experimental results such as these guide the building of an effective browsing applications.