The Impact of Contrastive Corpora for Term Relevance Measures

Estimate the relevance of extracted terms through indices based on contrastive corpora is acknowledged to be efficient. Unfortunately, there is no ground rules to help practitioners and researchers to choose adequate contrastive corpora. In this paper, we present an extensive analysis of different options of contrastive corpora for seven different target corpora. It is our goal to show that the impact of such choice is not as harmful as feared, and such impact tends to be diminished as the number of contrastive corpora increases.