Finding and Evaluating Sets of Nearest Neighbours
Julie Weeds, David James Weir · 2003
In this paper, we consider two applications of distributional similarity measures, probability estimation and prediction of semantic similarity. We investigate whether high performance in one application area is correlated with high performance in the other. This work also provides an evaluation of two state-of-the-art distributional similarity measures and introduces a variant of one. Further, we overcome statistical biases in the standard pseudo-disambiguation task and look at the effect of word and co-occurrence frequency on the performance of the measures.