A Hierarchical Semantics-Aware Distributional Similarity Scheme

The context type and similarity calculation are two essential features of a distributional similarity scheme (DSS). In this paper, we propose a hierarchical semanticaware DSS that exploits semantic relation words as extra context information to guide the similarity calculation. First, we define and extract five types of semantic relations, and then develop relation-based similarities from the distributional similarities among the top-ranked relation words. Finally, we integrate various similarities using learning-to-rank technique. Experiments show that semantic relations are beneficial to predicting accurate similarity. On 6904 pairwise similarity comparisons, the predictive accuracy of our approach reaches 83.9%, which significantly outperforms the baseline approaches. We also conduct intrinsic analysis by varying the quality of semantic relations and the usage of individual similarities.