Sentence similarity on structural representations

Abstract

Most previous approaches used various kinds of plain similarity features to represent the similarity of a sentence pair, and one of its limitations is its weak representation ability. This paper introduces the relational structures representation (shallow syntactic tree, dependency tree) to compute sentence similarity. Experimental results manifest that our approach achieves higher performance than that only uses plain features.