Sentence-Level Semantic Textual Similarity Using Word-Level Semantics

Abstract

Estimating semantic textual similarity between sentences is indispensable for many information retrieval tasks. Traditional lexical similarity measures cannot compute the similarity beyond a trivial level. Moreover, they only can capture textual similarity, but not semantic. Researchers proposed methods using a variety of approaches. In this paper, we propose a novel method for semantic textual similarity that leverages word-level semantics to compute the sentence-level semantic similarity. We introduced two new semantic similarity measures based on word-embedding models trained on two different corpora. Apart from these, another semantic similarity measure is also introduced using the word sense comparison. The similarity score between the sentence-pair is then computed by applying a linear ranking approach to all proposed measures with their importance estimated employing a linear regression model. We conducted experiments using the SemEval Semantic Textual Similarity (STS-2017) test collections. The experimental results demonstrated that our method is effective for measuring semantic textual similarity and outperformed some known related methods.