Semantic Textual Similarity as a Service

Abstract

Ensembling well performing models has been proved to outperform individual models in semantic textual similarity task; however, employing existing models still remains a challenge. In this paper, we tackle this issue by providing a service oriented system to index a text similarity model using RESTful services. We also propose a baseline approach, based on an effective penalty-award weighting schema and word-level edit distance, in which pairs of sentences are divided into two main categories based on the number of substitution, insert, and delete required to convert the first sentence to the second one. It is debated that, when the word-level edit distance is very small, it is wiser to measure dissimilarity than similarity. Using knowledge bases along with common natural language processing tools, the proposed method tries to enhance the accuracy of measuring similarity between two sentences. We compared the proposed method with existing approaches, and we found that it produces promising results. Our source code is freely available on GitLab.