Eigenvalue based features for semantic sentence similarity

Abstract

Due to its increasing importance, the semantic sentence similarity is getting more attention among natural language processing researchers during recent years. To the best of our knowledge, previous studies on the task have not exploited the eigenvalue analysis on their systems. In this paper we approach the sentence similarity task through eigenvalue analysis. We will propose a simple but efficient new aligner and introduce three new features for the task. Two of our proposed features are based on the eigenvalue analysis. Finally, we will show the significance of our proposed aligner and features through experiments. Specifically, we will show that our features outperform the STS2015 benchmarks for semantic sentence similarity.