Comparative Study of Techniques used for Word and Sentence Similarity

This study is intended to analyze the methods used to test resemblance of sentences. For many Natural Language Processing applications such as text grouping, information recovery, brief reaction reviewing, machine learning, passage summary and text categorization, measuring resemblance between sentences is a vital activity. In this paper, we classify the approaches to measuring the resemblance of sentences based on the methods implemented into three groups. The most frequently used methods to finding phrase resemblance are word-to-word based, structure-based, and vector-based. Centered on a particular viewpoint, each approach tests the interaction between short texts. Furthermore, to provide a full view of this problem, datasets that are often used as benchmarks for testing techniques in this field are added. Better outcomes are obtained through methods that incorporate more than one viewpoint. In addition, resemblance of sentences is based on the correspondence of their meanings that tests the semantic resemblance between two concepts, words or sentences needs further research.