A Sentence Similarity Measure Based on Conceptual Elements

There has always been a growing interest in sentence similarity measure for practical NLP tasks using various state-of-art NLP methods. Some of the widely used methods in measuring sentence similarity are lexical semantics, deep learning, neural networks, ontology, statistical models, graph based model and etc. Based on our findings, one of the main drawbacks in using these methods is not able to resolve word ambiguity where one word can have different interpretations in different sentences. In this paper, we present a sentence similarity measure by representing the sentences in conceptual elements to measure the semantic similarity between sentences. We used Microsoft Paraphrase Corpus (MSR) and Quora question pairs dataset to evaluate the performance. The study concludes that we were able to use conceptual elements to measure sentence similarity with the highest micro averaged precision of 0.71