COMBISIM: AN EFFECTIVE COMBINED MEASURE FOR THE WORD SIMILARITY PROBLEM

Abstract

Semantic word similarity measurement (WSM) is a fundamental research problem with numerous applications in natural language processing. However, accurately measuring semantic similarity between words remains a significant challenge. In addressing this issue, this paper introduces a novel technique named CombiSim. CombiSim represents a breakthrough by integrating two advanced WSM methods: word embeddings-based semantic vector space and WordNet-based approaches. By combining the strengths of both methods, CombiSim not only enhances the accuracy of semantic similarity measurement but also extends its applicability across multiple languages, including English and Vietnamese. Experimental results demonstrate that CombiSim achieves superior performance compared to traditional methods solely based on WordNet or word embeddings.