An Improved Semantic Similarity Measure for Word Pairs

The problem of measuring semantic similarity between word pairs has been considered as a fundamental operation in natural language processing, such as information retrieval, word sense disambiguation, etc. Nevertheless, developing a computational method capable of generating satisfactory results close to what humans would perceive is still a difficult task somewhat owed to the subjective nature of similarity. In this paper, we suggest an improved semantic similarity measure between words. It considers the structure of WordNet 3.0 based on DAG, and combines the improved distance-based measure and the information-based measure. The correlation value has been achieved between results by the proposed semantic similarity measure and human ratings reported by Miller and Charles for the dataset of 30 pairs of noun, which is higher than some other reported measures for the same dataset.