Measuring Semantic Similarity of Word Pairs Using Path and Information Content

Measuring semantic similarity of word pairs is a popular topic for many years. It is crucial in many applications, such as information extraction, semantic annotation, question answering system and so on. It is mandatory to design accurate metric for improving the performance of the bulk of applications relying on it. The paper presents a new metric for measuring word sense similarity using path and information content. Different from previous works, the new metric not only reflects the semantic density information, but also reflects the path information. It is evaluated on the dataset provided by Rubenstein and Goodenough. Experiments demonstrate that the coefficient based on our proposed metric with human judgment is 0.8817, which is significantly outperformed than other existing methods.