Skip to Main Content
This paper presents a new model of measuring semantic similarity in the taxonomy of WordNet. The model takes the path length between two concepts and IC value of each concept as its metric, furthermore, the weight of two metrics can be adapted artificially. In order to evaluate our model, traditional and widely used datasets are used. Firstly, coefficients of correlation between human ratings of similarity and six computational models are calculated, the result shows our new model outperforms their homologues. Then, the distribution graphs of similarity value of 65 word pairs are discussed our model having no faulted zone more centralized than other five methods. So our model can make up the insufficient of other methods which only using one metric(path length or IC value) in their model.