2015 International Symposium of Computer Application and Information Technology (IsCait 2015)

268 / 1971-01-01 00:00:00

Chinese Short Text Semantic Similarity Calculation Method Based On Hownet

HowNet; Semantic similarity; Short text

Current semantic similarity calculation methods which are based on HowNet have been proposed a lot. But these methods fail to give attention to the density of sememes divided and the feature of first basic sememe and following sememe described by Knowledge Database Mark-up Language in HowNet. To resolve these defects above, this article firstly takes into account the density of sememes divided in sememe treelike hierarchy system and proposes a new method which calculates the sub-nodes’ number of the sememe’s parent node. Meanwhile, a dynamic concept similarity calculation method which focuses on first basic sememe and following semem is also proposed in this article. At last, the methods above are applied to clustering algorithm by simplifying the text similarity calculation. The final experimental results show that the word similarity in this article is more reasonable than other existing methods. And the clustering results show well in recall rate and accuracy rate. All these verify the legitimacy and effectiveness of the method in this article.