The Analysis and Improvement about Word Similarity Computing Method Based on HowNet

Abstract:

Word similarity computing is widely used in many fields, such as question answer,text clustering and so on. This paper synthesizes previous research achievement and proposes an improved word similarity computing method based on Hownet. The improvement is mainly shown in the following aspects: We make a difference between the antonym and common words ,and define the value range of sememe similarity as [-1, 1]; what’s more we not only consider the hyponymy, but the depth and density as we compute the sememe similarity, then makes the result more distinctly; besides we introduce the concepts of weak and supplement sememes, and lower the effect of the weak and supplement sememes by reducing the weight of weak and supplement as we compute the concept similarity. Experiment shows that the improved method makes the word similarity more practically.

• Periodical: Applied Mechanics and Materials (Volumes 687-691)
• Pages: 1278-1281
• DOI: https://doi.org/10.4028/www.scientific.net/AMM.687-691.1278
• Online since: November 2014
• Authors: Hong Huang*, Shu Jie Cui
• Keywords: Density, HowNet, Word Similarity

References

[1] J. Peng, D.Q. Yang, S.W. Tang, T.J. Wang and J. Gao. A new similarity computing method based on concept similarity in Chinese text processing. Science in China Series F: Information Sciences, 2008, 519.

DOI: 10.1007/s11432-008-0103-4

[2] L.C. Yuan. Vari-gram language model based on word clustering. Journal of Central South University, 2012, 194.

[3] T. Xia. Chinese words semantic similarity computation of. Computer Engineering, 2007, 33 (6): 191-194.

[9] Lin D. An information-theoretic definition of similarity[C]/Proceedings of the 15th international conference on Machine Learning. 1998, 1: 296-304.

[4] B. Ge, F.F. Li, Guo Road, et al. Research on the application of. method, computing semantic similaritybased on HowNet 2010, 27 (9).