Infer the Semantic Orientation of Words by Optimizing Modularity

This paper proposes a novel algorithm, which attempts to attack the problem of word semantic orientation computing by optimizing the modularity of the word-to-word graph. Experimental results indicate that proposed method has two main advantages: (1) by spectral optimization of modularity, proposed approach displays a higher accuracy than other methods in inferring semantic orientation. For example, it achieves an accuracy of 88.8% on the HowNet-generated test set; (2) by effective usage of the global information, proposed approach is insensitive to the choice of paradigm words. In our experiment, only one pair of paradigm words is needed.