The SO-PMI-IR method proposed by  is a simple and effective method for predicting the polarity of words, but it suffers from three limitations: 1) polar paradigm words are selected by intuition; 2) few search engines nowadays officially support the NEAR operator; 3) the NEAR operator considers the co-occurrence within 10 words, which incurs some noises. In this paper, for predicting the polarity of Chinese adjectives automatically, we follow the framework of the SO-PMI-IR method in . However, by using only two polarity indicators, [bu](not) and [youdian](a bit), we overcome all the limitations listed above. To evaluate our method, a test set is constructed from two Chinese human-annotated polarity lexicons. We compare our method with Turney's in details and test our method on different settings. For Chinese adjectives, the performance of our method is satisfying. Furthermore, we perform noise analysis, and the relationship between the magnitude of SO-PMI-IR and accuracy is also analyzed. The results show that our method is more reliable than Turney's method in predicting the polarity of Chinese adjectives.
|Name||Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)|
|Conference||14th Workshop on Chinese Lexical Semantics, CLSW 2013|
|Period||10/05/13 → 12/05/13|
- Chinese adjective
- sentiment analysis
- Theoretical Computer Science
- Computer Science(all)