TY - GEN
T1 - Automatically predicting the polarity of Chinese adjectives: Not, a bit and a search engine
AU - Xu, Ge
AU - Huang, Chu-ren
AU - Wang, Houfeng
PY - 2013/12/1
Y1 - 2013/12/1
N2 - The SO-PMI-IR method proposed by [1] 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 [1]. 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.
AB - The SO-PMI-IR method proposed by [1] 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 [1]. 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.
KW - Chinese adjective
KW - polarity
KW - sentiment analysis
UR - http://www.scopus.com/inward/record.url?scp=84893420614&partnerID=8YFLogxK
U2 - 10.1007/978-3-642-45185-0_48
DO - 10.1007/978-3-642-45185-0_48
M3 - Conference article published in proceeding or book
SN - 9783642451843
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 453
EP - 465
BT - Chinese Lexical Semantics - 14th Workshop, CLSW 2013, Revised Selected Papers
T2 - 14th Workshop on Chinese Lexical Semantics, CLSW 2013
Y2 - 10 May 2013 through 12 May 2013
ER -