Stance analysis for debates on traditional chinese medicine at Tianya forum

Abstract

Internet and social media devices have created a new public space for debates on societal topics. This paper applies text mining methods to conduct stance analysis of on-line debates with the illustration of debates on traditional Chinese medicine (TCM) at one famous Chinese BBS Tianya Froum. After crawling and preprocessing data, logistic regression is adopted to get a domain lexicon. Words in the lexicon are taken as features to automatically distinguish stances. Furthermore a topic model latent Dirichlet allocation (LDA) is utilized to discover shared topics of different camps. Then further analysis is conducted to detect the focused technical terms of TCM and human names referred during the debates. The classification results reveal that using domain discriminating words as features of classifier outperforms taking nouns, verbs, adjectives and adverbs as features. The results of topic modeling and further analysis enable us to see how the different camps express their stances.