Stance Detection in Chinese Microblogs

[Purpose/significance] The paper introduces a new approach to automatically detect stance in Chinese microblogs by building a serial combination model based on Sentiment Weighted Algorithm and Naive Bayes (SWNB model). [Method/process] Firstly, this paper used the SWNB model to simplify complex sentences by using a defined complex sentence pattern, assigning a sentiment weight to each microblog according to calculation rules, and optimizing sentiment weight by detecting the presence of the target’s associated entities; thus, we could classify microblogs into those containing any stance or with no stance at all. Secondly, the SWNB model extracted some feature words and used Naive Bayes to classify the microblogs labeled as FAVOR or AGAINST. [Result/conclusion] Experiments show that this model can comprehensively process complex sentences, target-related entities and linguistic context.