Chinese-Vietnamese bilingual news unsupervised sentiment classification based on word-weighted JST algorithm

EasyChair Preprint 898

Abstract

Sentiment classification is an important part of text sentiment analysis. Usually sentiment classification methods are supervised models or semi-supervised models, but well-labeled corpora are often difficult to obtain which makes it difficult to classify emotions. In the Chinese-Vietnamese bilingualism, the lack of basic emotional resources leads to the accuracy of cross-language text-level sentiment classification is not as accurate as that in a single-language environment. Unsupervised sentiment classification can improve the accuracy of sentiment classification by constructing emotional dictionary and incorporating word weights. In this paper, the Gibbs sampling process is firstly guided by the emotional polarity of the words in the sentiment dictionary, which helps the model to predict the vocabulary emotion and subject distribution. Then this paper proposes an unsupervised sentiment classification method based on the word weighted Joint Sentiment/topic(JST) model algorithm. Through the linear combination of three kinds of weighted methods, the emotional weight was integrated into the vocabulary. This approach improves the impact of emotional words in the Gibbs sampling process, and ultimately improves the accuracy of sentiment classification. Experiments show that the word-weighted JST model combined with prior knowledge has greatly improved the accuracy of Vietnamese sentiment classification.

Keyphrases: Chinese-Vietnamese, Emotional Dictionary, sentiment classification, unsupervised, weighting method

@booklet{EasyChair:898,
  author    = {Siqi Lin and Zhengtao Yu and Shengxiang Gao and Junjun Guo and Yulong Wang},
  title     = {Chinese-Vietnamese bilingual news unsupervised sentiment classification based on word-weighted JST algorithm},
  howpublished = {EasyChair Preprint 898},
  year      = {EasyChair, 2019}}