Dimensional sentiment analysis of traditional Chinese words using pre-trained Not-quite-right Sentiment Word Vectors and supervised ensemble models

摘要

This work focuses on two specific types of sentimental information analysis for traditional Chinese words, i.e., valence represents the degree of pleasant and unpleasant feelings (i.e., sentiment orientation), and arousal represents the degree of excitement and calm (i.e., sentiment strength). To address it, we proposed supervised ensemble learning models to assign appropriate real valued ratings to each word on two sentimental dimensions, incorporating pre-trained semantic and sentiment word vectors into the models. Experimental results on IALP 2016 Shared Task data set showed that our method achieves desirable performance in predicting real valued ratings of given words in valence subtask and forecasting the order of words in arousal subtask. Specifically, for the valence subtask, our system ranks the first in terms of MAE measure.