Domain-dependent Dictionary Construction for Sentiment Analysis

The construction of a sentiment dictionary is essential for the text sentiment analysis. In this paper, news articles of CGTN were collected as the object of experimental data and pre-processed by removing stopping words and non-emotional words. TF-IDF was used to form a list of likely-to-be sentiment seed words. After determining the word sentiment polarity by human coding which were used as criteria to determine the polarity of other candidate emotion words of the whole texts, researchers used SO-PMI algorithm and finally constructed a customized sentiment dictionary containing 366 words in a specific domain. Experimental results show that the sentiment dictionary was feasible and the method employed in this paper may contribute to the future research on sentiment analysis.