Construction and Validation of a Fine-Grained Emotional Attitude Lexicon

Abstract: In response to the frequent occurrence of inflammatory incidents on social media, governments, and enterprises require more precise and in-depth analytical tools. Current sentiment lexicons suffer from issues of coarse granularity and limited keyword coverage. This paper initially expands the categories of sentiments based on cognitive-evaluative theory. Subsequently, experts screened and categorized keywords from a corpus of topical events on social media to compile the lexicon. Ultimately, a fine-grained sentiment lexicon comprising 50 categories was developed. We conducted manual testing and event analysis to evaluate the lexicon’s accuracy and efficiency in identifying emotions. Using manual classification as the reference standard, the average accuracy was 88%. We also analyzed public sentiments regarding the ‘Haitian Soy Sauce Double-Standard Incident’ and the ‘Chengdu Girl Bitten by Dog Incident,’ with results aligning with the valence of these events. Additionally, we observed heterogeneity in the temporal changes of fine-grained emotions. These findings support the utility of fine-grained sentiment analysis in understanding and responding to complex public opinion environments.