Chinese Psychological Counseling Corpus Construction for Valence-Arousal Sentiment Intensity Prediction

While several dimensional sentiment resources with valence-arousal annotations exist, they are not appropriate for domain-specific sentiment applications, especially psychological counseling texts which usually contain more high-negative and high-arousal words and phrases. This study (1) builds a Chinese valence-arousal resource in the psychological counseling domain called CVAT-PC, the first text corpus including 2,400 multi-sentence texts for psychological counseling with valence-arousal ratings manually annotated based on the Self-Assessment Manikin (SAM) rating scale, and (2) evaluates the proposed resource using different transformer-based models, including ELECTRA, XLNet, BERT, MacBERT and RoBERTa. We plan to release the entire CVAT-PC corpus with fully annotated valence-arousal ratings to facilitate future development in related research areas.