Fine-grained Speech Sentiment Analysis in Chinese Psychological Support Hotlines Based on Large-scale Pre-trained Model

• DOI: 10.48550/arxiv.2405.04128
• Publication Date: 2024-05-07
• Authors: Zhonglong Chen; Changwei Song; Yining Chen; Jianqiang Li; Guanghui Fu; Yongsheng Tong; Qing Zhao
• Keywords: Hotline; Sentiment Analysis

Suicide and suicidal behaviors remain significant challenges for public policy healthcare. In response, psychological support hotlines have been established worldwide to provide immediate help individuals in mental crises. The effectiveness of these largely depends on accurately identifying callers' emotional states, particularly underlying negative emotions indicative increased suicide risk. However, the high demand interventions often results a shortage professional operators, highlighting need an effective speech emotion recognition model. This model would automatically detect analyze emotions, facilitating integration into hotline services. Additionally, it enable large-scale data analysis interactions explore phenomena across populations. Our study utilizes from Beijing hotline, largest China. We analyzed 105 callers containing 20,630 segments categorized them 11 types emotions. developed fine-grained multi-label classification using pre-trained experiments indicate that achieves maximum F1-score 76.96%. shows limited efficacy task, with best achieving only 41.74% weighted F1-score. conducted error this discussed potential future improvements, considered clinical application possibilities our study. All codes are available.