Emotion Flow: Website URL
This study proposes a multi-stage empathetic AI system designed to support users experiencing mental distress by combining emotion classification with emotionally supportive response generation. We first evaluate a DistilBERT-based classifier for detecting emotional states—such as anxiety, depression, and suicidal ideation—in Reddit posts. Subsequently, we assess the effectiveness of Reinforcement Learn- ing from Human Feedback (RLHF) in enhancing the empathy and help- fulness of AI-generated responses. An optional component investigates the mood-regulating potential of personalized music recommendations. Through these experiments, we aim to validate the model’s classification performance, user preferences for empathetic interaction, and the thera- peutic potential of emotion-aware content delivery.
emotion classification, empathetic AI, RLHF, mental health, music recommendation, DistilBERT, large language models, Gemini
