PsyFusion-RAG: A Multi-source Knowledge Fusion Model for Psychological Dialogue Generation

Current mental health consultation systems are often limited to generic psychological question-answer interactions or basic diagnostic functions, failing to achieve systematic diagnosis and intervention based on professional counseling methodologies. This paper proposes PsyFusion-RAG, a multi-source heterogeneous knowledge–augmented framework for psychological consultation. The framework performs multi-source knowledge modeling and cross-structure fusion to integrate heterogeneous knowledge representations, thereby forming a collaborative retrieval mechanism for precise and semantically consistent knowledge integration. Furthermore, it optimizes retrieval performance through knowledge re-ranking and structured prompting, ensuring efficiency and coherence in multi-source knowledge utilization. Experimental results on multi-symptom psychological consultation datasets demonstrate that PsyFusion-RAG achieves substantially higher high-score rates and optimal response proportions, highlighting its robustness and potential for supporting complex and context-aware psychological decision-making.