Guided Intelligent Collection of Elements of Doctor-Patient Dispute Cases

Human-machine dialogue systems face sig-nificant challenges when handling doctor-patient dis-putes, primarily due to the brevity and informality of user input. Traditional statistical and machine learning models often struggle to perform well with incomplete syntax and colloquial language. To address this issue, this paper proposes a guided intelligent collection al-gorithm based on the RASA framework, for improving the identification and extraction of key information in medical dispute dialogues. To manage the colloquial expressions and grammat-ical omissions in user text, a multi-condition fusion BiLSTM-CRF joint model was developed, enabling simultaneous intent recognition and entity extraction with improved precision and efficiency. Furthermore, a multi-class classification model incorporating a self-attention mechanism was introduced to overcome the limitations of RASA's supervised learning-based dia-logue management, greatly improving performance and accuracy in managing complex doctor-patient dispute scenarios. Experimental results show significant enhancements in intent recognition accuracy and entity extraction Fl scores. The self-attention-based multi-turn manage-ment approach also improves action prediction accu-racy compared to RASA’ s default methods, demon-strating the proposed method's effectiveness in medical disputes.