话语
计算机科学
动机式访谈
变压器
人工智能
自然语言处理
心理学
语音识别
训练集
机器学习
作者
Chandankumar Johakhim Patel,Cassandra Mae Patel,Joshua C Patel
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-02-05
卷期号:: 1-6
标识
DOI:10.1109/acdsa67686.2026.11467751
摘要
Motivational Interviewing (MI) is a collaborative, goal-oriented communication method used in medical education and practice to enhance patient motivation and support behavior change. Accurate classification of clinician utterances—such as Open Questions, Closed Questions, Scaled Questions, and Reflections—is critical for evaluating MI quality and providing automated training feedback. However, linguistic complexity and class imbalance present challenges for automated analysis. This study compares the performance of advanced Transformer-based language models ALBERT, RoBERTa, Deberta, Electra and T5, for utterance classification in real-world clinician–patient transcripts. Models were trained on labeled utterances and evaluated using Precision, Recall, and F1 metrics. Results show that RoBERTa as the most reliable and effective model for MI utterance classification, followed closely by ALBERT which offers an optimal balance between performance and efficiency. DeBERTa and ELECTRA are viable alternatives for moderateperformance needs, while T5 is unsuitable for this task without further architectural adaptation. These insights highlight the importance of selecting architectures aligned with the discriminative nature of behavioral health analytics.
科研通智能强力驱动
Strongly Powered by AbleSci AI