The psychotherapy intervention technique is a multifaceted conversation between a therapist and a patient. Unlike general clinical discussions, psychotherapy's core components (viz. symptoms) are hard to distinguish, thus becoming a complex problem to summarize later. A structured counseling conversation may contain discussions about symptoms, history of mental health issues, or the discovery of the patient's behavior. It may also contain discussion filler words irrelevant to a clinical summary. We refer to these elements of structured psychotherapy as counseling components. In this paper, the aim is mental health counseling summarization to build upon domain knowledge and to help clinicians quickly glean meaning. We create a new dataset after annotating 12.9K utterances of counseling components and reference summaries for each dialogue. Further, we propose ConSum, a novel counseling-component guided summarization model. ConSum undergoes three independent modules. First, to assess the presence of depressive symptoms, it filters utterances utilizing the Patient Health Questionnaire (PHQ-9), while the second and third modules aim to classify counseling components. At last, we propose a problem-specific Mental Health Information Capture (MHIC) evaluation metric for counseling summaries. Our comparative study shows that we improve on performance and generate cohesive, semantic, and coherent summaries. We comprehensively analyze the generated summaries to investigate the capturing of psychotherapy elements. Human and clinical evaluations on the summary show that ConSum generates quality summary. Further, mental health experts validate the clinical acceptability of the ConSum. Lastly, we discuss the uniqueness in mental health counseling summarization in the real world and show evidences of its deployment on an online application with the support of mpathic.ai

本研究旨在通过基于领域知识建立心理健康辅导摘要来帮助顾问快速获取含义，我们创建了一个新的数据集，包括关于辅导组件的12.9K次言语注释及参考摘要，并提出了ConSum，一种新颖的辅导组件引导摘要模型，表现出更好的性能和生成的摘要具有连贯、语义和完整的特点。

运用心理健康知识引导话语过滤的辅导摘要