홈 › Papers

Counselling Summarization using Mental Health Knowledge guided Utterance Filtering

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 diagnose and become a complex problem to summarize later. A structured counselling conversation may contain discussions on symptoms or reasons for mental health issues, the discovery of the patient's behaviour and habits, and the rest of the conversation is generally chit-chat. We call these psychotherapy ingredients to be counselling components. We aim for the task of mental health counselling summarization exploiting domain knowledge. We curated a new dataset, MEMO, where we annotate 12.9K utterances for counselling components and reference summaries for each dialogue. Further, we propose ConSum: a novel counselling-component guided summarization model. The model works in three independent modules. First, it uses mental health domain knowledge to filter utterances by utilising the Patient Health Questionnaire (PHQ-9), while, the second and third modules work on the classification of counselling components. At last, we propose a problem specific Mental Health Information Capture (MHIC) evaluation metric for counselling summaries. Our extensive ablation study shows that we improve on the scores and better language semantics. We comprehensively analyse the generated summaries to investigate the capturing of core components of psychotherapy. In last, we discuss the uniqueness in mental health counselling summarization in contrast with medical conversation.

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반