Scholarlink

Depression stands as a major concern within the domain of mental health, underscoring the critical importance of assessing its severity. Timely detection and appropriate treatment of severe depression can significantly mitigate the risk of fatalities among affected individuals. Mental health professionals, including psychiatrists and psychologists, often conduct clinical interviews comprising structured yet simple conversations to determine depression levels in patients. Additionally, standardized psychometric questionnaires like PHQ-8 offer a quantifiable score directly correlating with depression severity. Unfortunately, the availability of psychiatrists and psychologists proficient in diagnosing and treating such cases remains limited. To tackle these challenges, this research work proposes the automation of the PHQ-8 questionnaire through artificial intelligence and Natural Language Processing (NLP) techniques, leveraging semi-clinical interviews with patients. Various machine learning (Ridge Regression, Random Forest, XGBoost, LightGBM, CatBoost) and deep learning methods (GRU and LSTMs) are evaluated and compared using textual transcripts sourced from the Extended Distress Analysis Interview Corpus (E-DAIC). Moreover, the proposed models demonstrate superior performance (lower RMSE) compared to baseline results as provided by the AVEC 2019 initiative utilizing the same dataset for the ‘Detecting Depression with AI’ sub-challenge.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Depression Detection by Extracting PHQ-8 Scores from Semi-clinical Interviews Using Machine Learning and Deep Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ishan Mangotra,
  • Amita Jain,
  • Minni Jain,
  • Mamta Sood,
  • Swati Kedia Gupta,
  • Nishtha Chawla,
  • Vaibhav Patil

摘要

Depression stands as a major concern within the domain of mental health, underscoring the critical importance of assessing its severity. Timely detection and appropriate treatment of severe depression can significantly mitigate the risk of fatalities among affected individuals. Mental health professionals, including psychiatrists and psychologists, often conduct clinical interviews comprising structured yet simple conversations to determine depression levels in patients. Additionally, standardized psychometric questionnaires like PHQ-8 offer a quantifiable score directly correlating with depression severity. Unfortunately, the availability of psychiatrists and psychologists proficient in diagnosing and treating such cases remains limited. To tackle these challenges, this research work proposes the automation of the PHQ-8 questionnaire through artificial intelligence and Natural Language Processing (NLP) techniques, leveraging semi-clinical interviews with patients. Various machine learning (Ridge Regression, Random Forest, XGBoost, LightGBM, CatBoost) and deep learning methods (GRU and LSTMs) are evaluated and compared using textual transcripts sourced from the Extended Distress Analysis Interview Corpus (E-DAIC). Moreover, the proposed models demonstrate superior performance (lower RMSE) compared to baseline results as provided by the AVEC 2019 initiative utilizing the same dataset for the ‘Detecting Depression with AI’ sub-challenge.