서브메뉴
검색
Generative, Human-Centered AI in Mental Health: Supporting Therapeutic Fidelity and Empathy in Interactions
Generative, Human-Centered AI in Mental Health: Supporting Therapeutic Fidelity and Empathy in Interactions
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105322
- ISBN
- 9798297672130
- DDC
- 615.8
- 서명/저자
- Generative, Human-Centered AI in Mental Health: Supporting Therapeutic Fidelity and Empathy in Interactions
- 발행사항
- [Sl] : The Pennsylvania State University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 173 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-04, Section: B.
- 주기사항
- Advisor: Abdullah, Saeed.
- 학위논문주기
- Thesis (Ph.D.)--The Pennsylvania State University, 2025.
- 초록/해제
- 요약The effective application of Artificial Intelligence (AI) in mental healthcare-particularly for supporting clinicians and enhancing therapy for conditions such as Post-Traumatic Stress Disorder (PTSD)-is impeded by several critical challenges. Earlier work leading up to this dissertation focused on patient-centric mental health assessment using privacypreserving machine learning techniques, including federated learning for depression detection and differential privacy for dementia risk prediction. While these approaches addressed data privacy concerns, they offered limited utility for clinician-facing tools and were constrained by restricted access to real-world datasets.Existing research in AI for mental health has largely prioritized patient diagnosis, with relatively little emphasis on supporting clinician workflows during therapy. This thesis responds to that gap by presenting a human-centered machine learning framework focused on Prolonged Exposure (PE) therapy for PTSD.It introduces three core contributions:1. Synthetic Data Generation: "Thousand Voices of Trauma," a large-scale dataset (3,000 conversations, 500 personas) simulating PE therapy. Validated by clinical experts, the dataset demonstrates high fidelity to real sessions across structural, linguistic, and therapeutic dimensions.2. Empathetic and Efficient Modeling: Fine-tuning small language models (SLMs, 0.5B-5B parameters) on the synthetic "Trauma-Informed Dialogue for Empathy (TIDE)" dataset to generate context-aware, empathetic responses for real-time, inthe-moment emotional support in PTSD-related dialogues-serving as companion tools, not therapy replacements-and achieving significant gains over zero-shot baselines.3. Temporal Localization of Therapeutic Elements: A parameter-efficient approach for fine-tuning multimodal audio-language models to identify key components of PE therapy using soft-supervised, clinician-verified annotations.Together, these contributions offer validated datasets, efficient models, and deployment strategies to build scalable, empathetic AI tools that augment therapist training, provide in-the-moment support for users, and enhance mental healthcare delivery-always as a supplement to, not a replacement for, clinical care.
- 일반주제명
- Therapists
- 일반주제명
- Trauma
- 일반주제명
- Therapy
- 일반주제명
- Gender
- 일반주제명
- Empathy
- 일반주제명
- Linguistics
- 일반주제명
- Mental health
- 일반주제명
- Error analysis
- 일반주제명
- Large language models
- 일반주제명
- Mathematics
- 일반주제명
- Sociology
- 기본자료저록
- Dissertations Abstracts International. 87-04B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017360211
■00520260202105322
■006m o d
■007cr#unu||||||||
■020 ▼a9798297672130
■035 ▼a(MiAaPQ)AAI32289613
■035 ▼a(MiAaPQ)PennState26342sxb701
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a615.8
■1001 ▼aNagaraj, Suhas Bettapalli.
■24510▼aGenerative, Human-Centered AI in Mental Health: Supporting Therapeutic Fidelity and Empathy in Interactions
■260 ▼a[Sl]▼bThe Pennsylvania State University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a173 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-04, Section: B.
■500 ▼aAdvisor: Abdullah, Saeed.
■5021 ▼aThesis (Ph.D.)--The Pennsylvania State University, 2025.
■520 ▼aThe effective application of Artificial Intelligence (AI) in mental healthcare-particularly for supporting clinicians and enhancing therapy for conditions such as Post-Traumatic Stress Disorder (PTSD)-is impeded by several critical challenges. Earlier work leading up to this dissertation focused on patient-centric mental health assessment using privacypreserving machine learning techniques, including federated learning for depression detection and differential privacy for dementia risk prediction. While these approaches addressed data privacy concerns, they offered limited utility for clinician-facing tools and were constrained by restricted access to real-world datasets.Existing research in AI for mental health has largely prioritized patient diagnosis, with relatively little emphasis on supporting clinician workflows during therapy. This thesis responds to that gap by presenting a human-centered machine learning framework focused on Prolonged Exposure (PE) therapy for PTSD.It introduces three core contributions:1. Synthetic Data Generation: "Thousand Voices of Trauma," a large-scale dataset (3,000 conversations, 500 personas) simulating PE therapy. Validated by clinical experts, the dataset demonstrates high fidelity to real sessions across structural, linguistic, and therapeutic dimensions.2. Empathetic and Efficient Modeling: Fine-tuning small language models (SLMs, 0.5B-5B parameters) on the synthetic "Trauma-Informed Dialogue for Empathy (TIDE)" dataset to generate context-aware, empathetic responses for real-time, inthe-moment emotional support in PTSD-related dialogues-serving as companion tools, not therapy replacements-and achieving significant gains over zero-shot baselines.3. Temporal Localization of Therapeutic Elements: A parameter-efficient approach for fine-tuning multimodal audio-language models to identify key components of PE therapy using soft-supervised, clinician-verified annotations.Together, these contributions offer validated datasets, efficient models, and deployment strategies to build scalable, empathetic AI tools that augment therapist training, provide in-the-moment support for users, and enhance mental healthcare delivery-always as a supplement to, not a replacement for, clinical care.
■590 ▼aSchool code: 0176.
■650 4▼aTherapists
■650 4▼aTrauma
■650 4▼aTherapy
■650 4▼aGender
■650 4▼aEmpathy
■650 4▼aLinguistics
■650 4▼aMulticulturalism & pluralism
■650 4▼aMental health
■650 4▼aError analysis
■650 4▼aLarge language models
■650 4▼aMathematics
■650 4▼aSociology
■690 ▼a0290
■690 ▼a0212
■690 ▼a0347
■690 ▼a0800
■690 ▼a0405
■690 ▼a0626
■71020▼aThe Pennsylvania State University.
■7730 ▼tDissertations Abstracts International▼g87-04B.
■790 ▼a0176
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360211▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


