Large Language Models for Mental Health Applications: A Systematic Review (Preprint)

JMIR Mental Health, Journal Year: 2024, Volume and Issue: 11, P. e57400 - e57400

Published: Sept. 3, 2024

Background Large language models (LLMs) are advanced artificial neural networks trained on extensive datasets to accurately understand and generate natural language. While they have received much attention demonstrated potential in digital health, their application mental particularly clinical settings, has generated considerable debate. Objective This systematic review aims critically assess the use of LLMs specifically focusing applicability efficacy early screening, interventions, settings. By systematically collating assessing evidence from current studies, our work analyzes models, methodologies, data sources, outcomes, thereby highlighting challenges present, prospects for use. Methods Adhering PRISMA (Preferred Reporting Items Systematic Reviews Meta-Analyses) guidelines, this searched 5 open-access databases: MEDLINE (accessed by PubMed), IEEE Xplore, Scopus, JMIR, ACM Digital Library. Keywords used were (mental health OR illness disorder psychiatry) AND (large models). study included articles published between January 1, 2017, April 30, 2024, excluded languages other than English. Results In total, 40 evaluated, including 15 (38%) conditions suicidal ideation detection through text analysis, 7 (18%) as conversational agents, 18 (45%) applications evaluations health. show good effectiveness detecting issues providing accessible, destigmatized eHealth services. However, assessments also indicate that risks associated with might surpass benefits. These include inconsistencies text; production hallucinations; absence a comprehensive, benchmarked ethical framework. Conclusions examines inherent risks. The identifies several issues: lack multilingual annotated experts, concerns regarding accuracy reliability content, interpretability due “black box” nature LLMs, ongoing dilemmas. clear, framework; privacy issues; overreliance both physicians patients, which could compromise traditional medical practices. As result, should not be considered substitutes professional rapid development underscores valuable aids, emphasizing need continued research area. Trial Registration PROSPERO CRD42024508617; https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=508617

Language: Английский

---

Prompt engineering for digital mental health: a short review

Frontiers in Digital Health, Journal Year: 2024, Volume and Issue: 6

Published: June 12, 2024

Prompt engineering, the process of arranging input or prompts given to a large language model guide it in producing desired outputs, is an emerging field research that shapes how these models understand tasks, information, and generate responses wide range natural processing (NLP) applications. Digital mental health, on other hand, becoming increasingly important for several reasons including early detection intervention, mitigate limited availability highly skilled medical staff clinical diagnosis. This short review outlines latest advances prompt engineering NLP digital health. To our knowledge, this first attempt discuss types, methods, tasks are used health We three types tasks: classification, generation, question answering. conclude, we challenges, limitations, ethical considerations, future directions believe contributes useful point departure

Language: Английский

---

A self-attention TCN-based model for suicidal ideation detection from social media posts

Expert Systems with Applications, Journal Year: 2024, Volume and Issue: 255, P. 124855 - 124855

Published: July 24, 2024

Early suicidal ideation detection has long been regarded as an important task that can benefit both society and individuals. In this regard, it shown that, very frequently, the first symptoms of problem be identified by analyzing contents shared on social media. Machine learning classification models have proven promising in capturing behavioral textual features from posts This study proposes a novel machine-learning model to detect risk suicide media posts, employing natural language processing state-of-the-art deep techniques. We propose ensemble LSTM-TCN benefits self-attention mechanism among users two well-known networks, Twitter (X) Reddit. Furthermore, we present comprehensive analysis data, examining statistically semantically, which provide rich knowledge about ideation. Our proposed (AL-BTCN) outperforms compared models, resulting over 94% accuracy, recall, F1-score. Researchers, mental health specialists, service providers all findings study.

Language: Английский

---

The good, the bad, and the GPT: Reviewing the impact of generative artificial intelligence on psychology

Current Opinion in Psychology, Journal Year: 2024, Volume and Issue: 59, P. 101872 - 101872

Published: Aug. 23, 2024

Language: Английский

---

Artificial Intelligence-Based Suicide Prevention and Prediction: A Systematic Review (2019-2023)

Published: Jan. 1, 2024

Language: Английский

---

ToxiLab: How Well Do Open-Source LLMs Generate Synthetic Toxicity Data?

Published: March 21, 2025

Effective toxic content detection relies heavily on high-quality and diverse data, which serve as the foundation for robust moderation models. Synthetic data has become a common approach training models across various NLP tasks. However, its effectiveness remains uncertain highly subjective tasks like hate speech detection, with previous research yielding mixed results. This study explores potential of open-source LLMs harmful synthesis, utilizing controlled prompting supervised fine-tuning techniques to enhance quality diversity. We systematically evaluated six open source five datasets, assessing their ability generate diverse, while minimizing hallucination duplication. Our results show that Mistral consistently outperforms other models, significantly enhances reliability further analyze trade-offs between prompt-based vs. fine-tuned discuss real-world deployment challenges, highlight ethical considerations. findings demonstrate provide scalable cost-effective solutions augment paving way more accessible transparent tools.

Language: Английский

---

Triaging Casual from Critical: Leveraging Machine Learning to Detect Self-Harm and Suicide Risks for Youth on Social Media (Preprint)

Published: April 15, 2025

This study aims to detect self-harm and/or suicidal ideation (SH/S) language used by youth (ages 13–21) in their private Instagram conversations. While automated mental health tools have shown promise, there remains a gap understanding how nuanced around SH/S can be effectively identified. Our work focuses on developing interpretable models that go beyond binary classification recognize the spectrum of expressions. We analyzed dataset conversations donated youth. A range traditional machine learning (SVM, Random Forest, Naive Bayes, XGBoost) and transformer-based architectures (BERT, DistilBERT) were trained evaluated. In addition raw text, we incorporated contextual, psycholinguistic (LIWC), sentiment (VADER), lexical (TF-IDF) features improve detection accuracy. further explored increasing conversational context—from message-level sub-conversation level—affected model performance. DistilBERT demonstrated good performance identifying presence behaviors within individual messages, achieving an accuracy 99%. However, when tasked with more fine-grained classification—differentiating among “Self” (personal accounts or suicide), “Other” (references experiences involving others), “Hyperbole” (sarcastic, humorous, exaggerated mentions not indicative genuine risk)—the model's declined 89%. Notably, expanding input window include broader context, these granular categories improved 91%, highlighting importance contextual distinguishing between subtle variations discourse. findings underscore designing automatic systems sensitive dynamic social media. Contextual sentiment-aware provide risk expression. research lays foundation for inclusive ethically grounded interventions, while also calling future validate across platforms populations.

Language: Английский

---

DiGrI: Distorted Greedy Approach for Human-Assisted Online Suicide Ideation Detection

Published: April 22, 2025

Language: Английский

---

Exploring Modular Prompt Design for Emotion and Mental Health Recognition

Published: April 24, 2025

Language: Английский

---

Transferring Natural Language Datasets Between Languages Using Large Language Models for Modern Decision Support and Sci-Tech Analytical Systems

Big Data and Cognitive Computing, Journal Year: 2025, Volume and Issue: 9(5), P. 116 - 116

Published: April 28, 2025

The decision-making process to rule R&D relies on information related current trends in particular research areas. In this work, we investigated how one can use large language models (LLMs) transfer the dataset and its annotation from another. This is crucial since sharing knowledge between different languages could boost certain underresourced directions target language, saving lots of effort data or quick prototyping. We experiment with English Russian pairs, translating DEFT (Definition Extraction Texts) corpus. corpus contains three layers dedicated term-definition pair mining, which a rare type for Russian. presence such beneficial natural processing methods trend analysis science terms definitions are basic blocks any scientific field. provide pipeline using LLMs. end, train BERT-based translated establish baseline.

Language: Английский

---