Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Self-Harm Detection for Mental Health Chatbots.

Saahil Deshpande1, Jim Warren1

  • 1University of Auckland, Auckland, New Zealand.

Studies in Health Technology and Informatics
|May 27, 2021
PubMed
Summary
This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Adolescents' and youths' perceived barriers and facilitators to engaging with digital mental health interventions for depression and anxiety: A scoping review.

Internet interventions·2025
Same author

The Digital Therapeutic Alliance With Mental Health Chatbots: Diary Study and Thematic Analysis.

JMIR mental health·2025
Same author

Personalised modelling of routine variability and affective states.

NPJ digital medicine·2025
Same author

Predicting Heart Rate Variability from Heart Rate and Step Count for University Student Weekdays.

Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference·2025
Same author

HCI-modelling for improving the clinical usability of digital health technologies.

Methods (San Diego, Calif.)·2024
Same author

Detecting Emotional Context for Safer Digital Mental Health Agents.

Studies in health technology and informatics·2024
Same journal

A GenAI Pipeline for Violinist Kinematic Data Management.

Studies in health technology and informatics·2026
Same journal

AMAL-For-Qatar: A Comprehensive AI Ecosystem for Fetal Ultrasound Analysis - Project Overview and Achievements.

Studies in health technology and informatics·2026
Same journal

Longitudinal Treatment-Aware Multimodal AI for Dermatology: A Scoping Review.

Studies in health technology and informatics·2026
Same journal

Predicting Postpartum Depression Using Imbalance-Aware Machine Learning.

Studies in health technology and informatics·2026
Same journal

Validation of Deep-Learning Models for Autosegmentation of Brain Metastases.

Studies in health technology and informatics·2026
Same journal

Delay-Dependent Gating in Modular RNNs.

Studies in health technology and informatics·2026
See all related articles

This study developed a self-harm classifier for mental health chatbots, achieving 97% accuracy in detecting user intent for self-harm. This advancement enhances chatbot safety and user support in mental healthcare.

Area of Science:

  • Artificial Intelligence
  • Mental Health Technology
  • Computational Linguistics

Background:

  • Chatbots offer potential solutions to healthcare workforce shortages and rising youth mental health concerns, including suicide.
  • Existing mental health chatbots face challenges in accurately identifying and responding to emergency situations like self-harm.
  • Developing reliable methods for detecting self-harm intent in user interactions is crucial for chatbot safety.

Purpose of the Study:

  • To design and evaluate a self-harm classifier for integration into mental health chatbots.
  • To improve the safety and efficacy of chatbots in addressing critical mental health issues.
  • To explore alternative data sources for training robust self-harm detection models.

Main Methods:

  • A self-harm classifier was designed to predict user intent from chatbot text input.
Keywords:
BERTChatbotLSTMMental healthSelf-harmsentiment analysis

Related Experiment Videos

  • Sentiment analysis and self-harm classifiers were trained on Twitter and Reddit data, respectively.
  • A Long Short-Term Memory-Recurrent Neural Network (LSTM-RNN) classifier with BERT encoding was utilized and models were combined for improved performance.

    • Main Results:

    • The combined model achieved an accuracy of 92.13% in detecting self-harm intent.
    • Testing on new Reddit data yielded an impressive accuracy of 97%.
    • The LSTM-RNN classifier with BERT encoding demonstrated superior performance.

    Conclusions:

    • The developed self-harm classifier shows significant promise for enhancing the safety of mental health chatbots.
    • Accurate detection of self-harm talk by users can lead to timely interventions.
    • This technology can improve user support and safety within digital mental health tools.