Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Advancing Depression Detection on Social Media Platforms Through Fine-Tuned Large Language Models

Shahid Munir Shah, Syeda Anshrah Gillani, Mirza Samad Ahmed Baig, Muhammad Aamer Saleem, Muhammad Hamzah Siddiqui

TL;DR

The study demonstrates that fine-tuning large language models on a depression-detection dataset yields state-of-the-art performance for identifying depressive content in social media posts, achieving 96.0% accuracy with GPT-3.5 Turbo 1106 and 84.0% with LLaMA2-7B. By leveraging PEFT (LoRA) for efficient adaptation and a rich, multi-faceted dataset, the approach surpasses several existing models while highlighting practical considerations for deployment. The work underscores the potential of LLM-based systems for early depression detection on social platforms, along with ethical, bias, and generalization considerations needed for real-world use. Overall, the findings point to a viable path for integrating depression monitoring into social media infrastructures, enabling earlier interventions and resource linkage.

Abstract

This study investigates the use of Large Language Models (LLMs) for improved depression detection from users social media data. Through the use of fine-tuned GPT 3.5 Turbo 1106 and LLaMA2-7B models and a sizable dataset from earlier studies, we were able to identify depressed content in social media posts with a high accuracy of nearly 96.0 percent. The comparative analysis of the obtained results with the relevant studies in the literature shows that the proposed fine-tuned LLMs achieved enhanced performance compared to existing state of the-art systems. This demonstrates the robustness of LLM-based fine-tuned systems to be used as potential depression detection systems. The study describes the approach in depth, including the parameters used and the fine-tuning procedure, and it addresses the important implications of our results for the early diagnosis of depression on several social media platforms.

Advancing Depression Detection on Social Media Platforms Through Fine-Tuned Large Language Models

TL;DR

The study demonstrates that fine-tuning large language models on a depression-detection dataset yields state-of-the-art performance for identifying depressive content in social media posts, achieving 96.0% accuracy with GPT-3.5 Turbo 1106 and 84.0% with LLaMA2-7B. By leveraging PEFT (LoRA) for efficient adaptation and a rich, multi-faceted dataset, the approach surpasses several existing models while highlighting practical considerations for deployment. The work underscores the potential of LLM-based systems for early depression detection on social platforms, along with ethical, bias, and generalization considerations needed for real-world use. Overall, the findings point to a viable path for integrating depression monitoring into social media infrastructures, enabling earlier interventions and resource linkage.

Abstract

This study investigates the use of Large Language Models (LLMs) for improved depression detection from users social media data. Through the use of fine-tuned GPT 3.5 Turbo 1106 and LLaMA2-7B models and a sizable dataset from earlier studies, we were able to identify depressed content in social media posts with a high accuracy of nearly 96.0 percent. The comparative analysis of the obtained results with the relevant studies in the literature shows that the proposed fine-tuned LLMs achieved enhanced performance compared to existing state of the-art systems. This demonstrates the robustness of LLM-based fine-tuned systems to be used as potential depression detection systems. The study describes the approach in depth, including the parameters used and the fine-tuning procedure, and it addresses the important implications of our results for the early diagnosis of depression on several social media platforms.
Paper Structure (26 sections, 9 figures, 5 tables)

This paper contains 26 sections, 9 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Traditional Models
  4. Limitations of the Traditional Approaches
  5. Capabilities of LLMs to Address Challenges of the Traditional Models
  6. Large Language Models
  7. Our Proposed Approach
  8. Methodology
  9. GPT-3.5 Turbo 1106
  10. LLaMA2-7B
  11. Comparison of GPT-3.5 Turbo 1106 and LLaMA2-7B models
  12. Experimentation
  13. Explanation of the Dataset's Origin, Structure, and Relevance to the Study
  14. Comprehensive overview of the fine-tuning process
  15. Fine-Tuning of GPT-3.5 Turbo 1106
  16. ...and 11 more sections

Figures (9)

  • Figure 1: Methodology adopted
  • Figure 2: GPT-3.5 Turbo 1106 vs LLaMA2-7B
  • Figure 3: GPT-3.5 Turbo 1106 Training Process and Model Inference.
  • Figure 4: LLaMA2-7B Training Process Model Inference.
  • Figure 5: Confusion matrices of fine-tuned GPT-3.5 Turbo 1106 and Fine-Tuned LLaMA2-7B
  • ...and 4 more figures