Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DepressionX: Knowledge Infused Residual Attention for Explainable Depression Severity Assessment

Yusif Ibrahimov, Tarique Anwar, Tommy Yuan

TL;DR

DepressionX addresses the challenge of predicting depression severity from social media while ensuring explainability. It fuses multi-level textual embeddings with a domain-specific knowledge graph, employing residual multihead attention and ordinal regression to generate calibrated severity predictions. The approach includes REBEL-based KG construction, two-layer GIN followed by a multihead GAT, and attention-driven explanations for both text and graph components. On Reddit datasets, DepressionX surpasses prior state-of-the-art by more than 7% in weighted $F_1$ on both imbalanced and balanced splits, demonstrating strong performance alongside transparency for trustworthy mental health analysis.

Abstract

In today's interconnected society, social media platforms have become an important part of our lives, where individuals virtually express their thoughts, emotions, and moods. These expressions offer valuable insights into their mental health. This paper explores the use of platforms like Facebook, $\mathbb{X}$ (formerly Twitter), and Reddit for mental health assessments. We propose a domain knowledge-infused residual attention model called DepressionX for explainable depression severity detection. Existing deep learning models on this problem have shown considerable performance, but they often lack transparency in their decision-making processes. In healthcare, where decisions are critical, the need for explainability is crucial. In our model, we address the critical gap by focusing on the explainability of depression severity detection while aiming for a high performance accuracy. In addition to being explainable, our model consistently outperforms the state-of-the-art models by over 7% in terms of $\text{F}_1$ score on balanced as well as imbalanced datasets. Our ultimate goal is to establish a foundation for trustworthy and comprehensible analysis of mental disorders via social media.

DepressionX: Knowledge Infused Residual Attention for Explainable Depression Severity Assessment

TL;DR

DepressionX addresses the challenge of predicting depression severity from social media while ensuring explainability. It fuses multi-level textual embeddings with a domain-specific knowledge graph, employing residual multihead attention and ordinal regression to generate calibrated severity predictions. The approach includes REBEL-based KG construction, two-layer GIN followed by a multihead GAT, and attention-driven explanations for both text and graph components. On Reddit datasets, DepressionX surpasses prior state-of-the-art by more than 7% in weighted on both imbalanced and balanced splits, demonstrating strong performance alongside transparency for trustworthy mental health analysis.

Abstract

In today's interconnected society, social media platforms have become an important part of our lives, where individuals virtually express their thoughts, emotions, and moods. These expressions offer valuable insights into their mental health. This paper explores the use of platforms like Facebook, (formerly Twitter), and Reddit for mental health assessments. We propose a domain knowledge-infused residual attention model called DepressionX for explainable depression severity detection. Existing deep learning models on this problem have shown considerable performance, but they often lack transparency in their decision-making processes. In healthcare, where decisions are critical, the need for explainability is crucial. In our model, we address the critical gap by focusing on the explainability of depression severity detection while aiming for a high performance accuracy. In addition to being explainable, our model consistently outperforms the state-of-the-art models by over 7% in terms of score on balanced as well as imbalanced datasets. Our ultimate goal is to establish a foundation for trustworthy and comprehensible analysis of mental disorders via social media.
Paper Structure (21 sections, 16 equations, 7 figures, 2 tables)

This paper contains 21 sections, 16 equations, 7 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem statement
  4. Proposed model
  5. Textual Representation
  6. Multi-level Encoding
  7. Residual Multihead Attention
  8. Knowledge Graph Representation
  9. Domain-specific knowledge graph construction
  10. Knowledge graph representation learning
  11. Knowledge Infusion and Output
  12. Explainability
  13. Experimental Results
  14. Datasets
  15. Experimental Settings
  16. ...and 6 more sections

Figures (7)

  • Figure 1: Proposed model DepressionX. It predicts depression severity of a post and generates explanation with word- and sentence-level attentions and a knowledge subgraph. The prediction and the explanatory components are highlighted in yellow.
  • Figure 2: Our domain-specific knowledge graph
  • Figure 3: Two-dimensional plots of post representations (generated by UMAP) before and after knowledge infusion
  • Figure 4: Explainability with sentence- and word-level importance
  • Figure 5: Attention heatmap of the most and the least important words
  • ...and 2 more figures

Theorems & Definitions (1)

  • Definition 4.1: Knowledge Graph