NarrationDep: Narratives on Social Media For Automatic Depression Detection

TL;DR

The paper addresses automatic depression detection from social media by modeling user narratives rather than treating posts individually. It introduces NarrationDep, a two-layer architecture combining a Hierarchical Attention Network (HAN) for per-tweet representations and a Hierarchical Attention-Based Clustering Network (HACN) for cluster-level semantics, with SBERT-based semantic embeddings and clustering via HDSBCAN. The model jointly learns $\boldsymbol{\theta} = \boldsymbol{\gamma} \bigoplus \boldsymbol{\beta}$ to produce a depression probability $\hat{y}$ while providing attention-based explanations that highlight narrative elements driving the prediction. Empirical results on Shen and related datasets show that NarrationDep outperforms baselines including DepressionNet and transformer-based summarization methods, and ablations confirm the benefit of integrating both local tweet information and global narrative clusters. The work advances narrative-aware depression detection with explainability and lays groundwork for future multi-modal extensions that incorporate visual and contextual cues for improved screening and intervention.

Abstract

Social media posts provide valuable insight into the narrative of users and their intentions, including providing an opportunity to automatically model whether a social media user is depressed or not. The challenge lies in faithfully modelling user narratives from their online social media posts, which could potentially be useful in several different applications. We have developed a novel and effective model called \texttt{NarrationDep}, which focuses on detecting narratives associated with depression. By analyzing a user's tweets, \texttt{NarrationDep} accurately identifies crucial narratives. \texttt{NarrationDep} is a deep learning framework that jointly models individual user tweet representations and clusters of users' tweets. As a result, \texttt{NarrationDep} is characterized by a novel two-layer deep learning model: the first layer models using social media text posts, and the second layer learns semantic representations of tweets associated with a cluster. To faithfully model these cluster representations, the second layer incorporates a novel component that hierarchically learns from users' posts. The results demonstrate that our framework outperforms other comparative models including recently developed models on a variety of datasets.

Figures (5)

• Figure 1: NarrationDep models user narratives to automatically detect depression-related narratives in user-generated text. By modelling attention weights, NarrationDep provides qualitative insights into the state of mind of potentially depressed users. It takes user tweets as input and produces a prediction of whether the user demonstrates signs of depression. Additionally, the learned attention weights reveal key elements captured by the model, aiding in interpretability. Notably, NarrationDep not only models user tweets as a whole but also considers different clusters of tweets and their corresponding cluster assignments.
• Figure 2: An illustration of our novel Hierarchical Attention Network for Clustering of Tweets (HACN) model.
• Figure 3: Narrative tweets captured by NarationDep.
• Figure 4: Analyzing narrative for a user per week. The x-axis represents the days of the week and the y-axis represents the attention weights.
• Figure 5: Analyzing narrative for a user during the 24 hours of the day. The x-axis represents the hours and the y-axis represents the attention weights.