MISE: Meta-knowledge Inheritance for Social Media-Based Stressor Estimation

TL;DR

This work introduces social media-based stressor estimation as a practical few-shot learning problem, where the goal is to identify specific stressors from user posts rather than broad stress categories. It presents MISE, a meta-learning framework enhanced with a meta-knowledge inheritance mechanism that preserves knowledge from prior tasks to prevent catastrophic forgetting while quickly adapting to new stressors with limited labeled data. The approach combines a RoBERTa-based encoder with CRF decoding and an optimization-based meta-learning loop, augmented by a knowledge inheritance loss that distills prior meta-model predictions into an inheritor-model. A public Weibo stressor dataset with 4,254 labeled posts is released, and experiments show that MISE achieves state-of-the-art performance (F1 up to 0.742) across 3-/5-/10-shot settings, with strong robustness and clear advantages over both traditional and few-shot baselines. The work highlights practical impact for targeted stress relief, supports broader wellbeing research, and provides a foundation for future stressor-aware mental health tools.

Abstract

Stress haunts people in modern society, which may cause severe health issues if left unattended. With social media becoming an integral part of daily life, leveraging social media to detect stress has gained increasing attention. While the majority of the work focuses on classifying stress states and stress categories, this study introduce a new task aimed at estimating more specific stressors (like exam, writing paper, etc.) through users' posts on social media. Unfortunately, the diversity of stressors with many different classes but a few examples per class, combined with the consistent arising of new stressors over time, hinders the machine understanding of stressors. To this end, we cast the stressor estimation problem within a practical scenario few-shot learning setting, and propose a novel meta-learning based stressor estimation framework that is enhanced by a meta-knowledge inheritance mechanism. This model can not only learn generic stressor context through meta-learning, but also has a good generalization ability to estimate new stressors with little labeled data. A fundamental breakthrough in our approach lies in the inclusion of the meta-knowledge inheritance mechanism, which equips our model with the ability to prevent catastrophic forgetting when adapting to new stressors. The experimental results show that our model achieves state-of-the-art performance compared with the baselines. Additionally, we construct a social media-based stressor estimation dataset that can help train artificial intelligence models to facilitate human well-being. The dataset is now public at \href{https://www.kaggle.com/datasets/xinwangcs/stressor-cause-of-mental-health-problem-dataset}{\underline{Kaggle}} and \href{https://huggingface.co/datasets/XinWangcs/Stressor}{\underline{Hugging Face}}.

Figures (4)

• Figure 1: Two posts. The first user's stress is caused by exams. The second user's stress is caused by writing paper. While prior studies may classify both posts as belonging to the broader category of school-related stress, our work aims to estimate the specific causes of stress (i.e., exams and writing paper), in order to provide more targeted support for stress relief.
• Figure 2: Schematic diagram of our practical scenario few-shot stressor estimation task, which focuses on the practical scenario that new stressors continue to emerge over time and some long-tailed stressor has scarce labeled data. Specifically, this scenario needs to train models with past data and effectively estimate latest stressor with a few adaption data.
• Figure 3: Overview of our proposed framework. (a) The basic estimator. (b) The meta-training process. (c) The meta-testing process with meta-knowledge inheritance mechanism.
• Figure 4: Parameter Study on $\lambda$ and $t$.