MISE: Meta-knowledge Inheritance for Social Media-Based Stressor Estimation
This work addresses the challenge of fine-grained stressor identification (e.g., “exams”, “thesis writing”) in social media text—characterized by a large number of categories, extremely limited per-class samples, and continual emergence of novel stressors—by formally introducing the few-shot stressor identification task. Methodologically, we propose a meta-learning framework based on a modified MAML algorithm, incorporating a semantic-enhanced text encoder and a transferable meta-knowledge memory module, along with a novel meta-knowledge inheritance mechanism to mitigate catastrophic forgetting during adaptation to unseen stressors. Evaluated on a newly constructed, publicly released dataset (hosted on Kaggle and Hugging Face), our model achieves state-of-the-art performance, demonstrating significant improvements in cross-stressor generalization and continual adaptability. This work establishes a new paradigm for mental health–oriented AI and provides a foundational benchmark resource.