TEAMMix: Taxonomy Enrichment Augmentation and Minority-augmented Mixing Strategy for LLM-enhanced Weak-Supervised Hierarchical Text Classification

2026-08-11

AI Summary

This work addresses the challenges in hierarchical text classification posed by complex label structures, long-tailed class distributions, and the tendency of large language models (LLMs) to produce verbose, structure-agnostic prompts under weak supervision. To mitigate these issues, the authors propose a weakly supervised framework that integrates semantic enhancement with confidence-based resampling. The approach enriches the label hierarchy through semantic augmentation, leverages LLMs to generate pseudo-samples for minority classes, and employs a Gaussian mixture model to estimate pseudo-label confidence and perform resampling, thereby improving data quality and label reliability. Experimental results demonstrate that the proposed framework significantly enhances classification performance in fine-grained and imbalanced settings by effectively combining structural priors with the generative capabilities of LLMs.

Abstract

Hierarchical Text Classification (HTC), as a critical text mining task, faces challenges such as complex label hierarchies and class imbalance. Existing methods based on large language models (LLMs) struggle to be efficiently applied to this task due to issues like lengthy prompts and loss of label structural information. To address these limitations, this paper proposes a weakly supervised HTC framework enhanced by LLM-based data augmentation. The framework first enriches the label hierarchy semantically through keyword generation and corpus mining, thereby enhancing the model's understanding of labels. Subsequently, it guides the LLM to generate pseudo-samples to mitigate the long-tail problem, and employs a Gaussian mixture model for confidence-based resampling to optimize the quality of generated data. Experimental results demonstrate that the proposed method effectively improves the reliability of LLM-generated pseudo-labels and significantly enhances classification performance on fine-grained and imbalanced datasets.

Problem

• Hierarchical Text Classification
• Class Imbalance
• Label Hierarchy
• Weak Supervision
• Large Language Models

Innovation

• Hierarchical Text Classification
• Large Language Models
• Data Augmentation
• Class Imbalance
• Weak Supervision