Hierarchical Text Classification Based on a Hierarchical Probability-Aware Balanced Model

This paper tackles the challenges of poor class separability, complex structural dependencies, and severe class imbalance in large-scale hierarchical text classification. To address these issues, we propose a Hierarchical Probability-Aware Balancing (HiPAB) model that jointly incorporates representation enhancement, structural dependency modeling, and probabilistic decision balancing. HiPAB improves fine-grained category discrimination by capturing hierarchical relationships and mitigating the impact of imbalanced data distributions. Extensive experiments show that HiPAB consistently outperforms strong baselines across multiple metrics, demonstrating superior robustness and generalization, particularly in deep-level nodes and long-tail categories. These results highlight HiPAB as an effective and scalable solution for real-world hierarchical text classification tasks.