TEAMMix Framework for Weakly Supervised Hierarchical Text Classification

TEAMMix uses LLM-based keyword generation and corpus mining to enrich label hierarchies, combined with Gaussian mixture model resampling of pseudo-samples to mitigate class imbalance. This approach improves the reliability of LLM-generated labels in long-tail hierarchical classification tasks.