Enhancing Hierarchical Text Classification through DynamicPath Pruning in Large Language Models

A novel hierarchical multi-label text classification model based on dynamic path pruning of large language models is proposed to address the limitations of existing algorithms in leveraging multi-level label samples and mining dependency relationships.This model utilizes the dynamic path pruning capability of large models to optimize hierarchical multi-granularity recognition and enhance the stability of graph neural networks.By harnessing the semantic perception capabilities of large models, the proposed model facilitates semantic feature encoding and extracts summary information and co-occurrence structures from domain-specific texts.Additionally, it employs graph neural networks for feature encoding, integrates text summaries with dependency relationship information, and merges original data through a semantic-guided attention mechanism to strengthen hierarchical feature analysis.Experimental results demonstrate that the proposed hierarchical text classification model based on dynamic path pruning performs exceptionally well on the WOS and RCV1-V2 datasets, exhibiting superior performance and greater resilience against data perturbations compared to other state-of-the-art models.