Research on industry label recognition method based on RoBERTa driven by data
You Wen, Xuan Fan, Pingyan Mo · International Journal of Low-Carbon Technologies · 2025
Abstract This study introduces an enhanced RoBERTa-based model, called Industry Aware RoBERTa (IA-RoBERTa), designed to improve the accuracy and generalization of industry label recognition. IA-RoBERTa innovatively integrates structured industry knowledge through a knowledge graph fusion approach, using multigranularity input representation and industry-aware self-attention mechanisms. Together, these features enhance the model’s ability to efficiently process and understand industry-specific information. In addition, IA-RoBERTa includes a layered industry classifier that expertly handles fine-grained and layered industry categories. Experimental evaluations of industry label recognition datasets show that IA-RoBERTa outperforms existing methods in terms of accuracy, F1 scores, and macro-average performance metrics.