DOI: 10.11799/ce202601023
Short text classification of the coal mine"three violations"data based on cBert-GCN
WU Xuyan
YANG Chaoyu
Abstract：To address three challenges in classifying coal mine"three violations"text data,including strong domain specificity,semantic ambiguity,and imbalanced sample distribution,a cBert-GCN classification model is proposed.Considering that the coal mine"three violations"texts are short yet contextually coherent and inherently ambiguous,alongside containing domain-specific information,pinyin and glyph embeddings are introduced to enhance text representation.Graph Convolutional Networks(GCN)are applied to"three violations"text classification by constructing text co-occurrence graphs to capture structural information and dependency relationships.The model integrates a Chinese pre-trained model with GCN for feature learning,combining character-level and word-level representations with tunable weights to achieve accurate classification.Experimental results show that the cBert-GCN model outperforms other models,achieving 97.03%accuracy on the training set and 93.17%on the test set,demonstrating strong generalization capability.Thus,the cBert-GCN model shows clear advantages for classifying coal mine"three violations"text data.
Keywords：the coal mine"three violations"GCNfeature learningtraining samplestext co-occurrence
Publication Date：2026-01-20
Online Publishing Date：2026-03-10(First online date of this platform, not the publication date of the document)
Pages：8( 184-191 )