Article: Multi-label text classification for power communication customer service system based on BR-GBDT with label correlation weighting Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.77 pp.40 - 56 Abstract: Aiming at the problems of inconsistent terminology, strong label coupling, insufficient generalisation, and difficult real-time deployment in power communication customer service texts, this paper proposes a multi-label classification method based on binary relevance (BR) and gradient boosting decision tree (GBDT) with adaptive label correlation weighting. We design an automated training set construction scheme using data cleaning, Jieba segmentation, TF-IDF, and Word2vec to build a standardised label system. The BR-GBDT model decomposes multi-label tasks into binary classification tasks and uses mutual information-based label correlation weights to overcome the label independence defect of traditional BR. Experiments on 12,800 real work orders show that the method achieves Hamming Loss 0.072, Subset Accuracy 82.3%, Macro-F1 89.5%, and Micro-F1 92.1%. The average single-sample prediction time is 18.3 ms under real system load, meeting the 30 ms real-time requirement. The model exhibits strong cross-domain adaptability, noise robustness, and low hardware dependence, and is interpretable via feature importance and per-label performance analysis. Comparisons with CNN-LSTM, BERT, and classic multi-label methods verify its superiority. This method supports efficient, real-time, and deployable multi-label classification for power communication customer service systems. Inderscience Publishers - linking academia, business and industry through research

Title: Multi-label text classification for power communication customer service system based on BR-GBDT with label correlation weighting

Authors: Wenting Zhang; Yanan Yang; Yunzhe Qian; Yuqi Zhou; Yanqiao Wang; Rui Huang

Addresses: State Grid Nanjing Power Supply Company, Nanjing, 210019, China ' State Grid Nanjing Power Supply Company, Nanjing, 210019, China ' State Grid Nanjing Power Supply Company, Nanjing, 210019, China ' State Grid Jiangsu Marketing Service Center, Nanjing, Jiangsu, 210019, China ' State Grid Nanjing Power Supply Company, Nanjing, 210019, China ' State Grid Nanjing Power Supply Company, Nanjing, 210019, China

Abstract: Aiming at the problems of inconsistent terminology, strong label coupling, insufficient generalisation, and difficult real-time deployment in power communication customer service texts, this paper proposes a multi-label classification method based on binary relevance (BR) and gradient boosting decision tree (GBDT) with adaptive label correlation weighting. We design an automated training set construction scheme using data cleaning, Jieba segmentation, TF-IDF, and Word2vec to build a standardised label system. The BR-GBDT model decomposes multi-label tasks into binary classification tasks and uses mutual information-based label correlation weights to overcome the label independence defect of traditional BR. Experiments on 12,800 real work orders show that the method achieves Hamming Loss 0.072, Subset Accuracy 82.3%, Macro-F1 89.5%, and Micro-F1 92.1%. The average single-sample prediction time is 18.3 ms under real system load, meeting the 30 ms real-time requirement. The model exhibits strong cross-domain adaptability, noise robustness, and low hardware dependence, and is interpretable via feature importance and per-label performance analysis. Comparisons with CNN-LSTM, BERT, and classic multi-label methods verify its superiority. This method supports efficient, real-time, and deployable multi-label classification for power communication customer service systems.

Keywords: power communication; multi-label text classification; binary relevance; BR; gradient boosting decision tree; GBDT; label correlation; model interpretability.

DOI: 10.1504/IJICT.2026.154959

International Journal of Information and Communication Technology, 2026 Vol.27 No.77, pp.40 - 56

Received: 04 Dec 2025
Accepted: 05 May 2026
Published online: 20 Jul 2026 *

Title: Multi-label text classification for power communication customer service system based on BR-GBDT with label correlation weighting

Keep up-to-date