A Multi-label Classification Approach for ICT Fault Text Analysis

The ICT customer service from State Grid has accumulated a large number of electric power fault reports and texts, which are difficult to classify and analyze because an occurrence of power fault may be triggered possibly by multiple different causes. In order to track different causes of electric power faults from historical fault texts, a multi-label based text classification approach is proposed for fault-assisted decision. Firstly, Chinese fault texts are sequentially preprocessed by word segmentation, stop words removal and feature representation. Then, we present a multi-label text classification by combining Binary Relevance and Gradient Boosting algorithm. The experimental results show that our method is better than BR+LR and ML-KNN for fault text classification.