Text Classification for Fault Knowledge Graph Construction Based on CNN-BiLSTM

In order to make full use of fault historical text data, knowledge graph is used to assist fault diagnosis. Text classification, a fundamental method for constructing knowledge graph, is prone to the overfitting issue when dealing with sparse and small sample data such as fault domain data. Therefore, an integrated neural network model based on CNN-BiLSTM (convolutional neural networks and bidirectional long and short-term memory networks) is proposed in this paper. In this model, CNN layers extract local semantic features while BiLSTM layers integrate contextual semantic information. Then, adversarial training and attention mechanism are introduced to enhance model performance and prevent overfitting. The proposed method is verified on two datasets, general text data set THUCNews and bearing fault dataset. Comparative experiments reveal that the model can push text classification accuracy in the bearing fault domain to 90%, thereby contributing to greater efficiency and lower labor costs for fault knowledge graph construction.