Few-shot Classification of Radar Equipment Fault Based on TF-IDF Feature Date Augmentation and BERT

In radar equipment fault text classification tasks, the existing fault data labeling sample size is small, and the cost of labeling a large number of unlabeled data is higher. Aiming at this practical problem, this paper proposes a data augmentation method based on TF-IDF features. First, new samples are synthesized by TF-IDF features of the corresponding fault category. Secondly, the new samples are expanded to the original samples and the fault text is vector-encoded using the BERT pretraining language model. Finally, the classification of radar equipment faults is achieved through the Softmax regression model. The experiment shows that, TEDA can be well combined with the big data training model, and the macro F1 value of the model has increased by 7 percentage points.