Short Text Semantic Matching Model based on Bert and Adversarial Network

Abstract

Short text semantic matching plays an important role in the fields of natural language processing such as fast retrieval, intelligent question answering, and information matching. Aiming at the problem of word polysemy that is difficult to solve by conventional models, this paper proposes a short text semantic matching model BERT-GAN based on the BERT (Bidirectional Encoder Representations from Transformers) pre-training model and combined with the adversarial network in the fine-tuning stage. The basic idea is as follows: Using BERT to extract text features, and then introducing an adversarial network in the fine-tuning stage of the task to add perturbation to the embedding layer to improve the generalization ability and robustness of the model. The experimental results show that the BERT-GAN short text semantic matching model is better than the comparison model, and the F1 value is improved by 10.5%, 6.6% and 0.9% respectively compared with the comparison model.