Research On Human-computer Dialogue Based On Improved Seq2seq Model

With the constant maturity of deep learning technology, human-computer dialogue has become a research hotspot in natural language processing. People in academia and industry are very concerned about it. The extensive use of artificial intelligence and deep learning technology in the human-machine dialogue system and the deep neural network modeling for text semantics are of great significance in promoting human-computer dialogue technologies and the application of human-computer dialogue to serve humanity better. Based on the above background, this paper focuses on the research of the human-computer dialogue system based on the improved seq2seq model, using the pre-trained Bert improved model as the codec modeling, and addressing the lack of Q&A data sets, the imbalance of category distribution, and the robustness of the model. These problems can be solved by adding disturbance structure adversarial sample training.