Chatbot is a type of program model that can interact with people through voice or text. The diversification of application scenarios and the effectiveness of information interaction will make the application of chatbots increasingly popular in the future. Therefore, research on chatbots has high social and economic benefits. We implemented a chatbot model based on the seq2seq architecture in this paper. Then, the attention mechanism is introduced to further optimize the model, and the teacher forcing mechanism is introduced to help the model converge faster throughout the training phase. Based on the LCCC-mini dataset and the Qingyun dataset, a series of dialogue experiments and BLEU metrics are evaluated on the model. The results show that the model incorporating the attention mechanism has significantly improved the response effect of long sequence dialogue compared with the single seq2seq model, and the BLEU-1 to BLEU-4 metrics on both datasets have been improved.