With the continuous development of artificial intelligence technology, dialogue system has attracted more and more attention because of its strong applicability and wide application scenarios. It has gradually entered all aspects of people's lives. The development of science and technology such as speech recognition and synthesis, natural language processing, machine learning, deep neural network and soon has also accelerated the transformation of this process, Make the machine closer to the goal of smooth dialogue with people. This paper mainly studies the task driven human-computer interaction system. Aiming at the sequence track of humancomputer conversation, we use the encoding and decoding model to capture human's dynamic intention. From the final system, this way can significantly improve the accuracy of machine response compared with the previous way of only recommending answers through single state conversation.