A generative chatbot based on deep learning method

Abstract

The invention discloses a generative chat robot based on a deep learning method, including historical dialogue encoding, knowledge selection, knowledge encoding and dialogue generation. The historical dialogue encoding first splices historical dialogues, then converts them into vector representations, Use the bidirectional gated neural unit to encode the historical dialogue, and obtain its representation through an attention layer, and then pass through a bidirectional gated neural unit for each dialogue representation in the historical dialogue, and also pass through an attention layer to obtain the final historical dialogue representations. On the basis of the traditional seq2seq model, the present invention improves the above problems by introducing external knowledge and a knowledge encoder. First, the knowledge encoder will save the dialogue topic when selecting knowledge, which is equivalent to saving the key information in the historical dialogue.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, in particular to a generative chat robot based on a deep learning method. Background technique [0002] With the development of the Internet, information communication and artificial intelligence technology, the inherent natural convenience of human-computer dialogue system makes it a new way of communicating with computing devices, which is considered to be a successor to mouse and keyboard tapping and screen touch. After control, a new generation of interaction paradigm in the future. Human-machine dialogue technology has been applied to various types of products and services by the industry. People are familiar with personal assistant systems such as Apple's Siri, Microsoft's Cortana, Google's Allo, and Baidu's Dubi, as well as Amazon's Echo smart home service system and Alibaba's Xiaomi e-commerce smart customer service system. These man-machine dialogue products have brough...

AI Technical Summary

Problems solved by technology

[0010] 1) It is easy to produce insignificant or unclear, generic and meaningless replies, such as "I don't know" and "haha" replies without practical meaning;

[0011] 2) Chatbots generally have multiple rounds of conversations. For multi-rounds of long conversations, it is difficult for the model to preserve previous memories, which may lead to inconsistencies in the topics of the conversations before and after;

[0012] 3) For different expressions of the same topic, different results may be obtained

[0013] The emergence of problems 1 and 3 is mainly because the model does not have certain prior knowledge like humans, and the problem 2 is because the traditional seq2seq is based on deep neural network (recurrent neural network (RNN)) for sequence encoding, in When there are too many rounds or the dialogue is too long, it is impossible to capture all the information of the dialogue

Images