Strong anthropomorphic intelligent dialogue robot with smooth and accurate response

Abstract

The intelligent dialogue robot based on deep learning has very obvious advantages, a dialogue template does not need to be made deliberately, only massive chat data is needed, manual intervention is not needed during training, the coverage of problems is very wide, a dialogue communication network in the intelligent dialogue robot is improved again based on the improved PerAttent communication model, and the communication efficiency is greatly improved. The calculation of the information PerAttent associated with the previous text is based on the output of the previous moment of anthropomorphic decoding instead of the original output at the moment, and then the adaptive variational encoder is added to generate a smooth output statement, so that the two problems of fixed word matching and cyclic output or loss of a new subject are solved, and the calculation efficiency is improved. According to the dialogue robot, the robot can not only answer a certain question of the user, but also communicate with the user in a humanized mode and know the requirements of the user, just like good friends of the user, the dialogue robot has the biggest advantage that answering is interesting, vivid and diverse instead of being dull and boring, and the dialogue robot has very good application prospects.

Description

technical field [0001] The present application relates to an intelligent dialogue robot with smooth and accurate responses, in particular to a strong anthropomorphic intelligent dialogue robot with smooth and accurate responses, which belongs to the technical field of intelligent dialogue robots. Background technique [0002] At present, in many occasions, dialogue robots have their important applications. For example, in a specific scenario, teachers and students can easily grasp the current book borrowing information and the opening and closing time of the library during the dialogue with the intelligent dialogue robot in the school library. In the website customer service system, manual service is a very complicated task. Many times there are many inquiries from customers during non-working hours. In order to solve this problem, intelligent dialogue robots can play a very important role. In some cases, the repetition rate is high. In terms of problems, dialogue robots ca...

AI Technical Summary

Problems solved by technology

[0009] First, the dialogue robots in the prior art generally need to deliberately create dialogue templates, and also need a large amount of chat data. Manual intervention is required during training, and the coverage of the problem is relatively narrow. The output sentences are not smooth enough, and there are often problems with fixed word collocation loop output, or missing new subjects. Some strong anthropomorphic questions have different answers, giving people a bad impression of "split personality", and it is impossible to judge the input sentences. Whether it is a question of personality tendency, the recognition accuracy rate of strong anthropomorphism is poor, and the problem of identity consistency in dialogue cannot be solved. Whether it is used as a daily leisure or a knowledge question and answer in the professional field, the effect is not ideal;

[0010] Second, for a dialogue robot, it is often regarded by users as a virtual person with personalized characteristics, such as often asking: "How old are you", "What are your hobbies", "Where are you from?" "Etc. If the dialogue robot is regarded as a virtual person, then the personalized information related to the virtual person, such as age, gender, hobbies, language style and other personality characteristics information, should maintain the consistency of answers, and the existing technology is trained through deep learning. It is often difficult for the model to maintain this kind of personalized answer, because from the huge amount of information, it is impossible to distinguish which is personal anthropomorphic information and which is general information, and cannot solve the problem of identity consistency, which eventually leads to a poor anthropomorphic image of the dialogue robot;

[0011] Third, the existing deep learning-based intelligent dialogue robot is prone to lose new subjects in the process of returning answers from the model. For example, it should answer "he said he is very good", but the answer is "he is very good". In addition, Sometimes the "fixed collocation" phrase is repeated continuously, for example, the continuous output of "he said he said he said", causing the problem of losing the new subject and the problem of repeatedly outputting the fixed phrase, the communication, retrieval and service of the dialogue robot The efficiency is very low, the experience of key links is poor, the application is greatly restricted, and the actual use value is lost in some formal scenarios;

[0012] Fourth, the technology used in the existing intelligent dialogue robots is based on the hybrid technology of artificial templates and retrieval. The construction process of the model is not end-to-end data-driven. Even if the training data is processed, the training effect cannot be achieved immediately. A possible dialogue robot adds a lot of complicated intermediate steps such as semantic analysis and syntactic analysis. These steps cannot be avoided in NLP; it is directly related to the specific dialogue language, and the scalability is poor, and it cannot handle different languages ​​at the same time. Dialogue, because the semantics and sentence structures of different languages ​​are very different, and the grammatical structure analysis between languages ​​is basically not universal. Only after a detailed analysis of each language can it be used, and a lot of repetitive work will be done. Efficiency Very low; dialogue robots are far from being interesting, lively, and rich in diversity. They are often boring and obviously not human-like conversations. The practical value of information retrieval and online quick answering is low, and the entertainment is also very low. , and the dialogue robustness is not guaranteed

Images