Intelligent robot chat context implementation method and system based on corpus annotation

Abstract

The invention discloses an effective method and system for context information transmission in intelligent robot chatting. The method can help the chat robot to realize accurate multi-round conversation. Context information stored in the system is updated in each round of chatting in real time by simulating a human dialogue process. Under the appropriate condition, the current round input of chatting combined with the current preceding text information, an instruction is generated, namely a function and parameters which can be executed by software are generated, and therefore the system synchronously updates the preceding text information which can be used by the chatting below while generating the current round reply. According to the method, a certain amount of training data is marked, namely part of training corpora is changed, a deep neural network model is indicated in an explicit mode, and when a user asks for questions, a robot replies and when there is the preceding text information, how to answer and reserve the preceding text information is achieved. The invention provides a plurality of practical data annotation schemes and a method for annotating corpora by using the data annotation schemes. Meanwhile, some sample application scenarios and related technical details are also provided.

Description

technical field [0001] The invention belongs to the field of intelligent chatting robots, and in particular relates to a method and system for realizing a chatting context of an intelligent robot based on corpus annotation. Background technique [0002] In recent years, artificial intelligence technology has continued to mature, especially deep learning technology has continued to develop, and the development and use of chatbots have gradually become popular. In theory, chatbots can not only communicate with humans emotionally, but can also be widely used in various customer service systems and consulting systems, saving a lot of human resources and improving work efficiency. [0003] There are many ways to implement chatbots, including the early rule-based ones, the later query-based or retrieval-based ones, and the later generative ones based on deep learning models. But no matter which way the chat robot is implemented, the capture and delivery of contextual information ...

AI Technical Summary

Problems solved by technology

[0005] (1) Since there is no explicit context information training, the accuracy of the model to grasp the context information is very low;

[0006] (2) Many systems that provide multi-round dialogue functions can only know the current main topic, such as knowing whether they are talking about food, emotion, or games, but they cannot communicate accurately in combination with current user questions or statements;

[0007] (3) The multi-round dialogue in the dialogue management method based on slot filling is completely limited to the preset rules, and cannot adapt to the questions outside the rules in the user chat;

[0009] (5) During training, with the increase in the number of multiple rounds of dialogue, the demand for the number of training corpora increases exponentially, and very high requirements are placed on the size and computing power of the model. The computational complexity of model reasoning will also be uncontrollable;

[0010] (6) For many chat systems, even if the dialogue in a specific scene is found to be unsatisfactory, it is difficult to optimize it in a targeted manner.

[0011] In addition, when some dialogue systems are faced with changing user expressions and need to output complex replies, for example, when replies need to call other models, access the background knowledge base, or query the network, and output multimedia information, the existing methods, For example, through keyword search, etc., it is also difficult to accurately understand user intentions and generate ideal replies

Images