AI digital human language training system and method

By introducing expression prediction neural network and ChatGPT language module in the AI ​​digital human language training system, combining self-confrontation learning and multi-domain training, the problem of inaccurate processing of key information by the ChatGPT model is solved, and AI digital human-to-natural language Understand and generate capabilities to ensure conversation quality and personalized replies.

CN118964553AInactive Publication Date: 2024-11-15NANJING INTEK DIGITAL TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410974248.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-11-15
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The ChatGPT model in AI digital virtual humans is inaccurate in processing key information, which may lead to user misleading.

Method used

An AI digital human language training system was designed, including expression prediction neural network module, ChatGPT language module and storage module. Through data set selection and preparation, language model design, self-confrontation learning and reinforcement learning, multi-domain training, combined with Transformer model Architecture and self-attention mechanism optimize the understanding and generation ability of language models, ensuring the accuracy of facial expressions and movements and the quality of conversations.

Benefits of technology

It realizes AI digital human understanding and generation of natural language, can have natural and smooth dialogue and communication with users, provide diversified services, and improves the model's adaptability and personalized dialogue response capabilities in different fields through multi-field training.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118964553A_ABST
    Figure CN118964553A_ABST
Patent Text Reader

Abstract

The invention discloses an AI digital human language training system and method, and the system comprises an expression prediction neural network module, a ChatGPT language module, and a storage module, and the expression prediction neural network module is used for extracting characters and languages for the ChatGPT language module, enabling an AI digital human to make corresponding facial expressions and font actions according to word meanings, and storing the characters and languages in the storage module. The expression prediction neural network module is used for expression prediction, the ChatGPT language module is used for selection and preparation of a data set, design of a language model and a network architecture, self-adversarial learning and reinforcement learning and multi-field training, and the storage module is used for storing data generated by the expression prediction neural network module and the ChatGPT language module. Through pre-training and fine tuning of large-scale data, understanding and generation of natural languages by the AI digital human are realized, and the AI digital human can carry out natural and smooth dialogue communication with the user.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of AI technology, and in particular to an AI digital human language training system and method. Background Art

[0002] Virtual human is a digital virtual character image generated based on artificial intelligence technology, which is often presented in the form of a combination of background environment and character screen. In the application scenario of digital human, the facial expressions and specific actions of the digital human are required to be driven in real time with the changes in the application scenario, and the background also changes in real time with the actions of the digital human. With the continuous development of artificial intelligence technology, virtual human has gradually become an increasingly important part of people's lives. ChatGPT, as a technology with revolutionary potential, provides new possibilities for the development of virtual human.

[0003] Although the development of AI digital virtual humans has broad prospects, the ChatGPT model in AI digital virtual humans has the problem of inaccurate processing of key information, which may mislead users. Therefore, an AI digital human language training system and method are provided to solve the above problems. Summary of the invention

[0004] In order to solve the problems in the above background technology, the technical solution adopted by the present invention to solve the technical problems is: an AI digital human language training system, which includes: an expression prediction neural network module, a ChatGPT language module, and a storage module. The expression prediction neural network module is used to extract text language for the ChatGPT language module, so that the AI ​​digital human makes corresponding facial expressions and font movements according to the meaning of the words. The ChatGPT language module is used for the selection and preparation of data sets, the design of language models and network architectures, self-adversarial learning and reinforcement learning, and multi-domain training. The storage module is used to store data generated by the expression prediction neural network module and the ChatGPT language module.

[0005] As a preferred technical solution of the present invention, the data set is used to help the AI ​​digital human language model better understand and learn the content and semantics of the conversation. The data set covers different topics and contexts, knowledge and context understanding capabilities.

[0006] As a preferred technical solution of the present invention, the expression prediction neural network module connects the network architecture and language model in the ChatGPT language module.

[0007] As a preferred technical solution of the present invention, the ChatGPT language module adopts the Transformer model architecture and uses a self-attention mechanism to enable the language model to better understand and utilize contextual information.

[0008] As a preferred technical solution of the present invention, the language model selection needs to balance the complexity and computational efficiency of the model to improve the speed and quality of dialogue generation, and the model parameters and hyperparameters suitable for the dialogue task are selected to achieve better performance and robustness.

[0009] As a preferred technical solution of the present invention, the self-adversarial learning is performed by allowing the AI ​​digital human to communicate with itself and adopting a reinforcement learning method to train the AI ​​digital human language model.

[0010] As a preferred technical solution of the present invention, the reinforcement learning helps the AI ​​digital human language model learn dialogue strategies and patterns faster by using high-quality responses generated during the training process as input.

[0011] As a preferred technical solution of the present invention, the multi-domain training classifies and divides the data into domains, and then performs independent training on each domain, so that the model can learn the differences and characteristics between domains.

[0012] A method for an AI digital human language training system, comprising the following steps:

[0013] Step 1: Create an expression prediction neural network module, ChatGPT language module, and storage module in the AI ​​digital human model;

[0014] Step 2: Use the data set in the ChatGPT language module to help the AI ​​digital human language model better understand and learn the content and semantics of the conversation, preprocess and clean the data, remove duplicate, noisy and unreasonable conversation samples, and improve the quality and reliability of the language model;

[0015] Step 3: Use the expression prediction neural network module to extract text language from the ChatGPT language module, so that the AI ​​digital human can make corresponding facial expressions and font movements according to the meaning of the words;

[0016] Step 4: ChatGPT language module, by using the self-attention mechanism, enables the language model to better understand and utilize contextual information. Language model selection needs to balance the complexity and computational efficiency of the model to improve the speed and quality of dialogue generation. Model parameters and hyperparameters suitable for dialogue tasks are selected to achieve better performance and robustness.

[0017] Step 5: Let the AI ​​digital human talk to itself through self-adversarial learning, and use reinforcement learning to train the AI ​​digital human language model, so that the model can optimize itself through continuous self-dialogue and improve the ability and quality of dialogue generation;

[0018] Step 6: Use the high-quality responses generated by reinforcement learning during the training process as input to help the AI ​​digital human language model learn dialogue strategies and patterns faster. Use progressive deconvolution and sampling strategies to gradually reduce the proportion of high-quality responses that are forced into the input, so that the AI ​​digital human language model can gradually generate dialogues independently.

[0019] Step 7: Classify and divide the data into domains through multi-domain training, and then perform independent training on each domain so that the model can learn the differences and characteristics between domains.

[0020] As a preferred technical solution of the present invention, in step seven, the data is classified and divided into fields through multi-field training, so that the AI ​​digital human language model can better adapt to the dialogue requirements of different fields, provide more personalized and professional dialogue responses, and improve the adaptability of the model in different fields.

[0021] The present invention has the following advantages: by pre-training and fine-tuning the ChatGPT language module in the AI ​​digital human through large-scale data, the AI ​​digital human can understand and generate natural language, and the AI ​​digital human can have natural and fluent conversations with users, providing users with diversified services and support;

[0022] Multi-domain training is used to classify and divide the data into domains, and then independent training is performed on each domain, so that the model can learn the differences and characteristics between domains, allowing the AI ​​digital human language model to better adapt to the dialogue requirements of different domains. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a schematic diagram of an AI digital human language training system according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present invention are described clearly and completely below. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Many specific details are explained in the following description to facilitate a full understanding of the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited by the specific embodiments disclosed below.

[0025] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] Please refer to Figure 1The present invention provides an AI digital human language training system, comprising: an expression prediction neural network module, a ChatGPT language module, and a storage module. The expression prediction neural network module is used to extract text language from the ChatGPT language module, so that the AI ​​digital human can make corresponding facial expressions and font movements according to the meaning of words. The ChatGPT language module is used for the selection and preparation of data sets, the design of language models and network architectures, self-adversarial learning and reinforcement learning, and multi-domain training. The storage module is used to store data generated by the expression prediction neural network module and the ChatGPT language module.

[0027] The dataset is used to help the AI ​​digital human language model better understand and learn the content and semantics of the conversation. The dataset covers different topics and contexts, knowledge and contextual understanding capabilities. The expression prediction neural network module connects the network architecture and language model in the ChatGPT language module. The ChatGPT language module adopts the Transformer model architecture. By using the self-attention mechanism, the language model can better understand and utilize contextual information.

[0028] Language model selection needs to balance the complexity of the model and computational efficiency to improve the speed and quality of dialogue generation. Model parameters and hyperparameters suitable for dialogue tasks are selected to achieve good performance and robustness. Self-adversarial learning allows AI digital humans to have dialogues with themselves and uses reinforcement learning to train the AI ​​digital human language model. This method enables the model to optimize itself through continuous self-dialogue and improve the ability and quality of dialogue generation. In addition, in order to prevent the AI ​​digital human model from generating unreasonable or inappropriate responses, the ChatGPT language module also introduces human supervision to guide and correct the model to ensure that the generated dialogue responses meet semantic and ethical norms. Reinforcement learning uses high-quality responses generated during the training process as input to help the AI ​​digital human language model learn dialogue strategies and patterns faster. Multi-domain training classifies and divides the data into domains, and then conducts independent training on each domain, so that the model can learn the differences and characteristics between domains.

[0029] A method for an AI digital human language training system, comprising the following steps:

[0030] Step 1: Create an expression prediction neural network module, ChatGPT language module, and storage module in the AI ​​digital human model;

[0031] Step 2: Use the data set in the ChatGPT language module to help the AI ​​digital human language model better understand and learn the content and semantics of the conversation, preprocess and clean the data, remove duplicate, noisy and unreasonable conversation samples, and improve the quality and reliability of the language model;

[0032] Step 3: Use the expression prediction neural network module to extract text language from the ChatGPT language module, so that the AI ​​digital human can make corresponding facial expressions and font movements according to the meaning of the words;

[0033] Step 4: ChatGPT language module, by using the self-attention mechanism, enables the language model to better understand and utilize contextual information. Language model selection needs to balance the complexity and computational efficiency of the model to improve the speed and quality of dialogue generation. Model parameters and hyperparameters suitable for dialogue tasks are selected to achieve better performance and robustness.

[0034] Step 5: Let the AI ​​digital human talk to itself through self-adversarial learning, and use reinforcement learning to train the AI ​​digital human language model, so that the model can optimize itself through continuous self-dialogue and improve the ability and quality of dialogue generation;

[0035] Step 6: Use the high-quality responses generated by reinforcement learning during the training process as input to help the AI ​​digital human language model learn dialogue strategies and patterns faster. Use progressive deconvolution and sampling strategies to gradually reduce the proportion of high-quality responses that are forced into the input, so that the AI ​​digital human language model can gradually generate dialogues independently.

[0036] Step 7: Classify and divide the data into domains through multi-domain training, and then conduct independent training in each domain so that the model can learn the differences and characteristics between domains, so that the AI ​​digital human language model can better adapt to the dialogue requirements of different domains, provide more personalized and professional dialogue responses, and improve the adaptability of the model in different domains.

[0037] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the attached claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved.

[0038] Other parts of the present invention not described in detail belong to the prior art and will not be described in detail here.

[0039] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An AI digital human language training system, characterized in that: include: An expression prediction neural network module, a ChatGPT language module, and a storage module. The expression prediction neural network module is used to extract text language from the ChatGPT language module, so that the AI ​​digital human can make corresponding facial expressions and font movements according to the meaning of words. The ChatGPT language module is used for the selection and preparation of data sets, the design of language models and network architectures, self-adversarial learning and reinforcement learning, and multi-domain training. The storage module is used to store data generated by the expression prediction neural network module and the ChatGPT language module.

2. The AI ​​digital human language training system according to claim 1, characterized in that: The dataset is used to help the AI ​​digital human language model better understand and learn the content and semantics of the conversation. The dataset covers different topics and contexts, knowledge and context understanding capabilities.

3. The AI ​​digital human language training system according to claim 1, characterized in that: The expression prediction neural network module connects the network architecture and language model in the ChatGPT language module.

4. The AI ​​digital human language training system according to claim 1, characterized in that: The ChatGPT language module adopts the Transformer model architecture and uses the self-attention mechanism to enable the language model to better understand and utilize contextual information.

5. The AI ​​digital human language training system according to claim 1, characterized in that: The language model selection needs to balance the complexity and computational efficiency of the model to improve the speed and quality of dialogue generation. Model parameters and hyperparameters suitable for dialogue tasks are selected to achieve better performance and robustness.

6. The AI ​​digital human language training system according to claim 1, characterized in that: The self-adversarial learning is achieved by allowing the AI ​​digital human to communicate with itself and using reinforcement learning to train the AI ​​digital human language model.

7. The AI ​​digital human language training system according to claim 1, characterized in that: The reinforcement learning helps the AI ​​digital human language model learn conversation strategies and patterns faster by using high-quality responses generated during the training process as input.

8. The AI ​​digital human language training system according to claim 1, characterized in that: The multi-domain training classifies and divides the data into domains and then performs independent training on each domain, so that the model can learn the differences and characteristics between domains.

9. A method for an AI digital human language training system as described in any one of claims 1 to 8, characterized in that: The following steps are involved: Step 1: Create an expression prediction neural network module, ChatGPT language module, and storage module in the AI ​​digital human model; Step 2: Use the data set in the ChatGPT language module to help the AI ​​digital human language model better understand and learn the content and semantics of the conversation, preprocess and clean the data, remove duplicate, noisy and unreasonable conversation samples, and improve the quality and reliability of the language model; Step 3: Use the expression prediction neural network module to extract text language from the ChatGPT language module, so that the AI ​​digital human can make corresponding facial expressions and font movements according to the meaning of the words; Step 4: ChatGPT language module, by using the self-attention mechanism, enables the language model to better understand and utilize contextual information. Language model selection needs to balance the complexity and computational efficiency of the model to improve the speed and quality of dialogue generation. Model parameters and hyperparameters suitable for dialogue tasks are selected to achieve better performance and robustness. Step 5: Let the AI ​​digital human talk to itself through self-adversarial learning, and use reinforcement learning to train the AI ​​digital human language model, so that the model can optimize itself through continuous self-dialogue and improve the ability and quality of dialogue generation; Step 6: Use the high-quality responses generated by reinforcement learning during the training process as input to help the AI ​​digital human language model learn dialogue strategies and patterns faster. Use progressive deconvolution and sampling strategies to gradually reduce the proportion of high-quality responses that are forced into the input, so that the AI ​​digital human language model can gradually generate dialogues independently. Step 7: Classify and divide the data into domains through multi-domain training, and then perform independent training on each domain so that the model can learn the differences and characteristics between domains.

10. The method of the AI ​​digital human language training system according to claim 9, characterized in that: In step seven, the data is classified and divided into domains through multi-domain training, so that the AI ​​digital human language model can better adapt to the dialogue requirements of different fields, provide more personalized and professional dialogue responses, and improve the adaptability of the model in different fields.