Man-machine conversation method based on AI intelligent large model

By constructing and pre-training AI intelligent big model, the problem of difficult analysis of word meaning and sentence meaning in human-computer dialogue in Chinese context is solved, and accurate understanding and response to different contexts and dialects is achieved, improving user experience and dialogue quality is improved. , supports cross-cultural communication.

CN118917437AInactive Publication Date: 2024-11-08南京千雪智能科技有限公司

Patent Information

Application Number
CN202411003178.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-25
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

In human-computer dialogue in Chinese context, the meanings and sentences are different in different contexts and dialect usage habits, which makes it difficult to accurately analyze human-computer dialogue and prone to incorrect answers.

Method used

By constructing and pre-training AI intelligent big models, learning the meaning of word in different contexts and dialects, combining multiple rounds of dialogue context information, generating natural language replies, and conducting targeted training based on the domain and role attributes selected by the user, increasing dialogue management and A secure filtering mechanism supports users' personalized interactive needs.

Benefits of technology

Improve the accuracy and personalized responsiveness of the dialogue system in different contexts and dialects, enhance user experience and satisfaction, support cross-cultural communication, promote social cohesion, and improve dialogue quality through continuous learning and optimization .

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses a man-machine dialogue method based on an AI intelligent large model, relates to the technical field of artificial intelligence man-machine dialogue methods, and aims to solve the problems that during man-machine dialogue of Chinese contexts, word meanings and sentences are different in different contexts and different dialect using habits, the man-machine dialogue is difficult to accurately analyze, and the man-machine dialogue effect is poor. The method comprises the following steps: step 1, constructing a large model, and carrying out pre-training through text data; 2, carrying out fine adjustment on a specific task or field; 3, understanding a historical dialogue; 4, generating a natural language reply; 5, setting a dialogue management capability; 6, learning different dialogue styles and emotions; 7, adding and implementing a filtering mechanism and setting a dialogue rule; 8, the user sets the required field and sets the type of the interaction role; 9, learning domain knowledge and corresponding expression modes; and step 10, replying and collecting and feeding back optimization performance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence human-computer dialogue methods, and specifically to a human-computer dialogue method based on an AI intelligent big model. Background Art

[0002] The human-computer dialogue method based on AI intelligent big model is a technology that simulates human dialogue behavior, which allows computer systems to communicate with humans in natural language. This technology combines advanced artificial intelligence technologies such as natural language processing, machine learning, and deep learning to achieve an efficient and intelligent dialogue system. With the advancement of technology, the human-computer dialogue system will become more intelligent, better able to understand complex language environments and user intentions, and provide a richer and more natural interactive experience. In addition, multimodal dialogue systems, combining multiple interaction methods such as text, voice, and images, are also an important direction for future development.

[0003] The Chinese patent publication number is CN206839959U, which discloses a human-computer dialogue method, device, electronic device and storage medium. The human-computer dialogue method includes: receiving a conversation voice of a user in a current round, and preprocessing the conversation voice to obtain text information, wherein the preprocessing includes text conversion and text error correction; processing the text information through a preset semantic analysis big model to obtain intention information, wherein the intention information at least includes the intention corresponding to the current round of the user; obtaining historical response information, and determining the conversation state of the current round according to the historical response information and the intention information, wherein the historical response information includes response information generated according to the conversation state of the previous round of dialogue; configuring the response information corresponding to the conversation state according to a preset response configuration big model, and generating a response voice corresponding to the response information, wherein the preset response configuration big model includes at least one of the following: a dialogue strategy learning big model, a knowledge base question and answer big model, which solves the problem of low dialogue efficiency and poor dialogue effect of the dialogue system in the related art, realizes the rapid and effective realization of the AI ​​robot outbound call function in various scenarios, reduces labor costs, and improves the beneficial effects of dialogue efficiency and dialogue effect.

[0004] The above-mentioned existing technical solutions have the following defects: in human-computer dialogue in the Chinese context, the meaning of words and sentences are different in different contexts and different dialect usage habits, and it is difficult for human-computer dialogue to accurately analyze this, so it is easy to give incorrect answers during the dialogue. Therefore, we propose a human-computer dialogue method based on an AI intelligent big model to solve the above-mentioned problems. Summary of the invention

[0005] The purpose of the present invention is to provide a human-computer dialogue method based on an AI intelligent large model to solve the problem raised in the above background technology that in human-computer dialogue in the Chinese context, the meaning of words and sentences are different in different contexts and different dialect usage habits, and it is difficult for human-computer dialogue to accurately analyze this, which makes it easy to give incorrect answers during the dialogue.

[0006] To achieve the above object, the present invention provides the following technical solution: a human-computer dialogue method based on an AI intelligent big model, comprising the following steps:

[0007] Step 1: Build a large model and pre-train it with text data. When analyzing text data, the meanings of words in different contexts and dialects are entered separately. The large model can learn the structure, semantics and contextual association of the language. The pre-training stage is the basis for the large model to acquire extensive knowledge and language comprehension capabilities.

[0008] Step 2: After pre-training, the large model needs to be fine-tuned for specific tasks or fields to adapt to specific tasks or fields and optimize its performance in specific dialogue scenarios. This usually involves training on dialogue datasets to allow the large model to learn more specific human-computer dialogue patterns and strategies.

[0009] Step 3: Add a program to allow the large model to understand the history of the conversation, that is, the content of the previous conversation, so as to generate coherent and appropriate responses. This requires the large model to be able to process and remember the context information of multiple rounds of conversations;

[0010] Step 4: Based on the understood context, the large model uses generative technology to generate natural language responses, making the generated responses consistent with the logic of the conversation and simulating human expression;

[0011] Step 5: Set up the dialogue management capability of the large model, that is, be able to adjust the dialogue strategy according to the progress and goals of the dialogue, which may include topic guidance, question clarification, information confirmation, etc.

[0012] Step 6: The large model edits the learning program so that it can adapt to different conversation styles and emotions. This may involve learning the tone, politeness, and sense of humor of different speakers.

[0013] Step 7: The big model adds analysis procedures to implement filtering mechanisms and set dialogue rules. When designing a human-computer dialogue system, it involves security and ethical issues. The big model will not generate harmful, misleading or inappropriate content.

[0014] Step 8: When interacting with the user, the user can set the required fields according to the usage requirements, and can also set the system interaction language to express the gender, age and personality of the character according to the user's preferences;

[0015] Step 9: Further learn the professional knowledge of the above user-selected field based on the field selected by the user, and learn the corresponding expression method based on the role attributes selected by the user;

[0016] Step 10: Analyze the questions raised by the user, determine the user's context and dialect usage habits through the word meaning and sentence intent of the analyzed sentences, and then analyze the user's question according to the corresponding situation, and then respond to it and collect user feedback on the big model's response to further optimize the performance of the big model. The big model can continuously learn and improve from the interaction with the user, thereby improving the quality of conversation and user experience.

[0017] Preferably, the pre-trained text data in step 1 is first pre-processed, and the sentences are divided into tokens by tokenization, and the tokens can be a word, multiple words, phrases or characters. Then each token is indexed to convert the word into a digital sequence to facilitate subsequent processing, and then the text data is converted into mathematical vectors and matrices. The preferred methods include bag-of-words and word embedding, the latter of which can retain the context information of the word and represent the relationship between semantics through distance.

[0018] Preferably, the generation technique in step 4 is one of a sequence-to-sequence large model and a transformer architecture.

[0019] Preferably, the optional fields in step 8 are: at least one of the fields of education, health consultation and entertainment games. The specific fields of education, health consultation and entertainment games can be expanded more specifically according to usage needs, such as educational subjects or student grades, departments in health consultation and categories of entertainment games.

[0020] Preferably, the conversation field, language expression role, conversation habits, and past conversation content selected by the user in steps 8, 9, and 10 are stored and directly called up during subsequent conversations. The customer can also change the conversation field and language expression role again according to his or her preferences.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1. During pre-training, the present invention enters the meanings of words in different contexts and dialects in the input text data, analyzes the questions raised by the user, and judges the context and dialect usage habits of the user through the word meaning and sentence meaning of the analyzed sentences, so as to analyze the user's questions according to the corresponding situation. Context is one of the key factors in understanding the meaning of language. If the dialogue system can judge different contexts, it can more accurately understand the user's speech intention and avoid misunderstanding and ambiguity. For example, the same words may have different meanings and tones in formal occasions and daily chats. At the same time, supporting dialects means that the system can identify and understand the regional language characteristics of users, thereby providing responses that are closer to user needs. When using the dialogue system, users expect to communicate in a natural and smooth way. When the system can accurately capture and respond to needs in different contexts and support the dialects that users are accustomed to using, the user's communication experience will be greatly improved. This personalized communication method makes users feel respected and understood, thereby enhancing user satisfaction and loyalty. Dialects are an important part of regional culture. Supporting dialects means that the dialogue system can cross the boundaries of language and culture and establish connections with a wider range of people. This inclusiveness not only helps to protect and inherit local culture, but also promotes mutual understanding and respect between different regions and different groups of people, enhances social cohesion and centripetal force, and solves the problem that in human-computer dialogue in the Chinese context, the meaning of words and sentences are different in different contexts and different dialect usage habits, making it difficult for human-computer dialogue to accurately analyze this, which makes it easy to give incorrect answers during dialogue.

[0023] 2. The present invention further learns the professional knowledge of the above user-selected field according to the field selected by the user, and learns the corresponding expression method according to the role attribute selected by the user. Through training for a specific field, the dialogue system can deeply learn and understand the professional terms, industry specifications and common problems in the field. This enables the system to provide more accurate and professional answers and suggestions when facing users in the field. Different fields have their own unique contexts and expressions. Through targeted training, the dialogue system can better adapt to these characteristics and reduce misunderstandings caused by context differences. Users can choose the field of the dialogue system according to their needs, so as to obtain a more personalized service experience. For example, users in the medical field can get professional medical consultation, and users in the education field can get help in learning. Field-specific training enables the dialogue system to understand the user's problems and needs more quickly, thereby providing a more efficient communication experience. Users do not need to spend extra time explaining background information or clarifying problems, and directly obtain the required answers or solutions. By allowing users to select fields for training, the dialogue system can be applied to a wider range of scenarios and fields. Whether it is finance, medical care, education or entertainment, the needs of different users can be met through targeted training. Exclusive dialogue system field training programs can be customized according to their own business needs and user group characteristics. This helps improve the service quality and brand image of enterprises and enhance user stickiness and loyalty. Field-specific training requires continuous optimization of algorithms and models to adapt to the characteristics and needs of different fields. This helps promote the innovation and development of dialogue system technology and improve the overall performance and stability of the system. By collecting and analyzing user data and feedback from different fields, it can provide strong support for the research and development of dialogue systems. Researchers can optimize algorithms and improve models based on these data, thereby promoting the continuous advancement of dialogue system technology. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments.

[0025] Embodiment 1: A human-computer dialogue method based on an AI intelligent big model, comprising the following steps:

[0026] Step 1: Build a large model and pre-train it with text data. When analyzing text data, the meanings of words in different contexts and dialects are entered separately. The large model can learn the structure, semantics and contextual association of the language. The pre-training stage is the basis for the large model to acquire extensive knowledge and language comprehension capabilities.

[0027] Step 2: After pre-training, the large model needs to be fine-tuned for specific tasks or fields to adapt to specific tasks or fields and optimize its performance in specific dialogue scenarios. This usually involves training on dialogue datasets to allow the large model to learn more specific human-computer dialogue patterns and strategies.

[0028] Step 3: Add a program to allow the large model to understand the history of the conversation, that is, the content of the previous conversation, so as to generate coherent and appropriate responses. This requires the large model to be able to process and remember the context information of multiple rounds of conversations;

[0029] Step 4: Based on the understood context, the large model uses generative technology to generate natural language responses, making the generated responses consistent with the logic of the conversation and simulating human expression;

[0030] Step 5: Set up the dialogue management capability of the large model, that is, be able to adjust the dialogue strategy according to the progress and goals of the dialogue, which may include topic guidance, question clarification, information confirmation, etc.

[0031] Step 6: The large model edits the learning program so that it can adapt to different conversation styles and emotions. This may involve learning the tone, politeness, and sense of humor of different speakers.

[0032] Step 7: The big model adds analysis procedures to implement filtering mechanisms and set dialogue rules. When designing a human-computer dialogue system, it involves security and ethical issues. The big model will not generate harmful, misleading or inappropriate content.

[0033] Step 8: When interacting with the user, the user can set the required fields according to the usage requirements, and can also set the system interaction language to express the gender, age and personality of the character according to the user's preferences;

[0034] Step 9: Further learn the professional knowledge of the above user-selected field based on the field selected by the user, and learn the corresponding expression method based on the role attributes selected by the user;

[0035] Step 10: Analyze the questions raised by the user, determine the user's context and dialect usage habits through the word meaning and sentence intent of the analyzed sentences, and then analyze the user's question according to the corresponding situation, and then respond to it and collect user feedback on the big model's response to further optimize the performance of the big model. The big model can continuously learn and improve from the interaction with the user, thereby improving the quality of conversation and user experience.

[0036] Furthermore, the pre-trained text data in step 1 is first pre-processed, and the sentences are divided into tokens by tokenization. The tokens can be a word, multiple words, phrases or characters. Each token is then indexed to convert the word into a digital sequence to facilitate subsequent processing. The text data is then converted into mathematical vectors and matrices. The preferred methods include bag-of-words and word embedding, the latter of which can retain the contextual information of the word and represent the relationship between semantics through distance.

[0037] Furthermore, the generation technique in step 4 is one of a sequence-to-sequence large model and a transformer architecture.

[0038] Furthermore, the optional fields in step 8 are: at least one of the fields of education, health consultation and entertainment games. The specific fields of education, health consultation and entertainment games can be expanded more specifically according to usage needs, such as educational subjects or student grades, departments in health consultation and categories of entertainment games.

[0039] Furthermore, the conversation areas, language expression roles, conversation habits, and past conversation contents selected by the user in steps 8, 9, and 10 are stored and directly called up during subsequent conversations. The customer can also change the conversation areas and language expression roles again according to his / her preferences.

[0040] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be considered exemplary and non-restrictive in all respects, and the scope of the present invention is defined by the appended 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 be included in the present invention.

Claims

1. A human-computer dialogue method based on AI intelligent big model, characterized in that: The following steps are involved: Step 1: Build a large model and pre-train it with text data. When analyzing text data, the meanings of words in different contexts and dialects are entered separately. The large model can learn the structure, semantics and contextual association of the language. The pre-training stage is the basis for the large model to acquire extensive knowledge and language comprehension capabilities. Step 2: After pre-training, the large model needs to be fine-tuned for specific tasks or fields to adapt to specific tasks or fields and optimize its performance in specific dialogue scenarios. This usually involves training on dialogue datasets to allow the large model to learn more specific human-computer dialogue patterns and strategies. Step 3: Add a program to allow the large model to understand the history of the conversation, that is, the content of the previous conversation, so as to generate coherent and appropriate responses. This requires the large model to be able to process and remember the context information of multiple rounds of conversations; Step 4: Based on the understood context, the large model uses generative technology to generate natural language responses, making the generated responses consistent with the logic of the conversation and simulating human expression; Step 5: Set up the dialogue management capability of the large model, that is, be able to adjust the dialogue strategy according to the progress and goals of the dialogue, which may include topic guidance, question clarification, information confirmation, etc. Step 6: The large model edits the learning program so that it can adapt to different conversation styles and emotions. This may involve learning the tone, politeness, and sense of humor of different speakers. Step 7: Large model. When designing a human-computer dialogue system, safety and ethical issues are involved. The large model will not generate harmful, misleading or inappropriate content. Step 8: When interacting with the user, the user can set the required fields according to the usage requirements, and can also set the system interaction language to express the gender, age and personality of the character according to the user's preferences; Step 9: Further learn the professional knowledge of the above user-selected field based on the field selected by the user, and learn the corresponding expression method based on the role attributes selected by the user; Step 10: Analyze the questions raised by the user, determine the user's context and dialect usage habits through the word meaning and sentence intent of the analyzed sentences, and then analyze the user's question according to the corresponding situation, and then respond to it and collect user feedback on the big model's response to further optimize the performance of the big model. The big model can continuously learn and improve from the interaction with the user, thereby improving the quality of conversation and user experience.

2. According to claim 1, a human-computer dialogue method based on AI intelligent big model is characterized in that: The pre-trained text data in step 1 is first pre-processed, and the sentences are divided into tokens by tokenization. The tokens can be a word, multiple words, phrases or characters. Each token is then indexed to convert the word into a digital sequence to facilitate subsequent processing, and then the text data is converted into mathematical vectors and matrices. The preferred methods include bag-of-words and word embedding, the latter of which can retain the contextual information of the word and represent the relationship between semantics through distance.

3. According to claim 1, a human-computer dialogue method based on AI intelligent big model is characterized in that: The generation technique in step 4 is one of a sequence-to-sequence large model and a transformer architecture.

4. The human-computer dialogue method based on AI intelligent big model according to claim 1 is characterized in that: The optional fields in step 8 are: at least one of the fields of education, health consultation and entertainment games. The specific fields of education, health consultation and entertainment games can be expanded more specifically according to usage needs, such as educational subjects or student grades, departments in health consultation and categories of entertainment games.

5. The human-computer dialogue method based on AI intelligent big model according to claim 1 is characterized in that: The dialogue domain, language expression role, dialogue habits, and past dialogue content selected by the user in steps 8, 9, and 10 are stored and directly called up during subsequent dialogues. The customer can also change the dialogue domain and language expression role again according to his or her preferences.

Citation Information

Patent Citations

  • Portable screw fastener

    CN206839959U

Cited By

  • Large model dialogue control method based on role enhancement

    CN119441442A

  • Intelligent installation and maintenance helmet and large model voice interaction method and system based on Bluetooth connection

    CN120783765A