CN117112760A - Intelligent education big model based on knowledge base - Google Patents

Abstract

The invention belongs to the technical field of intelligent education, in particular to an intelligent education big model based on a knowledge base, which comprises a big model training and fine tuning module, a voice recognition module, a text preprocessing module and an intelligent dialogue module, wherein the voice recognition module and the text preprocessing module cooperatively process user input, and the intelligent dialogue module takes the big model as a base, cooperates with the voice recognition and text preprocessing module, responds to a user request to conduct intelligent reasoning and replies based on prior information provided by a self-research knowledge base. Specifically, the voice recognition module is used for processing voice input of a user in a complex noise environment, and then the voice input is converted into text information by the text preprocessing module and is input into the subsequent intelligent dialogue module. According to the invention, the safety and controllability of the model are enhanced, the safety shielding design is carried out in the process of training corpus and outputting the model, the safety and privacy of educational users can be better protected, and the model is prevented from misleading students or generating harmful information.

Description

A large model of intelligent education based on knowledge base

Technical field

The invention belongs to the field of intelligent education technology, and is specifically a large intelligent education model based on a knowledge base.

Background technique

In the development of artificial intelligence, the creation and application of large-scale pre-trained models have made remarkable achievements, especially in data extraction and decision support. However, although these models have achieved great progress, they still have to face the challenges of security and reliability. There are still significant problems and limitations when it comes to controllability and its application in vertical industries such as education.

First of all, from the perspective of security and controllability, the development of large-scale models originated abroad. There may be security risks when they are commercially applied in China, especially in scenarios involving confidential information. In addition, training corpus for large models is often lacking. Continuous learning and adaptation to Chinese and domestic semantic environments have resulted in relatively few large models based on Chinese. In vertical application scenarios, such as the intelligent education industry, large models are currently mainly used in conversational question and answer, but in specific application scenario research There are still some shortcomings.

Secondly, the existing pre-training model training data mostly comes from the Internet. This data knowledge is not equivalent to professional knowledge in the education field. Therefore, when existing large models face the transfer of knowledge in professional fields such as physics, chemistry, English, etc., it may Difficulties will arise, which may lead to users in the education industry receiving misleading information, and there is not yet an effective solution strategy or solution to these problems.

Therefore, a large intelligent education model based on knowledge base is needed to solve the above-mentioned problems.

Contents of the invention

In view of the above situation, in order to overcome the shortcomings of the existing technology, the present invention provides a large intelligent education model based on a knowledge base, which effectively solves the problems raised in the above background technology.

In order to achieve the above objectives, the present invention provides the following technical solution: a large intelligent education model based on a knowledge base, including a large model training and fine-tuning module, a speech recognition module, a text preprocessing module and an intelligent dialogue module. The speech recognition module and The text preprocessing module collaboratively processes user input. The intelligent dialogue module is based on a large model and collaborates with the speech recognition and text preprocessing modules to respond to user requests, perform intelligent reasoning and reply. Specifically, the speech recognition module is used to process the user's voice input in a complex noise environment, and then the text preprocessing module converts it into text information and inputs it into the subsequent intelligent dialogue module, which is handed over to the large model for inference prediction and reply. At the same time, by enhancing the security and controllability of the model, the present invention performs a security shielding design on the model training corpus and output post-processing, which can better protect the security and privacy of education users and prevent the model from misleading students or generating harmful information.

Speech recognition module: This module is mainly used to process the user's voice input in complex and noisy environments, and is responsible for a series of audio preprocessing tasks such as voice denoising. Its semi-transcribed speech signal will then be sent to subsequent text preprocessing. Module for further speech-to-text process. The design and workflow of the speech recognition module are:

S1. Use audio data sets containing various background sounds/noises to fine-tune the speech recognition module to improve its robustness;

S2. Collect the user's pronunciation of the specified text to determine the user's pronunciation accuracy and pronunciation preference, and further fine-tune the speech recognition module to improve its recognition accuracy;

S3. During the use of the system, preprocessing steps such as noise reduction and correction are performed on the audio input by the user, and the output is output to subsequent modules;

Text preprocessing module: This module is used to receive data from the speech recognition module and complete tasks such as speech transcription, word segmentation, and text vectorization. The design and workflow of the text preprocessing module are:

S1. For the output from the speech recognition module, perform speech-to-text transcription; perform word segmentation on the converted text and segment it into a sequence of words or sub-words;

S2. Analyze whether the words in the sentence contain non-learning related vocabulary and illegal content. If so, send a corresponding prompt message to remind the user and ask the user to re-enter or refuse to respond;

S3. Perform part-of-speech tagging for each word;

S4. Identify and mark named entities in text;

S5. Split the text into independent sentences as input to the intelligent dialogue module;

Intelligent dialogue module that collaborates with the self-developed knowledge base: This module is designed to be further fine-tuned based on the domestic large model, and collaborates with the speech recognition and text preprocessing modules to respond to user requests and perform intelligent intelligent dialogue based on the prior information provided by the self-developed knowledge base. Reason and reply. The workflow of the intelligent dialogue module is:

S1. Based on machine learning and deep learning algorithms, use the domestic Chinese base model to fine-tune on high-quality educational data, so that it has the ability to identify grammatical errors, correct confusing words, etc., and is more suitable for educational scenarios, obtaining a dedicated large-scale education-oriented Model;

S2. Users set preferred scenarios for voice practice, such as business conversations, English speeches, daily conversations, etc., to fine-tune the usage scenarios of the intelligent dialogue module;

S3. Update and maintain the self-researched knowledge base, and promptly update and expand the knowledge reserve of the model;

S4. The fine-tuned large language model generates the response text required by the user based on its background knowledge, scene-adaptive prompt words and highly time-sensitive knowledge text;

S5. Use the voice conversion sub-module in the intelligent dialogue module to generate the voice corresponding to the response text and feed it back to the user.

Further, based on the machine learning and deep learning algorithms, the domestic Chinese base model is used to fine-tune S1 on high-quality education data. The workflow is as follows:

S1-1. Use crawler and other technologies to construct tens of millions of scale fine-tuning data sets from Internet data to conduct supervised training of the base model to enhance its responsiveness and reasoning performance for education-related knowledge;

S1-2. Facing the education industry, collect highly relevant high-quality data in a targeted manner for further fine-tuning of parameters to enhance the knowledge understanding of the model in detailed professional fields;

S1-3. Based on the feedback of professional education practitioners on the model output content, the model can learn the language style of professionals and the analysis process of specific problems to enhance the professionalism of the model output.

Further, update and maintain the self-researched knowledge base, and timely update and expand the model's knowledge reserve S3. The workflow is as follows:

S3-1. Based on the data collection sub-module, capture and update document knowledge related to each subject in the database in real time from the Internet, such as textbooks, lesson plans, text information in video media, etc., forming a massive and timely database;

S3-2. Based on the data segmentation sub-module, it is classified and segmented according to the characteristics of knowledge to build a vector database for model analysis and use;

S3-3. Based on the user query vectorization module, project the structured text information into the embedding space, extract the feature vector of the text, and combine it with vector retrieval technology to obtain and return the most relevant feature vector from the database, thereby realizing knowledge Quick Search.

Compared with the prior art, the beneficial effects of the present invention are:

1. This invention collects tens of millions of real data in the industry through self-developed large models dedicated to education, and completes efficient intelligent technology deployment and practical application in the education industry.

2. The present invention has application expansion capabilities. By enhancing the proprietary training data of the model, it can realize the migration of these technical applications from English to more educational subjects, such as Chinese, mathematics, physics, chemistry, etc., to achieve wider social benefits. .

3. By enhancing the security and controllability of the model, the present invention performs a security shielding design on the model training corpus and output post-processing, which can better protect the security and privacy of education users and prevent the model from misleading students or generating harmful information.

Description of drawings

The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. In the attached picture:

Figure 1 is a schematic diagram of the module of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, not all of them; based on The embodiments of the present invention and all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

Embodiment 1, as shown in Figure 1, the present invention discloses a large intelligent education model based on a knowledge base. A large intelligent education model based on a knowledge base includes a large model training and fine-tuning module, a speech recognition module, and a text pre-processing module. processing module and intelligent dialogue module. The speech recognition module and the text preprocessing module cooperate to process user input. The intelligent dialogue module uses a large model as a base to cooperate with the speech recognition and text preprocessing module to respond to user requests, perform intelligent reasoning and reply. . Specifically, the speech recognition module is used to process the user's voice input in a complex noise environment, and then the text preprocessing module converts it into text information and inputs it into the subsequent intelligent dialogue module, which is handed over to the large model for inference prediction and reply. At the same time, by enhancing the security and controllability of the model, the present invention performs a security shielding design on the model training corpus and output post-processing, which can better protect the security and privacy of education users and prevent the model from misleading students or generating harmful information.

Speech recognition module: This module is mainly used to process the user's voice input in complex and noisy environments, and is responsible for a series of audio preprocessing tasks such as voice denoising. Its semi-transcribed speech signal will then be sent to subsequent text preprocessing. Module for further speech-to-text process. The design and workflow of the speech recognition module are:

S1. Use audio data sets containing various background sounds/noises to fine-tune the speech recognition module to improve its robustness;

S2. Collect the user's pronunciation of the specified text to determine the user's pronunciation accuracy and pronunciation preference, and further fine-tune the speech recognition module to improve its recognition accuracy;

S3. During the use of the system, preprocessing steps such as noise reduction and correction are performed on the audio input by the user, and the output is output to subsequent modules;

Text preprocessing module: This module is used to receive data from the speech recognition module and complete tasks such as speech transcription, word segmentation, and text vectorization. The design and workflow of the text preprocessing module are:

S1. For the output from the speech recognition module, perform speech-to-text transcription; perform word segmentation on the converted text and segment it into a sequence of words or sub-words;

S2. Analyze whether the words in the sentence contain non-learning related vocabulary and illegal content. If so, send a corresponding prompt message to remind the user and ask the user to re-enter or refuse to respond;

S3. Perform part-of-speech tagging for each word;

S4. Identify and mark named entities in text;

S5. Split the text into independent sentences as input to the intelligent dialogue module;

Intelligent dialogue module that collaborates with the self-developed knowledge base: This module is designed to be further fine-tuned based on the domestic large model, and collaborates with the speech recognition and text preprocessing modules to respond to user requests and perform intelligent intelligent dialogue based on the prior information provided by the self-developed knowledge base. Reason and reply. The workflow of the intelligent dialogue module is:

S1. Based on machine learning and deep learning algorithms, use the domestic Chinese base model to fine-tune on high-quality educational data, so that it has the ability to identify grammatical errors, correct confusing words, etc., and is more suitable for educational scenarios, obtaining a dedicated large-scale education-oriented Model;

S2. Users set preferred scenarios for voice practice, such as business conversations, English speeches, daily conversations, etc., to fine-tune the usage scenarios of the intelligent dialogue module;

S3. Update and maintain the self-researched knowledge base, and promptly update and expand the knowledge reserve of the model;

S4. The fine-tuned large language model generates the response text required by the user based on its background knowledge, scene-adaptive prompt words and highly time-sensitive knowledge text;

S5. Use the voice conversion sub-module in the intelligent dialogue module to generate the voice corresponding to the response text and feed it back to the user.

Further, based on the machine learning and deep learning algorithms, the domestic Chinese base model is used to fine-tune S1 on high-quality education data. The workflow is as follows:

S1-1. Use crawler and other technologies to construct tens of millions of scale fine-tuning data sets from Internet data to conduct supervised training of the base model to enhance its responsiveness and reasoning performance for education-related knowledge;

S1-2. Facing the education industry, collect highly relevant high-quality data in a targeted manner for further fine-tuning of parameters to enhance the knowledge understanding of the model in detailed professional fields;

S1-3. Based on the feedback of professional education practitioners on the model output content, the model can learn the language style of professionals and the analysis process of specific problems to enhance the professionalism of the model output.

Further, update and maintain the self-researched knowledge base, and timely update and expand the model's knowledge reserve S3. The workflow is as follows:

S3-1. Based on the data collection sub-module, capture and update document knowledge related to each subject in the database in real time from the Internet, such as textbooks, lesson plans, text information in video media, etc., forming a massive and timely database;

S3-2. Based on the data segmentation sub-module, it is classified and segmented according to the characteristics of knowledge to build a vector database for model analysis and use;

S3-3. Based on the user query vectorization module, project the structured text information into the embedding space, extract the feature vector of the text, and combine it with vector retrieval technology to obtain and return the most relevant feature vector from the database, thereby realizing knowledge Quick Search.

It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that these entities or operations are mutually exclusive. any such actual relationship or sequence exists between them. Furthermore, the terms "comprises," "comprises," or any other variations thereof are intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus that includes a list of elements includes not only those elements, but also those not expressly listed other elements, or elements inherent to the process, method, article or equipment.

Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art will understand that various changes, modifications, and substitutions can be made to these embodiments without departing from the principles and spirit of the invention. and modifications, the scope of the invention is defined by the appended claims and their equivalents.

Claims (3)

1. A large intelligent education model based on a knowledge base, including a speech recognition module, a text processing module and an intelligent dialogue module, characterized in that: the speech recognition module and the text preprocessing module collaboratively process user input, and the intelligent dialogue module Using the large model as the base, it collaborates with the speech recognition and text preprocessing modules, and based on the prior information provided by the self-developed knowledge base, responds to user requests for intelligent reasoning and replies; Speech recognition module: This module is mainly used to process the user's voice input in complex and noisy environments, and is responsible for a series of audio preprocessing tasks such as voice denoising. Its semi-transcribed speech signal will then be sent to subsequent text preprocessing. module for further speech-to-text process; the design and workflow of the speech recognition module are: S1. Use audio data sets containing various background sounds/noises to fine-tune the speech recognition module to improve its robustness; S2. Collect the user's pronunciation of the specified text to determine the user's pronunciation accuracy and pronunciation preference, and further fine-tune the speech recognition module to improve its recognition accuracy; S3. During the use of the system, preprocessing steps such as noise reduction and correction are performed on the audio input by the user, and the output is output to subsequent modules; Text preprocessing module: This module is used to receive data from the speech recognition module and complete tasks such as speech transcription, word segmentation, and text vectorization; the design and workflow of the text preprocessing module are: S1. For the output from the speech recognition module, perform speech-to-text transcription; perform word segmentation on the converted text and segment it into a sequence of words or sub-words; S2. Analyze whether the words in the sentence contain non-learning related vocabulary and illegal content. If so, send a corresponding prompt message to remind the user and ask the user to re-enter or refuse to respond; S3. Perform part-of-speech tagging for each word; S4. Identify and mark named entities in text; S5. Split the text into independent sentences as input to the intelligent dialogue module; Intelligent dialogue module that collaborates with the self-developed knowledge base: This module is designed to be further fine-tuned based on the domestic large model, and collaborates with the speech recognition and text preprocessing modules to respond to user requests and perform intelligent intelligent dialogue based on the prior information provided by the self-developed knowledge base. Reason and reply; the workflow of the intelligent dialogue module is: S1. Based on machine learning and deep learning algorithms, use the domestic Chinese base model to fine-tune on high-quality educational data, so that it has the ability to identify grammatical errors, correct confusing words, etc., and is more suitable for educational scenarios, obtaining a dedicated large-scale education-oriented Model; S2. Users set preferred scenarios for voice practice, such as business conversations, English speeches, daily conversations, etc., to fine-tune the usage scenarios of the intelligent dialogue module; S3. Update and maintain the self-researched knowledge base, and promptly update and expand the knowledge reserve of the model; S4. The fine-tuned large language model generates the response text required by the user based on its background knowledge, scene-adaptive prompt words and highly time-sensitive knowledge text; S5. Use the voice conversion sub-module in the intelligent dialogue module to generate the voice corresponding to the response text and feed it back to the user. 2. A large intelligent education model based on knowledge base according to claim 1, characterized in that: the domestic Chinese base model is used to fine-tune S1 on high-quality education data, and its work flow is as follows: S1-1. Use crawler and other technologies to construct tens of millions of scale fine-tuning data sets from Internet data to conduct supervised training of the base model to enhance its responsiveness and reasoning performance for education-related knowledge; S1-2. Facing the education industry, collect highly relevant high-quality data in a targeted manner for further fine-tuning of parameters to enhance the knowledge understanding of the model in detailed professional fields; S1-3. Based on the feedback of professional education practitioners on the model output content, the model can learn the language style of professionals and the analysis process of specific problems to enhance the professionalism of the model output. 3. A large intelligent education model based on knowledge base according to claim 1, characterized in that: updating and maintaining the self-developed knowledge base, timely updating and expanding the knowledge reserve S3 of the model, and its work flow is as follows: S3-1. Based on the data collection sub-module, capture and update document knowledge related to each subject in the database in real time from the Internet, such as textbooks, lesson plans, text information in video media, etc., forming a massive and timely database; S3-2. Based on the data segmentation sub-module, it is classified and segmented according to the characteristics of knowledge to build a vector database for model analysis and use; S3-3. Based on the user query vectorization module, project the structured text information into the embedding space, extract the feature vector of the text, and combine it with vector retrieval technology to obtain and return the most relevant feature vector from the database, thereby realizing knowledge Quick Search.