An AI big data-based education knowledge base content generation method and system

Abstract

The application relates to the technical field of knowledge base generation, and particularly discloses an education knowledge base content generation method and system based on AI big data. The method comprises the following steps: comparing user answer content with answer information to obtain a matching degree, calibrating positive samples and negative samples to train an evaluation model, and determining a reference standard score according to the quantity relationship between the positive samples and the negative samples; based on the reference standard score and a tolerance threshold, a plurality of candidate knowledge contents are screened to calibrate them as stable knowledge contents or to-be-corrected knowledge contents, and are output. By constructing an evaluation system based on user real answer data and an evaluation standard reflecting the actual application effect of the knowledge content, and automatically calibrating the evaluation standard according to the data distribution, objective screening and continuous iteration of the knowledge content are realized, and the accuracy and reliability of the knowledge base content are improved.

Description

Technical Field

[0001] This invention belongs to the field of knowledge base generation technology, specifically relating to a method and system for generating educational knowledge base content based on AI big data. Background Technology

[0002] With the development of information technology, in the field of education, especially in personalized learning and intelligent teaching, the educational knowledge base, as a structured data warehouse that carries and organizes knowledge, constitutes the content foundation of the intelligent education system. A high-quality educational knowledge base is a key content support for realizing adaptive learning path recommendation, intelligent question answering, and accurate assessment of academic level, and it is also the cornerstone for accumulating and accumulating high-quality teaching resources and building a new digital education ecosystem.

[0003] The current construction of educational knowledge bases generally relies on manual organization, compilation, and input by domain experts, resulting in lengthy construction cycles and high labor costs. Moreover, the current educational knowledge bases have poor scalability, and in the context of the current explosive growth of knowledge, the coverage and update frequency of the knowledge bases are difficult to meet actual needs. In terms of content quality and depth, the manual construction of current educational knowledge bases is subjective and inconsistent, making it difficult to guarantee the refinement and systematization of knowledge, and lacking timely supplementation of subtle concepts, relationships, and cutting-edge developments.

[0004] Furthermore, due to the lack of objective and unified evaluation standards for current educational knowledge bases, errors, redundancies, or logically confused content are easily mixed into the knowledge base, damaging its authority and usability. Traditional knowledge bases, through static listing of knowledge, lack the mining and expression of deep semantic relationships between knowledge, making it difficult to form a dynamically evolving knowledge network. This makes it difficult for the knowledge base to support intelligent means such as reasoning and analysis, affecting the development and effectiveness of upper-level intelligent applications.

[0005] In view of this, this application provides a method and system for generating educational knowledge base content based on AI big data. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for generating educational knowledge base content based on AI big data, in order to solve the problems of long cycles and unreliable overall quality of knowledge base content in existing technologies.

[0007] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0008] A method for generating educational knowledge base content based on AI and big data includes the following steps:

[0009] Generate positive and negative samples;

[0010] Determine the reference standard score;

[0011] In response to the acquired raw information, the knowledge generation model is invoked to generate multiple candidate knowledge contents;

[0012] Based on a pre-set evaluation model and reference standard scores, multiple candidate knowledge contents are evaluated to identify each candidate knowledge content as stable knowledge content or knowledge content to be corrected, and then output.

[0013] The process of generating positive and negative samples includes: acquiring question information, answer information, and user responses from the question-answering system; comparing user responses with answer information to obtain a matching degree; and classifying the corresponding answer results into real answer results and fictitious answer results based on the matching degree, where real answer results are labeled as positive samples and fictitious answer results are labeled as negative samples.

[0014] The process of determining the reference standard score includes: training an evaluation model based on positive and negative samples obtained from the answering system; applying the evaluation model to the positive and negative samples respectively to obtain a set of positive sample evaluation scores and a set of negative sample evaluation scores; if the number of positive samples is less than or equal to the number of negative samples, the reference standard score is determined based on the set of positive sample evaluation scores; if the number of positive samples is greater than the number of negative samples, the reference standard score is determined based on the set of negative sample evaluation scores.

[0015] The original information obtained includes at least one of the following: course syllabus, textbook chapters, and definitions of core concepts.

[0016] Preferably, the evaluation of multiple candidate knowledge contents based on a preset evaluation model and reference standard scores includes:

[0017] Multiple candidate knowledge contents are input into the evaluation model to obtain the content score corresponding to each candidate knowledge content. The content score is compared with the reference standard score. If the absolute value of the difference between the two is less than the tolerance threshold, the corresponding candidate knowledge content is marked as stable knowledge content. If the absolute value of the difference is greater than or equal to the tolerance threshold, the corresponding candidate knowledge content is marked as knowledge content to be corrected.

[0018] Preferably, classifying the corresponding answer results into real answer results and fictitious answer results based on the matching degree includes:

[0019] If the matching degree is greater than the matching degree threshold, the answer is determined as a real answer; if the matching degree is less than or equal to the matching degree threshold, the answer is determined as a fictitious answer.

[0020] Preferably, the method further includes:

[0021] A verification sample set is divided from the negative samples according to a preset ratio. The remaining negative samples are merged with all positive samples to form the test sample set. The test sample set and the verification sample set are respectively input into the evaluation model to obtain the test evaluation output and the verification evaluation output. The knowledge generation model is updated based on the consistency between the test evaluation output and the verification evaluation output.

[0022] Preferably, after identifying the stable knowledge content and the knowledge content to be modified, the method further includes:

[0023] Compare the number of knowledge contents to be corrected with the number of stable knowledge contents; if the number of knowledge contents to be corrected is greater than the number of stable knowledge contents, sort the knowledge contents to be corrected in descending order according to their content scores, and re-label the knowledge contents to be corrected with the highest scores as stable knowledge contents; if the number of stable knowledge contents is greater than the number of knowledge contents to be corrected, sort the stable knowledge contents in ascending order according to their content scores, and take the stable knowledge contents with the lowest scores as the correction result, replacing the corresponding knowledge contents to be corrected.

[0024] An AI-based big data-driven educational knowledge base content generation system, used to execute the aforementioned AI-based big data-driven educational knowledge base content generation method, includes the following modules:

[0025] The evaluation criteria determination module is used to train the evaluation model and determine the reference standard score based on positive and negative samples obtained from the answering system.

[0026] The knowledge generation module is used to generate multiple candidate knowledge contents in response to the acquired raw information content;

[0027] The knowledge filtering module, in response to multiple candidate knowledge contents generated by the knowledge generation module, calls the evaluation model trained by the evaluation criteria determination module and the determined reference standard scores to evaluate the multiple candidate knowledge contents, so as to label each candidate knowledge contents as stable knowledge contents or knowledge contents to be corrected, and outputs them.

[0028] The generative model update module is used to update the knowledge generation model applied by the knowledge generation module based on the verification results of positive and negative samples.

[0029] Preferably, obtaining positive and negative samples from the question-answering system includes:

[0030] The system acquires question information, answer information, and user responses from the question-answering system; compares user responses with answer information to obtain a matching degree; and classifies the corresponding answer results into real answer results and fictitious answer results based on the matching degree, where real answer results are labeled as positive samples and fictitious answer results are labeled as negative samples.

[0031] Preferably, classifying the corresponding answer results into real answer results and fictitious answer results based on the matching degree includes:

[0032] If the matching degree is greater than the matching degree threshold, the answer is determined as a real answer; if the matching degree is less than or equal to the matching degree threshold, the answer is determined as a fictitious answer.

[0033] Preferably, based on positive and negative samples obtained from the question-answering system, an evaluation model is trained, and a reference standard score is determined, including:

[0034] The evaluation model is trained based on positive and negative samples obtained from the question-answering system. The evaluation model is applied to positive and negative samples respectively to obtain a set of evaluation scores for positive samples and a set of evaluation scores for negative samples. If the number of positive samples is less than or equal to the number of negative samples, a reference standard score is determined based on the set of evaluation scores for positive samples. If the number of positive samples is greater than the number of negative samples, a reference standard score is determined based on the set of evaluation scores for negative samples.

[0035] Preferably, evaluating multiple candidate knowledge contents by calling the evaluation model trained by the evaluation criteria module and the determined reference standard score includes:

[0036] Multiple candidate knowledge contents are input into the evaluation model to obtain the content score corresponding to each candidate knowledge content. The content score is compared with the reference standard score. If the absolute value of the difference between the two is less than the tolerance threshold, the corresponding candidate knowledge content is marked as stable knowledge content. If the absolute value of the difference is greater than or equal to the tolerance threshold, the corresponding candidate knowledge content is marked as knowledge content to be corrected.

[0037] Beneficial effects

[0038] This invention obtains the matching degree by comparing user answers with answer information, and labels the answer results as real or fictitious based on the matching degree. Then, positive and negative samples are labeled to train the evaluation model, so that the evaluation model is derived from real user behavior data and the evaluation criteria can reflect the actual application effect of knowledge content. This overcomes the subjective defects of manual review and improves the accuracy and reliability of knowledge base content evaluation.

[0039] This invention utilizes an evaluation model to obtain a set of positive sample evaluation scores and a set of negative sample evaluation scores. It then determines a reference standard score based on the set of scores of the smaller sample. Based on the reference standard score and a tolerance threshold, multiple candidate knowledge contents are screened to identify them as stable knowledge contents or knowledge contents to be corrected. The evaluation standard is automatically calibrated according to the data distribution to avoid the problem of inaccurate evaluation results caused by using a fixed threshold. This ensures the fairness and flexibility of the screening process, thereby guaranteeing the quality and consistency of the final knowledge contents stored in the database.

[0040] This invention uses the output of the evaluation model to update the knowledge generation model, and compares the amount of stable knowledge content with the amount of knowledge content to be corrected before outputting. Then, it performs optimization operations such as recalibration or triggering regeneration, so that the knowledge generation model can be continuously iterated, improving the quality of content from the source, ensuring the balance and usability of the delivered results, and enhancing the stability and intelligence of the content generation process. Attached Figure Description

[0041] Figure 1 This is a flowchart of the method of the present invention;

[0042] Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely for explaining the invention and are not intended to limit the scope of protection of the invention.

[0044] Example 1

[0045] Please refer to Figure 1 This embodiment provides a method for generating educational knowledge base content based on AI big data, including the following steps:

[0046] The system acquires question and answer information from the quiz, collects user responses to the questions, and generates quiz results. It also collects raw data from real educational scenarios, specifically from the online learning platform's quiz module. This involves batch acquiring question stems, options, relevant knowledge points, and standard answer information from past exams, as well as collecting massive amounts of actual responses submitted by different users.

[0047] The system compares user responses with answer information and obtains a quantitative matching degree through preset text processing and quantitative comparison rules. Specifically, it performs text normalization processing such as removing punctuation and converting capitalization on user responses and answer information, extracts keywords or semantic feature vectors of each, and calculates the matching degree value by calculating the similarity between features. This value is used to characterize the degree of semantic closeness between the user response and the standard answer. The similarity between features can be calculated based on a quantitative comparison method of keyword co-occurrence frequency and weight distribution or by calculating the cosine of the angle between the vectors of the text.

[0048] Based on the matching degree, the answer results are classified into real answer results and fictitious answer results. Here, a matching degree threshold is set. If the matching degree of a certain answer result is greater than the threshold, it is determined that it is basically consistent with the standard answer in semantics and is identified as a real answer result. Conversely, if the matching degree is less than or equal to the threshold, it is determined that it deviates significantly from the standard answer in semantics or fact and is identified as a fictitious answer result.

[0049] Real answers are labeled as positive samples, and fictitious answers are labeled as negative samples, constructing a supervised reference set for subsequent evaluation criteria; where positive samples represent high-quality and reliable content examples, while negative samples represent low-quality and unreliable content examples.

[0050] Based on positive and negative samples, a quantitative evaluation model is constructed. The specific process of constructing the quantitative evaluation model is as follows: analyze the statistical differences between positive and negative samples in terms of text features. Text features may include, but are not limited to, keyword distribution, syntactic structure complexity, information entropy, and the strength of association with specific knowledge points.

[0051] Based on the differences, an evaluation standard and calculation process that can effectively distinguish between the two types of samples are established. After receiving any text input, the calculation process can output an evaluation score that reflects the content quality or credibility through preset weighted calculation and logical judgment steps.

[0052] The functional relationship of the quantitative evaluation model is:

[0053]

[0054] In the formula, This represents the credibility score, which indicates the level of confidence in the input text content. The quantitative assessment results of quality or credibility;

[0055] Represents the feature weight, its meaning is the first... The importance coefficient of each text feature in the comprehensive evaluation is determined based on the statistical difference between positive and negative samples;

[0056] This represents the feature function value, which means the input text content... Calculate its first Quantitative values ​​of text features such as keyword distribution and syntactic complexity;

[0057] This represents the total number of features, which is the total number of text features used to evaluate content quality.

[0058] The evaluation model is dynamically optimized through a feedback update process. Specifically, the feedback update process includes: dividing the negative samples into a validation sample set according to a preset ratio, and merging the remaining negative samples with all positive samples to form the test sample set, where the preset ratio can be set to 10%.

[0059] The test sample set and the validation sample set are respectively input into the evaluation model to obtain a set of scores as the test evaluation output and another set of scores as the validation evaluation output. By calculating the statistical deviation between the distribution of scores such as the mean and variance of the test evaluation output and the distribution of scores of the validation evaluation output, it is determined whether there is a drift in the evaluation standard of the mechanism.

[0060] If the deviation exceeds the preset stability threshold, it indicates that the consistency of the mechanism's judgment has decreased. Based on the comparison result, the weighting coefficients or judgment logic in the evaluation model will be adjusted to form a closed-loop adaptive optimization.

[0061] The constructed evaluation model is applied to process the contents of the positive and negative sample sets one by one, thereby obtaining a set of positive sample evaluation scores and a set of negative sample evaluation scores, which establishes a data-driven quality benchmark for subsequent screening of newly generated content.

[0062] The reference standard score is determined by a dynamic and adaptive approach, based on the relationship between the number of positive and negative samples: if the number of positive samples is less than or equal to the number of negative samples, the reference standard score is obtained by taking the arithmetic mean or median of the set of positive sample evaluation scores; conversely, if the number of positive samples is greater than the number of negative samples, the reference standard score is calculated based on the set of negative sample evaluation scores. This ensures that the category with a smaller sample size or more concentrated features is always selected as the benchmark to establish a more stable and reliable evaluation anchor.

[0063] The system acquires original information content and generates multiple candidate knowledge content through a knowledge generation model. The original information content can be course outlines, textbook chapters, core concept definitions, etc. The knowledge generation model, based on the semantic core and text structure of the original information content, creates multiple candidate knowledge content that differ in content or form through preset content transformation rules. The preset content transformation rules can include synonym replacement, sentence structure reorganization, supplementary explanatory cases, or reverse generation of questions, etc.

[0064] Multiple candidate knowledge contents are input into the evaluation model one by one to obtain a numerical score representing the quality of each candidate knowledge content.

[0065] Content scores are compared with reference standard scores to filter out stable knowledge content, and a dynamically determined tolerance threshold is introduced to increase the flexibility of filtering stable knowledge content.

[0066] The process of determining the tolerance threshold includes: obtaining the correction ratio through the correction ratio acquisition process; and obtaining the tolerance threshold by calculating the product of the reference standard score and the correction ratio based on the reference standard score and the correction ratio.

[0067] The process of obtaining the correction ratio includes: providing one or more benchmark review samples; automatically generating a test content set based on the benchmark review samples through preset text perturbation operations that do not change the core semantics, including synonym replacement, word order adjustment, and addition or deletion of non-critical information; inputting the benchmark review samples and the test content set into the evaluation model to obtain the benchmark sample score and the test content score set; calculating the average of the absolute values ​​of the differences between the benchmark sample score and the scores in the test content score set, and determining this average value as the correction ratio. This correction ratio quantifies the evaluation mechanism's sensitivity to scoring of non-substantial changes in content.

[0068] Based on the comparison results, candidate knowledge content is calibrated. If the absolute value of the difference between the content score of a candidate knowledge content and the reference standard score is less than the tolerance threshold, it indicates that the quality of the candidate knowledge content is comparable to the benchmark level, and it is calibrated as stable knowledge content. If the absolute value of the difference is greater than or equal to the tolerance threshold, it indicates that its quality does not meet the standard, and it is calibrated as knowledge content to be corrected.

[0069] The system identifies and outputs stable knowledge content and knowledge content that needs to be corrected. Different tags or metadata can be attached to the output for subsequent or manual processing.

[0070] After output, perform additional optimization operations: compare the number of knowledge content to be corrected with the number of stable knowledge content. If the number of knowledge content to be corrected is greater than the number of stable knowledge content, it indicates that the quality of this generation is generally low. Sort the knowledge content to be corrected in descending order according to its content score, re-label the knowledge content to be corrected with the highest score as stable knowledge content, and trigger the regeneration of the original information content corresponding to the other knowledge content to be corrected.

[0071] If the number of stable knowledge contents is greater than or equal to the number of knowledge contents to be corrected, the stable knowledge contents with the lowest score can be used as correction examples to guide the subsequent manual correction or automatic adjustment process of the knowledge contents to be corrected.

[0072] Example 2

[0073] Please refer to Figure 2 This embodiment provides an educational knowledge base content generation system based on AI big data, including the following modules:

[0074] The evaluation criteria determination module is used to train the evaluation model and determine the reference standard score from positive and negative samples obtained from the question-answering system; it also obtains historical data from the associated question-answering system, which includes question information, answer information, and user answers; and compares the user answers with the corresponding answer information to calculate the matching degree; wherein, the matching degree can be determined by text similarity algorithms such as cosine similarity and Jaccard similarity or semantic understanding models.

[0075] The matching degree is compared with a preset matching degree threshold. If the matching degree is greater than the threshold, the user's answer is considered to be highly consistent with the answer information, representing a valid and genuine answering behavior. Therefore, the answer result is classified as a genuine answer result and labeled as a positive sample. The answer result is a data pair consisting of question information, user answer content, etc.

[0076] Conversely, if the matching degree is less than or equal to the threshold, it is considered that the user's answer content differs greatly from the answer information, and may be a guess, error or irrelevant answer, representing a fictitious or invalid answering behavior. Therefore, the answer result is classified as a fictitious answer result and labeled as a negative sample.

[0077] After obtaining a sufficient number of positive and negative samples, these samples are used to train a pre-defined evaluation model. The evaluation model can be a classification model or a regression model, specifically logistic regression, support vector machine, or deep neural network. Its goal is to learn the features that distinguish between high-quality content represented by positive samples and low-quality content represented by negative samples.

[0078] After training, the trained evaluation model is applied to all positive and negative samples respectively to obtain the evaluation score set of positive samples and the evaluation score set of negative samples. The number of positive samples and the number of negative samples are counted. If the number of positive samples is less than or equal to the number of negative samples, it indicates that the high-quality content samples are relatively sparse and their score distribution is more representative. Therefore, based on the evaluation score set of positive samples, the average, median or a certain quantile of the scores in the set is used to determine the reference standard score.

[0079] Conversely, if the number of positive samples is greater than the number of negative samples, the reference standard score is determined based on the set of negative sample evaluation scores. This ensures that the reference standard score is determined based on a smaller set of samples with more concentrated information, thereby improving its robustness as an evaluation benchmark.

[0080] The knowledge generation module is used to generate multiple candidate knowledge contents in response to the acquired raw information content. When a generation task is received, if it is necessary to generate explanatory content for specific knowledge points such as the definition of photosynthesis, the raw information content such as the definition of photosynthesis is transmitted to one or more knowledge generation models, which can be based on the large language model of the Transformer architecture.

[0081] By calling the knowledge generation model, taking the original information content as input or prompts, and running it multiple times by adjusting temperature parameters and using different sampling methods, a set of multiple candidate knowledge contents with slight differences in content but all related to the topic are generated.

[0082] The knowledge filtering module is used to respond to multiple candidate knowledge contents generated by the knowledge generation module, call the evaluation model trained by the evaluation criteria determination module and the determined reference standard score, evaluate the multiple candidate knowledge contents, and label each candidate knowledge contents as stable knowledge contents or knowledge contents to be corrected, and output them.

[0083] The multiple candidate knowledge contents produced by the knowledge generation module are input one by one into the evaluation model trained by the evaluation standard determination module, and a quantitative content score is calculated for each candidate knowledge content; the content score of each candidate knowledge content is compared with the reference standard score determined by the evaluation standard determination module; the absolute value of the difference between the two is calculated and compared with the preset tolerance threshold.

[0084] If the absolute value of the difference is less than the tolerance threshold, it indicates that the score of the candidate knowledge content is very close to the benchmark score representing high-quality content, and its content quality is considered reliable and stable. The candidate knowledge content is then marked as stable knowledge content. If the absolute value of the difference is greater than or equal to the tolerance threshold, it indicates that the score of the candidate knowledge content deviates far from the benchmark score, and its content quality is questionable. Further manual review or correction is required, and it is marked as knowledge content to be corrected.

[0085] After the initial calibration is completed, an optional optimization operation is performed to compare the number of knowledge contents that are calibrated as knowledge contents to be corrected with the number of stable knowledge contents. If the number of knowledge contents to be corrected is greater than the number of stable knowledge contents, it indicates that the overall quality of the generated results is low. In order to ensure high-quality output, all knowledge contents to be corrected are sorted in descending order according to their content scores, and the knowledge contents to be corrected with the highest score are recalibrated as stable knowledge contents.

[0086] If the number of stable knowledge content is greater than the number of knowledge content to be corrected, it indicates that the overall quality of the generated results is high. In order to use high-quality content to optimize low-quality content, all stable knowledge content is sorted in ascending order according to its content score, and the stable knowledge content with the lowest score is used as the correction result to replace the corresponding knowledge content to be corrected, or it is output as a correction suggestion along with the knowledge content to be corrected.

[0087] Output all calibrated and potentially recalibrated stable knowledge content and knowledge content to be corrected for subsequent database entry or manual review.

[0088] The model update module is used to update the knowledge generation model applied by the knowledge generation module based on the verification results of positive and negative samples; the update process is executed periodically or under specific triggering conditions to achieve continuous optimization of the knowledge generation model.

[0089] Specifically, from the negative samples generated by the evaluation criteria determination module, a portion is randomly divided into a verification sample set according to a preset ratio. The verification sample set represents unseen and identified low-quality content. The remaining negative samples are then merged with all the positive samples to form the test sample set.

[0090] The test sample set and the validation sample set are respectively input into the evaluation model to obtain the test evaluation output and the validation evaluation output. The generalization ability and stability of the evaluation model are judged based on the consistency between the test evaluation output and the validation evaluation output. The validation result of the validation evaluation output is the verification of the reliability of the evaluation model. Based on the validation result, a feedback signal is generated to update the knowledge generation model.

[0091] If the knowledge generation model frequently produces content similar to the features of the verification sample set, the probability of generating such low-quality content can be reduced by using a penalty mechanism in reinforcement learning or by fine-tuning the parameters of the knowledge generation model, thereby ensuring that the feedback signals used to update the generation model are accurate.

[0092] Those skilled in the art will understand that the system described in the above embodiments can be implemented by computer program instructions, can be deployed on servers, cloud computing platforms, or dedicated educational equipment, and can interact with external data sources and content libraries such as question-answering systems through corresponding application programming interfaces.

[0093] The above description is merely a preferred embodiment of this application and is not intended to limit this application. For those skilled in the art, this application can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for generating educational knowledge base content based on AI big data, characterized in that, Includes the following steps: Generate positive and negative samples; Determine the reference standard score; In response to the acquired raw information, the knowledge generation model is invoked to generate multiple candidate knowledge contents; Based on a pre-set evaluation model and reference standard scores, multiple candidate knowledge contents are evaluated to identify each candidate knowledge content as stable knowledge content or knowledge content to be corrected, and then output. The process of generating positive and negative samples includes: acquiring question information, answer information, and user responses from the question-answering system; comparing user responses with answer information to obtain a matching degree; and classifying the corresponding answer results into real answer results and fictitious answer results based on the matching degree, where real answer results are labeled as positive samples and fictitious answer results are labeled as negative samples. The process of determining the reference standard score includes: training an evaluation model based on positive and negative samples obtained from the answering system; applying the evaluation model to the positive and negative samples respectively to obtain a set of positive sample evaluation scores and a set of negative sample evaluation scores; if the number of positive samples is less than or equal to the number of negative samples, the reference standard score is determined based on the set of positive sample evaluation scores; if the number of positive samples is greater than the number of negative samples, the reference standard score is determined based on the set of negative sample evaluation scores. The original information obtained includes at least one of the following: course syllabus, textbook chapters, and definitions of core concepts. The evaluation of multiple candidate knowledge contents based on a pre-set evaluation model and reference standard scores includes: inputting multiple candidate knowledge contents one by one into the evaluation model to obtain a numerical score representing the quality of each candidate knowledge content; comparing the content score with the reference standard score to screen out stable knowledge contents, and introducing a dynamically determined tolerance threshold to increase the flexibility of screening stable knowledge contents; calibrating the candidate knowledge contents according to the comparison results; if the absolute value of the difference between the content score and the reference standard score of a candidate knowledge content is less than the tolerance threshold, it indicates that the quality of the candidate knowledge content is comparable to the benchmark level, and it is calibrated as stable knowledge content; if the absolute value of the difference is greater than or equal to the tolerance threshold, it indicates that its quality does not meet the standard, and it is calibrated as knowledge content to be corrected; and identifying and outputting the stable knowledge contents and the knowledge contents to be corrected. The process of determining the tolerance threshold includes: obtaining the correction ratio through the correction ratio acquisition process; and obtaining the tolerance threshold by calculating the product of the reference standard score and the correction ratio based on the reference standard score and the correction ratio. The process of obtaining the correction ratio includes: providing one or more benchmark review samples; automatically generating a test content set based on the benchmark review samples through preset text perturbation operations that do not change the core semantics; inputting the benchmark review samples and the test content set into the evaluation model respectively to obtain the benchmark sample score and the test content score set; calculating the average of the absolute values ​​of the differences between the benchmark sample score and the test content score set, and determining the average value as the correction ratio.

2. The method for generating educational knowledge base content based on AI big data according to claim 1, characterized in that, The classification of corresponding answer results into real answer results and fictitious answer results based on the matching degree includes: If the matching degree is greater than the matching degree threshold, the answer is determined as a real answer; if the matching degree is less than or equal to the matching degree threshold, the answer is determined as a fictitious answer.

3. The method for generating educational knowledge base content based on AI big data according to claim 1, characterized in that, The method further includes: A verification sample set is divided from the negative samples according to a preset ratio. The remaining negative samples are merged with all positive samples to form the test sample set. The test sample set and the verification sample set are respectively input into the evaluation model to obtain the test evaluation output and the verification evaluation output. The knowledge generation model is updated based on the consistency between the test evaluation output and the verification evaluation output.

4. The method for generating educational knowledge base content based on AI big data according to claim 1, characterized in that, After identifying stable knowledge content and knowledge content to be modified, the following is also included: Compare the number of knowledge contents to be corrected with the number of stable knowledge contents; if the number of knowledge contents to be corrected is greater than the number of stable knowledge contents, sort the knowledge contents to be corrected in descending order according to their content scores, and re-label the knowledge contents to be corrected with the highest scores as stable knowledge contents; if the number of stable knowledge contents is greater than the number of knowledge contents to be corrected, sort the stable knowledge contents in ascending order according to their content scores, and take the stable knowledge contents with the lowest scores as the correction result, replacing the corresponding knowledge contents to be corrected.

5. An AI-based big data-driven educational knowledge base content generation system, used to execute the AI-based big data-driven educational knowledge base content generation method as described in any one of claims 1-4, characterized in that, Includes the following modules: The evaluation criteria determination module is used to train the evaluation model and determine the reference standard score based on positive and negative samples obtained from the answering system. The knowledge generation module is used to generate multiple candidate knowledge contents in response to the acquired raw information content; The knowledge filtering module, in response to multiple candidate knowledge contents generated by the knowledge generation module, calls the evaluation model trained by the evaluation criteria determination module and the determined reference standard scores to evaluate the multiple candidate knowledge contents, so as to label each candidate knowledge contents as stable knowledge contents or knowledge contents to be corrected, and outputs them. The generative model update module is used to update the knowledge generation model applied by the knowledge generation module based on the verification results of positive and negative samples.

6. The educational knowledge base content generation system based on AI big data according to claim 5, characterized in that, Based on obtaining positive and negative samples from the question-answering system, including: The system acquires question information, answer information, and user responses from the question-answering system; compares user responses with answer information to obtain a matching degree; and classifies the corresponding answer results into real answer results and fictitious answer results based on the matching degree, where real answer results are labeled as positive samples and fictitious answer results are labeled as negative samples.

7. The educational knowledge base content generation system based on AI big data according to claim 6, characterized in that, The classification of corresponding answer results into real answer results and fictitious answer results based on the matching degree includes: If the matching degree is greater than the matching degree threshold, the answer is determined as a real answer; if the matching degree is less than or equal to the matching degree threshold, the answer is determined as a fictitious answer.

8. The educational knowledge base content generation system based on AI big data according to claim 5, characterized in that, Based on positive and negative samples obtained from the question-answering system, an evaluation model is trained, and reference standard scores are determined, including: The evaluation model is trained based on positive and negative samples obtained from the question-answering system. The evaluation model is applied to positive and negative samples respectively to obtain a set of evaluation scores for positive samples and a set of evaluation scores for negative samples. If the number of positive samples is less than or equal to the number of negative samples, a reference standard score is determined based on the set of evaluation scores for positive samples. If the number of positive samples is greater than the number of negative samples, a reference standard score is determined based on the set of evaluation scores for negative samples.

9. The educational knowledge base content generation system based on AI big data according to claim 5, characterized in that, The evaluation model, trained by the evaluation criteria and using the determined reference standard scores, is invoked to evaluate multiple candidate knowledge contents, including: Multiple candidate knowledge contents are input into the evaluation model to obtain the content score corresponding to each candidate knowledge content. The content score is compared with the reference standard score. If the absolute value of the difference between the two is less than the tolerance threshold, the corresponding candidate knowledge content is marked as stable knowledge content. If the absolute value of the difference is greater than or equal to the tolerance threshold, the corresponding candidate knowledge content is marked as knowledge content to be corrected.

Images