Online education course management method based on artificial intelligence

Abstract

The embodiment of the invention provides an online education course management method based on artificial intelligence, and the method comprises the steps: carrying out the semantic analysis of a question of a user through a natural language understanding module based on a deep learning model, so as to improve the accuracy of an intelligent question answering system; a knowledge base of education content is constructed through the knowledge graph to enhance the question answering ability of the system; the feedback information of the user is used as training data, and the natural language understanding module is updated and optimized regularly; according to the learning progress and preference of the user, the course content and difficulty are dynamically adjusted; the answering effect of the user is monitored in real time, a manual intervention mechanism is automatically triggered, and understanding errors are reduced. Through the scheme of the embodiment of the invention, the problem of understanding errors caused by limited natural language processing capability in the intelligent question answering system can be solved.

Description

Technical Field

[0001] The present application relates to the field of online education technology, and specifically to an online education course management method based on artificial intelligence. Background Art

[0002] With the development of artificial intelligence technology, online education course management systems have realized intelligent management of online education courses by integrating artificial intelligence technology, including adaptive learning path recommendation, intelligent evaluation and feedback, personalized homework push and other functions. However, this method still has some challenges in the intelligent question-answering system, especially due to the limited natural language processing capabilities, the system may have misunderstandings or errors when understanding and answering questions raised by students, affecting students' learning experience and the effectiveness of question-answering. Summary of the invention

[0003] In view of this, the embodiments of the present disclosure provide an online education course management method based on artificial intelligence, which at least partially solves the problems existing in the prior art.

[0004] An online education course management method based on artificial intelligence, comprising:

[0005] The natural language understanding module based on the deep learning model performs semantic analysis on user questions to improve the accuracy of the intelligent question answering system;

[0006] Build a knowledge base of educational content through knowledge graphs to enhance the system's question-answering capabilities;

[0007] Use user feedback as training data to regularly update and optimize the natural language understanding module;

[0008] Dynamically adjust course content and difficulty based on user's learning progress and preferences;

[0009] Monitor the effectiveness of users' Q&A in real time, automatically trigger the manual intervention mechanism, and reduce misunderstandings.

[0010] According to one embodiment, the natural language understanding module based on the deep learning model performs semantic analysis on the user's question, specifically comprising the following steps:

[0011] Get the user's original question input;

[0012] Convert the user's original question input into a standard text format;

[0013] Perform word segmentation on questions in standard text format;

[0014] Use pre-trained deep learning models for semantic analysis to improve the accuracy of question understanding and reduce misunderstandings caused by limited natural language processing capabilities.

[0015] According to one embodiment, the natural language understanding module based on the deep learning model performs semantic analysis on the user's question further comprising the following steps:

[0016] Extract relevant background information based on the user's question context information;

[0017] Perform feature extraction on the extracted background information;

[0018] Combine the feature extraction results with the problems in the standard text format for comprehensive analysis;

[0019] If the semantic complexity of the user's question exceeds a preset threshold, the multimodal information fusion mechanism is triggered. The semantic complexity is expressed by the formula C=αL+βF, where L is the vocabulary length of the question, F is the feature vector length, and α and β are weight parameters, which are used to measure the impact of vocabulary length and feature vector on semantic complexity.

[0020] According to one embodiment, the natural language understanding module based on the deep learning model performs semantic analysis on the user's question further comprising the following steps:

[0021] Generate a set of similar questions based on historical question-answering records;

[0022] Conduct cluster analysis on sets of similar questions;

[0023] Prioritize the cluster center issues;

[0024] If the similarity between the user's current question and the historical question exceeds the preset threshold T, the historical answer is directly returned. The similarity calculation formula is S = 1 / (1+exp(d(P1,P2))), where P1 and P2 represent the user's current question and historical question respectively, and d represents the distance measure between the two questions.

[0025] According to one embodiment, the natural language understanding module based on the deep learning model performs semantic analysis on the user's question further comprising the following steps:

[0026] Build user portraits based on users’ historical behavior data;

[0027] Use user portraits to predict users’ question-answering needs;

[0028] Dynamically adjust the response priority of the Q&A system;

[0029] If the user's question-answering request frequency f exceeds a preset threshold F, the user is marked as a high-priority user. The question-answering request frequency calculation formula is f=N / T, where N is the number of question-answering requests of the user within time T.

[0030] According to one embodiment, the natural language understanding module based on the deep learning model performs semantic analysis on the user's question further comprising the following steps:

[0031] Based on the multimodal information fusion mechanism, it integrates information from multiple data sources;

[0032] Conduct secondary verification on the integrated information;

[0033] Dynamically adjust the parameters of the verification model;

[0034] If the understanding confidence C of the user's question is lower than the threshold τ, manual intervention is triggered. The understanding confidence calculation formula is C = 1 / (1+exp(z)), where z is the predicted output score of the model.

[0035] According to one embodiment, the natural language understanding module based on the deep learning model performs semantic analysis on the user's question further comprising the following steps:

[0036] Generate improvement plans based on user feedback;

[0037] Incorporate improvements into the training dataset;

[0038] Regularly evaluate model performance and optimize it;

[0039] If the number of times N of the same type of errors appearing in user feedback exceeds the threshold M, targeted training is performed on this type of error. The calculation formula is E=sum(Err) / T, where Err is the set of error records reported by users and T is the time period.

[0040] According to one embodiment, the natural language understanding module based on the deep learning model performs semantic analysis on the user's question further comprising the following steps:

[0041] Adjust the model’s training strategy based on real-time monitoring results;

[0042] Regularly check the accuracy and reliability of the Q&A system;

[0043] Mark and archive low-quality Q&A;

[0044] If the system's overall performance score P is lower than the preset threshold Q, the emergency plan is activated. The overall performance score calculation formula is P=W1*A1+W2*A2+W3*A3, where W1, W2, W3 are weight parameters, and A1, A2, A3 are various performance indicators, including accuracy, response time and user experience.

[0045] The disclosed embodiment provides an online education course management method based on artificial intelligence, including: a natural language understanding module based on a deep learning model to perform semantic analysis on user questions to improve the accuracy of the intelligent question-answering system; a knowledge base of educational content is constructed through a knowledge graph to enhance the system's question-answering ability; user feedback information is used as training data to regularly update and optimize the natural language understanding module; course content and difficulty are dynamically adjusted according to the user's learning progress and preferences; real-time monitoring of the user's question-answering effect, automatic triggering of a manual intervention mechanism, and reduction of misunderstandings. The solution of the disclosed embodiment can solve the problem of misunderstandings caused by limited natural language processing capabilities in the intelligent question-answering system. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the multiple drawings represent the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments disclosed in the present application and should not be regarded as limiting the scope of the present application.

[0047] Figure 1 It is a flowchart of an online education course management method based on artificial intelligence;

[0048] Figure 2 is a flow chart of a natural language understanding module based on a deep learning model performing semantic analysis on a user's question in an embodiment;

[0049] Figure 3 is a flow chart of a natural language understanding module based on a deep learning model performing semantic analysis on a user's question in an embodiment;

[0050] Figure 4 is a flow chart of a natural language understanding module based on a deep learning model performing semantic analysis on a user's question in an embodiment;

[0051] Figure 5 is a flow chart of a natural language understanding module based on a deep learning model performing semantic analysis on a user's question in an embodiment;

[0052] Figure 6 is a flow chart of a natural language understanding module based on a deep learning model performing semantic analysis on a user's question in an embodiment;

[0053] Figure 7 is a flow chart of a natural language understanding module based on a deep learning model performing semantic analysis on a user's question in an embodiment;

[0054] Figure 8The present invention is a flowchart of an embodiment of a natural language understanding module based on a deep learning model performing semantic analysis on a user's question. DETAILED DESCRIPTION

[0055] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0056] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.

[0057] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.

[0058] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The drawings only show components related to the present disclosure rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0059] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0060] Next, refer to Figure 1 , describing an online education course management method based on artificial intelligence of the present invention.

[0061] S101: The method first performs semantic analysis on the user's questions through a natural language understanding module based on a deep learning model to improve the accuracy of the intelligent question-answering system. The core function of this module is to accurately understand and process the questions raised by the user, and even vague or colloquial expressions can be effectively converted into standardized question forms. In one embodiment, when a user encounters difficulties in learning mathematics courses and enters a vague question such as "Why can't I always find the area of ​​a circle?", the module can quickly identify the specific mathematical problem hidden behind it - that is, the method and formula application for finding the area of ​​a circle. Then, it will convert this problem into a standard form, such as how to calculate the area S of a circle if the radius R is known, and match the corresponding solution through the background algorithm.

[0062] S102: In addition, the method also includes constructing a knowledge base of educational content through a knowledge graph to enhance the system's question-answering ability. The key to this link is to build an interconnected, hierarchical structured network among a large amount of teaching materials and knowledge points, so that the content of each subject can be systematically integrated. In a specific embodiment, for history courses, the knowledge graph will connect major historical events with the time and place of their occurrence, the people involved, and the background information to form a complete historical knowledge network. Suppose that when a student asks the question of when World War II ended, he can not only directly get the specific time September 2, 1945, but also learn about a series of other key information points related to this, such as changes in the world pattern after the war, important conference resolutions, etc., so as to deepen his understanding and memory.

[0063] S103: The third step is to use user feedback as training data to regularly update and optimize the natural language understanding module. This process helps the machine to continuously learn and improve so that it can more accurately identify new types of questions and provide high-quality answers. For example, in a certain version, the natural language processing model may have deficiencies in parsing terminology in a specific field, such as the accurate translation of physics professional vocabulary; by collecting a large number of such questions from students and their correct answers, the technical team can conduct targeted training on the model to improve its recognition rate and accuracy in this regard.

[0064] S104: Next, the system will dynamically adjust the course content and difficulty based on the user's learning progress and preferences. This means that the platform can recommend more personalized learning paths or resources based on the student's performance, ensuring that each student can get the most suitable educational experience for him or her. For example, if the system detects that a user performs well in the genetics part of biology but is unfamiliar with the concept of cell division, it can automatically recommend more basic knowledge exercises and auxiliary materials on cell division, and reduce the repetition of the content that the user has already mastered.

[0065] S105: Finally, in order to further ensure the teaching quality and user experience, the system will also monitor the user's question-answering effect in real time, automatically trigger the manual intervention mechanism, and reduce misunderstandings caused by limited natural language processing capabilities. For example, during the monitoring process, if a certain question is found to be unsuccessful in being resolved by the machine or the user repeatedly raises the same or similar questions, the backend support team will be immediately prompted to respond in a timely manner, and the problem can be solved quickly and effectively through direct communication between real teachers and users, while also accumulating valuable case materials for subsequent technical improvements. In this way, even the most complex problems can be effectively and quickly handled.

[0066] Next, refer to Figure 2 , describing that the natural language understanding module based on the deep learning model of the present invention performs semantic parsing on the user's question specifically includes the following steps:

[0067] S201: First, obtain the original question input from the user. This step refers to the system receiving the natural language question or instruction raised by the user, usually obtained through a user interface such as a web form or a voice input device. For example, on an online education platform, a user may enter the question "How to solve a quadratic equation?" through a course forum or a chat window.

[0068] S202: Secondly, the user's original question input is converted into a standard text format. This step mainly formats the input content to make it conform to the standard format processed by the system. For example, redundant spaces are removed, punctuation marks are standardized, and the text is converted to lowercase.

[0069] S203: Third, the questions in the standard text format are segmented. This step is to segment the standard text into words, breaking long sentences into separate vocabulary units. The word segmentation tool can more accurately understand and process the key information in the sentence. For example, "How to solve a quadratic equation?" becomes a word list like "How to solve a quadratic equation?" after word segmentation. Each part is a separate vocabulary, which helps to better understand the meaning of the sentence.

[0070] S204: Finally, use the pre-trained deep learning model to perform semantic analysis to improve the accuracy of question understanding and reduce the problem of misunderstanding caused by limited natural language processing capabilities. This step uses a trained deep learning model to perform semantic understanding and intent recognition on the word sequence after word segmentation. Through deep neural networks, the model can capture complex language structures and contextual relationships to more accurately understand the user's intentions. Specifically, the model may infer that the user's purpose is to ask about mathematical problem solving methods based on solving a quadratic equation, and generate corresponding answers. In addition, the deep learning model has accumulated a large amount of data and experience during the training process, which can improve the accuracy of problem understanding in a variety of situations. For example, if the user asks "I don't know how to start solving this equation, can you help me?", the model can also correctly parse the user's intentions and provide relevant guidance.

[0071] Next, refer to Figure 3 , describing that the natural language understanding module based on the deep learning model of the present invention performs semantic parsing on the user's question also includes the following steps:

[0072] S301: Extract relevant background information based on the user's question context information; perform feature extraction on the extracted background information; conduct comprehensive analysis based on the feature extraction results and questions in standard text format; if the semantic complexity of the user's question exceeds a preset threshold, trigger a multimodal information fusion mechanism, where the semantic complexity is represented by the formula C=αL+βF, where L is the vocabulary length of the question, F is the feature vector length, and α and β are weight parameters, respectively, used to measure the impact of vocabulary length and feature vector on semantic complexity.

[0073] Extract relevant background information based on the user's question context information. This step involves obtaining information from the user's conversation history, context environment, and possible external data sources that can help understand the user's question. For example, if a user asks a question on an online education platform: "How is the first law of thermodynamics mentioned in the previous lesson applied in chemical reactions?", the system will find relevant materials from previous courses through the user's course history records to better understand the user's question context.

[0074] S302: Extract features from the extracted background information. Feature extraction aims to convert the extracted background information into a form that can be processed by a computer, such as a numerical vector or key feature words. Specifically, the system can identify features such as key concepts, noun phrases, and verb phrases through natural language processing technology. These features are used for subsequent semantic analysis to help the model understand the user's intention more accurately. For example, the system extracts keywords such as the first law of thermodynamics, chemical reaction, and conservation of energy from the background information.

[0075] S303: Combine the feature extraction results with the questions in standard text format for comprehensive analysis. This step is to fuse the feature information extracted in the previous step with the standard text of the user's question to generate a more comprehensive understanding. Through this comprehensive analysis, the system can more accurately identify the user's actual needs and provide more precise answers. For example, when analyzing the question "How is the first law of thermodynamics applied in chemical reactions?" mentioned in the previous lesson, the system combines the feature words with the keywords in the question, the first law of thermodynamics and chemical reactions, to generate a detailed answer that explains the application of the first law of thermodynamics in chemical reactions.

[0076] S304: If the semantic complexity of the user's question exceeds the preset threshold, the multimodal information fusion mechanism is triggered. The semantic complexity here is calculated by the formula C=αL+βF, where L is the vocabulary length of the question, F is the feature vector length, and α and β are weight parameters, which are used to measure the impact of vocabulary length and feature vector on semantic complexity. For example, assuming L=10 words, F=5 features, α=0.7, β=0.3, then C=0.7*10+0.3*5=8.5. If the preset threshold is set to 9, C is less than the threshold, so multimodal information fusion is not triggered. If C is greater than the preset threshold, the multimodal information fusion mechanism is triggered, and the system will combine multiple information sources such as text, images, and audio to answer the user's questions. The purpose of setting the formula in this way is to balance the surface complexity (word length) and deep complexity (feature vector) of the problem, so as to comprehensively evaluate the difficulty of the problem and ensure that complex problems are more accurately parsed and answered. In terms of optimal values, the specific values ​​of α and β can be adjusted based on a large amount of experimental data. The common range is 0 to 1. Usually, α may be set between 0.6 and 0.8, and β between 0.2 and 0.4.

[0077] Next, refer to Figure 4 , describing the steps of semantically parsing the user's question by the natural language understanding module based on the deep learning model of the present invention:

[0078] S401: Generate a similar question set based on historical question-answering records; this step aims to build a question library for subsequent processing by mining similar questions in the user's historical question records. By collecting and classifying these historical questions, the background of the question raised by the current user can be better understood, thereby providing more accurate answers. For example, in one embodiment, when the system receives a question about programming knowledge "How to reverse a list in Python?", the system will search the historical question-answering records to find all relevant questions about Python operations, such as "How to reverse a string in Python?", "How to create a copy of a list in Python?", and classify them into a similar question set.

[0079] S402: Cluster analysis of similar question sets; this refers to grouping the similar questions collected above into several clusters through a specific algorithm. This can help locate and process specific types of questions more accurately. The clustering method can use common K-means or DBSCAN algorithms. In one embodiment, the above questions may be divided into two main categories: Python data structure operations and Python advanced programming skills.

[0080] S403: Prioritize the problems at the center of the cluster; that is, prioritize the problems that are identified as the center in the cluster analysis, because they usually represent the most common or most critical parts of similar problems. This can improve the response efficiency and accuracy of the system. For example, the question "How to reverse a list in Python?" is considered to be the most typical case in the Python data structure operation cluster, so the system will focus on processing this question and its variants.

[0081] S404: If the similarity between the user's current question and the historical question exceeds the preset threshold T, the historical answer is directly returned. The similarity calculation formula is S = 1 / (1+exp(d(P1, P2))), where P1 and P2 represent the user's current question and historical question respectively, and d represents the distance measure between the two questions. Here, the parameter T sets a judgment standard. When the similarity between the two reaches a certain degree (usually the value of T is set between 0.85 and 0.9, and the optimal value can be determined by experiment), it is considered that the current problem can be quickly solved by referring to the past answer. The d function is used to calculate the degree of grammatical and semantic differences between two sentences, which can be implemented based on technologies such as word vector distance or edit distance. The reason for choosing this formula is that the Sigmoid function 1 / (1+exp(-x)) is often used to output probability values ​​in machine learning. The deformed formula here can effectively map the value range of d to the [0,1] interval, thereby reasonably evaluating the relative closeness between the two questions. For example, if the system detects that the new question "How to reverse an integer array in Python?" is extremely similar to a previous question (assuming the similarity calculated by the S formula is as high as 0.95), the system will directly retrieve the corresponding answer from the historical answer database and quickly reply to the user, avoiding duplication of work.

[0082] Next, refer to Figure 5 , describing the steps of semantically parsing the user's question by the natural language understanding module based on the deep learning model of the present invention:

[0083] S501: Build a user portrait based on the user's historical behavior data; use the user portrait to predict the user's question-answering needs; dynamically adjust the response priority of the question-answering system; if the user's question-answering request frequency f exceeds the preset threshold F, mark the user as a high-priority user. The formula for calculating the question-answering request frequency is f=N / T, where N is the number of question-answering requests from the user within time T.

[0084] First, build a user profile based on the user's historical behavior data. This process involves collecting various activity records of users on the platform, including but not limited to login frequency, browsing path, length of stay, and number of visits to specific resources. By analyzing these behavioral data, the system can generate a multi-dimensional user profile that includes the user's learning preferences, interests, and learning efficiency. For example, a user who often watches math content in online courses and submits homework early before the deadline may be marked as a student with a high interest in math and high learning efficiency.

[0085] S502: Secondly, use the user portrait to predict the user's question-answering needs. Once the user portrait is established, the system can predict the problems or needs that a specific user may encounter through machine learning algorithms. For example, an online education platform can predict the user's knowledge of related fields and common problems based on the content that a user frequently searches and consults. This helps to prepare relevant content in advance and improve user experience and satisfaction.

[0086] S503: Furthermore, the response priority of the Q&A system is dynamically adjusted. The system's processing speed and service quality of the questions asked by the user are dynamically adjusted according to their importance and urgency. For example, if the system detects that a student has recently prepared for an exam and has accessed exam-related tutoring materials many times, the student's consulting questions will be given a higher priority to ensure that they receive the necessary support in a timely manner.

[0087] S504: Finally, if the user's question-answering request frequency f exceeds the preset threshold F, the user is marked as a high-priority user. Here f = N / T, N represents the total number of question-answering requests made by users within a period of time T, and T is the statistical request time span, which generally ranges from a few hours to a few weeks. The specific optimal value depends on the characteristics of the application scenario and the resource allocation. When the f value is greater than the preset threshold F, it indicates that the user's current demand for question answering is more urgent and requires special attention. The purpose of setting this formula is to effectively screen out the users who need help the most, and to ensure that limited resources can efficiently serve students who really need it, such as students who are highly active in situations such as final review or on the eve of project deadlines. In this context, the above formula can accurately reflect the user's behavior pattern of seeking help in a concentrated manner in a short period of time, and provide strong support for the precise provision of services.

[0088] Next, refer to Figure 6 , describing the steps of semantically parsing the user's question by the natural language understanding module based on the deep learning model of the present invention:

[0089] S601: Based on the multimodal information fusion mechanism, integrate information from multiple data sources; perform secondary verification on the fused information; dynamically adjust the parameters of the verification model; if the understanding confidence C of the user's question is lower than the threshold τ, trigger manual intervention. The understanding confidence calculation formula is C = 1 / (1+exp(z)), where z is the predicted output score of the model.

[0090] Based on the multimodal information fusion mechanism, information from multiple data sources is integrated. This step aims to combine data from different sources, such as text, images, and audio information, to provide richer and more comprehensive input information so that the model can understand and parse the user's questions more accurately. For example, in an online education platform, students may ask questions about a text description while watching a course video and upload related pictures. These texts and pictures are simultaneously input into the deep learning model to assist it in better parsing the background and context information of the question.

[0091] S602: Perform secondary verification on the fused information. This step is to further process the multimodal fusion information to ensure that the model's interpretation of the information is accurate. By using different verification methods or algorithms, such as cross-validation or multi-model integration, further confirm whether the model's understanding of the information is reliable. For example, after the natural language understanding module of the online education platform initially parses the user's picture and text questions, it will use different pre-trained models for re-verification to ensure that the understanding of the questions is accurate.

[0092] S603: Dynamically adjust the parameters of the verification model. During the model operation, the parameters of the verification model are dynamically adjusted according to the types of questions and data characteristics of different users to keep the performance of the model in the best state. Specifically, the system will evaluate the performance of the model after each training cycle, and adjust the key parameters of the model such as the learning rate and regularization parameters according to the evaluation results to improve the generalization ability and accuracy of the model. For example, by analyzing a large number of similar questions raised by students over a period of time, the system finds that the performance is better under certain parameter configurations, and it will automatically adjust these parameters to optimize the processing of similar questions in the future.

[0093] S604: If the confidence C in understanding the user's question is lower than the threshold τ, manual intervention is triggered. The confidence C is calculated by the formula C=1 / (1+exp(z)), where z is the predicted output score of the model. This formula is a form of the sigmoid function, which can map z values ​​in any real number range to the interval (0,1), indicating the model's confidence in understanding the user's question. When the understanding confidence C is lower than the predetermined threshold τ (generally between 0.5 and 0.8), it means that the model cannot be sure that it has correctly understood the user's question. At this time, the system transfers the question to manual customer service for processing. For example, on an online education platform, if the model's confidence in solving a student's math problem is lower than 0.6, the question will be automatically submitted to a human expert to ensure the accuracy of the answer.

[0094] Next, refer to Figure 7 , describing the steps of semantically parsing the user's question by the natural language understanding module based on the deep learning model of the present invention:

[0095] S701: Generate improvement plans based on user feedback information; incorporate the improvement plans into the training data set; regularly evaluate model performance and optimize; if the number of errors of the same type N in user feedback exceeds the threshold M, conduct targeted training on this type of error, and the calculation formula is E=sum(Err) / T, where Err is the set of error records fed back by users, and T is the time period.

[0096] Generate improvement plans based on user feedback information. This means that the system collects user feedback, identifies deficiencies or errors in the system when handling certain specific problems, and formulates corresponding improvement plans based on these feedbacks. For example, on an online education platform, if multiple users report that submission failures often occur when submitting homework, the system can analyze specific error logs to determine that the problem lies in unstable network connection or long server response time, and thus generate improvement plans, such as increasing server resources or optimizing client requests.

[0097] S702: Incorporate the improvement plan into the training data set. This refers to converting the generated improvement plan into training data to enhance the learning ability of the model. For example, in the problem mentioned in the previous step, the system can add the specific error description and scenario of the submission failure provided by the user to the training data set, so that the model can better understand and handle similar problems. In this way, in future interactions, when the user encounters similar problems again, the system can respond faster and more accurately.

[0098] S703: Regularly evaluate model performance and optimize. This step refers to regularly evaluating the model, checking the performance of the model in different aspects, such as accuracy, response time, and user experience, and making necessary adjustments and optimizations based on the evaluation results. For example, in one embodiment, the system automatically generates a model evaluation report every three months, which includes indicators such as user satisfaction, correct parsing rate, and error type distribution. If it is found that the model performs poorly when parsing certain specific grammatical structures, the model parameters can be adjusted or relevant training data can be added to improve the overall performance.

[0099] S704: If the number of errors of the same type N in user feedback exceeds the threshold M, targeted training is performed on this type of error. This step emphasizes the importance of focusing on frequently occurring errors. For example, if the system records show that more than 10 (i.e., M=10) user feedbacks mentioned that a complex sentence could not be parsed in the past month, the system will trigger a targeted training process. The formula E=sum(Err) / T involved here is used to calculate the average frequency of occurrence of the same type of errors within a specific period of time. In this formula, Err represents the set of all error records in user feedback, and T represents the time period of evaluation. Specifically, if T is one month and Err is 50, then E=50 / 1=50. Generally speaking, the higher the E value, the more common such errors are, and the system needs to pay more attention to them. Through this mechanism, the system can dynamically adjust the optimization strategy to ensure that the model is always in the best state. In the context of the present invention, the initial value of M is set to 10 in order to balance system resources and actual needs. When the number of errors reaches this threshold, it means that such problems have seriously affected the user experience and need to be resolved immediately.

[0100] Next, refer to Figure 8 , describing the steps of semantically parsing the user's question by the natural language understanding module based on the deep learning model of the present invention:

[0101] S801: Adjust the model training strategy based on real-time monitoring results;

[0102] S802: Regularly check the accuracy and reliability of the Q&A system;

[0103] S803: Mark and archive low-quality Q&A;

[0104] S804: If the overall performance score P of the system is lower than the preset threshold Q, the emergency plan is activated. The overall performance score calculation formula is P=W1*A1+W2*A2+W3*A3, where W1, W2, W3 are weight parameters, and A1, A2, A3 are various performance indicators, including accuracy, response time and user experience.

[0105] Based on the real-time monitoring results, the model's training strategy is adjusted. This means that by continuously monitoring the user's interaction data, such as the frequency of questions and answers, user feedback, etc., the system can identify which parts of the model are performing poorly, thereby optimizing the training data or updating the model architecture to improve accuracy and response speed. For example, in one embodiment, if it is found that certain types of questions frequently cause user dissatisfaction, the system can strengthen training in these areas by adding data of similar questions.

[0106] Regularly check the accuracy and reliability of the Q&A system. This process involves running automated test scripts regularly or having professionals manually evaluate the performance of the system. Specifically, the accuracy and reliability of the system can be evaluated by comparing the answers generated by the system with the standard answers provided by human experts. In one embodiment, every other month, the development team will sample all user questions and answers collected in the previous month, evaluate the performance of the system, and record improvement measures.

[0107] Mark and archive low-quality questions and answers. When it is detected that the answers provided by the system do not meet the standards or the user feedback is poor, these questions will be specially marked and stored in a special database for subsequent analysis and improvement. For example, a knowledge point in a specific field may cause confusion to multiple users due to unclear expression. Such questions will be marked as to be optimized to encourage the team to focus on improving related content.

[0108] If the overall performance score P of the system is lower than the preset threshold Q, the emergency plan will be activated. The overall performance score calculation formula is P = W1*A1+W2*A2+W3*A3, where W1, W2, and W3 are weight parameters with a value range of 0 to 1, and the optimal value depends on the needs of the specific application scenario; A1, A2, and A3 are various performance indicators, representing accuracy, response time, and user experience, respectively, with a value range of 0 to 100. The P score combines the performance of these three aspects and reflects the overall performance of the system. For example, assuming that the threshold Q set by an education platform is 85 points, when the P value is lower than this standard, the system will automatically trigger a series of measures, such as notifying the technical team, switching to a backup model, or urgently launching a patch to restore service quality as soon as possible. The design of this formula takes into account the importance of different performance indicators in practical applications, and adjusts the emphasis of various aspects by assigning weight parameters to ensure comprehensive optimization and stable operation of the system.

[0109] The present invention discloses an online education course management method based on artificial intelligence, comprising: first, using a natural language understanding module based on a deep learning model to perform semantic analysis on the user's questions, the module extracts and understands the semantic information of the user's questions through a deep neural network, and improves the accuracy of the intelligent question-answering system. Secondly, the system constructs a knowledge graph to form a knowledge base of educational content, which not only contains a large amount of subject knowledge, but also links these knowledge through relationships and attributes, so that the intelligent question-answering system can provide correct and detailed answers more effectively. Thirdly, the user's feedback information during use is collected and used as training data to regularly update and optimize the natural language understanding module, further improving the system's understanding and answering capabilities. In addition, the system dynamically adjusts the course content and difficulty according to each user's learning progress and preferences, thereby achieving a personalized learning experience. Finally, by real-time monitoring of the user's question-answering effect, the system can automatically detect the user's understanding deviation or misunderstanding, and automatically trigger the manual intervention mechanism when necessary, further reducing the occurrence of misunderstandings.

[0110] This method mainly solves the problem of misunderstanding caused by limited natural language processing capabilities in intelligent question-answering systems. Specifically, the natural language understanding module combines the advantages of deep learning models and knowledge graphs, which not only improves the semantic understanding of user questions, but also provides users with answers from multiple angles and dimensions. At the same time, user feedback information is promptly incorporated into the model training process to ensure that the system can continuously optimize and improve its performance. When a problem is discovered, the system can respond quickly and take measures, such as automatically triggering a manual intervention mechanism, thereby effectively reducing misunderstandings caused by insufficient natural language processing capabilities and improving user learning efficiency and satisfaction.

[0111] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium.

[0112] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0113] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0114] Any process or method description in the flow chart or otherwise described herein can be understood to represent a module, fragment or portion of a code including one or more executable instructions for implementing the steps of a specific logical function or process. And the scope of the preferred embodiment of the present application includes other implementations, in which the functions may not be performed in the order shown or discussed, including in a substantially simultaneous manner or in a reverse order according to the functions involved.

[0115] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by an instruction execution system, apparatus or device (such as a computer-based system, a system including a processor or other system that can fetch instructions from an instruction execution system, apparatus or device and execute instructions), or used in combination with these instruction execution systems, apparatuses or devices.

[0116] It should be understood that the various parts of the present application can be implemented with hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented with software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the above embodiment method can be completed by instructing the relevant hardware through a program, which can be stored in a computer-readable storage medium, and when the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0117] In addition, each functional unit in each embodiment of the present application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into one module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of a software functional module. If the above-mentioned integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a disk or an optical disk, etc.

[0118] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An online education course management method based on artificial intelligence, characterized in that: include: The natural language understanding module based on the deep learning model performs semantic analysis on user questions to improve the accuracy of the intelligent question answering system; Build a knowledge base of educational content through knowledge graphs to enhance the system's question-answering capabilities; Use user feedback as training data to regularly update and optimize the natural language understanding module; Dynamically adjust course content and difficulty based on user's learning progress and preferences; Monitor the effectiveness of users' Q&A in real time, automatically trigger the manual intervention mechanism, and reduce misunderstandings.

2. The method for managing online education courses based on artificial intelligence according to claim 1, characterized in that: The natural language understanding module based on the deep learning model performs semantic analysis on the user's question, specifically comprising the following steps: Get the user's original question input; Convert the user's original question input into a standard text format; Perform word segmentation on questions in standard text format; Use pre-trained deep learning models for semantic analysis to improve the accuracy of question understanding and reduce misunderstandings caused by limited natural language processing capabilities.

3. The online education course management method based on artificial intelligence according to claim 2 is characterized in that: The natural language understanding module based on the deep learning model performs semantic analysis on the user's question and further comprises the following steps: Extract relevant background information based on the user's question context information; Perform feature extraction on the extracted background information; Combine the feature extraction results with the problems in the standard text format for comprehensive analysis; If the semantic complexity of the user's question exceeds a preset threshold, the multimodal information fusion mechanism is triggered. The semantic complexity is expressed by the formula C=αL+βF, where L is the vocabulary length of the question, F is the feature vector length, and α and β are weight parameters, which are used to measure the impact of vocabulary length and feature vector on semantic complexity.

4. The online education course management method based on artificial intelligence according to claim 3 is characterized in that: The natural language understanding module based on the deep learning model performs semantic analysis on the user's question and further comprises the following steps: Generate a set of similar questions based on historical question-answering records; Conduct cluster analysis on sets of similar questions; Prioritize the cluster center issues; If the similarity between the user's current question and the historical question exceeds the preset threshold T, the historical answer is directly returned. The similarity calculation formula is S = 1 / (1+exp(d(P1,P2))), where P1 and P2 represent the user's current question and historical question respectively, and d represents the distance measure between the two questions.

5. The method for managing online education courses based on artificial intelligence according to claim 4, characterized in that: The natural language understanding module based on the deep learning model performs semantic analysis on the user's question and further comprises the following steps: Build user portraits based on users’ historical behavior data; Use user portraits to predict users’ question-answering needs; Dynamically adjust the response priority of the Q&A system; If the user's question-answering request frequency f exceeds a preset threshold F, the user is marked as a high-priority user. The question-answering request frequency calculation formula is f=N / T, where N is the number of question-answering requests of the user within time T.

6. The method for managing online education courses based on artificial intelligence according to claim 5, characterized in that: The natural language understanding module based on the deep learning model performs semantic analysis on the user's question and further comprises the following steps: Based on the multimodal information fusion mechanism, it integrates information from multiple data sources; Conduct secondary verification on the integrated information; Dynamically adjust the parameters of the verification model; If the understanding confidence C of the user's question is lower than the threshold τ, manual intervention is triggered. The understanding confidence calculation formula is C = 1 / (1+exp(z)), where z is the predicted output score of the model.

7. The method for managing online education courses based on artificial intelligence according to claim 6, characterized in that: The natural language understanding module based on the deep learning model performs semantic analysis on the user's question and further comprises the following steps: Generate improvement plans based on user feedback; Incorporate improvements into the training dataset; Regularly evaluate model performance and optimize it; If the number of times N of the same type of errors appearing in user feedback exceeds the threshold M, targeted training is performed on this type of error. The calculation formula is E=sum(Err) / T, where Err is the set of error records reported by users and T is the time period.

8. The method for managing online education courses based on artificial intelligence according to claim 7, characterized in that: The natural language understanding module based on the deep learning model performs semantic analysis on the user's question and further comprises the following steps: Adjust the model’s training strategy based on real-time monitoring results; Regularly check the accuracy and reliability of the Q&A system; Mark and archive low-quality Q&A; If the system's overall performance score P is lower than the preset threshold Q, the emergency plan is activated. The overall performance score calculation formula is P=W1*A1+W2*A2+W3*A3, where W1, W2, W3 are weight parameters, and A1, A2, A3 are various performance indicators, including accuracy, response time and user experience.

Images