Method, apparatus, program product, and electronic device for optimizing online courses

Abstract

The present disclosure provides an online course optimization method, device, program product and electronic equipment, relating to the technical field of computer. The method is: determining a to-be-optimized course; configuring a set of virtual student agents of the to-be-optimized course; the set of virtual student agents includes a plurality of virtual student agents with different knowledge absorption capacity parameters; determining the cognitive assimilation information of the set of virtual student agents to the to-be-optimized course, and generating a student confusion heat map according to the cognitive assimilation information; generating an optimization strategy according to the student confusion heat map and the cognitive assimilation information, so as to obtain a target course after the to-be-optimized course is optimized based on the optimization strategy. The present disclosure optimizes the to-be-optimized course before the online course is published through the virtual student agent, and optimizes the to-be-optimized course based on the cognitive assimilation situation, thereby improving the optimization effect of the online course.

Description

Technical Field

[0001] The embodiments of this disclosure relate to the field of computer technology, and more specifically, the embodiments of this disclosure relate to an online course optimization method, an online course optimization apparatus, a computer program product, and an electronic device. Background Technology

[0002] With the deep integration of digital technology and the education industry, online education has comprehensively covered diverse scenarios such as K-12 online classes, vocational skills training, remote teaching, and digital teacher training, becoming an important supplement and extension to traditional offline education. This teaching model, unrestricted by time and space, not only breaks down geographical barriers to high-quality educational resources but also provides learners with flexible and independent learning options and offers teachers an efficient and convenient empowerment channel, driving the education industry towards rapid development towards large-scale and personalized education. Therefore, the market demand for online education continues to expand, and the requirements for teaching quality and delivery effectiveness are also increasing.

[0003] To address the lack of interaction in remote teaching, some technologies have attempted to incorporate artificial intelligence-assisted instructional analysis, such as collecting and analyzing external information like teacher behavior and students' visual characteristics. However, these technologies largely remain at the level of analyzing external behavioral features, and the feedback results are determined after the course has been completed. This leads to a significant lag in feedback and a lack of interpretable support for understanding the causes of teaching problems, making it difficult to provide effective technical guidance for online courses. Summary of the Invention

[0004] In view of this, the present disclosure provides an online course optimization method, an online course optimization device, a computer program product, and an electronic device to improve the optimization effect of online courses to a certain extent.

[0005] According to a first aspect of this disclosure, an optimization method for online courses is provided, the method comprising: Identify courses that need optimization; Configure the set of virtual student intelligent agents for the course to be optimized; the set of virtual student intelligent agents includes multiple virtual student intelligent agents with different knowledge absorption ability parameters; Determine the cognitive assimilation information of the virtual student intelligent body for the course to be optimized, and generate a heatmap of student confusion based on the cognitive assimilation information; Based on the student confusion heatmap and the cognitive assimilation information, an optimization strategy is generated to obtain a target course that optimizes the course to be optimized based on the optimization strategy.

[0006] In one possible implementation, configuring the set of virtual student intelligent agents for the course to be optimized includes: Determine the multimodal learning data and construction instructions for the course to be optimized; The multimodal learning data is subjected to retrieval enhancement processing to determine the knowledge set of cognitive gaps and the knowledge set of high-frequency cognitive doubts; the knowledge set of cognitive gaps is subjected to mapping processing to determine the knowledge mask range; and the knowledge set of high-frequency cognitive doubts is subjected to mapping processing to determine the logical deviation rules. The constructed instruction information is subjected to feature parsing and dimension mapping to determine the learner feature profile; Based on the knowledge mask range, logical deviation rules, and learner feature profiles, a set of virtual student intelligent entities is obtained.

[0007] In one possible implementation, determining the cognitive assimilation information of the virtual student intelligence set regarding the course to be optimized includes: Force alignment is performed on the audio of the course to be optimized to determine the absolute timeline at the millisecond level; The virtual student agents within the virtual student agent set are controlled to scan the course to be optimized one by one according to the millisecond-level absolute time axis to determine cognitive assimilation information.

[0008] In one possible implementation, the virtual student agents within the set of virtual student agents are controlled to scan the course to be optimized one by one according to the millisecond-level absolute time axis to determine cognitive assimilation information, including: For the virtual student agents within the aforementioned set of virtual student agents, perform the following operations: The system retrieves whether the knowledge base within the virtual student agent contains new knowledge points; these new knowledge points are obtained by extracting entities from the current teaching content, where the current teaching content is the teaching content corresponding to the current time slice. If it is determined that the new knowledge point is not found in the knowledge base inside the student agent, then the first teaching text is determined; the first teaching text is the teaching text corresponding to the adjacent time slice of the current time slice; Based on the first teaching text, determine whether the virtual student agent generates comprehension impairment information; When it is determined that the virtual student agent has the comprehension-impeding information, the comprehension-impeding information is classified into cognitive assimilation information.

[0009] In one possible implementation, determining whether the virtual student agent exhibits comprehension difficulties based on the first instruction text includes: Determine whether the first teaching text includes words from a preset vocabulary set; the preset vocabulary set is determined based on real-life analogies and example content; When it is determined that the first teaching text does not include words from the preset word set, the virtual student agent is identified as having comprehension difficulties.

[0010] In one possible implementation, determining whether the virtual student agent exhibits comprehension difficulties based on the first instruction text includes: Determine the number of logical reasoning steps for adjacent sentences within the first teaching text; When it is determined that the number of logical reasoning steps is greater than the first threshold, it is determined that the virtual student agent generates comprehension impairment information.

[0011] In one possible implementation, determining whether the virtual student agent exhibits comprehension difficulties based on the instruction text includes: Determine the formula content within the first teaching text; If it is determined that the first teaching text does not include variable introduction information containing the formula, then it is determined that the virtual student agent has generated comprehension impairment information.

[0012] In one possible implementation, the method further includes: When it is determined that the cognitive assimilation information includes comprehension impairment information, causal tracing processing is initiated for the course to be optimized to determine the causal diagnosis result.

[0013] In one possible implementation, a causal attribution mechanism is initiated for the course to be optimized to determine the causal diagnosis results, including: Determine the current moment when the virtual student agent generates the information about comprehension blockage; The offline tracing engine is triggered to perform a backtracking scan on the second teaching text corresponding to the current moment on the millisecond-level absolute time axis based on the knowledge graph of the teaching syllabus of the course to be optimized, in order to determine the causal diagnosis result.

[0014] In one possible implementation, an offline tracing engine is triggered to perform a backtracking scan of the second teaching text corresponding to the current moment on the millisecond-level absolute timeline, based on the knowledge graph of the syllabus of the course to be optimized, to determine the causal diagnosis result, including: Query the prerequisite dependencies of the current teaching text in the knowledge graph; For the aforementioned prerequisite content and the second teaching text, keyword matching and semantic similarity calculation are performed to determine the processing result information; When the processing result information of the first historical moment meets the first preset condition, it is determined that the course to be optimized lacks key preparatory information; the first preset condition is that the keyword matching degree is lower than the second threshold and the semantic similarity is lower than the third threshold; Based on the content of the course to be optimized corresponding to the current moment and the first historical moment, a first causal chain is constructed.

[0015] In one possible implementation, an offline tracing engine is triggered to perform a backtracking scan of the teaching texts along the historical timeline based on the knowledge graph of the syllabus of the course to be optimized, to determine the causal diagnosis results, including: If the processing result information of the second historical moment does not meet the first preset condition, determine the attention half-life of the virtual student agent; When it is determined that the time difference between the second historical moment and the current moment exceeds the attention half-life, and the content of the course to be optimized within the time difference does not include the preceding dependent content, then a second causal chain is constructed based on the content of the course to be optimized corresponding to the current moment and the second historical moment.

[0016] In one possible implementation, an offline tracing engine is triggered to perform a backtracking scan of the teaching texts along the historical timeline based on the knowledge graph of the syllabus of the course to be optimized, to determine the causal diagnosis results, including: Determine the formula content within the second lecture text at the third historical moment; When it is determined that the quantity of information in the formula exceeds the fourth threshold, a third causal chain is constructed based on the content of the course to be optimized corresponding to the current time and the third historical time.

[0017] In one possible implementation, the method further includes: Based on the duration and severity of the comprehension impairment information included in the cognitive assimilation information, the comprehension score corresponding to the moment on the millisecond-level absolute time axis is determined. The comprehension curve of the virtual student agent is determined based on the comprehension score and the millisecond-level absolute time axis.

[0018] In one possible implementation, generating a student confusion heatmap based on the cognitive assimilation information includes: Determine the first comprehension curve of the virtual student intelligence set; Based on the first comprehension curve of the virtual student intelligence set, determine the comprehensive confusion information at each moment on the millisecond-level absolute time axis; Determine the causal chain corresponding to the virtual student agents within the set of virtual student agents, determine the causal nodes, and determine the confusion enhancement moments corresponding to the causal nodes; Based on the comprehensive confusion information at each moment on the millisecond-level absolute time axis and the source-tracing penalty weight corresponding to the confusion enhancement moment, the target confusion information at each moment on the millisecond-level absolute time axis is determined. Heatmap mapping processing is performed on the target confusion information at each moment on the millisecond-level absolute time axis to determine the student confusion heatmap.

[0019] In one possible implementation, an optimization strategy is generated based on the student confusion heatmap and the cognitive assimilation information, including: Based on the student confusion heatmap and the cognitive assimilation information, the high confusion interval is determined; Determine the third instruction text corresponding to the high confusion interval, the learner profile of the virtual student agent, and the target causal chain; The third teaching text, the learner profile of the virtual student agent, and the target causal chain are input into a preset large language model to determine optimization suggestions, and challenge propositions are generated based on the optimization suggestions.

[0020] In one possible implementation, determining high-confusion intervals based on the student confusion heatmap and the cognitive assimilation information includes: The first high confusion interval is determined based on the first region within the student confusion heatmap where the confusion index exceeds the fifth threshold. Identify the second region of the causal node within the student confusion heatmap; When the result node corresponding to the causal node is located within the first high confusion interval, and the confusion index of the second region does not exceed the fifth threshold but exceeds the sixth threshold, then the second high confusion interval is determined based on the second region. The high confusion interval is determined based on the first high confusion interval and / or the second high confusion interval.

[0021] In one possible implementation, the method further includes: Receive the optimized curriculum based on the aforementioned challenge proposition; The optimized course is retested based on the virtual student intelligence set to determine the second comprehension curve corresponding to the virtual student intelligence set. When it is determined that the comparison result between the second comprehension curve and the first comprehension curve meets the second preset condition, the optimized course is taken as the target course.

[0022] According to a second aspect of this disclosure, an optimization apparatus for online courses is provided, the apparatus comprising: Identify units to determine the courses to be optimized; A configuration unit is used to configure the set of virtual student intelligent agents for the course to be optimized; the set of virtual student intelligent agents includes multiple virtual student intelligent agents with different knowledge absorption ability parameters; The processing unit is used to determine the cognitive assimilation information of the virtual student intelligent body to the course to be optimized, and to generate a student confusion heatmap based on the cognitive assimilation information. An optimization unit is used to generate an optimization strategy based on the student confusion heatmap and the cognitive assimilation information, so as to obtain a target course after optimizing the course to be optimized based on the optimization strategy.

[0023] In one possible implementation, the configuration unit is configured to: Determine the multimodal learning data and construction instructions for the course to be optimized; The multimodal learning data is subjected to retrieval enhancement processing to determine the knowledge set of cognitive gaps and the knowledge set of high-frequency cognitive doubts; the knowledge set of cognitive gaps is subjected to mapping processing to determine the knowledge mask range; and the knowledge set of high-frequency cognitive doubts is subjected to mapping processing to determine the logical deviation rules. The constructed instruction information is subjected to feature parsing and dimension mapping to determine the learner feature profile; Based on the knowledge mask range, logical deviation rules, and learner feature profiles, a set of virtual student intelligent entities is obtained.

[0024] In one possible implementation, the processing unit is configured to: Force alignment is performed on the audio of the course to be optimized to determine the absolute timeline at the millisecond level; The virtual student agents within the virtual student agent set are controlled to scan the course to be optimized one by one according to the millisecond-level absolute time axis to determine cognitive assimilation information.

[0025] In one possible implementation, the processing unit is configured to: For the virtual student agents within the aforementioned set of virtual student agents, perform the following operations: The system retrieves whether the knowledge base within the virtual student agent contains new knowledge points; these new knowledge points are obtained by extracting entities from the current teaching content, where the current teaching content is the teaching content corresponding to the current time slice. If it is determined that the new knowledge point is not found in the knowledge base inside the student agent, then the first teaching text is determined; the first teaching text is the teaching text corresponding to the adjacent time slice of the current time slice; Based on the first teaching text, determine whether the virtual student agent generates comprehension impairment information; When it is determined that the virtual student agent has the comprehension-impeding information, the comprehension-impeding information is classified into cognitive assimilation information.

[0026] In one possible implementation, the processing unit is configured to: Determine whether the first teaching text includes words from a preset vocabulary set; the preset vocabulary set is determined based on real-life analogies and example content; When it is determined that the first teaching text does not include words from the preset word set, the virtual student agent is identified as having comprehension difficulties.

[0027] In one possible implementation, the processing unit is configured to: Determine the number of logical reasoning steps for adjacent sentences within the first teaching text; When it is determined that the number of logical reasoning steps is greater than the first threshold, it is determined that the virtual student agent generates comprehension impairment information.

[0028] In one possible implementation, the processing unit is configured to: Determine the formula content within the first teaching text; If it is determined that the first teaching text does not include variable introduction information containing the formula, then it is determined that the virtual student agent has generated comprehension impairment information.

[0029] In one possible implementation, the processing unit is configured to: When it is determined that the cognitive assimilation information includes comprehension impairment information, causal tracing processing is initiated for the course to be optimized to determine the causal diagnosis result.

[0030] In one possible implementation, the processing unit is configured to: Determine the current moment when the virtual student agent generates the information about comprehension blockage; The offline tracing engine is triggered to perform a backtracking scan on the second teaching text corresponding to the current moment on the millisecond-level absolute time axis based on the knowledge graph of the teaching syllabus of the course to be optimized, in order to determine the causal diagnosis result.

[0031] In one possible implementation, the processing unit is configured to: Query the prerequisite dependencies of the current teaching text in the knowledge graph; For the aforementioned prerequisite content and the second teaching text, keyword matching and semantic similarity calculation are performed to determine the processing result information; When the processing result information of the first historical moment meets the first preset condition, it is determined that the course to be optimized lacks key preparatory information; the first preset condition is that the keyword matching degree is lower than the second threshold and the semantic similarity is lower than the third threshold; Based on the content of the course to be optimized corresponding to the current moment and the first historical moment, a first causal chain is constructed.

[0032] In one possible implementation, the processing unit is configured to: If the processing result information of the second historical moment does not meet the first preset condition, determine the attention half-life of the virtual student agent; When it is determined that the time difference between the second historical moment and the current moment exceeds the attention half-life, and the content of the course to be optimized within the time difference does not include the preceding dependent content, then a second causal chain is constructed based on the content of the course to be optimized corresponding to the current moment and the second historical moment.

[0033] In one possible implementation, the processing unit is configured to: Determine the formula content within the second lecture text at the third historical moment; When it is determined that the quantity of information in the formula exceeds the fourth threshold, a third causal chain is constructed based on the content of the course to be optimized corresponding to the current time and the third historical time.

[0034] In one possible implementation, the processing unit is further configured to: Based on the duration and severity of the comprehension impairment information included in the cognitive assimilation information, the comprehension score corresponding to the moment on the millisecond-level absolute time axis is determined. The comprehension curve of the virtual student agent is determined based on the comprehension score and the millisecond-level absolute time axis.

[0035] In one possible implementation, the optimization unit is configured to: Determine the first comprehension curve of the virtual student intelligence set; Based on the first comprehension curve of the virtual student intelligence set, determine the comprehensive confusion information at each moment on the millisecond-level absolute time axis; Determine the causal chain corresponding to the virtual student agents within the set of virtual student agents, determine the causal nodes, and determine the confusion enhancement moments corresponding to the causal nodes; Based on the comprehensive confusion information at each moment on the millisecond-level absolute time axis and the source-tracing penalty weight corresponding to the confusion enhancement moment, the target confusion information at each moment on the millisecond-level absolute time axis is determined. Heatmap mapping processing is performed on the target confusion information at each moment on the millisecond-level absolute time axis to determine the student confusion heatmap.

[0036] In one possible implementation, the optimization unit is configured to: Based on the student confusion heatmap and the cognitive assimilation information, the high confusion interval is determined; Determine the third instruction text corresponding to the high confusion interval, the learner profile of the virtual student agent, and the target causal chain; The third teaching text, the learner profile of the virtual student agent, and the target causal chain are input into a preset large language model to determine optimization suggestions, and challenge propositions are generated based on the optimization suggestions.

[0037] In one possible implementation, the optimization unit is configured to: The first high confusion interval is determined based on the first region within the student confusion heatmap where the confusion index exceeds the fifth threshold. Identify the second region of the causal node within the student confusion heatmap; When the result node corresponding to the causal node is located within the first high confusion interval, and the confusion index of the second region does not exceed the fifth threshold but exceeds the sixth threshold, then the second high confusion interval is determined based on the second region. The high confusion interval is determined based on the first high confusion interval and / or the second high confusion interval.

[0038] In one possible implementation, the optimization unit is configured to: Receive the optimized curriculum based on the aforementioned challenge proposition; The optimized course is retested based on the virtual student intelligence set to determine the second comprehension curve corresponding to the virtual student intelligence set. When it is determined that the comparison result between the second comprehension curve and the first comprehension curve meets the second preset condition, the optimized course is taken as the target course.

[0039] According to a third aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method of the first aspect described above and possible implementations thereof.

[0040] According to a fourth aspect of this disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the method of the first aspect and possible implementations thereof by executing the executable instructions.

[0041] The technical solution disclosed herein has the following beneficial effects: In this embodiment, a course to be optimized can be determined, and then a set of virtual student intelligent agents for the course to be optimized can be configured. The set of virtual student intelligent agents includes multiple virtual student intelligent agents with different knowledge absorption capacity parameters. Therefore, in this embodiment, an intelligent class group, i.e., a set of virtual student intelligent agents, matching the course to be optimized can be generated. This set of virtual student intelligent agents includes virtual student intelligent agents with poor course absorption capacity, virtual student intelligent agents with average course absorption capacity, and virtual student intelligent agents with high course absorption capacity. The number of these three types of intelligent agents can be determined based on the learning situation of the targeted class in the actual teaching of the course to be optimized. In this way, a set of virtual student intelligent agents precisely adapted to the course to be optimized can be generated.

[0042] In this embodiment of the disclosure, after determining the course to be optimized and the virtual student agent, the cognitive assimilation information of the virtual student agent set on the course to be optimized can be determined, and a student confusion heatmap can be generated based on the cognitive assimilation information; and, an optimization strategy can be generated based on the student confusion heatmap and the cognitive assimilation information to obtain a target course after optimization of the course to be optimized based on the optimization strategy.

[0043] It is evident that a heatmap of student confusion can be determined based on the cognitive assimilation information of a virtual student agent, and then the course to be optimized can be optimized based on this heatmap to obtain an optimized target course. In other words, in this embodiment, a virtual student agent can be used to pre-optimize the course before its release, thereby clearly identifying specific defects and causes in aspects such as course content, structure, interaction, and knowledge point design. This provides an explainable, traceable, and implementable basis for improving course instruction. Furthermore, it allows for the identification and correction of causes before problems arise, enhancing the interpretability of teaching problem analysis, overcoming the lag in feedback results, and achieving the goal of improving course quality and teaching effectiveness from the source.

[0044] Other features and advantages of this disclosure will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the disclosure. The objects and other advantages of this disclosure may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings used in the embodiments of this disclosure will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1A schematic diagram of an application scenario in this exemplary embodiment is shown.

[0047] Figure 2 A schematic diagram of an online course optimization system according to this exemplary embodiment is shown.

[0048] Figure 3 This diagram illustrates a flowchart of an online course optimization method according to this exemplary embodiment.

[0049] Figure 4 This illustration shows a process for determining a virtual student agent in this exemplary embodiment.

[0050] Figure 5 This illustration shows a process for obtaining a heatmap of student confusion in this exemplary embodiment.

[0051] Figure 6 This illustration shows a process for obtaining a challenging proposition in one of the exemplary embodiments of the present invention.

[0052] Figure 7 A schematic diagram of the structure of an online course optimization device in this exemplary embodiment is shown.

[0053] Figure 8 A schematic diagram of the structure of an electronic device in this exemplary embodiment is shown. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this disclosure clearer, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure. Unless otherwise specified, the embodiments and features in the embodiments of this disclosure can be arbitrarily combined with each other. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0055] The term "comprising" and any variations thereof in the specification and claims of this disclosure are intended to cover non-exclusive protection. For example, a process, method, system, product, or apparatus that comprises a series of steps or units is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0056] In this disclosure, there are one or more embodiments; "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0057] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order, sequence, size, or priority. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0058] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, which are schematic illustrations of this disclosure and are not necessarily drawn to scale. Some block diagrams shown in the drawings may be functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in hardware modules or integrated circuits, or in networks, processors, or microcontrollers. Implementations can be carried out in various forms and should not be construed as limited to the examples set forth herein. The features, structures, or characteristics described in this disclosure can be combined in any suitable manner in one or more implementations. Numerous specific details are provided in the following description to give a thorough description of the embodiments of this disclosure. However, those skilled in the art will recognize that one or more specific details may be omitted when implementing the technical solutions of this disclosure, or other methods, components, devices, steps, etc., may be used to replace one or more specific details. It should be noted that in the embodiments of this disclosure, the collection, dissemination, use, and display of data all comply with relevant national laws and regulations. Invention Overview

[0060] Currently, in scenarios such as K-12 online classes, vocational training, remote teaching, and digital teacher training, teachers generally face the practical dilemma of "teaching remotely." Because "teaching remotely" makes it impossible to perceive the students' attention and understanding in real time, problems such as a mismatch between the difficulty of the lecture and the students' ability to absorb the information, and an imbalance in the pace of the class can easily arise.

[0061] Therefore, how to test and improve the effectiveness of "remote teaching" has become an urgent technical problem to be solved.

[0062] To address the aforementioned issues, most related technologies rely on post-course quality control through manual sampling, which not only results in significant delays in feedback but also fails to provide objective and actionable evidence for teaching improvement. However, attempting to conduct quality verification before course release and organizing real students for trial lectures presents challenges due to organizational difficulties and high implementation costs. Furthermore, the evaluation sample cannot cover students with varying knowledge bases and learning abilities, resulting in ineffective pre-course quality control. Additionally, while some solutions propose evaluation techniques based on external behavioral recognition during lectures—using computer vision to monitor student viewing status, facial expressions, and other classroom behaviors, or analyzing teacher speech rate, volume, and frequency of classroom interaction through voice analysis—these approaches largely remain at the level of judging external teaching behaviors and classroom performance, or checking for sensitive information, visual anomalies, or audio malfunctions in the course video.

[0063] It is evident that the technical solutions provided in these related technologies fail to truly address the core pain points of remote teaching. They are only effective during or after the course, lacking low-cost, high-efficiency pre-class preparation and refinement mechanisms. They cannot optimize course design and teaching plans from the outset, nor can they accurately assess students' true understanding of knowledge points. Teachers struggle to identify gaps in their teaching logic or the clarity and accessibility of their explanations. Furthermore, most feedback is generated only after the course ends, often presented in general grades, lacking precise interpretation and explainability regarding specific problems and their causes, resulting in weak guidance. Additionally, existing grading methods largely employ uniform standards, failing to adapt to the teaching characteristics of different levels and types of classes, thus lacking specificity.

[0064] In view of this, this disclosure provides a method for optimizing online courses. This method identifies the course to be optimized and then configures a set of virtual student intelligent agents for that course. The set of virtual student intelligent agents includes multiple virtual student intelligent agents with different knowledge absorption capacity parameters. Therefore, this disclosure can generate an intelligent class group, i.e., a set of virtual student intelligent agents, that matches the course to be optimized. This set of virtual student intelligent agents includes virtual student intelligent agents with poor course absorption capacity, virtual student intelligent agents with average course absorption capacity, and virtual student intelligent agents with high course absorption capacity. The number of these three types of intelligent agents can be determined based on the learning situation of the targeted class in the actual teaching of the course to be optimized. In this way, a set of virtual student intelligent agents that is precisely adapted to the course to be optimized can be generated.

[0065] In this embodiment of the disclosure, after determining the course to be optimized and the virtual student agent, the cognitive assimilation information of the virtual student agent set on the course to be optimized can be determined, and a student confusion heatmap can be generated based on the cognitive assimilation information; and, an optimization strategy can be generated based on the student confusion heatmap and the cognitive assimilation information to obtain a target course after optimization of the course to be optimized based on the optimization strategy.

[0066] It is evident that a heatmap of student confusion can be determined based on the cognitive assimilation information of a virtual student agent, and then the course to be optimized can be optimized based on this heatmap to obtain an optimized target course. In other words, in this embodiment, a virtual student agent can be used to pre-optimize the course before its release, thereby clearly identifying specific defects and causes in aspects such as course content, structure, interaction, and knowledge point design. This provides an explainable, traceable, and implementable basis for improving course instruction. Furthermore, it allows for the identification and correction of causes before problems arise, enhancing the interpretability of teaching problem analysis, overcoming the lag in feedback results, and achieving the goal of improving course quality and teaching effectiveness from the source.

[0067] After introducing the basic principles of this disclosure, various non-limiting embodiments of this disclosure will be described in detail below.

[0068] Application Scenarios Overview

[0069] To better understand the technical solutions provided in the embodiments of this disclosure, the following is a brief introduction to the application scenarios applicable to the technical solutions provided in the embodiments of this disclosure. It should be noted that the application scenarios described below are only for illustrating the embodiments of this disclosure and are not intended to limit the scope. In specific implementation, the technical solutions provided in the embodiments of this disclosure can be flexibly applied according to actual needs.

[0070] In this embodiment of the disclosure, the online course optimization technology can be applied to various business scenarios for course optimization, such as business scenarios for optimizing K12 online courses, business scenarios for optimizing vocational training online courses, and business scenarios for optimizing digital teacher training. This embodiment of the disclosure does not limit this.

[0071] Please see Figure 1 As shown, Figure 1 This is an application scenario to which the technical solution of the present disclosure embodiment can be applied. The schematic diagram of this scenario includes a terminal device 110 and a service device 120. The terminal device 110 can be one or more. Figure 1 One example is shown below. Terminal device 110 and service device 120 are connected via network 130.

[0072] In this embodiment of the disclosure, the terminal device 110 may be deployed with a platform (such as an application, website, etc.) that provides course optimization, which can be logged in and used by different users with corresponding accounts and passwords.

[0073] In this embodiment, a user can log in to the recommendation system through terminal device 110, determine the course A to be optimized through terminal device 110, and send the data of course A to be optimized to service device 120. Then, service device 120 can determine the course to be optimized; configure a set of virtual student intelligent agents for the course to be optimized; the set of virtual student intelligent agents includes multiple virtual student intelligent agents with different knowledge absorption ability parameters; determine the cognitive assimilation information of the course to be optimized by the set of virtual student intelligent agents, and generate a student confusion heatmap based on the cognitive assimilation information; generate an optimization strategy based on the student confusion heatmap and the cognitive assimilation information to obtain a target course after optimization based on the optimization strategy.

[0074] In this embodiment of the disclosure, Figure 1 The service device 120 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server or cloud server cluster that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms, but is not limited to these.

[0075] In this embodiment of the disclosure, the service device 120 is equipped with an online course optimization system, which can provide course optimization services for the platform in the terminal device 110.

[0076] For example, see Figure 2As shown, the online course optimization system includes an input module, a preprocessing module, a virtual class generation engine, a full-time comprehension causal tracing module, and an adaptive teaching adjustment generation module.

[0077] In this exemplary embodiment, the input module receives the course video to be optimized (i.e., online course video), learning data, and instructions (such as instructions for building learner profiles) from the teacher. The input module is connected to a virtual class generation engine module, which generates heterogeneous student agents to obtain a set of virtual student agent instances. The input module is also connected to a preprocessing module, which integrates Iterated Sequence Estimation (ISE) and Automatic Speech Recognition (ASR) technologies. ASR converts the audio in the course video to be optimized into text, while ISE forces timeline alignment between audio frames and text, resulting in a timestamped, sentence-by-sentence aligned teaching content stream.

[0078] In this exemplary embodiment, the full-time comprehension causal attribution module receives the teaching content stream and a set of virtual student agent instances, drives the agents to perform cognitive simulation and causal attribution, and obtains comprehension curves, causal chains, and student confusion heatmaps. The obtained comprehension curves, causal chains, and student confusion heatmaps are then input into the adaptive teaching adjustment generation module to obtain optimization suggestions and challenge propositions for the course to be optimized, and these suggestions are fed back to the terminal device, i.e., to the teacher. In this way, the teacher can optimize the course to be optimized based on the optimization suggestions and challenge propositions to generate an optimized course. When it is determined that the optimized course meets the requirements after retesting, the optimized course can be published as the target course.

[0079] In this embodiment of the disclosure, Figure 1 The terminal device 110 can be a mobile phone, tablet computer (PAD), personal computer, smart TV, smart watch, smart speaker, smart vehicle device, and wearable device, but is not limited to these.

[0080] Of course, the methods provided in this disclosure are not limited to... Figure 1 The application scenarios shown can also be used in other possible application scenarios, and this disclosure does not limit the scope of the embodiments.

[0081] Exemplary methods

[0082] To further illustrate the technical solutions provided by the embodiments of this disclosure, a detailed description is provided below in conjunction with the accompanying drawings and specific implementation methods. Although the embodiments of this disclosure provide method operation steps as shown in the following embodiments or drawings, the method may include more or fewer operation steps based on conventional or non-inventive methods. In steps where there is no logically necessary causal relationship, the execution order of these steps is not limited to the execution order provided by the embodiments of this disclosure. In actual processing or when the device executes the method, it may be executed sequentially or in parallel according to the method shown in the embodiments or drawings.

[0083] Please see Figure 3 , Figure 3 This is a flowchart illustrating an online course optimization method according to an embodiment of this disclosure. The method flow can be, for example, carried out by an electronic device, such as... Figure 1 The service device 120 executes the method, and the specific implementation process is as follows: Step 301: Identify the courses to be optimized.

[0084] In this embodiment, the electronic device can receive an optimization request triggered by a user's terminal device. This optimization request carries information about the course to be optimized, allowing the electronic device to identify the course. Alternatively, the electronic device can automatically retrieve the course based on optimization cycle information; this embodiment does not limit this approach. The courses mentioned in this disclosure are all presented in the form of online courses. These courses do not rely on offline classroom teaching but are opened and delivered to learners through digital methods such as remote teaching and online courses; therefore, they can be uniformly defined as online courses.

[0085] For example, teacher A selects "lesson preparation mode" for course A through a platform that provides course optimization deployed on the corresponding terminal device, so that the terminal device can send the information of course A to electronic devices.

[0086] For example, if Teacher B launches a course on data analysis for Grade 11 students every Friday during the 2026 academic year and optimizes the course on Wednesday, then electronic devices can automatically obtain the course information sent by the terminal devices every Wednesday, thereby determining Teacher B's courses to be optimized.

[0087] Step 302: Configure the set of virtual student intelligent agents for the course to be optimized; the set of virtual student intelligent agents includes multiple virtual student intelligent agents with different knowledge absorption parameters.

[0088] In this embodiment of the disclosure, after determining the course to be optimized, learners who attend the class can be simulated to determine a virtual student class that is compatible with the course to be optimized, that is, a virtual student intelligent body that is configured with the course to be optimized.

[0089] In one alternative implementation, the set of virtual student intelligent agents can be determined based on the following: Step A: Determine the multimodal learning data and construction instructions for the course to be optimized.

[0090] In this embodiment of the disclosure, the electronic device can receive multimodal learning data and construction instruction information sent by the user based on the terminal device.

[0091] In this embodiment of the disclosure, multimodal learning data refers to a collection of multiple types of data acquired through multi-source sensing devices and intelligent analysis technology during online teaching, remote classrooms, and digital learning processes. This data comprehensively reflects learners' learning status, cognitive level, and behavioral characteristics. Specifically, multimodal learning data includes visual modal data such as learners' facial expressions, body movements, and concentration levels (video and image information); auditory modal data such as classroom voice interaction, responses, and pronunciation fluency (audio information); textual modal data such as course content, quiz results, and interactive messages (textual information); and behavioral modal data such as learning progress, operation trajectory, quiz duration, and frequency of knowledge point review (interactive behavioral data).

[0092] In this embodiment of the disclosure, constructing instruction information refers to extracting and clarifying the characteristic profile and learning requirements of the learner group that the course optimization target group targets, and forming a set of standardized instructions to guide course optimization. For example, if the course optimization target group of Course 1 is Teacher F, then constructing instruction information means extracting and clarifying the characteristic profile and learning requirements of the learner group that Teacher F targets, and forming a set of standardized instructions to guide course optimization.

[0093] In this embodiment of the disclosure, the multimodal learning data and construction instruction information may be adapted courses to be optimized and sent by the user based on the terminal device.

[0094] For example, teacher A teaches the same subject to four classes: Class 1, Class 2, Class 3, and Class 4. There are four sets of multimodal learning data corresponding to the course to be optimized: Multimodal Learning Data 1, Multimodal Learning Data 2, Multimodal Learning Data 3, and Multimodal Learning Data 4, as well as construction instruction information 1, Construction Instruction Information 2, Construction Instruction Information 3, and Construction Instruction Information 4 corresponding to different classes. Thus, when teacher A wants to optimize the course, they can either input all four sets of multimodal learning data and their corresponding construction instruction information for simultaneous optimization, or they can input only one set of multimodal learning data and its corresponding construction instruction information for optimization. After optimizing the course for one class, they can then optimize the course for the next class. This embodiment does not limit this approach. The following description uses the determination of a virtual student intelligent body set based on one set of multimodal data and its corresponding construction instruction information as an example.

[0095] Step B: Perform retrieval enhancement processing on the multimodal learning data to identify the knowledge set not covered by cognition and the knowledge set of high-frequency cognitive doubts; perform mapping processing on the knowledge set not covered by cognition to determine the knowledge mask range; and perform mapping processing on the knowledge set of high-frequency cognitive doubts to determine the logical deviation rules.

[0096] In this embodiment of the disclosure, the obtained multimodal learning data can first be cleaned to obtain cleaned multimodal learning data, and then the cleaned multimodal learning data can be converted into vector embeddings and stored in a vector database. Optionally, the cleaned multimodal learning data can be converted into vector embeddings and stored in a vector database using an embedding model (such as text-embedding-ada-002).

[0097] In this embodiment, Retrieval-Augmented Generation (RAG) technology can be used to retrieve the knowledge nodes with the highest error rate or the most frequent questions in the class targeted by the course, using the knowledge point tree of the syllabus of the course to be optimized as the query. These nodes are defined as the cognitive gap set, or "knowledge blind spot". This cognitive gap set includes concepts that students have not yet mastered or have mastered incorrectly. At the same time, expressions or easily confused concepts that students repeatedly use when asking questions are extracted and defined as the high-frequency cognitive doubt knowledge set, or "high-frequency confusion point". This high-frequency cognitive doubt knowledge set includes the specific logical obstacles where cognitive biases occur.

[0098] In this embodiment, the online course optimization system (hereinafter referred to as the system) incorporates a structured "profile parameter" system, mainly including: Knowledge Masking Range (i.e., which concepts are mastered); Cognitive Style (e.g., preference for formula derivation or life analogy); Attention Decay Rate, etc. The system directly maps the "knowledge blind spots" extracted in the previous step to the Agent's "knowledge masking range" (i.e., depriving the Agent of prior knowledge of these concepts), and maps "high-frequency confusion points" to the Agent's cognitive bias Prompt rules. Thus, it can map the knowledge set not covered by cognition to determine the knowledge masking range; and map the knowledge set of high-frequency cognitive confusion points to determine the logical bias rules.

[0099] Step C: Perform feature parsing and dimension mapping on the construction instruction information to determine the learner feature profile.

[0100] In this embodiment of the disclosure, the system can receive construction instruction information input by the user through the control panel of the terminal device that provides optimization services. The construction instruction information includes multiple sub-information, and each sub-information can be processed as follows to obtain a corresponding learner feature profile, thereby obtaining multiple learner feature profiles.

[0101] For example, the system can obtain the sub-information "create a student with extremely poor attention and a preference for life analogies". Then, based on the built-in structured parameter parser, the system performs feature parsing and dimension mapping on the construction instruction information, mapping the input natural language construction instruction information sub-information to preset model dimension values, thereby determining the learner feature profile based on the model dimension values.

[0102] In this embodiment of the disclosure, the user (e.g., a teacher creating or teaching a course) can determine the actual student situation of a specific class in the course to be optimized, and then determine the construction instruction information based on the actual student situation, i.e., natural language instructions describing the actual student situation. This natural language instruction can be a detailed description or a general description, and this embodiment of the disclosure does not limit this. For example, a general description could be: learners are divided into four groups according to their learning characteristics, with proportions of 10%, 50%, 30%, and 10%, respectively. These four groups are: a first group with sufficient knowledge reserves and long sustained attention span; a second group with moderate knowledge reserves and short sustained attention span; a third group with relatively low knowledge reserves and reliance on concrete examples for understanding; and a fourth group with a phased decline in attention.

[0103] In this way, a corresponding learner feature profile can be generated based on the construction instruction information.

[0104] Step D: Obtain the virtual student intelligent entity set based on the knowledge mask range, logical deviation rules, and learner feature profile.

[0105] In this embodiment of the disclosure, after obtaining the knowledge mask range, logical deviation rules and learner feature profile, the corresponding virtual student intelligent agent can be generated based on the knowledge mask range, logical deviation rules and learner feature profile, that is, a virtual student with specific cognitive defects is obtained, thereby obtaining a set of virtual student intelligent agents, i.e., an intelligent class group.

[0106] For example, by combining extracted blind spots, points of confusion, and pre-set or teacher-adjusted profile parameters, the system uses a large model to generate a specific System Prompt, such as: "You are a third group of learners with limited knowledge reserves and a reliance on concrete examples for understanding. Your knowledge mask includes [knowledge blind spots A and B]. When you encounter [high-frequency point of confusion C], you will experience logical confusion..." This instantiates virtual student agents with specific cognitive deficiencies.

[0107] As can be seen, this embodiment abandons the "expert scoring" perspective and instead uses a pre-set Large Language Model (LLM) Agent to construct a virtual student with "knowledge deficiencies" (Knowledge Masking). This virtual student can simulate the process of knowledge assimilation, reasoning obstacles (inability to understand), and attention loss (distraction) that students experience during lectures.

[0108] In this embodiment of the disclosure, for a better explanation of the process of determining the virtual student intelligent agent, please refer to... Figure 4 As shown, the Parametric Class Customization Engine in the system processes input data and instantiates virtual students through the following dual-thread parallel and then fused mechanism.

[0109] exist Figure 4 The left-hand side handles the learning data stream processing. The input to this process is multimodal learning data for a specific class uploaded by teachers (e.g., images / text containing incorrect question stems and student errors in their answers, or chat logs and question-asking logs from previous Q&A sessions). The specific processing techniques involve cleaning the data and then converting it into vector embeddings using an embedding model before storing it in a vector database; and using RAG (Retrieval Enhancement Generation) technology to perform retrieval using the knowledge point tree from the teaching syllabus as the query. The output of this process is the extraction of the "knowledge blind spots" and "high-frequency confusion points" sets for a specific class.

[0110] exist Figure 4 The right-hand side handles the profile parameter stream processing. The input to this processing is parameter commands entered by the teacher through the control panel (e.g., "Create a student with extremely poor attention span and a preference for everyday analogies"), or by directly accessing a pre-set parameter package from the system. The specific processing technique involves using the system's built-in structured parameter parser to map the teacher's natural language commands into pre-set model dimension values. The output of this processing is a standardized set of profile parameters (including cognitive style, attention half-life, etc.).

[0111] exist Figure 4In this system, once the sets of "knowledge blind spots," "high-frequency confusion points," and standardized profile parameters for a specific class are obtained, LLM is used as the instantiation engine. The system maps the "knowledge blind spots" extracted from the left path to "knowledge mask ranges," maps the "high-frequency confusion points" to logical deviation rules, and combines them with parameters such as "cognitive style" from the right path to automatically generate a structured System Prompt. For example, the System Prompt might be: "You are a high school student. Your knowledge mask includes [function continuity] (from the left-path blind spot), you prefer [life analogies] (from the right-path parameters), and you experience logical confusion when encountering [derivative derivation] (from the left-path confusion point)..." This allows the system to output virtual students with specific cognitive deficiencies.

[0112] As can be seen, in this embodiment of the disclosure, teachers can customize "virtual classes". That is, by setting different proportions of the first type of virtual student intelligent agents, the second type of virtual student intelligent agents, and the third type of virtual student intelligent agents, different sets of virtual student intelligent agents can be determined, or different sets of virtual student intelligent agents can be determined directly based on the "multimodal learning data and construction instruction information" input by the teacher. This meets the needs of teachers for differentiated teaching for different classes and differentiated teaching for different regions.

[0113] Step 303: Determine the cognitive assimilation information of the virtual student intelligent entity set for the course to be optimized, and generate a heatmap of student confusion based on the cognitive assimilation information.

[0114] In this embodiment of the disclosure, after obtaining the virtual student intelligent body set and the course to be optimized, the cognitive assimilation information of the virtual student intelligent body set on the course to be optimized can be determined.

[0115] In one possible implementation, the audio of the course to be optimized can be forcibly aligned to determine a millisecond-level absolute timeline. Alternatively, the ISE engine can be used to forcibly align the audio of the course to be optimized, obtaining millisecond-level subtitle timestamps to determine a millimeter-level absolute timeline. Furthermore, the virtual student agents within the virtual student agent set can be controlled to scan the course to be optimized one by one according to the millisecond-level absolute timeline to determine cognitive assimilation information.

[0116] In this embodiment, the virtual student agent can be controlled to scan the course content of the course to be optimized in slices according to the aforementioned determined millimeter-level absolute timeline, thereby determining the cognitive assimilation information of the virtual student agent regarding the course to be optimized. Cognitive assimilation information refers to the entire process of the virtual student agent connecting, integrating, and absorbing new knowledge and content with its existing knowledge structure during the learning process, primarily reflecting the student's comprehension of the course content and any cognitive obstacles.

[0117] In one possible implementation, after obtaining the set of virtual student intelligent agents and the course to be optimized, a two-way coupling analysis is conducted on the course to be optimized, combining the course teaching content with the information processing capabilities of the virtual student intelligent agents, to accurately simulate the cognitive assimilation basis and information acceptance level of the virtual student intelligent agents regarding the course to be optimized.

[0118] In the analysis process, each virtual student agent is taken as the smallest research unit, and the personalized cognitive assimilation information of each virtual student agent is quantitatively defined one by one. Then, based on the learning distribution characteristics of the virtual agent set, the personalized cognitive assimilation information of all students is weighted and summed, and finally the overall cognitive assimilation level of the virtual student agent set is obtained.

[0119] In this exemplary embodiment, the overall simulation system is supported by two-dimensional information comparison, where the two-dimensional information is information density flow and cognitive throughput.

[0120] In this exemplary embodiment, the Information Density Flow is based on the teaching text of the course to be optimized. It combines the results of reasoning and content extraction from the teaching text to filter and extract micro-learning factors such as the volume of new concepts, newly added knowledge points, and the number of logical reasoning steps associated with the course. These factors are then calculated through a multi-factor weighted composite calculation. Its core function is to quantitatively represent the scale of new concepts, knowledge content, and logical complexity carried by the teaching text per unit time.

[0121] In this example embodiment, the cognitive throughput rate is determined by a combination of learning state and cognitive boundary parameters, including the virtual student agent's stable classroom attention, the coverage of its knowledge recognition mask, and its knowledge reception boundaries. Its core function is to quantify the threshold of the virtual student agent's ability to process information per unit of time.

[0122] In this embodiment, the system compares the information density flow and cognitive throughput at each time slice on a millimeter-level absolute time axis, objectively measuring the cognitive fit between the virtual student agent and the course to be optimized. If the "information density flow" of the current time slice exceeds the agent's "cognitive throughput," for example, when a large number of concepts within the agent's knowledge blind spot emerge in a short period and lack common explanations, the comparison result is determined to be a "comprehension blockage" (i.e., the student is disconnected or cannot understand). The following section will use a single virtual student agent as an example to specifically explain the comparison logic of the two pieces of information.

[0123] In one possible implementation, the following operations can be performed on virtual student agents within a set of virtual student agents: Step a: Search the knowledge base within the virtual student agent to see if it contains new knowledge points; where new knowledge points are obtained by entity extraction of the current teaching content, and the current teaching content is the teaching content corresponding to the current time slice.

[0124] In this exemplary embodiment, the Agent's own knowledge base can be queried, which is restricted by the aforementioned "knowledge mask". If the knowledge falls within the mask range, the retrieval fails, indicating that the virtual student lacks the "prerequisite knowledge" to understand the concept.

[0125] Step b: If it is determined that the knowledge base inside the student agent does not retrieve the current teaching content, then determine the first teaching text; wherein, the first teaching text is the teaching text corresponding to the adjacent time slice of the current time slice.

[0126] In this exemplary embodiment, when it is determined that the Agent's internal retrieval fails, the Agent will determine the teaching text of the current time slice and the adjacent slices before and after it, which is the first teaching text.

[0127] Step c: Based on the first teaching text, determine whether the virtual student agent generates comprehension-impeding information.

[0128] In this exemplary embodiment, once the Agent determines the first teaching text, it can evaluate whether the course to be optimized effectively explains the new terminology based on the first teaching text, thereby determining whether there is any information that hinders understanding.

[0129] Optionally, it can be determined whether the first teaching text includes words from a preset vocabulary set; the preset vocabulary set is determined based on real-life analogies and example content; if it is determined that the first teaching text does not include words from the preset vocabulary set, it is determined that the virtual student agent generates comprehension impairment information.

[0130] In this exemplary embodiment, it can be determined that the first teaching text does not explain the new knowledge point based on the fact that the first teaching text does not include words from the preset word set, and considering that the knowledge base inside the virtual student agent does not retrieve the new knowledge point, it can be determined that the virtual student agent has generated comprehension impairment information.

[0131] It is evident that when the virtual student agent suffers from a "lack of prior knowledge" problem—that is, when the virtual student agent's internal knowledge base fails to retrieve new knowledge points, and the first teaching text lacks real-life analogies or concrete examples, such as the absence of words like "for example" or "like" in the first teaching text—in other words, the first teaching text suffers from a "lack of clear explanation" problem, it can be determined that the virtual student agent suffers from a "lack of prior knowledge" problem, and the first teaching text being taught suffers from a "clear explanation" problem. Thus, it can be determined that the virtual student agent experiences comprehension difficulties in the current time slice.

[0132] Optionally, the number of logical reasoning steps between adjacent sentences in the first teaching text can be determined; when the number of logical reasoning steps is determined to be greater than a first threshold, it is determined that the virtual student agent generates comprehension-impeding information.

[0133] In this exemplary embodiment, the number of logical reasoning steps is an indicator used to quantify the complexity of textual logical deduction. This indicator is objectively calculated primarily through the following two methods: The first method: Using syntactic analysis techniques, the sentences and paragraphs in the first lecture text are structurally deconstructed to accurately identify and count the number of conjunctions and correlation markers that carry logical relationships such as cause and effect, progression, contrast, condition, and parallelism. The number of conjunctions and correlation markers is used as the number of logical reasoning steps in explicit logic. In this way, the number of explicit logical conjunctions can reflect the basic reasoning steps.

[0134] The second method: The first lecture text is input into a preset large language model. The preset large language model extracts the reasoning links and deduction processes from the first lecture text. According to the rules of content progression, layered argumentation, and step-by-step deduction, the reasoning levels are divided and the link depth is counted. The level depth of the reasoning links is used as the number of logical reasoning steps of implicit logic.

[0135] In this exemplary embodiment, after obtaining the number of logical reasoning steps for explicit logic and implicit logic, a weighted sum can be performed on the two numbers to determine the total number of logical reasoning steps. In this way, the number of logical reasoning steps can be used to comprehensively measure the number of complete logical deduction steps from premises, conditions to conclusion in the first teaching text. A higher number of logical reasoning steps indicates a longer logical deduction chain and a more complex thought process structure in the text.

[0136] In this exemplary embodiment, it can be determined that the logical explanation of the first teaching text is insufficient based on the fact that the number of logical reasoning steps of adjacent sentences in the first teaching text is greater than a preset threshold. In addition, considering the situation that no new knowledge points were retrieved from the knowledge base inside the virtual student agent, it can be determined that the virtual student agent has generated comprehension blockage information.

[0137] It is evident that when the virtual student agent suffers from a "lack of prior knowledge" problem—that is, when the virtual student agent's internal knowledge base fails to retrieve new knowledge points—and the logical reasoning steps between adjacent sentences in the first teaching text exceed a first threshold, indicating excessive logical jumps between adjacent sentences in the first teaching text, the first teaching text suffers from a "lack of prior knowledge" problem, or "unclear explanation" problem. Thus, it can be determined that the virtual student agent suffers from a "lack of prior knowledge" problem, and the first teaching text being taught suffers from an "unclear explanation" problem. Therefore, it can be determined that the virtual student agent experiences comprehension difficulties in the current time slice. The first threshold can be determined based on actual implementation; this embodiment does not limit its specific value.

[0138] Optionally, the formula content within the first teaching text can be determined; if the variable description information that does not include formula content is determined within the first teaching text, then the virtual student agent is determined to have comprehension impairment information.

[0139] In this exemplary embodiment, based on the variable description information in the first teaching text that does not include formula content, it can be determined that the representation of the new knowledge point in the first teaching text is not complete and clear enough. In addition, considering the aforementioned situation that the knowledge base inside the virtual student agent does not retrieve the new knowledge point, it can be determined that the virtual student agent has comprehension obstacles.

[0140] It is evident that when the virtual student agent suffers from a "lack of prior knowledge" problem—meaning that the knowledge base within the virtual student agent does not retrieve any new knowledge points, and the first teaching text directly presents formulas without variable explanations, indicating that the first teaching text suffers from an "unclear explanation" problem (i.e., the representation is not fully or clearly explained)—it can be determined that the virtual student agent suffers from a "lack of prior knowledge" problem, and the first teaching text being taught suffers from an "unclear explanation" problem. Thus, it can be determined that the virtual student agent experiences comprehension hindrance information in the current time slice.

[0141] Step d: When it is determined that the virtual student agent generates information that hinders comprehension, the information that hinders comprehension is classified into cognitive assimilation information.

[0142] In one possible implementation, when it is determined that cognitive assimilation information includes information on comprehension blockade, causal attribution processing is initiated for the course to be optimized to determine the causal diagnosis result.

[0143] In this embodiment of the disclosure, when a “comprehension block” occurs at time t, the system does not simply mark the current time, but triggers the offline tracing engine to automatically backtrack to the second teaching text of the historical timeline (tn to t) based on the knowledge graph of the teaching syllabus of the course to be optimized, to investigate the root cause of the current block, that is, to perform root cause analysis, thereby obtaining the causal diagnosis result.

[0144] In this embodiment of the disclosure, the current moment corresponding to the virtual student agent generating comprehension impairment information can be determined. Then, the offline tracing engine is triggered to perform a backtracking scan on the second teaching text corresponding to the current moment on the millisecond-level absolute timeline based on the knowledge graph of the syllabus of the course to be optimized, to determine the causal diagnosis result.

[0145] Optionally, the preceding dependencies of the current teaching text can be queried in the knowledge graph. Then, keyword matching and semantic similarity calculation are performed on the preceding dependencies and the second teaching text to determine the processing result information. When the processing result information at the first historical moment meets the first preset condition, it is determined that the course to be optimized lacks key preparatory information. The first preset condition is that the keyword matching degree is lower than the second threshold and the semantic similarity is lower than the third threshold. Further, a first causal chain can be constructed based on the content of the course to be optimized corresponding to the current moment and the first historical moment.

[0146] For example, the system queries the knowledge graph for the "preceding dependent nodes" (e.g., "stack") of the currently blocked concept (e.g., "stack overflow"). Then, using keyword matching and semantic similarity calculation, it scans the historical text from tn to t, which is the second teaching text. If the frequency of related terms in the preceding dependent node is 0 or below a set threshold, it is determined as "missing relational groundwork," and a first causal chain is established: "Because 'stack' was not explained in the 5th minute, 'stack overflow' could not be understood in the 12th minute."

[0147] Optionally, if the processing result information of the second historical moment does not meet the first preset condition, the attention half-life of the virtual student agent is determined; if it is determined that the time difference between the second historical moment and the current moment exceeds the attention half-life, and the content of the course to be optimized within the time difference does not include the pre-dependent content, then a second causal chain is constructed based on the content of the course to be optimized corresponding to the current moment and the second historical moment.

[0148] For example, if the second teaching text mentions a prerequisite concept, but the time tn of the occurrence is more than the current time t of the virtual student agent's "attention half-life" parameter, such as a time difference of 20 minutes, and the virtual student agent's "attention half-life" parameter is 15 minutes, and no review or reminder is performed within this period of 20 minutes, the system determines that the knowledge point has been forgotten by the student, thus causing the current blockage, and then constructs a second causal chain.

[0149] Optionally, the formula content within the second teaching text at the third historical moment can be determined. If the amount of determined formula content exceeds a fourth threshold, a third causal chain is constructed based on the content of the course to be optimized corresponding to the current moment and the third historical moment.

[0150] For example, if the system scans the second teaching text and finds that during the period from tn to t, the second teaching text continuously uses "abstract formula derivation" that exceeds the threshold that the virtual student agent can tolerate, which does not conform to the style of analogy preferred by the third type of group with learning characteristics of having less knowledge reserves and relying on concrete cases for understanding, the system will determine that the accumulation of prior cognitive fatigue has led to the complete disconnection at the current moment, and thus construct a third causal chain based on this.

[0151] In this embodiment of the disclosure, a causal diagnostic result, including a causal chain, can be determined based on the root cause leading to the current blockage. The causal diagnostic result (also known as a knowledge gap causal chain) refers to a discrete data structure generated by the causal tracing engine, which records the directed graph edges from "cause nodes (historical moments)" to "result nodes (current moments of confusion)".

[0152] In this embodiment of the disclosure, the comprehension score corresponding to a moment on the millisecond-level absolute time axis can also be determined based on the duration and severity of the comprehension hindrance information included in the cognitive assimilation information. Optionally, in the cognitive assimilation simulation, if the agent assimilates information normally, the score remains high; if "comprehension hindrance" occurs, the score drops sharply according to the severity of the hindrance (e.g., the number of logical jumps), thereby determining the comprehension score corresponding to a moment on the millisecond-level absolute time axis. Then, based on the comprehension score and the millisecond-level absolute time axis, the comprehension curve of the virtual student agent is determined. The comprehension curve refers to a set of continuous line data for a single type of virtual student agent. The horizontal axis of the coordinate system of the comprehension curve is the "absolute time axis" (accurate to seconds / milliseconds), and the vertical axis is the "comprehension score" (0-100 points).

[0153] In one possible implementation, considering that the comprehension curve and causal diagnosis results data are too fragmented, once the causal diagnosis results and comprehension curves including the causal chain are obtained, a student confusion heatmap can be generated based on the causal diagnosis results and comprehension curves of the causal chain.

[0154] Optionally, a first comprehension curve for the virtual student intelligence set is determined, and then, based on the first comprehension curve, the comprehensive confusion information at each moment on the millisecond-level absolute time axis is determined. That is, at the same moment on the time axis, the system can perform a weighted average of the "comprehension curve" scores of all types of agents (such as the first group with sufficient knowledge reserves and long sustained attention duration; the second group with moderate knowledge reserves and short sustained attention duration; the third group with less knowledge reserves and comprehension process dependent on concrete examples; and the fourth group with a phased decline in attention) to calculate the comprehensive confusion information at that moment, thus obtaining the overall confusion index for the entire class.

[0155] Furthermore, the system determines the causal chain corresponding to each virtual student agent within the set of virtual student agents, identifies causal nodes, and determines the moments of heightened confusion corresponding to these causal nodes. It is evident that the system extracts "knowledge gap causal chain" data. If a certain moment is identified as the "culprit" (i.e., the cause node) for multiple subsequent moments of confusion, the system assigns that moment an additional "source-tracing penalty weight," causing the overall confusion index for the entire class at that moment to spike further.

[0156] In this way, the target confusion information at each moment on the millisecond-level absolute time axis can be determined by weighted summation based on the comprehensive confusion information at each moment and the source penalty weight corresponding to the confusion amplification moment. The source penalty weight is proportional to the number of subsequent comprehension-impeding events pointed to by that causal node. For example, if causal node A points to 5 subsequent comprehension-impeding events and causal node B points to 6, then the source penalty weight corresponding to causal node A is less than the source penalty weight corresponding to causal node B.

[0157] Furthermore, a heatmap mapping process can be performed on the target confusion information at each moment on the millisecond-level absolute time axis to determine the student confusion heatmap. This heatmap mapping process can utilize color mapping, mapping the final target confusion information to RGB color values. For example, low target confusion (i.e., smooth comprehension) is mapped to green, and high target confusion (i.e., severe comprehension difficulties) is mapped to dark red.

[0158] In this embodiment of the disclosure, to better illustrate the process of obtaining the causal chain of the comprehension change trajectory (comprehension curve) and the student confusion heatmap, please refer to... Figure 5As shown, the Holistic Comprehension Causal Tracer module in the system reconstructs the trajectory of virtual students' comprehension of course content based on the full set of teaching videos, in order to obtain causal chains and student confusion heatmaps.

[0159] exist Figure 5 In the middle, we can first focus on optimizing the full set of teaching videos for the course (i.e. Figure 5 The course videos in the document are fully transcribed to obtain the teaching content stream. Furthermore, the entire teaching video stream can be forcibly aligned using the ISE engine to obtain a millimeter-level absolute timeline (i.e.,...). Figure 5 (Millisecond-level timeline). Then, based on the virtual student agent, the teaching content flow, and the millisecond-level timeline, cognitive assimilation simulation, detection of comprehension obstacles, and causal attribution analysis can be performed to generate causal chains and student confusion heatmaps.

[0160] As can be seen, the input of the full-time comprehension causal tracing module is the full set of teaching videos and virtual student agents; the output is the "comprehension curve", the "knowledge gap causal chain", and the final product "student confusion heatmap" generated by the aggregation of the former two.

[0161] Step 304: Based on the student confusion heatmap and cognitive assimilation information, generate an optimization strategy to obtain the target course after optimization based on the optimization strategy.

[0162] In this embodiment of the disclosure, after obtaining the student confusion heatmap, it is not only convenient for users to understand the students' cognitive status regarding the optimized course, but also to determine the high confusion interval based on the student confusion heatmap and cognitive assimilation information.

[0163] Optionally, the first high-confusion zone can be determined based on the first area within the student confusion heatmap where the confusion index exceeds the fifth threshold. For example, scan the "overall confusion index" of the student confusion heatmap. If the confusion index exceeds the set warning line, i.e., the second threshold, for example, the second threshold is 80 / 100, it will be reflected as a dark red area on the heatmap, and this area will be automatically identified as the high-confusion zone.

[0164] Optionally, when the result node corresponding to the causal node is located within the first high confusion interval, and the confusion index of the second region does not exceed the fifth threshold but exceeds the sixth threshold, then the second high confusion interval is determined based on the second region.

[0165] For example, if the confusion index of a certain time period (such as the 3rd minute) does not exceed the fifth threshold but exceeds the sixth threshold, such as 50 / 100, the heatmap will be yellow. However, the causal chain data shows that this time period is the "root cause node" of multiple subsequent high confusion intervals, meaning the result node is located in the high confusion zone. For example, if the teaching content corresponding to this causal node did not clearly explain the "stack," it would cause students to experience disconnection or comprehension difficulties in three places within the high confusion zone. Therefore, the system will assign a high penalty weight to this causal node and forcibly mark the second area corresponding to this causal node as a "high-risk, high-confusion interval."

[0166] In this way, the high confusion interval can be determined based on the first high confusion interval and / or the second high confusion interval.

[0167] In this embodiment of the disclosure, after determining the high confusion interval, the third teaching text, the learner profile of the virtual student agent, and the target causal chain corresponding to the high confusion interval can be determined; then, the third teaching text, the learner profile of the virtual student agent, and the target causal chain are input into a preset large language model to determine optimization suggestion information, and challenge propositions are generated based on the optimization suggestion information.

[0168] In an exemplary embodiment, the input of the predefined large language model is a structured Prompt package containing three elements: the original teaching text corresponding to the defined "high confusion interval", the specific virtual student profile that is missing in the interval (such as "a third group with learning characteristics of having less knowledge reserves and relying on concrete cases for understanding"), and the "causal chain" corresponding to the interval, namely the target causal chain (such as "because the 'stack' was missing in the 3rd minute").

[0169] First, based on the above input, the pre-set large language model infers the root cause of the confusion and selects the most suitable strategy from the pre-set "teaching strategy library" (including: analogy mapping, prior knowledge supplementation, step breakdown, visualization assistance, etc.).

[0170] Subsequently, the pre-defined large language model generates specific "optimization suggestions" (i.e., guiding text on "how to change", such as: "It is recommended to add the concept of 'stack' in the 3rd minute and use a real-life analogy to explain it").

[0171] Furthermore, to prevent teachers from simply looking at the suggestions without actually practicing them, the system will further transform the aforementioned "optimization suggestions" into a specific, actionable "challenge proposition." For example, if the optimization suggestion is "use analogy to explain," then the generated challenge proposition would be: "[Analogy Challenge] In the next 60 seconds, please try to re-explain the difference between 'queue' and 'stack' to a third group of learners who have limited prior knowledge and rely on concrete examples for understanding, using the example of 'queueing to buy tickets at a restaurant.' Please begin recording."

[0172] In one possible implementation, the optimized course based on the challenge proposition can be received; then the optimized course can be retested based on the virtual student intelligence set to determine the second comprehension curve corresponding to the virtual student intelligence set; and when it is determined that the comparison result between the second comprehension curve and the first comprehension curve meets the second preset condition, the optimized course can be used as the target course.

[0173] As can be seen, teachers record patch segments or modify their lectures based on the optimization suggestions generated by the system, and submit them to the system for re-listening and testing. The system then re-selects all types of virtual student agents for full cognitive assimilation simulation, including the first group with sufficient knowledge reserves and long sustained attention spans; the third group with less knowledge reserves and reliance on concrete examples for comprehension; and the fourth group with a tendency for attention to decline in stages. Although the problem may initially be triggered by a particular group (such as the third group with less knowledge reserves and reliance on concrete examples for comprehension), the retesting process must ensure global verification to guarantee that optimizations to the third group (e.g., adding lengthy analogies) do not cause redundancy or attention decline in the first group (with sufficient knowledge reserves and long sustained attention). Furthermore, the system compares the "comprehension curves" of the old and new versions in the retest results. If, within the modified timeframe, the comprehension curve of the third group of agents—characterized by limited knowledge reserves and reliance on concrete examples for understanding—significantly rebounds (eliminating the previous precipitous drop), and the curves of other student types do not show a significant decline, the system determines that the "teaching challenge" has been passed. Teachers continuously repeat this iterative process until the class heatmap returns to a "smooth (green)" state.

[0174] In this embodiment of the disclosure, for a better explanation of the process of obtaining the challenging proposition, please refer to [link to relevant documentation]. Figure 6As shown, the Adaptive Teaching Challenge Generator module in the system can automatically generate targeted challenge propositions based on the causal chain generated from the source and the heat map of student confusion.

[0175] exist Figure 6 In the adaptive learning challenge generation module, the inputs are causal chains and student confusion heatmaps (i.e., Figure 6 The process involves using a heatmap of confusion to identify high-confusion regions, selecting an optimization strategy for each high-confusion region, generating optimization suggestions based on these strategies, and finally outputting a challenge proposition generated based on these optimization suggestions (i.e., a challenge proposition generated from the optimization suggestions). Figure 6 Teaching challenges in China.

[0176] In this embodiment of the disclosure, in order to better illustrate the optimization scheme for online courses, a specific example is described below.

[0177] In this exemplary embodiment, high school math teacher M directly begins explaining the "proof steps from k to k+1" to obtain a micro-lesson on "mathematical induction," thus completing the initial recording.

[0178] In this exemplary embodiment, high school math teacher M can upload a micro-lesson on "Mathematical Induction" to a platform that provides course optimization services, and then select "Lesson Preparation Mode" within that platform. When the general mode within "Lesson Preparation Mode" is selected, the system generates a standard class (e.g., a first group of 20% with sufficient knowledge reserves and long sustained attention duration + a second group of 60% with moderate knowledge reserves and short sustained attention duration + a third group of 20% with less knowledge reserves and comprehension relying on concrete examples). When the customized mode within "Lesson Preparation Mode" is selected, teacher M uploads multimodal learning data and construction instructions, thereby generating a specific class.

[0179] In this exemplary embodiment, the system can extract millisecond-level timestamps based on the "micro-lesson on mathematical induction" to construct a precise teaching timeline (i.e., the aforementioned millisecond-level absolute timeline). Then, it performs virtual listening and causal diagnosis on the virtual student agents within the class according to this teaching timeline. For example, the first group of agents, characterized by ample knowledge reserves and long periods of sustained focus, demonstrates good comprehension. However, the third group of agents, characterized by less knowledge reserves and reliance on concrete examples for understanding, becomes confused starting in the second minute, questioning, "Why can the derivation be made by assuming n=k?" The system then points out that the third group, characterized by less knowledge reserves and reliance on concrete examples for understanding, failed to grasp the principle of the "domino effect," leading to doubts about the core logic.

[0180] In this exemplary embodiment, a heatmap of student confusion can be generated based on the causal chain included in the causal diagnosis results, highlighting in red the section on "the collective confusion of the third group, characterized by limited knowledge and reliance on concrete examples for understanding." Based on this, the system can generate optimization suggestions, such as displaying a pop-up on the platform suggesting "Add an animated demonstration and analogy of 'dominoes' at the beginning." Additionally, a pop-up on the platform can display: "[Teaching Challenge] In the next 30 minutes, please try to explain the example of 'dominoes' to the third group, characterized by limited knowledge and reliance on concrete examples for understanding. Please start your recording."

[0181] In this exemplary embodiment, high school math teacher M can record patch segments based on system-generated optimization suggestions and submit them to the system for further "lesson preparation." The system obtains the comprehension curve determined in this "lesson preparation." It can be determined that after adding analogies, the comprehension curve of the third group of agents—those with limited knowledge and whose understanding relies on concrete examples—significantly rebounds, while the comprehension curves of other agent groups do not show a significant decline. If this is the case, the system determines that the "teaching challenge" has been passed. Furthermore, when the student confusion heatmap indicates that all virtual student agents do not experience comprehension difficulties, the system determines that the "micro-lesson on 'mathematical induction'" has been completed.

[0182] In this embodiment, the abstract audience configuration is transformed into specific class learning data migration, enabling the system to better adapt to real teaching scenarios. This allows teachers to conduct lesson preparation and refinement based on specific learning situations (such as the distribution of incorrect answers in a test, weak knowledge points, etc.), significantly improving the relevance and practicality of course design. Simultaneously, by introducing prompt word engineering and knowledge masking mechanisms, the output of the large language model is reasonably constrained, allowing it to realistically simulate the cognitive state of students who have not yet understood relevant knowledge, rather than directly presenting complete knowledge. This more realistically reproduces the cognitive assimilation process of students in learning, providing more practical feedback for course optimization. Furthermore, leveraging the advantages of offline full-data analysis, the system can achieve in-depth teaching logic diagnosis. It can not only pinpoint students' current learning difficulties but also trace the underlying causes of insufficient prior knowledge, thereby enabling precise optimization of course content, explanation logic, and pacing, significantly improving the teaching adaptability and overall quality of online courses.

[0183] Exemplary device

[0184] Exemplary embodiments of this disclosure also provide an optimization apparatus for online courses. (See reference...) Figure 7 As shown, the online course optimization device 700 includes the following program units: Unit 701 is used to identify courses to be optimized. Configuration unit 702 is used to configure the set of virtual student intelligent agents for the course to be optimized; the set of virtual student intelligent agents includes multiple virtual student intelligent agents with different knowledge absorption ability parameters; Processing unit 703 is used to determine the cognitive assimilation information of the virtual student intelligent body to the course to be optimized, and generate a student confusion heatmap based on the cognitive assimilation information. The optimization unit 704 is used to generate an optimization strategy based on the student confusion heatmap and the cognitive assimilation information, so as to obtain a target course after optimizing the course to be optimized based on the optimization strategy.

[0185] In one possible implementation, the configuration unit 702 is configured to: Determine the multimodal learning data and construction instructions for the course to be optimized; The multimodal learning data is subjected to retrieval enhancement processing to determine the knowledge set of cognitive gaps and the knowledge set of high-frequency cognitive doubts; the knowledge set of cognitive gaps is subjected to mapping processing to determine the knowledge mask range; and the knowledge set of high-frequency cognitive doubts is subjected to mapping processing to determine the logical deviation rules. The constructed instruction information is subjected to feature parsing and dimension mapping to determine the learner feature profile; Based on the knowledge mask range, logical deviation rules, and learner feature profiles, a set of virtual student intelligent entities is obtained.

[0186] In one possible implementation, the processing unit 703 is configured to: Force alignment is performed on the audio of the course to be optimized to determine the absolute timeline at the millisecond level; The virtual student agents within the virtual student agent set are controlled to scan the course to be optimized one by one according to the millisecond-level absolute time axis to determine cognitive assimilation information.

[0187] In one possible implementation, the processing unit 703 is configured to: For the virtual student agents within the aforementioned set of virtual student agents, perform the following operations: The system retrieves whether the knowledge base within the virtual student agent contains new knowledge points; these new knowledge points are obtained by extracting entities from the current teaching content, where the current teaching content is the teaching content corresponding to the current time slice. If it is determined that the new knowledge point is not found in the knowledge base inside the student agent, then the first teaching text is determined, and the first teaching text is the teaching text corresponding to the adjacent time slice of the current time slice; Based on the first teaching text, determine whether the virtual student agent generates comprehension impairment information; When it is determined that the virtual student agent has the comprehension-impeding information, the comprehension-impeding information is classified into cognitive assimilation information.

[0188] In one possible implementation, the processing unit 703 is configured to: Determine whether the first teaching text includes words from a preset vocabulary set; the preset vocabulary set is determined based on real-life analogies and example content; When it is determined that the first teaching text does not include words from the preset word set, the virtual student agent is identified as having comprehension difficulties.

[0189] In one possible implementation, the processing unit 703 is configured to: Determine the number of logical reasoning steps for adjacent sentences within the first teaching text; When it is determined that the number of logical reasoning steps is greater than the first threshold, it is determined that the virtual student agent generates comprehension impairment information.

[0190] In one possible implementation, the processing unit 703 is configured to: Determine the formula content within the first teaching text; If it is determined that the first teaching text does not include variable introduction information containing the formula, then it is determined that the virtual student agent has generated comprehension impairment information.

[0191] In one possible implementation, the processing unit 703 is configured to: When it is determined that the cognitive assimilation information includes comprehension impairment information, causal tracing processing is initiated for the course to be optimized to determine the causal diagnosis result.

[0192] In one possible implementation, the processing unit 703 is configured to: Determine the current moment when the virtual student agent generates the information about comprehension blockage; The offline tracing engine is triggered to perform a backtracking scan on the second teaching text corresponding to the current moment on the millisecond-level absolute time axis based on the knowledge graph of the teaching syllabus of the course to be optimized, in order to determine the causal diagnosis result.

[0193] In one possible implementation, the processing unit 703 is configured to: Query the prerequisite dependencies of the current teaching text in the knowledge graph; For the aforementioned prerequisite content and the second teaching text, keyword matching and semantic similarity calculation are performed to determine the processing result information; When the processing result information of the first historical moment meets the first preset condition, it is determined that the course to be optimized lacks key preparatory information; the first preset condition is that the keyword matching degree is lower than the second threshold and the semantic similarity is lower than the third threshold; Based on the content of the course to be optimized corresponding to the current moment and the first historical moment, a first causal chain is constructed.

[0194] In one possible implementation, the processing unit 703 is configured to: If the processing result information of the second historical moment does not meet the first preset condition, determine the attention half-life of the virtual student agent; When it is determined that the time difference between the second historical moment and the current moment exceeds the attention half-life, and the content of the course to be optimized within the time difference does not include the preceding dependent content, then a second causal chain is constructed based on the content of the course to be optimized corresponding to the current moment and the second historical moment.

[0195] In one possible implementation, the processing unit 703 is configured to: Determine the formula content within the second lecture text at the third historical moment; When it is determined that the quantity of information in the formula exceeds the fourth threshold, a third causal chain is constructed based on the content of the course to be optimized corresponding to the current time and the third historical time.

[0196] In one possible implementation, the processing unit 703 is further configured to: Based on the duration and severity of the comprehension impairment information included in the cognitive assimilation information, the comprehension score corresponding to the moment on the millisecond-level absolute time axis is determined. The comprehension curve of the virtual student agent is determined based on the comprehension score and the millisecond-level absolute time axis.

[0197] In one possible implementation, the optimization unit 704 is configured to: Determine the first comprehension curve of the virtual student intelligence set; Based on the first comprehension curve of the virtual student intelligence set, determine the comprehensive confusion information at each moment on the millisecond-level absolute time axis; Determine the causal chain corresponding to the virtual student agents within the set of virtual student agents, determine the causal nodes, and determine the confusion enhancement moments corresponding to the causal nodes; Based on the comprehensive confusion information at each moment on the millisecond-level absolute time axis and the source-tracing penalty weight corresponding to the confusion enhancement moment, the target confusion information at each moment on the millisecond-level absolute time axis is determined. Heatmap mapping processing is performed on the target confusion information at each moment on the millisecond-level absolute time axis to determine the student confusion heatmap.

[0198] In one possible implementation, the optimization unit 704 is configured to: Based on the student confusion heatmap and the cognitive assimilation information, the high confusion interval is determined; Determine the third instruction text corresponding to the high confusion interval, the learner profile of the virtual student agent, and the target causal chain; The third teaching text, the learner profile of the virtual student agent, and the target causal chain are input into a preset large language model to determine optimization suggestions, and challenge propositions are generated based on the optimization suggestions.

[0199] In one possible implementation, the optimization unit 704 is configured to: The first high confusion interval is determined based on the first region within the student confusion heatmap where the confusion index exceeds the fifth threshold. Identify the second region of the causal node within the student confusion heatmap; When the result node corresponding to the causal node is located in the first high confusion interval, and the confusion index of the second region does not exceed the fifth threshold but exceeds the sixth threshold, then the second high confusion interval is determined based on the second region. The high confusion interval is determined based on the first high confusion interval and / or the second high confusion interval.

[0200] In one possible implementation, the optimization unit 704 is configured to: Receive the optimized curriculum based on the aforementioned challenge proposition; The optimized course is retested based on the virtual student intelligence set to determine the second comprehension curve corresponding to the virtual student intelligence set. When it is determined that the comparison result between the second comprehension curve and the first comprehension curve meets the second preset condition, the optimized course is taken as the target course.

[0201] The specific details of each part of the above-mentioned device have been described in detail in the method section of the implementation plan. For any undisclosed details, please refer to the implementation plan of the method section, and therefore will not be repeated here.

[0202] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to exemplary embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0203] Exemplary program product

[0204] Exemplary embodiments of this disclosure also provide a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the aforementioned optimized method for online courses.

[0205] In one implementation, the computer program product can be a tangible product containing a computer program, such as a computer-readable storage medium storing the computer program. The readable storage medium can be a storage medium based on electrical, magnetic, optical, electromagnetic, infrared, or other signals, including but not limited to: random access memory (RAM), read-only memory (ROM), magnetic tape, floppy disk, flash memory, hard disk drive (HDD), solid-state drive (SSD), etc. For example, the computer program product can be implemented as a non-volatile storage medium storing a computer program, such as read-only memory, NAND flash memory, etc.

[0206] In one implementation, the computer program product can be an intangible product containing a computer program. For example, the computer program product can be implemented as a virtual digital product, such as an executable file, installation package, or other digital file storing the computer program.

[0207] Computer program code can be written in one or more programming languages. Examples of programming languages ​​include C, Java, and C++. Program code can execute entirely on the user's computing device, partially on the user's computing device, or as a standalone software package. It can also execute partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, such as a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via an internet connection provided by a mobile network operator).

[0208] Computer programs can be carried or transmitted via signals such as electricity, magnetism, light, electromagnetic fields, and infrared radiation. Electronic devices can convert the signals carrying computer programs into digital signals, thereby running the computer programs. When a computer program runs on an electronic device, its code is used to cause the electronic device to execute (more specifically, to be executed by the processor of the electronic device) the method steps of various exemplary embodiments of this disclosure, such as the above-described online course optimization method, which includes the following steps: Step 301: Determine the course to be optimized; Step 302: Configure a set of virtual student intelligent agents for the course to be optimized; the set of virtual student intelligent agents includes multiple virtual student intelligent agents with different knowledge absorption ability parameters; Step 303: Determine the cognitive assimilation information of the virtual student intelligent agents for the course to be optimized, and generate a student confusion heatmap based on the cognitive assimilation information; Step 304: Generate an optimization strategy based on the student confusion heatmap and the cognitive assimilation information to obtain a target course after optimization based on the optimization strategy.

[0209] By implementing the above method steps through a computer program, the course to be optimized can be determined, and then a set of virtual student intelligent agents for the course to be optimized can be configured. The set of virtual student intelligent agents includes multiple virtual student intelligent agents with different knowledge absorption capacity parameters. Therefore, in this embodiment of the disclosure, an intelligent class group, i.e., a set of virtual student intelligent agents, that matches the course to be optimized can be generated. This set of virtual student intelligent agents includes virtual student intelligent agents with poor course absorption capacity, virtual student intelligent agents with average course absorption capacity, and virtual student intelligent agents with high course absorption capacity. The number of these three types of intelligent agents can be determined based on the learning situation of the targeted class in the actual teaching of the course to be optimized. In this way, a set of virtual student intelligent agents that is precisely adapted to the course to be optimized can be generated.

[0210] In this embodiment of the disclosure, after determining the course to be optimized and the virtual student agent, the cognitive assimilation information of the virtual student agent set on the course to be optimized can be determined, and a student confusion heatmap can be generated based on the cognitive assimilation information; and, an optimization strategy can be generated based on the student confusion heatmap and the cognitive assimilation information to obtain a target course after optimization of the course to be optimized based on the optimization strategy.

[0211] It is evident that a heatmap of student confusion can be determined based on the cognitive assimilation information of a virtual student agent, and then the course to be optimized can be optimized based on this heatmap to obtain an optimized target course. In other words, in this embodiment, a virtual student agent can be used to pre-optimize the course before its release, thereby clearly identifying specific defects and causes in aspects such as course content, structure, interaction, and knowledge point design. This provides an explainable, traceable, and implementable basis for improving course instruction. Furthermore, it allows for the identification and correction of causes before problems arise, enhancing the interpretability of teaching problem analysis, overcoming the lag in feedback results, and achieving the goal of improving course quality and teaching effectiveness from the source.

[0212] Exemplary electronic devices

[0213] Exemplary embodiments of this disclosure also provide an electronic device. The electronic device may include a processor and a memory. The memory stores executable instructions for the processor, such as computer programs. The processor executes these executable instructions to perform the method steps of various exemplary embodiments of this disclosure.

[0214] The following is for reference. Figure 8 The electronic device is illustrated by way of a general-purpose computing device. It should be understood that... Figure 8 The electronic device 800 shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments disclosed herein.

[0215] like Figure 8 As shown, the electronic device 800 may include: a processor 810, a memory 820, a bus 830, an I / O (input / output) interface 840, and a network adapter 850.

[0216] The memory 820 may include volatile memory, such as RAM 821 and cache unit 822, and may also include non-volatile memory, such as ROM 823. The memory 820 may also include one or more program modules 824, including but not limited to: an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. For example, program module 824 may include the units described above.

[0217] The processor 810 may include one or more processing units, such as an AP (Application Processor), a modem processor, a GPU (Graphics Processing Unit), an ISP (Image Signal Processor), a controller, an encoder, a decoder, a DSP (Digital Signal Processor), a baseband processor, and / or an NPU (Neural-Network Processing Unit).

[0218] The processor 810 can be used to execute executable instructions stored in the memory 820, such as the above-mentioned online course optimization method, which includes the following steps: Step 301: Determine the course to be optimized; Step 302: Configure a set of virtual student intelligent agents for the course to be optimized; the set of virtual student intelligent agents includes multiple virtual student intelligent agents with different knowledge absorption ability parameters; Step 303: Determine the cognitive assimilation information of the virtual student intelligent agents for the course to be optimized, and generate a student confusion heatmap based on the cognitive assimilation information; Step 304: Generate an optimization strategy based on the student confusion heatmap and the cognitive assimilation information to obtain a target course after optimization based on the optimization strategy.

[0219] By executing the above method steps through processor 810, the course to be optimized can be determined, and then a set of virtual student intelligent agents for the course to be optimized can be configured. The set of virtual student intelligent agents includes multiple virtual student intelligent agents with different knowledge absorption capacity parameters. Therefore, in this embodiment of the disclosure, an intelligent class group, i.e., a set of virtual student intelligent agents, that matches the course to be optimized can be generated. This set of virtual student intelligent agents includes virtual student intelligent agents with poor course absorption capacity, virtual student intelligent agents with average course absorption capacity, and virtual student intelligent agents with high course absorption capacity. The number of these three types of intelligent agents can be determined based on the learning situation of the targeted class in the actual teaching of the course to be optimized. In this way, a set of virtual student intelligent agents that is precisely adapted to the course to be optimized can be generated.

[0220] In this embodiment of the disclosure, after determining the course to be optimized and the virtual student agent, the cognitive assimilation information of the virtual student agent set on the course to be optimized can be determined, and a student confusion heatmap can be generated based on the cognitive assimilation information; and, an optimization strategy can be generated based on the student confusion heatmap and the cognitive assimilation information to obtain a target course after optimization of the course to be optimized based on the optimization strategy.

[0221] It is evident that a heatmap of student confusion can be determined based on the cognitive assimilation information of a virtual student agent, and then the course to be optimized can be optimized based on this heatmap to obtain an optimized target course. In other words, in this embodiment, a virtual student agent can be used to pre-optimize the course before its release, thereby clearly identifying specific defects and causes in aspects such as course content, structure, interaction, and knowledge point design. This provides an explainable, traceable, and implementable basis for improving course instruction. Furthermore, it allows for the identification and correction of causes before problems arise, enhancing the interpretability of teaching problem analysis, overcoming the lag in feedback results, and achieving the goal of improving course quality and teaching effectiveness from the source.

[0222] Bus 830 is used to connect different components of electronic device 800 and may include data bus, address bus and control bus.

[0223] Electronic device 800 can communicate with one or more external devices 900 (such as keyboard, mouse, external controller, etc.) through I / O interface 840.

[0224] Electronic device 800 can communicate with one or more networks via network adapter 850. For example, network adapter 850 can provide mobile communication solutions such as 3G / 4G / 5G, or wireless communication solutions such as wireless LAN, Bluetooth, and near-field communication. Network adapter 850 can communicate with other modules of electronic device 800 via bus 830.

[0225] although Figure 8 As not shown in the diagram, other hardware and / or software modules may also be configured in the electronic device 800, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0226] As can be seen from the above, the technical solutions disclosed herein can be implemented as methods, apparatus, systems, computer program products, storage media, electronic devices, etc. Those skilled in the art will understand that various aspects of this disclosure can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or an implementation combining hardware and software aspects, which may be referred to as "circuit," "module," or "system," respectively.

[0227] It should be understood that this disclosure is not limited to the specific methods, steps, or structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. Those skilled in the art will readily conceive of other embodiments based on the specific implementations provided in this disclosure. Therefore, the specific implementations provided in this disclosure are merely exemplary, and the scope and spirit of this disclosure are indicated by the claims, and should cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary technical means in the art not disclosed in this disclosure.

Claims

1. A method for optimizing online courses, characterized in that, The method includes: Identify courses that need optimization; Configure the set of virtual student intelligent agents for the course to be optimized; the set of virtual student intelligent agents includes multiple virtual student intelligent agents with different knowledge absorption ability parameters; Determine the cognitive assimilation information of the virtual student intelligent body for the course to be optimized, and generate a heatmap of student confusion based on the cognitive assimilation information; Based on the student confusion heatmap and the cognitive assimilation information, an optimization strategy is generated to obtain a target course that optimizes the course to be optimized based on the optimization strategy.

2. The method according to claim 1, characterized in that, Configure the virtual student intelligent body set of the course to be optimized, including: Determine the multimodal learning data and construction instructions for the course to be optimized; The multimodal learning data is subjected to retrieval enhancement processing to determine the knowledge set of cognitive gaps and the knowledge set of high-frequency cognitive doubts; the knowledge set of cognitive gaps is subjected to mapping processing to determine the knowledge mask range; and the knowledge set of high-frequency cognitive doubts is subjected to mapping processing to determine the logical deviation rules. The constructed instruction information is subjected to feature parsing and dimension mapping to determine the learner feature profile; Based on the knowledge mask range, logical deviation rules, and learner feature profiles, a set of virtual student intelligent entities is obtained.

3. The method according to claim 1, characterized in that, Determining the cognitive assimilation information of the virtual student intelligence set for the course to be optimized includes: Force alignment is performed on the audio of the course to be optimized to determine the absolute timeline at the millisecond level; The virtual student agents within the virtual student agent set are controlled to scan the course to be optimized one by one according to the millisecond-level absolute time axis to determine cognitive assimilation information.

4. The method according to claim 3, characterized in that, The virtual student agents within the set of virtual student agents are controlled to scan the course to be optimized one by one according to the millisecond-level absolute time axis to determine cognitive assimilation information, including: For the virtual student agents within the aforementioned set of virtual student agents, perform the following operations: The system retrieves whether the knowledge base within the virtual student agent contains new knowledge points; these new knowledge points are obtained by extracting entities from the current teaching content, where the current teaching content is the teaching content corresponding to the current time slice. If it is determined that the new knowledge point is not found in the knowledge base inside the student agent, then the first teaching text is determined; the first teaching text is the teaching text corresponding to the adjacent time slice of the current time slice; Based on the first teaching text, determine whether the virtual student agent generates comprehension impairment information; When it is determined that the virtual student agent has the comprehension-impeding information, the comprehension-impeding information is classified into cognitive assimilation information.

5. The method according to claim 4, characterized in that, Based on the first instruction text, determine whether the virtual student agent generates comprehension impairment information, including: Determine whether the first teaching text includes words from a preset vocabulary set; the preset vocabulary set is determined based on real-life analogies and example content; When it is determined that the first teaching text does not include words from the preset word set, the virtual student agent is identified as having comprehension difficulties.

6. The method according to claim 4, characterized in that, Based on the first instruction text, determine whether the virtual student agent generates comprehension impairment information, including: Determine the number of logical reasoning steps for adjacent sentences within the first teaching text; When it is determined that the number of logical reasoning steps is greater than the first threshold, it is determined that the virtual student agent generates comprehension impairment information.

7. The method according to claim 4, characterized in that, Based on the teaching text, determine whether the virtual student agent generates comprehension impairment information, including: Determine the formula content within the first teaching text; If it is determined that the first teaching text does not include variable introduction information containing the formula, then it is determined that the virtual student agent has generated comprehension impairment information.

8. An optimization device for online courses, characterized in that, The device includes: Identify units to determine the courses to be optimized; A configuration unit is used to configure the set of virtual student intelligent agents for the course to be optimized; the set of virtual student intelligent agents includes multiple virtual student intelligent agents with different knowledge absorption ability parameters; The processing unit is used to determine the cognitive assimilation information of the virtual student intelligent body to the course to be optimized, and to generate a student confusion heatmap based on the cognitive assimilation information. An optimization unit is used to generate an optimization strategy based on the student confusion heatmap and the cognitive assimilation information, so as to obtain a target course after optimizing the course to be optimized based on the optimization strategy.

9. An electronic device, characterized in that, include: processor; Memory for storing the executable instructions of the processor; The processor is configured to perform the method of any one of claims 1-7 by executing the executable instructions.

10. A computer program product storing a computer program thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-7.

Images