Intelligent educational system based on search augmentation generation

Abstract

The application relates to the technical field of online education and discloses an intelligent education system based on retrieval enhancement generation, which comprises a user input module configured to send received input information and student identity information to a large language model module; the large language model module is configured to generate corresponding answers according to first target teaching content and inquiry information, and send the answers to a virtual digital human module; a retrieval enhancement generation module is configured to obtain course progress and cognitive level; according to the inquiry information, the cognitive level and the course progress, the first target teaching content is obtained by searching in a preset teaching resource library; and the first target teaching content is fed back to the large language model module; the virtual digital human module is configured to convert the answers into first voice data; and generate output instructions containing the first voice data to control the digital human instructor to output the first voice data. The system can generate more accurate and personalized teaching answers.

Description

Technical Field

[0001] This application relates to the field of online education technology, and for example to an intelligent education system based on retrieval-enhanced generation. Background Technology

[0002] With the popularization of online education, improving participation and teaching quality in remote learning has become a core challenge. Meanwhile, with the rapid development of artificial intelligence technology, intelligent educational assistants have shown great potential in areas such as learning guidance and automated Q&A, gradually becoming a research hotspot in the field of online education technology. Existing intelligent educational systems typically employ conversational interaction based on Large Language Models (LLMs), which can understand students' questions to a certain extent and generate corresponding answers, providing immediate learning support.

[0003] However, in practical applications, the relevant technologies have the following drawbacks: First, traditional online education systems, upon receiving student inquiries, often directly call a general-purpose large language model to generate answers. Since the training data for this model comes from a vast amount of internet text, its knowledge system is complex and lacks specificity. This makes the model prone to "factual hallucination," generating information that contradicts the course content, deviates from the syllabus, or is even completely erroneous. Second, traditional online education systems typically answer all students' questions in a uniform manner. However, for the same subject, different students may be at different stages of understanding (e.g., beginner, consolidation, review). The answers provided may be too profound, exceeding the student's current comprehension; or too superficial, failing to meet the student's need for in-depth exploration. Therefore, they cannot provide personalized learning support for students.

[0004] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This summary is not intended as a general commentary, nor is it intended to identify key / important components or describe the scope of protection of these embodiments, but rather as a prelude to the detailed description that follows.

[0006] This disclosure provides an intelligent education system based on retrieval-enhanced generation, which can generate more accurate and personalized teaching answers.

[0007] In some embodiments, the intelligent education system based on retrieval enhancement generation includes: a user input module configured to, upon receiving input information from a student, obtain the student's identity information and send the input information and the identity information to a large language model module; the large language model module configured to, when the input information is a query, send the identity information, the query information, a preset subject corresponding to the query information, and a preset knowledge point corresponding to the query information to a retrieval enhancement generation module, triggering the retrieval enhancement generation module to provide feedback on first target teaching content; generate an answer corresponding to the query information based on the first target teaching content and the query information; and send the answer corresponding to the query information to a virtual digital human module; the query information is used to represent a question about preset knowledge points of a preset subject; the retrieval enhancement generation... The module is configured to receive the identity information, the query information, the preset subject corresponding to the query information, and the preset knowledge point corresponding to the query information sent by the large language model module; obtain the student's course progress regarding the preset subject based on the identity information and the preset subject corresponding to the query information; obtain the student's cognitive level regarding the preset knowledge point based on the identity information and the preset knowledge point corresponding to the query information; search in a preset teaching resource library based on the query information, the cognitive level, and the course progress to obtain a first target teaching content; and feed back the first target teaching content to the large language model module; the virtual digital human module is configured to convert the answer corresponding to the query information into first voice data; and generate an output instruction containing the first voice data to control the digital human tutor to output the first voice data.

[0008] The intelligent education system based on retrieval-enhanced generation provided in this disclosure can achieve the following technical effects: When the large language model module receives a student's inquiry about a preset subject, the retrieval-enhanced generation module obtains the student's cognitive level and course progress regarding the preset subject. Based on the inquiry information, cognitive level, and course progress, it searches the teaching resource database. The large language model module then integrates and generates the answer based on the first target teaching content, which has been filtered and fed back from the retrieval-enhanced generation module, combined with the original inquiry information. In this way, since the source material for the answer, i.e., the first target teaching content, is obtained from the teaching resource database using retrieval-enhanced generation technology, the possibility of the model fabricating knowledge points that contradict the teaching syllabus is reduced, greatly reducing the generation of "factual illusions." Furthermore, by using the student's cognitive level and course progress regarding the subject as necessary conditions for retrieval, the obtained answer is more tailored to the student. Therefore, the large language model can ultimately generate more accurate and personalized teaching answers.

[0009] The above general description and the description below are exemplary and illustrative only and are not intended to limit this application. Attached Figure Description

[0010] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations and drawings do not constitute a limitation on the embodiments. Elements having the same reference numerals in the drawings are shown as similar elements. The drawings are not to be scaled. And wherein: Figure 1 This is a schematic diagram of the structure of an intelligent education system with enhanced retrieval provided in an embodiment of this disclosure. Detailed Implementation

[0011] To provide a more detailed understanding of the features and technical content of the embodiments of this disclosure, the implementation of the embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. The accompanying drawings are for illustrative purposes only and are not intended to limit the embodiments of this disclosure. In the following technical description, for ease of explanation, several details are used to provide a full understanding of the disclosed embodiments. However, one or more embodiments may still be implemented without these details. In other cases, well-known structures and devices may be simplified in their depiction to simplify the drawings.

[0012] 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 or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this disclosure described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0013] Unless otherwise stated, the term "multiple" means two or more.

[0014] In this embodiment of the disclosure, the character " / " indicates that the objects before and after it are in an "or" relationship. For example, A / B means: A or B.

[0015] The term "and / or" describes an association between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or A and B.

[0016] The term "correspondence" can refer to an association or binding relationship. The correspondence between A and B means that there is an association or binding relationship between A and B.

[0017] This disclosure provides an intelligent education system based on retrieval-augmented generation. It deeply integrates the generation capabilities of a large language model with a teaching resource database, utilizing retrieval-augmented generation (RAG) technology to ensure that the answers output by the large language model conform to the teaching syllabus requirements, eliminating model illusion. This ensures accurate alignment between the answers and the students' learned knowledge. Furthermore, by using the students' cognitive level regarding the subjects involved in the consultation questions and their course progress as necessary conditions for retrieval, the final answers generated by the large language model are more tailored to the students and possess personalization.

[0018] The intelligent education system based on retrieval enhancement provided in this disclosure is applicable to online education platforms, smart classroom systems, lightweight remote teaching software, virtual simulation teaching environments, etc.

[0019] Combination Figure 1 As shown, this disclosure provides an intelligent education system, including: a user input module 101, a large language model module 102, a retrieval enhancement generation module 103, and a virtual digital human module 104.

[0020] The user input module 101 is configured to obtain the student's identity information upon receiving input from the student. It then sends the input information and the identity information to the large language model module. The input information can be text or voice. The identity information includes the student's student ID, user ID, mobile phone number, or other identifiers that uniquely identify the student. In one embodiment, the student logs in by entering an account (e.g., student ID, user ID, mobile phone number) and password. After successful system verification, a token containing the identity information is generated. The user input module then parses this token to obtain the identity information.

[0021] The large language model module 102 is configured to receive input information and identity information sent by the user input module. When the input information is a query, it sends the identity information, the query information, the corresponding preset subject, and the corresponding preset knowledge point to the retrieval enhancement generation module, triggering the retrieval enhancement generation module to provide the first target teaching content. Based on the first target teaching content and the query information, it generates the answer corresponding to the query information and sends it to the virtual digital human module. The student's query information represents the question the student is asking about a preset knowledge point within a preset subject.

[0022] In one embodiment, the large language model module, upon receiving input information from the user input module, performs intent recognition on the input information. If the input information is identified as a question inquiring about a preset subject (i.e., an inquiry), the module sends the inquiry information, the student's identification information, and the corresponding preset subject to the retrieval enhancement generation module. The preset subjects include Chinese, mathematics, English, physics, chemistry, history, geography, biology, or archaeology. For example, if the student's input is "What are the dating characteristics of stone carvings?" the large language model module recognizes that this input is a question inquiring about the dating characteristics of archaeological stone carvings, i.e., the input is an inquiry, the corresponding preset subject is "archaeology," and the corresponding preset knowledge point is "dating characteristics of stone carvings."

[0023] The retrieval enhancement generation module 103 is configured to receive identity information, query information, a preset subject corresponding to the query information, and a preset knowledge point corresponding to the query information sent by the large language model module. Based on the identity information and the preset subject corresponding to the query information, it obtains the student's course progress regarding the preset subject. Based on the identity information and the preset knowledge point corresponding to the query information, it obtains the student's cognitive level regarding the preset knowledge point. Based on the query information, cognitive level, and course progress, it searches in a preset teaching resource database to obtain the first target teaching content. The first target teaching content is then fed back to the large language model module.

[0024] The virtual digital human module 104 is configured to receive the answer corresponding to the query information sent by the large language model module, and use text-to-speech (TTS) technology to convert the answer corresponding to the query information into first speech data. It then generates an output instruction containing the first speech data to control the digital human tutor to output the first speech data.

[0025] The intelligent education system based on retrieval-enhanced generation provided in this embodiment, when the large language model module receives a student's inquiry about a preset subject, obtains the student's cognitive level and course progress regarding that subject through the retrieval-enhanced generation module. Based on the inquiry information, cognitive level, and course progress, it searches the teaching resource database. The large language model module then integrates and generates the answer based on the filtered first target teaching content returned by the retrieval-enhanced generation module, combined with the original inquiry information. In this way, because the source material for the answer, i.e., the first target teaching content, is obtained from the teaching resource database using retrieval-enhanced generation technology, the possibility of the model fabricating knowledge points that contradict the teaching syllabus is reduced, greatly minimizing the occurrence of "factual illusions." Furthermore, by using the student's cognitive level and course progress regarding the subject as necessary conditions for retrieval, the obtained answer is more tailored to the student. Therefore, the large language model can ultimately generate more accurate and personalized teaching answers.

[0026] Optionally, the retrieval enhancement generation module is configured to obtain the student's course progress for a preset subject based on the preset subject corresponding to the identity information and query information in the following way: obtain the student's course progress for the preset subject from a preset unified context based on the preset subject corresponding to the identity information and query information.

[0027] Optionally, the retrieval enhancement generation module is configured to obtain the student's cognitive level about preset knowledge points based on the preset knowledge points corresponding to the identity information and query information in the following way: obtain the student's cognitive level about preset knowledge points from a preset unified context based on the preset knowledge points corresponding to the identity information and query information.

[0028] The unified context is a data set stored in a pre-defined Model Context Protocol (MCP) server. The unified context includes at least the correspondence between student identification information and state parameters. The state parameters include course progress across multiple subjects and cognitive levels of multiple knowledge points. In some embodiments, a student's cognitive level regarding knowledge points includes four levels, from lowest to highest: Level 1 (L1), Level 2 (L2), Level 3 (L3), and Level 4 (L4). Level 1 (L1) is the memorization level, representing the ability to recognize basic terminology related to knowledge points and recite core definitions. Level 2 (L2) is the application level, representing the ability to use formulas or principles in familiar contexts. Level 3 (L3) is the analysis level, representing the ability to decompose complex problems and associate multiple knowledge points. Level 4 (L4) is the evaluation and innovation level, representing the ability to judge the merits of solutions and design according to requirements.

[0029] A student's course progress in a given subject is the number of chapters they have already studied in that subject. For example, if a student has already studied Chapter 3, Section 2 of the first semester of the third year of junior high school chemistry, then their course progress in chemistry is Chapter 3, Section 2 of the first semester of the third year of junior high school.

[0030] In one embodiment, the intelligent education system based on retrieval enhancement further includes a Model Context Protocol (MCP) server, which is configured to store and update a preset unified context. This unified context is organized using the Model Context Protocol (MCP) data structure.

[0031] The retrieval enhancement generation module performs a search operation within a predefined unified context based on the student's identity information and the preset subject. It finds the course progress that corresponds to both the identity information and the preset subject, and identifies this course progress as the student's course progress for that subject. If no course progress corresponding to both the identity information and the preset knowledge point is found in the unified context, the student's course progress for the preset subject is determined to be Chapter 1, Section 1. Here, the preset subject can be one of multiple subjects.

[0032] The retrieval enhancement generation module performs a search operation within a pre-defined unified context based on the student's identity information and pre-defined knowledge points, finding the knowledge level that corresponds to both the identity information and the pre-defined knowledge points. This found knowledge level is determined as the student's cognitive level regarding the pre-defined knowledge points. If no knowledge level corresponding to both the identity information and the pre-defined knowledge points is found within the unified context, the student's cognitive level regarding the pre-defined knowledge points is determined to be Level 1 (L1).

[0033] In one embodiment, the intelligent education system based on retrieval enhancement further includes: a course progress recording module, configured to record the number of chapters of the learned courses, establish a correspondence between the student's identity information, the corresponding subject and the number of chapters of the learned courses, and send the correspondence to the MCP server, triggering the MCP server to store the correspondence and update it in the unified context.

[0034] In one embodiment, the intelligent education system based on retrieval enhancement further includes: a cognitive level acquisition module, configured to perform a lookup operation on a preset first data table using the student's test results for each knowledge point, and determine the found cognitive level as the cognitive level of the corresponding knowledge point. The first data table stores the correspondence between test results and cognitive levels. A correspondence is established between the cognitive level of the knowledge point and the student's identity information and the corresponding subject, and this correspondence is sent to the MCP server, triggering the MCP server to store the correspondence and update it in the unified context.

[0035] Optionally, the cognitive level acquisition module is further configured to perform a lookup operation on a preset first data table to determine the cognitive level of the corresponding knowledge point based on the student's accuracy rate in answering questions about each knowledge point. The first data table stores the correspondence between answer accuracy rate and cognitive level.

[0036] Optionally, the cognitive level acquisition module is further configured to perform a lookup operation on a preset first data table based on the frequency of students' queries for each knowledge point, and determine the found cognitive level as the cognitive level for the corresponding knowledge point. The first data table stores the correspondence between query frequency and cognitive level.

[0037] For example, in the virtual field trip course on the Dazu Rock Carvings, the cognitive level acquisition module of the intelligent education system detected that when a student consulted the basic terminology table more than a preset number of times to answer the question "What are the dating characteristics of the rock carvings?", the student's cognitive level on the knowledge point of "dating characteristics of the rock carvings" was determined to be L1 (memorization level), and this quantitative result was immediately written into the unified context of the MCP server.

[0038] In some embodiments, the pre-defined teaching resource base includes textbooks, syllabi, question banks, etc. The teaching resource base is structured Educational Atomic Units (EAUs). Each EAU includes: a Unique Identifier (UID) for each knowledge point; a Document Chunk for each knowledge point; and Knowledge Graph Tags for each knowledge point. The Document Chunk is plain text and / or a formula, representing the smallest unit of knowledge extracted from the teaching resource base. The Knowledge Graph Tags for each knowledge point include a parent node and prerequisite knowledge points. The parent node represents the chapter and / or subject to which the knowledge point belongs; prerequisite knowledge points are other knowledge points that must be mastered before learning the knowledge point. For example, before learning the knowledge point "lever balance," one must master the knowledge point "force balance."

[0039] Optionally, the retrieval enhancement generation module is configured to retrieve the primary target teaching content from a pre-defined teaching resource database based on the query information, cognitive level, and course progress by: obtaining the target score for each knowledge unit segment in the teaching resource database based on the query information, cognitive level, and course progress; sorting the knowledge unit segments in descending order of their target scores; and determining the top pre-defined number of knowledge unit segments as the primary target teaching content.

[0040] In this way, by using a multi-dimensional evaluation function based on inquiry information, cognitive level, and course progress, and employing retrieval-enhanced generative techniques, knowledge unit fragments that both match the student's current "cognitive level" and align with the "course progress" are selected from a massive structured teaching resource database—this is the primary target teaching content. Inputting this primary target teaching content along with the student's inquiry information into the large language model reduces the possibility of the model fabricating knowledge points that contradict the syllabus. This enables the large language model to generate more accurate and rigorously syllabus-compliant adaptive learning responses.

[0041] Optionally, based on the query information, the cognitive level, and the course progress, the target score for each knowledge unit segment in the teaching resource database is obtained, including: obtaining the semantic similarity score between the query information and each knowledge unit segment; obtaining the cognitive matching score between the student's cognitive level regarding the preset knowledge points and each knowledge unit segment; obtaining the progress alignment score between the student's course progress regarding the preset subject and each knowledge unit segment; and obtaining the target score for each knowledge unit segment based on the semantic similarity score, the cognitive matching score, and the progress alignment score.

[0042] Furthermore, the semantic similarity scores between the query information and each knowledge unit fragment are obtained, including: using a pre-trained semantic model to vectorize the query information and generate corresponding query feature vectors; using the pre-trained semantic model to vectorize each knowledge unit fragment and generate corresponding knowledge unit fragment feature vectors; and calculating the cosine distance between the query feature vector and the feature vectors of each knowledge unit fragment, determining each cosine distance as the semantic similarity score between the query information and the corresponding knowledge unit fragment. Each semantic similarity score is greater than or equal to 0 and less than or equal to 1.

[0043] Furthermore, the cognitive matching scores of students regarding preset knowledge points and each knowledge unit segment are obtained, including: performing a lookup operation on a preset second data table to find the cognitive level score corresponding to the cognitive level; performing a lookup operation on each knowledge unit segment in the preset second data table to find the difficulty score corresponding to each knowledge unit segment. The preset second data table stores the cognitive level score and the difficulty score corresponding to each knowledge unit segment. The absolute value of the difference between each difficulty score and the cognitive level score is calculated. The absolute value of each difference is then looked up in a preset third data table to obtain the difference score corresponding to each absolute value. Each difference score is determined as the cognitive matching score between the cognitive level and the corresponding knowledge unit segment. The third data table stores the difference score corresponding to each absolute value of the difference. The smaller the absolute value of the difference, the higher the difference score. Each cognitive matching score is greater than or equal to 0 and less than or equal to 1.

[0044] Furthermore, the system obtains the alignment scores between the student's course progress for a preset subject and the progress of each knowledge unit segment. This includes: determining whether the chapter to which each knowledge unit segment belongs is a chapter the student has already studied; if so, determining the chapter distance between the chapter to which the knowledge unit segment belongs and the course progress; performing a lookup operation on the chapter distance in a preset fourth data table to find the corresponding distance score; and determining this distance score as the alignment score between the course progress and the corresponding knowledge unit segment; otherwise, determining the alignment score between the course progress and the corresponding knowledge unit segment as 0. The fourth data table stores the distance scores corresponding to the chapter distances. For example, if the number of chapters to which a knowledge unit segment belongs is greater than the course progress, then the chapter to which the knowledge unit segment belongs is determined to be one that the student has not studied. If the number of chapters to which a knowledge unit segment belongs is equal to the course progress, then the alignment score between the course progress and the knowledge unit segment is determined to be 1. Each alignment score is greater than or equal to 0 and less than or equal to 1.

[0045] Optionally, the target score of each knowledge unit segment is obtained based on each semantic similarity score, each cognitive matching score, and each progress alignment score, including: using a preset algorithm to calculate each semantic similarity score, each cognitive matching score, and each progress alignment score to obtain the target score of each knowledge unit segment.

[0046] Furthermore, through calculation Obtain the target score for each knowledge unit segment. Among them, For the first d The target score for each knowledge unit segment For the first d The semantic similarity score corresponding to each knowledge unit fragment. For the first d Progress alignment score corresponding to each knowledge unit segment. For the first d The cognitive matching score corresponding to each knowledge unit fragment. For semantic similarity weights, For progress alignment weights, For cognitive matching weights, and .

[0047] Optionally, the intelligent education system based on retrieval enhancement also includes a response module. The large language model module is further configured to perform intent recognition on the student's input information, and if the student's input information is recognized as an operation command, generate a corresponding function call command based on the operation command; and send the function call command to the response module.

[0048] The response module is configured to receive function call instructions sent by the large language model module, determine the corresponding target teaching tool based on the function call instructions, call the target teaching tool to execute the operation instructions, obtain the execution results, and send the execution results to the preset Model Context Protocol (MCP) server.

[0049] In one embodiment, the student's input is "Help me adjust the balance to a level position." The large language model module performs intent recognition on this input, identifying it as an operation instruction to "adjust the experimental balance to a level position." The large language model module then generates a corresponding function call instruction and sends it to the response module. The response module invokes the corresponding experimental balance according to the received function call instruction, executes the operation instruction to "adjust the experimental balance to a level position," and sends the execution result, "The experimental balance has been adjusted to a level position," to the MCP server. This triggers the MCP server to store the received execution result in the unified context.

[0050] The unified context is a data set stored on the MCP server, accessible to all functional modules. This data set includes at least three types of information. The first type of context data characterizes student status parameters, including one or more of the following: course progress, cognitive level, execution results of operational instructions, teaching status, behavior logs, and identification information of teaching interest areas. The second type of context data defines information about external tools that can be invoked, including tool names, interface parameters, and invocation methods. The third type of context data provides system role information to guide model behavior, including one or more of the following: teaching strategies, response rules, and personality settings. All functional modules, such as the virtual digital human module, the large language model module, and the response module, collaborate by accessing the same context data. Updates to the context data by any functional module are synchronized to other functional modules in real time, ensuring that all functional modules make decisions and respond based on a consistent global state.

[0051] The MCP server is configured to store the received execution results in a unified context, retrieve the teaching status based on the execution results, and send the teaching status to the virtual digital human module. The MCP server is also configured to store the latest retrieved teaching status in the unified context for updating the teaching status.

[0052] Furthermore, the teaching status is obtained based on the execution result, including: performing a lookup operation in a pre-defined fifth data table to find the teaching status corresponding to the execution result. The fifth data table stores the correspondence between execution results and teaching statuses.

[0053] The virtual digital human module is configured to receive the teaching status sent by the MCP server, obtain the corresponding guidance text according to the teaching status, convert the guidance text into second voice data, and generate an output instruction containing the second voice data to control the digital human tutor to output the second voice data.

[0054] Optionally, obtaining the corresponding guidance text based on the teaching status includes: performing a lookup operation in a preset sixth data table based on the teaching status to find the guidance text corresponding to the teaching status. The sixth data table stores the correspondence between teaching statuses and guidance texts.

[0055] In one embodiment, the teaching status includes four states: theoretical explanation state, operation preparation state, experiment in progress state, and result analysis state. For example, if the execution result received by the MCP server is "Balance scale zeroing completed," the MCP server stores this execution result in the unified context, specifically in the ContextRes.tool_status field of the corresponding MCP protocol. It then looks up the corresponding teaching status as "Experiment in Progress" in the fifth data table and stores this status in the unified context, specifically in the ContextRes.pedagogical_state field of the corresponding MCP protocol. The virtual digital human module listens to changes in the ContextRes.pedagogical_state field to find the corresponding guidance text based on the teaching status. For example, based on the teaching status "Experiment in Progress," the virtual digital human module looks up the corresponding guidance text "Experiment started, please observe the pointer changes" in the sixth data table. This guidance text is converted into second voice data; and an output instruction containing this second voice data is generated to control the digital human tutor to output the second voice data.

[0056] The intelligent education system based on retrieval-enhanced generation provided in this disclosure introduces the MCP protocol into the educational scenario, serving as the communication backbone for collaboration among various functional modules (such as the virtual digital human module, the large language model module, the response module, and external tools). This unifies information transmission formats and session state management. Furthermore, by introducing a Pedagogical State Machine (PSM) into the MCP protocol, it further ensures that the virtual digital human tutor's responses remain logically consistent with changes in the teaching scenario and the student's real-time state, thereby providing logical guidance and enabling real-time interaction between the virtual digital human tutor and teaching environment tools. This high responsiveness and logical consistency significantly enhances student immersion, making the interactive experience more natural and fluid. Simultaneously, state synchronization based on the MCP context ensures that even under network latency, each agent can still obtain a consistent teaching environment state through the MCP server, further avoiding state fragmentation in multi-agent systems.

[0057] Optionally, the intelligent education system based on retrieval enhancement further includes a visual positioning module. The visual positioning module is configured to obtain the student's current viewpoint coordinates in the panoramic teaching scene; if the current viewpoint coordinates belong to a preset teaching interest area, it obtains the identification information of that teaching interest area; and sends the identification information of the teaching interest area to the Model Context Protocol (MCP) server, triggering the MCP server to store the identification information of the teaching interest area in a unified context. Here, the panoramic teaching scene is the front-end interaction layer of the intelligent education system. The current viewpoint coordinates are the positioning coordinates of the mouse's current location when the student uses the mouse to manipulate the panoramic teaching scene in the intelligent education system.

[0058] In one embodiment, an interactive 360° panoramic teaching scene is rendered on the browser using Web technology, serving as the front-end interaction layer of the intelligent education system. This panoramic teaching scene allows students to zoom, drag, and explore. The intelligent education system automatically switches teaching scenes according to the teaching narrative flow. Each teaching scene establishes a three-dimensional logical coordinate system. Multiple teaching interest areas are pre-set within the panoramic teaching scene. Each teaching interest area includes multiple teaching points of interest (POIs), and each teaching interest area is assigned a unique identifier, Anchor_ID. A pre-set seventh data table stores the correspondence between the image coordinates and identifier information of the teaching interest points within each teaching interest area.

[0059] The MCP server is configured to receive the identification information of the teaching interest area, store the received identification information of the teaching interest area in the unified context, and send the identification information of the teaching interest area to the virtual digital human module.

[0060] The virtual digital human module is configured to receive identification information of teaching interest areas, obtain corresponding explanation content based on the identification information, convert the explanation content into third-party audio data, and generate an output command containing the third-party audio data to control the digital human instructor to output the third-party audio data. In this way, by using the MCP protocol to link the viewpoint trajectory with the digital human instructor's explanation content, the visual presentation and audio content are synchronized.

[0061] Optionally, the corresponding explanation content can be obtained based on the identification information of the teaching interest area, including: performing a table lookup operation in a preset eighth data table based on the identification information of the teaching interest area to find the explanation content corresponding to the identification information. The preset eighth data table stores the correspondence between identification information and explanation content.

[0062] In one embodiment, the visual positioning module determines that the current viewpoint coordinates belong to the teaching interest region corresponding to any teaching interest point if the current viewpoint coordinates match the image coordinates of that teaching interest point. It then performs a lookup operation in a first data table using the image coordinates of that teaching interest point to find the corresponding identification information of the teaching interest region. The current viewpoint coordinates and the corresponding identification information of the teaching interest region are then sent to the MCP server. The MCP server stores the received current viewpoint coordinates and the corresponding identification information of the teaching interest region in a unified context, specifically in the ContextRes.scene_status field of the corresponding MCP protocol. The virtual digital human module monitors changes in the ContextRes.scene_status field, retrieves the corresponding explanation content based on the identification information of the teaching interest region corresponding to the current viewpoint coordinates, and controls the digital human instructor to output the audio data of that explanation content. For example, in the virtual field trip course on the Dazu Rock Carvings, when a student's current viewpoint coordinates fall within the teaching interest area of ​​the "base inscription," the MCP server updates its unified context and sends the identification information corresponding to the teaching interest area of ​​the "base inscription" to the virtual digital human module. This triggers the virtual digital human module to retrieve the corresponding explanation content: "Now we see the base inscription, which reflects the information of the patrons at that time..." The virtual digital human module converts this explanation content into the corresponding third-party audio data and controls the digital human tutor to output this third-party audio data.

[0063] In this way, the student's viewpoint trajectory is synchronized to a unified context in real time via the MCP protocol, enabling digital human instructors to provide targeted explanations based on the student's observation position, achieving a deep coupling between panoramic browsing and teaching semantics. This achieves an immersive experience that can run in a regular browser without requiring high-performance VR hardware.

[0064] Optionally, the intelligent education system based on retrieval enhancement also includes: a behavior log recording module and an assessment module. The behavior log recording module is configured to record students' behavior logs during the teaching process and send the behavior logs to the assessment module.

[0065] In one embodiment, the behavior log includes test questions for students on preset knowledge points, the number of questions, the answering time for each question, and the answering status for each question, including whether the answer is correct or incorrect.

[0066] The behavior log includes student interaction and click data within the panoramic teaching scenario. This data includes the sequence of steps involved in operating virtual experimental equipment and the number of times students ask questions to the digital human tutor.

[0067] The behavior log also includes student interaction trajectory data and learning flow control data within the panoramic teaching scenario. Interaction trajectory data includes: viewpoint dwell hotspots, zoom operation frequency, etc. Learning flow control data includes: video replay counts, video speed adjustments, etc.

[0068] The behavior log recording module uses web front-end tracking technology to send students' behavior logs to the MCP server in real time, triggering the MCP server to store them in a unified context, that is, in the ContextRes field of the corresponding MCP protocol.

[0069] The assessment module is configured to obtain students' mastery of preset knowledge points based on behavior logs; and send the mastery of preset knowledge points to the retrieval enhancement generation module.

[0070] Optionally, the behavior log includes the student's test questions on preset knowledge points, the number of questions, and the student's answers to each question. Based on the behavior log, the student's mastery of the preset knowledge points is obtained, including: performing a lookup operation in a preset ninth data table based on the student's identification information and the preset knowledge points to find the corresponding ability parameters. The ninth data table stores the correspondence between the identification information, the knowledge points, and the ability parameters. The ability parameters characterize the student's comprehensive ability level in that knowledge point area. A higher ability parameter indicates a higher degree of mastery of the corresponding knowledge point.

[0071] Based on the test questions for each preset knowledge point, a lookup operation is performed in the preset tenth data table to find the difficulty parameter corresponding to each test question. The tenth data table stores the correspondence between test questions and difficulty parameters.

[0072] Students with ability parameters The log-likelihood function under the given conditions is: in, nThe number of questions. For students' ability parameters, For the first i The answer to the question; when the answer is correct... =1, when the answer is incorrect =0, The student's ability parameters are In the case of answering the correct question, i The probability of the question.

[0073] Among them, through calculation The student's ability parameters are obtained as follows In the case of answering the correct question, i The probability of the question. For the first i The difficulty parameter of the question.

[0074] Through calculation To achieve a certain level of mastery over the pre-defined knowledge points. This refers to students' mastery of the pre-defined knowledge points. It's the ability parameter that determines when the log-likelihood function reaches its maximum value. This is determined as the student's mastery of the pre-set knowledge point. .

[0075] In some embodiments, whenever a student completes a new test question, the system will recalculate the log-likelihood function based on the new answer and dynamically update the ability parameters, thereby reflecting the student's mastery of the knowledge points in real time.

[0076] Optionally, the behavior log includes a sequence of operation steps for the virtual experimental equipment. Based on the behavior log, the student's mastery of the preset knowledge points is obtained, including: comparing the student's sequence of experimental operation steps corresponding to the preset knowledge points with the preset standard operation sequence using the longest common subsequence (LCS) to obtain the sequence similarity, and taking the obtained sequence similarity as the mastery of the corresponding preset knowledge points.

[0077] The retrieval enhancement generation module is configured to search a preset teaching resource database based on the mastery level of preset knowledge points to obtain secondary target teaching content. This secondary target teaching content is then sent to the display module. The display module is configured to receive and display the secondary target teaching content sent by the retrieval enhancement generation module.

[0078] Furthermore, based on the mastery level of the preset knowledge points, a search is performed in a preset teaching resource database to obtain the second target teaching content. This includes: calculating the semantic similarity between the preset knowledge points and each knowledge unit segment, and identifying knowledge unit segments with semantic similarity greater than or equal to a set threshold as candidate knowledge unit segments. A lookup operation is performed in a preset eleventh data table to find the difficulty range corresponding to each candidate knowledge unit segment. The eleventh data table stores the correspondence between knowledge unit segments and difficulty ranges. The difficulty range to which the mastery level of the preset knowledge points belongs is determined as the target difficulty range, and the candidate knowledge unit segments corresponding to the target difficulty range are determined as the first target knowledge unit segments. Based on the graph tags, the parent node and predecessor knowledge points corresponding to the first target knowledge unit segments are obtained, and the knowledge unit segments corresponding to the parent nodes are determined as the second target knowledge unit segments; the knowledge unit segments corresponding to the predecessor knowledge points are determined as the third target knowledge unit segments. All three target knowledge unit segments are then identified as the second target teaching content.

[0079] In this way, students' mastery of knowledge points can be assessed based on their behavior logs, and personalized teaching content can be pushed according to the assessment results.

[0080] Optionally, the intelligent education system based on retrieval enhancement also includes a home visit record acquisition module. This module is configured to acquire students' home visit records and send them to the MCP server. The MCP server stores the received home visit records in a unified context and sends them to the large language model module. The large language model module generates a corresponding summary report based on the home visit records and sends the summary report to the virtual digital human module. The virtual digital human module updates the teaching behavior logic of the digital human tutor based on the received summary report.

[0081] In some embodiments, home visit records include: student's home learning status, parental observations and feedback, or special concerns. The home visit record acquisition module receives the home visit record input by the parent and writes it into the home visit summary field of Contextprompts via the MCP protocol. The large language model module generates a personalized summary report based on the home visit record, and the virtual digital human module updates the subsequent teaching behavior logic of the digital human tutor based on this summary report.

[0082] The technical solutions of this disclosure can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in this disclosure. The aforementioned storage medium can be a non-transitory storage medium, including: a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, and other media capable of storing program code; it can also be a transient storage medium.

[0083] The foregoing description and accompanying drawings fully illustrate embodiments of this disclosure to enable those skilled in the art to practice them. Other embodiments may include structural, logical, electrical, procedural, and other changes. The embodiments represent only possible variations. Individual components and functions are optional unless explicitly required, and the order of operation may vary. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. Moreover, the terminology used in this application is for describing embodiments only and is not intended to limit the claims. As used in the description of embodiments and claims, the singular forms “a,” “an,” and “the” are intended to equally include the plural forms unless the context clearly indicates otherwise. Similarly, the term “and / or” as used in this application means including one or more of the associated listed items and all possible combinations thereof. Additionally, when used in this application, the term "comprise" and its variations "comprises" and / or "comprising" refer to the presence of stated features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof. Without further limitations, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes said element. In this document, each embodiment may focus on the differences from other embodiments, and similar or identical parts between embodiments can be referred to mutually. For methods, products, etc., disclosed in the embodiments, if they correspond to the method section disclosed in the embodiments, the relevant parts can be referred to the description of the method section.

[0084] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of this disclosure. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0085] The block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of a system according to embodiments of this disclosure. In this regard, each block in the block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description; sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved. Each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

Claims

1. An intelligent education system based on retrieval-enhanced generation, characterized in that, include: The user input module is configured to obtain the student's identity information upon receiving the student's input information; The input information and the identity information are then sent to the large language model module. The large language model module is configured to send the identity information, the query information, the preset subject corresponding to the query information, and the preset knowledge point corresponding to the query information to the retrieval enhancement generation module when the input information is query information, thereby triggering the retrieval enhancement generation module to provide feedback on the first target teaching content. Based on the first target teaching content and the inquiry information, generate the answer corresponding to the inquiry information; And send the answers corresponding to the query information to the virtual digital human module; The query information is used to characterize the questions asked about preset knowledge points in a preset subject. The retrieval enhancement generation module is configured to receive the identity information, the query information, the preset subject corresponding to the query information, and the preset knowledge point corresponding to the query information sent by the large language model module; Based on the preset subject corresponding to the identity information and the query information, obtain the student's course progress for the preset subject; Based on the identity information and the preset knowledge points corresponding to the inquiry information, the student's cognitive level regarding the preset knowledge points is obtained; based on the inquiry information, the cognitive level, and the course progress, a search is performed in a preset teaching resource database to obtain the first target teaching content; The first target teaching content is then fed back to the large language model module. The virtual digital human module is configured to convert the answer corresponding to the query information into first voice data; It generates an output instruction containing the first voice data to control the digital human tutor to output the first voice data.

2. The intelligent education system based on retrieval enhancement generation according to claim 1, characterized in that, The search enhancement generation module is configured to obtain the student's course progress for a preset subject based on the identity information and the preset subject corresponding to the query information in the following manner: Based on the identity information and the preset subject corresponding to the query information, the student's course progress for the preset subject is obtained from a preset unified context; wherein, the unified context is a data set stored in a preset Model Context Protocol (MCP) server, and the unified context includes at least the correspondence between the student's identity information and status parameters; wherein, the status parameters include the course progress of multiple subjects.

3. The intelligent education system based on retrieval enhancement generation according to claim 2, characterized in that, The retrieval enhancement generation module is configured to obtain the student's cognitive level regarding the preset knowledge points based on the identity information and the query information in the following manner: Based on the preset knowledge points corresponding to the identity information and the inquiry information, the student's cognitive level regarding the preset knowledge points is obtained from a preset unified context; wherein, the state parameter also includes the cognitive level of multiple knowledge points.

4. The intelligent education system based on retrieval enhancement generation according to claim 1, characterized in that, The retrieval enhancement generation module is configured to retrieve the first target teaching content by searching a preset teaching resource database based on the query information, the cognitive level, and the course progress in the following manner: Based on the inquiry information, the cognitive level, and the course progress, obtain the target score for each knowledge unit segment in the teaching resource database; The knowledge unit segments are sorted in descending order of target score, and the top-ranked number of knowledge unit segments are determined as the primary target teaching content.

5. The intelligent education system based on retrieval enhancement generation according to claim 4, characterized in that, Based on the inquiry information, the cognitive level, and the course progress, obtain the target score for each knowledge unit segment in the teaching resource database, including: Obtain the semantic similarity score between the query information and each knowledge unit segment; obtain the cognitive matching score between the cognitive level and each knowledge unit segment; obtain the progress alignment score between the course progress and each knowledge unit segment. The target score for each knowledge unit segment is obtained based on the semantic similarity score, cognitive matching score, and progress alignment score.

6. The intelligent education system based on retrieval enhancement generation according to claim 5, characterized in that, Based on the semantic similarity scores, cognitive matching scores, and progress alignment scores, the target scores for each knowledge unit segment are obtained, including: The target score for each knowledge unit segment is obtained by calculating the semantic similarity score, cognitive matching score and progress alignment score using a preset algorithm.

7. The intelligent education system based on retrieval enhancement generation according to claim 1, characterized in that, The large language model module is further configured to, when the input information is an operation instruction, generate a corresponding function call instruction based on the operation instruction; and send the function call instruction to the response module. The response module is configured to determine the corresponding target teaching tool based on the function call instruction, call the target teaching tool to execute the operation instruction, and obtain the execution result; The execution result is sent to the preset Model Context Protocol (MCP) server; The MCP server is configured to obtain the teaching status based on the execution result and send the teaching status to the virtual digital human module. The virtual digital human module is configured to acquire corresponding guidance text according to the teaching status, convert the guidance text into second voice data, and generate an output instruction containing the second voice data to control the digital human tutor to output the second voice data.

8. The intelligent education system based on retrieval enhancement generation according to claim 1, characterized in that, Also includes: The visual positioning module is configured to obtain the student's current viewpoint coordinates in the panoramic teaching scene; if the current viewpoint coordinates belong to a preset teaching interest area, obtain the identification information of the teaching interest area; and send the identification information of the teaching interest area to the Model Context Protocol (MCP) server; wherein, the panoramic teaching scene is the front-end interaction layer of the intelligent education system. The MCP server is configured to receive the identification information of the teaching interest area and send the identification information of the teaching interest area to the virtual digital human module; The virtual digital human module is configured to obtain corresponding explanation content based on the identification information of the teaching interest area, convert the explanation content into third voice data, and generate an output instruction containing the third voice data to control the digital human tutor to output the third voice data.

9. The intelligent education system based on retrieval enhancement generation according to claim 1, characterized in that, Also includes: The behavior log recording module is configured to record students' behavior logs during the teaching process; The behavior log is then sent to the evaluation module. The assessment module is configured to obtain the student's mastery of preset knowledge points based on the behavior log; and send the mastery of preset knowledge points to the retrieval enhancement generation module. The retrieval enhancement generation module is configured to search in a preset teaching resource database based on the mastery of the preset knowledge points to obtain the second target teaching content; Send the second target teaching content to the display module; The display module is configured to receive the second target teaching content sent by the retrieval enhancement generation module and display the second target teaching content.

Images