Optimization Method for Dynamically Generating Learning Paths of Large Language Models Based on RAG

Abstract

The present invention discloses an optimization method for dynamically generating a learning path of a large language model based on RAG, which relates to the technical field of data processing and includes: obtaining a predetermined learning path in a predetermined learning syllabus; obtaining an RAG model; performing information retrieval and analysis on a target dynamic learning portrait to obtain a target real-time learning state, and performing a correlation comparison analysis with a predetermined knowledge base to obtain a candidate knowledge sequence; performing knowledge fusion analysis on the target predetermined knowledge and the candidate knowledge sequence to obtain a target knowledge sequence, and dynamically adjusting the predetermined learning path in combination with a target period to generate a target dynamic learning path for a target student. The present invention solves the technical problem that the prior art cannot perform intelligent dynamic adjustment of the subsequent learning path according to the real-time learning situation of students, resulting in a lack of pertinence in the learning process of students, unsatisfactory learning quality and effect, and low efficiency, and achieves the technical effect of improving the personalization and flexibility of the learning path and enhancing the overall learning effect of students.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to an optimization method for dynamically generating a learning path of a large language model based on RAG. Background Art

[0002] With the rapid development of artificial intelligence and online education technologies, personalized learning and the generation of adaptive learning paths have been widely applied in the field of education. However, with the continuous increase in the complexity and scale of learning content, traditional learning path setting methods face many challenges. Traditional learning paths are often static and cannot be dynamically adjusted according to the real-time learning situation of students, resulting in low learning efficiency and an inability to meet the needs of personalized learning. Summary of the Invention

[0003] The present application provides an optimization method for dynamically generating a learning path of a large language model based on RAG, which is used to solve the technical problem that the prior art cannot intelligently and dynamically adjust the subsequent learning path according to the real-time learning situation of students, resulting in a lack of pertinence in the learning process of students, unsatisfactory learning quality and effects, and low efficiency.

[0004] In view of the above problems, the present application provides an optimization method for dynamically generating a learning path of a large language model based on RAG.

[0005] The present application provides an optimization method for dynamically generating a learning path of a large language model based on RAG, and the method includes:

[0006] Obtain a predetermined learning path in a predetermined learning syllabus; obtain a RAG model, where the RAG model includes a multimodal retriever and an integrated fusion device; perform multimodal information retrieval and analysis on the target dynamic learning portrait of a target student through the multimodal retriever to obtain a target real-time learning state; perform a correlation comparison and analysis on the target real-time learning state and a predetermined knowledge base to obtain a candidate knowledge sequence; perform knowledge fusion analysis on the target predetermined knowledge of a target period in the predetermined learning path and the candidate knowledge sequence through the integrated fusion device to obtain a target knowledge sequence; combine the target knowledge sequence and the target period to dynamically adjust the predetermined learning path to generate a target dynamic learning path for the target student.

[0007] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0008] This application obtains a predetermined learning path in a predetermined learning syllabus; obtains a RAG model, which includes a multimodal retriever and an integrated fusion engine; performs multimodal information retrieval and analysis on the target dynamic learning portrait of the target student through the multimodal retriever to obtain the target real-time learning status; conducts a correlation comparison and analysis between the target real-time learning status and a predetermined knowledge base to obtain a candidate knowledge sequence; performs knowledge fusion analysis on the target predetermined knowledge and the candidate knowledge sequence in the target period of the predetermined learning path through the integrated fusion engine to obtain a target knowledge sequence; combines the target knowledge sequence and the target period to dynamically adjust the predetermined learning path, and generates the target dynamic learning path of the target student. The present invention solves the technical problems in the prior art that the subsequent learning path cannot be intelligently and dynamically adjusted according to the real-time learning situation of the student, resulting in a lack of pertinence in the learning process of the student, unsatisfactory learning quality and effect, and low efficiency. By obtaining the predetermined learning path and the RAG model, using the multimodal retriever to analyze the dynamic learning portrait of the student, obtaining the real-time learning status, then comparing the learning status with the predetermined knowledge base to generate a candidate knowledge sequence, performing fusion on the predetermined learning path and the candidate knowledge through the integrated fusion engine to generate a target knowledge sequence, and dynamically adjusting the learning path, finally generating a personalized dynamic learning path, achieving the technical effect of improving the personalization and flexibility of the learning path and enhancing the overall learning effect of the student. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following-described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0010] Figure 1 Schematic flowchart of the method for dynamically generating and optimizing the learning path of a large language model based on RAG provided by an embodiment of this application;

[0011] Figure 2 Schematic flowchart of the process for obtaining the candidate knowledge sequence in the method for dynamically generating and optimizing the learning path of a large language model based on RAG provided by an embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0012] The present application provides an optimization method for dynamically generating the learning path of a large language model based on RAG, which is used to solve the technical problem that the prior art cannot perform intelligent dynamic adjustment of the subsequent learning path according to the real-time learning situation of the students, resulting in a lack of pertinence in the learning process of the students, unsatisfactory learning quality and effect, and low efficiency. By obtaining a predetermined learning path and a RAG model, using a multi-modal retriever to analyze the dynamic learning portrait of the students, obtaining the real-time learning status, then comparing the learning status with a predetermined knowledge base to generate a candidate knowledge sequence, and fusing the predetermined learning path and the candidate knowledge through an integration fuser to generate a target knowledge sequence, and dynamically adjusting the learning path, finally generating a personalized dynamic learning path, so as to achieve the technical effect of improving the personalization and flexibility of the learning path and enhancing the overall learning effect of the students.

[0013] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0014] It should be noted that any variations of the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or server that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products, or devices.

[0015] As Figure 1 shown, the present application provides an optimization method for dynamically generating the learning path of a large language model based on RAG, and the method includes:

[0016] Step S100: Obtain a predetermined learning path in a predetermined learning syllabus.

[0017] In the embodiment of the present application, first, the preset learning syllabus that has been set, that is, the predetermined learning syllabus, is loaded from a preset database. The predetermined learning syllabus is pre-planned according to the learning objectives, course requirements, and knowledge fields of the students, and includes all necessary learning contents, modules, and knowledge points. Then, the predetermined learning syllabus is parsed to identify and extract each learning module, chapter, and knowledge point, and arranged in a logical order.

[0018] After parsing the predetermined learning syllabus, the predetermined learning path therein is extracted, which includes the learning contents that the students need to complete according to the predetermined plan, and is planned according to the learning order, difficulty level, and progress requirements.

[0019] Step S200: Obtain a RAG model, where the RAG model includes a multimodal retriever and an integrated fusion unit.

[0020] In the embodiments of the present application, first, the RAG model is extracted from a predefined model library. The core of the RAG model includes a multimodal retriever and an integrated fusion unit. Among them, the RAG (Retrieval-Augmented Generation) technology is a technical method that combines retrieval and generation, aiming to enhance the capabilities of large language models (LLMs) when dealing with knowledge-intensive tasks.

[0021] The main function of the multimodal retriever is to retrieve information from multiple data sources, covering data in different modalities such as text, images, audio, and video. The multimodal retriever is used to perform cross-domain analysis in multimodal information to find knowledge points related to the current learning state of the trainee. Through the multimodal retriever, the RAG model can not only perform single-text information retrieval but also extract useful information from videos or audio.

[0022] The function of the integrated fusion unit is to integrate and fuse-analyze the multiple types of information obtained by the multimodal retriever. Through semantic analysis and feature extraction, the integrated fusion unit deeply fuses the content from different modalities to ensure that the generated knowledge content has logical consistency and coherence. For example, the integrated fusion unit integrates different information on the same topic extracted from text and video into a complete knowledge sequence, making the generated learning path or knowledge recommendation more targeted and practical.

[0023] By obtaining and loading the RAG model and combining the functions of the multimodal retriever and the integrated fusion unit, it is possible to achieve intelligent dynamic learning path generation and provide comprehensive and personalized learning support for trainees.

[0024] Step S300: Perform multimodal information retrieval and analysis on the target dynamic learning profile of the target trainee through the multimodal retriever to obtain the target real-time learning state.

[0025] In the embodiments of the present application, first, multi-dimensional data of the target trainee during the learning process is collected, including the target trainee's historical learning records, completed courses, learning duration, mastery of learning content, and the results of exams or quizzes, etc. Based on the collected data, the target dynamic learning profile of the target trainee is constructed.

[0026] Next, the multi-modal retriever performs multi-modal information retrieval and analysis on the target dynamic learning portrait of the target student. First, the knowledge content that the target student has learned is identified, which includes the target learned text sequence and the target learned video sequence. Then, through the text retrieval channel of the multi-modal retriever, the text content of multiple learned class hours is retrieved, and the first target learned content corresponding to the class hours is extracted, such as text materials like course notes and book chapters. At the same time, through the non-text retrieval channel, multiple course videos watched by the student are retrieved, and the second target learned content related to these courses is extracted, such as important knowledge points and explanations in the videos. After completing the retrieval of text and video content, the first target learned content and the second target learned content are integrated and analyzed, that is, the text information and video information are integrated to generate the target real-time learning status of the target student.

[0027] Furthermore, the method provided by the application embodiment further includes:

[0028] The multi-modal retriever includes a text retrieval channel and a non-text retrieval channel; analyzing the target dynamic learning portrait to obtain the target learned knowledge of the target student, where the target learned knowledge includes a target learned text sequence and a target learned video sequence; performing content retrieval on multiple segments of text corresponding to multiple class hours in the target learned text sequence through the text retrieval channel to obtain the first target learned content; performing content retrieval on multiple videos corresponding to multiple courses in the target learned video sequence through the non-text retrieval channel to obtain the second target learned content; using the first target learned content and the second target learned content as the target real-time learning status.

[0029] In the embodiment of the present application, the multi-modal retriever is composed of a text retrieval channel and a non-text retrieval channel, and is used to comprehensively analyze the learning data of the student. Specifically, the text retrieval channel is responsible for extracting and analyzing the target learned text sequence from the target dynamic learning portrait of the student, and the non-text retrieval channel is used to analyze the target learned video sequence. The multi-modal retriever constructs the target learned knowledge of the student through the text retrieval channel and the non-text retrieval channel.

[0030] In specific operations, first, content retrieval is performed on multiple class hours in the target learned text sequence through the text retrieval channel. The text retrieval uses natural language processing techniques, such as TF-IDF, to perform vectorization analysis on the text content, extracts the core knowledge points that the student has learned in multiple class hours, and generates the first target learned content.

[0031] Meanwhile, content retrieval is performed on multiple course videos in the target learned video sequence through a non-text retrieval channel. By using video content analysis techniques such as automatic speech recognition, the audio part in the video is converted into text through automatic speech recognition technology, and then the TF-IDF method same as that of the text retrieval channel is used to perform content analysis on the converted text to extract the knowledge points learned by the learner from the video, generating the second target learned content.

[0032] Finally, the first target learned content obtained from the text retrieval channel and the second target learned content obtained from the non-text retrieval channel are integrated to obtain the target real-time learning status of the target learner.

[0033] Furthermore, the method provided by the application embodiment further includes:

[0034] The non-text retrieval channel stores a format processing strategy; match the first video corresponding to the first course in the multiple videos, where the first course is any one of the multiple courses; parse the first video according to the format processing strategy to obtain a first video parsing result, where the first video parsing result includes a first image time series and a first audio; convert the first audio into a first text, and form a formatted text with the first contained text of the first image in the first image time series; perform content retrieval on the formatted text to obtain the second target learned content.

[0035] In the embodiment of the present application, a format processing strategy is stored in the non-text retrieval channel, and its function is to convert the audio and image content in the video into a retrievable text format.

[0036] Specifically, first, through the keyword matching method, the first video corresponding to the course learned by the student is found from multiple course videos. When performing keyword matching, by retrieving the metadata of the video, such as the course title or course number, and comparing this information, the video that best matches the content learned by the student is identified. The first course is any one of the multiple courses learned by the student. After matching, according to the format processing strategy, the first video is parsed in terms of both audio and image time series. In audio parsing, the automatic speech recognition (ASR) technology is used to convert the audio part in the video into text. The ASR technology directly converts speech to text by analyzing the speech content in the audio, generating the first text corresponding to the audio, and this part of the text reflects the language content described in the video. At the same time, the image part of the video is parsed, and the OCR technology, such as Tesseract, is used to analyze the video frame by frame to extract the text content in the image frames. Through this process, the text information displayed in the video frames is extracted, generating the first image time series corresponding to the image time series, which records the text and its time information extracted from each frame of the image. After generating the first text and the first image time series, the audio text and the image text are integrated through the timestamp alignment method. According to the timestamps of the video frames, the audio content is accurately matched with the text information in the image frames in terms of time and integrated into a structured formatted text.

[0037] Finally, content retrieval is performed on the generated formatted text. By using the TF-IDF method, the content in the formatted text is segmented and the word frequencies are counted to extract the keywords related to the student's learning progress. Through this step, the knowledge points learned by the student through the video are identified, and finally the second target learned content is generated, that is, the learning results obtained by the student by watching the video.

[0038] Step S400: Perform a correlation comparison and analysis between the target real-time learning state and the predetermined knowledge base to obtain a candidate knowledge sequence.

[0039] In the embodiment of the present application, the predetermined knowledge base is a knowledge system covering multiple fields, including various knowledge points and conceptual structures, and these fields include different disciplines and topics.

[0040] When generating the candidate knowledge sequence of the student, a correlation comparison and analysis is performed to compare the target real-time learning state of the student with the knowledge points in the predetermined knowledge base to determine which content in the knowledge base is relevant to the knowledge currently mastered by the student. Specifically, first, the learning state of the student and the knowledge points in the knowledge base are converted into vector forms, and then, through the cosine similarity method, the similarity between the learning state of the student and each knowledge point in the knowledge base is calculated. Based on the results of this correlation analysis, the knowledge points most relevant to the current learning of the student are selected from the predetermined knowledge base to generate a candidate knowledge sequence.

[0041] Further, as Figure 2 shown, the method provided by the application embodiment further includes:

[0042] Extracting the first predetermined text of the first predetermined knowledge in the predetermined knowledge base, and obtaining the first predetermined word vector of the first predetermined text; obtaining the target real-time word vector of the target real-time learning state; when the first cosine similarity between the first predetermined word vector and the target real-time word vector reaches the first similarity threshold, adding the first predetermined knowledge to the candidate knowledge sequence; wherein, the calculation expression of the first cosine similarity between the first predetermined word vector and the target real-time word vector is as follows: ; refers to the first predetermined word vector and the target real-time word vector of the first cosine similarity, refers to the first predetermined word vector and the target real-time word vector of the dot product, and respectively refer to the Euclidean norm of the first predetermined word vector and the Euclidean norm of the target real-time word vector , and respectively refer to the first predetermined word vector and the target real-time word vector in the th element, refers to the first predetermined word vector and the target real-time word vector of the dimension.

[0043] In the embodiment of the present application, first, a first predetermined knowledge is randomly extracted from the predetermined knowledge base by a random selection method, and the first predetermined text related to this knowledge point is obtained, where the first predetermined text is a specific content description of the first predetermined knowledge. Then, a word vector model, such as Word2Vec, is used to process the first predetermined text to generate the first predetermined word vector related to the first predetermined knowledge. At the same time, the target real-time learning state of the student is obtained, and the target real-time learning state represents the knowledge currently mastered by the student. For comparative analysis, the same word vector model is used to convert the text data in the student's learning state, such as learning notes, completed course content, etc. into vectors to generate the target real-time word vector.

[0044] Subsequently, the similarity between the first predetermined word vector and the target real-time word vector is calculated by cosine similarity, and the specific calculation process is as follows:

[0045] ;

[0046] Wherein, refers to the first cosine similarity between the first predetermined word vector and the target real-time word vector ; refers to the dot product of the first predetermined word vector and the target real-time word vector ; and respectively refer to the Euclidean norms of the first predetermined word vector and the target real-time word vector ; and respectively refer to the -th element in the first predetermined word vector and the target real-time word vector ; refers to the dimension of the first predetermined word vector and the target real-time word vector . If the calculated cosine similarity reaches a preset first similarity threshold, i.e., a preset similarity threshold, it indicates that this predetermined knowledge point has a high correlation with the learning state of the student, and the first predetermined knowledge is added to the candidate knowledge sequence.

[0047] Repeat the above process, continuously and randomly extract other knowledge points from the predetermined knowledge base, and calculate their similarities with the learning state of the student in turn. When all knowledge points have been calculated for similarity, the traversal of the predetermined knowledge base is completed, and finally a candidate knowledge sequence containing multiple relevant knowledge points is generated.

[0048] Further, the method provided by the application embodiment further includes:

[0049] Dividing the target real-time learning state to obtain N learned contents in N cycles, wherein the N cycles correspond to N weight coefficients, and 2 ≤ N < 4, N is an integer; extracting the first cycle among the N cycles, and matching the first learned content of the first cycle among the N learned contents, the first learned content includes multiple contents of multiple nodes; randomly collecting the multiple nodes to obtain a first node set, and matching the first content set corresponding to the first node set among the multiple contents; performing fusion analysis on the first content set to obtain a first content vector, and weighting the first weight coefficient to obtain a first target content vector; analyzing the first target content vector to obtain the target real-time word vector.

[0050] In the embodiment of the present application, first, a time series partitioning technique is used to partition the target real-time learning state of the trainee, and the entire learning process is divided into N cycles, where 2 ≤ N < 4. Each cycle represents the knowledge content mastered by the trainee at different time periods. For each cycle, a preset weight coefficient is assigned, and these weight coefficients are preset to reflect the influence degree of each cycle on the current learning state. Generally, the earlier learning stage has a smaller influence, so the weight coefficient is lower; while the most recent learning stage has a greater influence on the current learning state, so the weight coefficient is higher. For example, the weight is set such that the coefficient of the first cycle is 0.2, the coefficient of the second cycle is 0.3, and the weight of the most recent cycle is 0.5.

[0051] Next, the first cycle is extracted from the N cycles obtained by partitioning. This cycle represents the knowledge mastered by the trainee in the earliest learning stage. Through a data extraction method with time markers, the first learned content related to the first cycle is extracted. This content consists of multiple nodes, and each node represents a unit learned by the trainee in a course or class period. Then, these nodes are randomly sampled, and a part of the nodes are randomly selected from multiple class periods or courses through random sampling technology to form the first node set. Based on node labels, such as course numbers or class period names, the first content set related to these nodes is matched, representing the specific content learned by the trainee at these nodes.

[0052] After obtaining the first content set, a word vector model, such as Word2Vec, is used to convert this content into a vector representation for fusion analysis. By mapping the text content of each node into a vector, a comprehensive vector representing the trainee's first learning stage, that is, the first content vector, is generated. Next, the first content vector is weighted according to the preset weight coefficient. Since the influence of the first cycle is smaller, its weight coefficient is set to 0.2. Through the weighted summation method, the weighted result of this cycle is calculated, that is, the first target content vector.

[0053] This process is repeated to weight the content vectors of each cycle. Assume that the weight coefficient of the second cycle is 0.3 and the weight of the most recent cycle is 0.5. The weighted results of each cycle will successively generate the second target content vector and the third target content vector. Finally, the weighted results of all cycles are summarized, and the target real-time word vector of the trainee is generated through weighted combination.

[0054] Step S500: Through the integration and fusion device, perform knowledge fusion analysis on the target predetermined knowledge in the target cycle of the predetermined learning path and the candidate knowledge sequence to obtain the target knowledge sequence.

[0055] In the embodiment of the present application, when the knowledge fusion analysis is performed on the target predetermined knowledge and the candidate knowledge sequence in the target period of the predetermined learning path through the integrated fuser, first, the candidate knowledge sequence is sorted in descending order by cosine similarity, and a candidate knowledge descending list is generated. Next, the first adjustment ratio is read, and the target predetermined knowledge is adjusted to delete some knowledge points to obtain a first adjustment result. Then, the first candidate knowledge is extracted from the candidate knowledge descending list and added to the adjustment result to generate a first knowledge fusion result. Next, the vector of the first knowledge fusion result is calculated through the word vector model, and the similarity is calculated with the target real-time word vector of the student to determine whether the second similarity threshold is reached. If not, the adjustment and fusion steps are repeated until the similarity requirement is met or the predetermined number of iterations is reached, and finally, the fused target knowledge sequence is output.

[0056] Further, the method provided by the application embodiment further includes:

[0057] Step a: Sort the candidate knowledge sequence in descending order by cosine similarity to obtain a candidate knowledge descending list; Step b: Read the first adjustment ratio; Step c: Perform deletion adjustment on the target predetermined knowledge based on the first adjustment ratio to obtain a first adjustment result; Step d: Extract the first candidate knowledge from the candidate knowledge descending list based on the first adjustment ratio; Step e: Add the first candidate knowledge to the first adjustment result to obtain a first knowledge fusion result; Step f: Obtain the first fusion word vector of the first knowledge fusion result; Step g: Calculate the first fusion cosine similarity between the first fusion word vector and the target real-time word vector; Step h: Determine whether the first fusion cosine similarity reaches the second similarity threshold; Step i: If not, repeat Steps b to i until the predetermined number of iterations is reached, and output the knowledge fusion result at that time as the target knowledge sequence.

[0058] In the embodiment of the present application, first, the candidate knowledge sequence is sorted in descending order by cosine similarity, and the candidate knowledge points are sorted from high to low by similarity to generate a candidate knowledge descending list. Next, the preset first adjustment ratio is called. The first adjustment ratio is used to determine the ratio of deletion and replacement required for the knowledge points in the predetermined learning path. The adjustment ratio is an internal parameter used to find a balance between the predetermined knowledge and the candidate knowledge. For example, if the adjustment ratio is set to 0.5, 50% of the predetermined knowledge will be deleted and replaced with candidate knowledge.

[0059] According to the first adjustment ratio read, start the deletion process for the target predetermined knowledge. Use a deletion algorithm based on the adjustment ratio to delete a certain proportion of knowledge points from the predetermined knowledge to generate a first adjustment result. For example, if the adjustment ratio is 0.5, half of the predetermined knowledge points are deleted, and the remaining part is retained for subsequent fusion. Based on the same first adjustment ratio, extract the most relevant first candidate knowledge from the previously sorted descending list of candidate knowledge. Next, add the first candidate knowledge to the first adjustment result to obtain a first knowledge fusion result.

[0060] After the fusion is completed, use a word vector model, such as Word2Vec, to convert the fused first knowledge fusion result into a numerical representation to generate a first fused word vector. Subsequently, calculate the similarity between the first fused word vector and the target real-time word vector through cosine similarity to obtain a first fused cosine similarity. Next, determine whether the first fused cosine similarity reaches a second similarity threshold. The second similarity threshold is a preset similarity threshold used to determine whether the knowledge fusion result meets the learning requirements. If the calculated similarity reaches or exceeds this threshold, the fusion result is considered valid.

[0061] If the first fused cosine similarity does not reach the second similarity threshold, then repeat the foregoing fusion process until the preset number of iterations is reached. Finally, use the knowledge fusion result at this time as the target knowledge sequence.

[0062] Step S600: Dynamically adjust the predetermined learning path by combining the target knowledge sequence and the target period to generate the target dynamic learning path of the target learner.

[0063] In the embodiment of the present application, first, combine the target knowledge sequence and the target period. The target knowledge sequence contains the knowledge points that the learner currently needs to learn, and the target period represents the current stage of the learner in the learning path. Next, perform a similarity match between the knowledge points in the predetermined learning path and the knowledge points in the target knowledge sequence. Specifically, by comparing the semantic similarity of two knowledge points, calculate the similarity score between them. Use cosine similarity to quantify the matching degree of each knowledge point. According to the obtained similarity matching result, start to adjust the order of the knowledge points in the predetermined learning path. Rearrange the knowledge points in the learning path according to the similarity level. Specifically, give priority to placing the knowledge points with higher similarity in the front position of the learning path to ensure that the learner first learns the most relevant content. For those knowledge points with lower similarity or no longer relevant, move them backward. After completing the order adjustment, generate the target dynamic learning path through a direct path generation method.

[0064] Furthermore, the method provided in the embodiment of the application further includes:

[0065] The predetermined learning syllabus further includes a predetermined learning acceptance and a predetermined learning objective; obtaining a target learning result of the target student learning based on the target dynamic learning path; determining whether the target learning result meets the predetermined learning acceptance; if so, issuing a quantitative evaluation instruction; based on the quantitative evaluation instruction, combining the predetermined learning objective to perform a dynamic adjustment effect evaluation on the target dynamic learning path.

[0066] In an embodiment of the present application, the predetermined learning syllabus further includes a predetermined learning acceptance and a predetermined learning objective. The predetermined learning acceptance refers to the minimum standard that a student must achieve, measured by the student's academic performance, task completion, or skill mastery level. The predetermined learning objective is the specific result that a student needs to achieve throughout the learning path.

[0067] Obtain the target learning result of the target student learning based on the target dynamic learning path through online tests, assignment submissions, or progress reports. These learning results reflect the specific results achieved by the student during the learning process, including the student's mastery of knowledge points, performance in completing tasks, etc. Compare the target learning result of the target student with the predetermined learning acceptance criteria. If the student's learning result meets or exceeds these acceptance criteria, it indicates that the student has mastered the learning content, meets the predetermined learning acceptance, and issues a quantitative evaluation instruction through an automatic trigger mechanism. This instruction collects the student's learning data based on the student's performance, such as learning duration, test scores, completion of learning tasks, etc. Through data analysis techniques, these data are converted into quantitative evaluation results for evaluating the student's learning performance.

[0068] Finally, based on the quantitative evaluation results and combined with the predetermined learning objectives, evaluate the adjustment effect of the target dynamic learning path. Analyze whether the trainees achieve the predetermined learning experience objectives, learning domain objectives, and learning knowledge quantity objectives in the adjusted learning path. The specific evaluation methods include judging whether the trainees feel that their learning experience has been improved in the adjusted learning path through the trainees' learning feedback, such as satisfaction surveys or learning participation data. For example, trainees may feedback that the adjusted learning tasks are easier, the pressure is reduced, or the interest is enhanced. At the same time, analyze the learning duration and frequency of the trainees to observe whether the trainees invest more time and energy in the adjusted path. If the learning experience of the trainees becomes more positive, confirm that the learning experience objective is achieved. Evaluate whether the trainees have comprehensively mastered the predetermined domain knowledge according to the test scores or task completion situations of the trainees in different learning domains. For example, if a trainee performs poorly in a certain knowledge domain in the previous learning path and supplements it by increasing the learning tasks in this domain after adjustment, and the evaluation results show that the trainee's performance in this domain has been significantly improved, it will be confirmed that the learning domain objective is achieved. By analyzing the trainees' mastery of knowledge points before and after adjustment, judge whether the trainees have mastered more knowledge. For example, compare the knowledge point coverage rate in the tests before and after adjustment or the completion situation of learning tasks of the trainees. If the evaluation results show that the amount of knowledge mastered by the trainees has increased from 60% before adjustment to more than 80% after adjustment, it will be confirmed that the learning knowledge quantity objective has been successfully achieved.

[0069] By comprehensively analyzing the achievement situations of the above three objectives, complete the evaluation of the dynamic adjustment effect of the target dynamic learning path.

[0070] Furthermore, the method provided by the application embodiment further includes:

[0071] The predetermined learning objectives include learning experience objectives, learning domain objectives, and learning knowledge quantity objectives.

[0072] In the embodiment of the present application, the predetermined learning objectives include learning experience objectives, learning domain objectives, and learning knowledge quantity objectives. Among them, the learning experience objective refers to the subjective feelings of the trainees about the learning process, and evaluates the emotional experiences such as interest and pressure of the trainees in learning. The learning domain objective refers to the knowledge domains involved by the trainees in the learning process, ensuring that the trainees can cover the predetermined disciplines or knowledge categories. The learning knowledge quantity objective refers to the total amount of knowledge that the trainees need to master in the learning process, and is used to measure the number of knowledge points obtained by the trainees in the entire learning path.

[0073] In the embodiment of the present application, in summary, the embodiment of the present application has at least the following technical effects:

[0074] This application obtains a predetermined learning path in a predetermined learning syllabus; obtains a RAG model, where the RAG model includes a multimodal retriever and an integrated fusion engine; performs multimodal information retrieval and analysis on the target dynamic learning profile of the target learner through the multimodal retriever to obtain the target real-time learning status; conducts a correlation comparison and analysis between the target real-time learning status and a predetermined knowledge base to obtain a candidate knowledge sequence; performs knowledge fusion analysis on the target predetermined knowledge in the target period of the predetermined learning path and the candidate knowledge sequence through the integrated fusion engine to obtain a target knowledge sequence; combines the target knowledge sequence and the target period to dynamically adjust the predetermined learning path, and generates a target dynamic learning path for the target learner. The present invention solves the technical problem that the prior art cannot perform intelligent dynamic adjustment of the subsequent learning path according to the learner's real-time learning situation, resulting in a lack of pertinence in the learner's learning process, unsatisfactory learning quality and effect, and low efficiency. By obtaining the predetermined learning path and the RAG model, using the multimodal retriever to analyze the learner's dynamic learning profile, obtaining the real-time learning status, then comparing the learning status with the predetermined knowledge base to generate a candidate knowledge sequence, using the integrated fusion engine to fuse the predetermined learning path and the candidate knowledge to generate a target knowledge sequence, and dynamically adjusting the learning path, finally generating a personalized dynamic learning path, achieving the technical effect of improving the personalization and flexibility of the learning path and enhancing the overall learning effect of the learner.

[0075] It should be noted that the above-mentioned order of the embodiments of this application is only for description and does not represent the superiority or inferiority of the embodiments. And the above describes specific embodiments of this specification. The processes depicted in the drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0076] The above are only the preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of this application shall be included within the protection scope of this application.

[0077] This specification and the drawings are only exemplary descriptions of this application and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Obviously, those skilled in the art can make various changes and modifications to this application without departing from the scope of this application. Thus, if these modifications and variations of this application fall within the scope of this application and its equivalent technologies, this application is intended to include these changes and modifications.

Claims

1. A RAG-based large language model learning path dynamic generation optimization method, characterized in that: include: Get the scheduled learning path in the scheduled learning outline; Acquire a RAG model, wherein the RAG model includes a multimodal retriever and an integrated fusion device; The multimodal retriever performs multimodal information retrieval analysis on the target dynamic learning profile of the target student to obtain the target real-time learning status; Performing correlation comparison analysis on the target real-time learning state and the predetermined knowledge base to obtain a candidate knowledge sequence; Performing knowledge fusion analysis on the target predetermined knowledge of the target cycle in the predetermined learning path and the candidate knowledge sequence by the integrated fusion device to obtain a target knowledge sequence; Dynamically adjusting the predetermined learning path in combination with the target knowledge sequence and the target period to generate a target dynamic learning path for the target learner; Extracting a first predetermined text of the first predetermined knowledge in the predetermined knowledge base, and obtaining a first predetermined word vector of the first predetermined text; Obtain a target real-time word vector of the target real-time learning state; When the first cosine similarity between the first predetermined word vector and the target real-time word vector reaches a first similarity threshold, adding the first predetermined knowledge to the candidate knowledge sequence; The calculation expression of the first cosine similarity between the first predetermined word vector and the target real-time word vector is as follows: ； refers to the first predetermined word vector With the target real-time word vector The first cosine similarity, refers to the first predetermined word vector With the target real-time word vector The dot product of and They refer to the first predetermined word vectors The Euclidean norm of the target real-time word vector The Euclidean norm of , and They refer to the first predetermined word vectors With the target real-time word vector The elements, refers to the first predetermined word vector With the target real-time word vector Dimensions; Obtaining a target real-time word vector of the target real-time learning state includes: Dividing the target real-time learning state to obtain N learned contents of N periods, wherein the N periods correspond to N weight coefficients, and 2≤N<4, where N is an integer; Extracting a first cycle from the N cycles, and matching first learned content of the first cycle in the N learned content, where the first learned content includes multiple contents of multiple nodes; Randomly collect the multiple nodes to obtain a first node set, and match a first content set corresponding to the first node set in the multiple contents; Performing fusion analysis on the first content set to obtain a first content vector, and weighting it with a first weight coefficient to obtain a first target content vector; The first target content vector is analyzed to obtain the target real-time word vector.

2. According to the RAG-based large language model learning path dynamic generation optimization method according to claim 1, it is characterized in that: include: The multimodal retriever includes a text retrieval channel and a non-text retrieval channel; Analyzing the target dynamic learning portrait to obtain the target learned knowledge of the target student, wherein the target learned knowledge includes a target learned text sequence and a target learned video sequence; Performing content retrieval on multiple text segments corresponding to multiple lessons in the target learned text sequence through the text retrieval channel to obtain a first target learned content; Performing content retrieval on a plurality of videos corresponding to a plurality of courses in the target learned video sequence through the non-text retrieval channel to obtain a second target learned content; The first target learned content and the second target learned content are used as the target real-time learning status.

3. According to the RAG-based large language model learning path dynamic generation optimization method according to claim 2, it is characterized in that: include: The non-text retrieval channel stores a format processing strategy; Matching a first video corresponding to a first course among the multiple videos, where the first course is any one of the multiple courses; Parsing the first video according to the format processing strategy to obtain a first video parsing result, wherein the first video parsing result includes a first image timing and a first audio; Convert the first audio into a first text, and form a formatted text with the first included text of the first image in the first image sequence; Performing content retrieval on the format text to obtain the second target learned content.

4. According to claim 1, the RAG-based large language model learning path dynamic generation optimization method is characterized in that: include: Step a: Arrange the candidate knowledge sequence in descending order of cosine similarity to obtain a candidate knowledge descending list; Step b: reading the first adjustment ratio; Step c: deleting and adjusting the target predetermined knowledge based on the first adjustment ratio to obtain a first adjustment result; Step d: extracting the first candidate knowledge in the candidate knowledge descending list based on the first adjustment ratio; Step e: adding the first candidate knowledge to the first adjustment result to obtain a first knowledge fusion result; Step f: obtaining a first fused word vector of the first knowledge fusion result; Step g: Calculate the first fused cosine similarity between the first fused word vector and the target real-time word vector; Step h: determining whether the first fusion cosine similarity reaches a second similarity threshold; Step i: If not reached, repeat steps b to i until the predetermined number of iterations is reached, and output the knowledge fusion result at that time as the target knowledge sequence.

5. According to the RAG-based large language model learning path dynamic generation optimization method according to claim 1, it is characterized in that: Also includes: The predetermined learning outline also includes predetermined learning acceptance and predetermined learning objectives; Obtaining a target learning result of the target learner based on the target dynamic learning path; Determining whether the target learning result meets the predetermined learning acceptance; If it meets the requirements, a quantitative assessment instruction will be issued; Based on the quantitative evaluation instruction and in combination with the predetermined learning goal, the target dynamic learning path is dynamically adjusted and evaluated for its effect.

6. According to claim 5, the RAG-based large language model learning path dynamic generation optimization method is characterized in that: The predetermined learning goals include learning experience goals, learning field goals, and learning knowledge amount goals.

Images