A Course Recommendation Method and System Based on Decoupling Modeling of Large Language Models

Abstract

This invention relates to a course recommendation method and system based on decoupled modeling using a large language model. The method includes: defining a generation task and a required output format prompt template; generating a course sequence from the format prompt template and course names using an LLM model; extracting course names and corresponding relationship descriptions from the course sequence, constructing a course relationship graph, inputting the course relationship graph into the LLM model, and obtaining knowledge-enhanced embedding vectors; obtaining learners' historical record sequences, converting the historical record sequences into historical feature sequences, calculating the dot product between the historical feature embeddings of the target learner and the historical embeddings of other learners based on the historical feature sequences, and obtaining the set of learners with the highest scores; obtaining the historical feature sequences of similar learners based on the set of learners with the highest scores, integrating them with the interaction sequences of the target learner, and inputting the integration result into the LLM model to obtain future course embedding vectors.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine learning, and particularly relates to a course recommendation method and system based on large language model decoupling modeling. BACKGROUND

[0002] The goal of a course recommendation system is to help learners efficiently filter massive courses based on historical interactions and behavioral characteristics, thereby providing personalized course selection recommendations. This not only alleviates the decision-making burden caused by information overload, but also improves learning efficiency and promotes continuous learning. To meet the growing demand for accurate course prediction, more and more researchers are committed to developing course recommendation frameworks for intelligent tutoring systems and MOOC platforms. Current course recommendation models can be roughly divided into three categories. (1) Collaborative filtering-based models use user interaction matrices to infer learner preference patterns through neighborhood-based similarity calculations. (2) Deep learning-based architectures use complex neural networks to model multi-relational dependencies between learners and course resources. (3) Reinforcement learning-based models use experiential learning to iteratively improve and optimize their recommendation strategies, thereby gradually improving their ability to provide personalized and effective course recommendations. Despite these significant research achievements, there are still some challenges in actual online education applications that are worth further exploring.

[0003] First, online education systems face ongoing challenges caused by the long-tail distribution and cold start problem, mainly due to insufficient historical behavior data. The effectiveness of personalized course recommendations fundamentally depends on accurate modeling of user interaction patterns to predict future engagement. However, in practical applications, most learners only interact with a limited number of courses, resulting in inherent interaction sparsity. The scarcity of historical data severely hinders the capture of learner preferences, making effective model training challenging.

[0004] Second, most existing recommendation models mainly rely on course identifiers to model user history, which limits the effective extraction and utilization of course relationships in practical education applications. This limitation arises because courses often exhibit various structural relationships that help to more comprehensively understand user behavior and play a key role in improving recommendation effectiveness. For example, before learning "Data Structures", it is recommended to first learn "Discrete Mathematics", "Computer Fundamentals", and "Programming Language Fundamentals" to establish a solid theoretical and practical foundation. Therefore, overly simple course identifier representations ignore course relevance and structural relationships, thereby weakening the predictive ability of future course recommendation models.

[0005] Recent advances in large language models (LLMs) have demonstrated superior semantic representation capabilities and cross-domain knowledge retention, revolutionizing the learning paradigm in the field of AI research. However, the comprehensive potential of LLMs in educational recommendation systems has not been fully explored, particularly in the context of course recommendation architectures. Furthermore, while LLMs perform exceptionally well in natural language processing tasks, their direct application to real-world course recommendation scenarios seems impractical. On one hand, LLMs fail to adequately meet the individualized needs of learners. They typically default to recommending popular courses rather than providing truly personalized suggestions. On the other hand, the effectiveness of LLM-based recommendations is further hindered by the sparsity of interaction data. SUMMARY

[0006] To solve the above problems existing in the prior art, the purpose of the present application is to provide a course recommendation method and system based on large language model decoupling modeling, a dual-channel architecture for course recommendation: LLM-enhanced learner and course decoupling modeling framework. It has two synergistic components: (1) knowledge-enhanced LLM-based course modeling, integrating the potential relationships between different courses to obtain enhanced embedding vectors; (2) interaction-enhanced LLM-based learner modeling, utilizing more historical interactions from similar learners to simulate cold-start learner interactions.

[0007] To achieve the above purpose, the present application provides the following scheme:

[0008] The course recommendation method based on large language model decoupling modeling comprises:

[0009] Defining a generation task and a required output format prompt template, generating a course sequence from the format prompt template and the course name using an LLM model;

[0010] Extracting the course name and corresponding relationship description in the course sequence, constructing a course relationship graph, inputting the course relationship graph into the LLM model, and obtaining a knowledge-enhanced embedding vector;

[0011] Obtaining a learner's historical record sequence, converting the historical record sequence into a historical feature sequence, calculating the dot product between the target learner's historical feature embedding and other learners' historical embeddings based on the historical feature sequence, and obtaining a set of highest-scoring learners;

[0012] Based on the set of highest-scoring learners, obtaining a similar learner's historical feature sequence and integrating it with the target learner's interaction sequence, inputting the integration result into the LLM model, and obtaining a future course embedding vector.

[0013] Optionally, generating the course sequence comprises:

[0014] S i = LLMP,c i = c i , s i ;

[0015] where c i denotes the name of a course, s i denotes the generated course sequence, i.e., the course relationship description includes prerequisites and advancement paths, and P denotes the format prompt template.

[0016] Optionally, constructing the course relationship graph comprises:

[0017] treating each course in the course name as a center node, constructing a directed graph around the center node, and taking the directed graph as the course relationship graph where, denotes a set of courses related to each course, denotes that a directed edge is established between courses based on the relationship description, forming an edge set.

[0018] Optionally, an edge in the directed graph from node v i points to node v j is represented as: where node v i is a prerequisite course of node v j , and node v j is an advanced course of node v i .

[0019] Optionally, obtaining the knowledge-enhanced embedding vector comprises:

[0020]

[0021] where M c is a specific merging strategy on the final hidden state of the course LLM to generate an enhanced representation of the course, T i is a simplified text representation of the course relationship graph, is a real number field, and d is the dimension of the hidden layer.

[0022] Optionally, obtaining the set of top-scoring learners comprises:

[0023]

[0024] where j = 1, 2, …, |U|, K is a hyperparameter, T is a transpose, H i denotes the historical feature sequence of user u i , and H j denotes the historical feature sequence of user u j .

[0025] Optionally, obtaining the future course embedding vector comprises:

[0026]

[0027] wherein, denotes the historical feature sequence of K learners similar to the target learner u i denotes the historical feature sequence of K learners similar to the target learner u i,n+1 denotes the embedding vector of the course predicted by the LLM for the future interaction of u i denotes the embedding vector of the course predicted by the LLM for the future interaction of u u denotes the user LLM.

[0028] Optionally, the method further comprises:

[0029] Generating the embedding of the next recommended course based on the historical interaction embedding of the target learner as a training target, obtaining the predicted interest score of each course:

[0030]

[0031] wherein, W and b are trainable parameters;

[0032] According to the predicted interest score, determining the minimum predicted response:

[0033]

[0034] Optimizing the learning process of the model through the cross-entropy loss between the minimum predicted response and the real response:

[0035]

[0036] wherein, y i is the real response.

[0037] To achieve the above purpose, the present application also provides a course recommendation system based on large language model decoupling modeling, comprising:

[0038] A course modeling unit is configured to define a generation task and a required output format prompt template, and generate a course sequence from the format prompt template and a course name by using an LLM model;

[0039] Extract the course name and corresponding relationship description in the course sequence, construct a course relationship graph, input the course relationship graph into the LLM model, and obtain a knowledge-enhanced embedding vector;

[0040] a learner modeling unit configured to obtain a history record sequence of a learner, convert the history record sequence into a history feature sequence, calculate a dot product between a history feature embedding of a target learner and history embeddings of other learners based on the history feature sequence, and obtain a set of top-scoring learners;

[0041] a preference prediction unit configured to obtain a history feature sequence of a similar learner based on the set of top-scoring learners, integrate the history feature sequence of the similar learner with an interaction sequence of the target learner, input the integration result into the LLM model, and obtain a future course embedding vector.

[0042] The present application has the following beneficial effects:

[0043] The model proposed in the present application always achieves significant performance improvement on MOOCCube and MOOCCourse datasets. The most significant improvement is seen on the MOOCCube dataset, with a performance improvement of up to 7.3% on the HR@5 indicator, up to 8.9% on the NDCG@5 indicator, and up to 12.1% on the NDCG@10 indicator compared to the most advanced baseline. In comparison, on the MOOCCourse dataset, the performance improvement on the NDCG@5 indicator reaches 11.6%, and on the NDCG@10 indicator reaches 6.8%.

[0044] Compared with traditional recommendation models, the LLM-based model of the present application performs well on both datasets. Furthermore, the model of the present application exhibits better improvement compared to the most advanced baseline HHCoR. This enhancement is attributed to the effective utilization of the rich world knowledge, semantic representation capability, and reasoning ability of LLM.

[0045] Compared with zero-shot LLMs, the model of the present application is always superior to the most advanced GPT-4o on both datasets. Notably, on the MOOCCube dataset, the model of the present application achieves a performance improvement of up to 151% on the HR@5 indicator, up to 140% on the HR@10 indicator, up to 104% on the NDCG@5 indicator, and up to 80% on the NDCG@10 indicator. This confirms that the model of the present application can effectively address the challenges inherent in directly applying LLMs to course recommendation tasks, specifically their inability to meet the needs of individual learners and the adverse effects of sparse interaction data.

[0046] Compared with fine-tuned LLMs, the model of the present application exhibits clear advantages. This further validates the effectiveness of the dual-channel course and learner modeling approach, which fully leverages the comprehensive potential of LLMs to enhance individualized recommendation capabilities.

[0047] Therefore, the results demonstrate that the model of the present application effectively alleviates the limitations brought by the sparsity of learner interaction data and the isolation of courses, thereby achieving more accurate and effective course recommendations. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only represent some of the embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0049] Figure 1 The flowchart of the course recommendation method based on large language model decoupling modeling of the embodiments of the present application. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only represent some 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 skilled in the art without creative labor are within the scope of protection of the present application.

[0051] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0052] The rapid development of artificial intelligence (AI) is driving changes in various fields of society. Similarly, in the field of education and learning, artificial intelligence is redefining traditional paradigms by providing innovative tools to enhance learning experiences and outcomes. For example, the emergence of massive open online courses (MOOCs) has facilitated the creation of a vast repository of shared educational content, effectively addressing the issue of educational inequality. However, such a wealth of resources also presents a significant challenge: how to effectively identify the most suitable courses for individual needs and goals among the numerous courses available. This challenge has prompted researchers to explore the development of personalized course recommendation systems using artificial intelligence technology, aiming to promote more efficient and customized learning experiences and ultimately establish a student-centered intelligent education ecosystem. Although previous course recommendation models have achieved encouraging results, they are still limited in practical applications due to the sparsity of learner interaction data and the isolation of course projects. The remarkable success of large language models (LLMs) in various fields has inspired the exploration of integrating them into course recommendation systems to further optimize personalized recommendation functions.

[0053] As Figure 1As shown, the embodiment discloses a course recommendation method based on large language model decoupling modeling, comprising: defining a generation task and a required output format prompt template, generating a course sequence from the format prompt template and a course name using an LLM model; extracting the course name and the corresponding relationship description in the course sequence, constructing a course relationship graph, inputting the course relationship graph into the LLM model, and obtaining a knowledge-enhanced embedding vector; obtaining a historical record sequence of a learner, converting the historical record sequence into a historical feature sequence, calculating the dot product between the historical feature embedding of the target learner and the historical embedding of other learners based on the historical feature sequence, and obtaining a set of highest-score learners; based on the set of highest-score learners, obtaining the historical feature sequence of similar learners and integrating it with the interaction sequence of the target learner, inputting the integration result into the LLM model, and obtaining a future course embedding vector.

[0054] Specifically: Figure 1 The overall framework of the LLM-enhanced learner and course decoupling modeling model is shown, which contains two key components: LLM-based knowledge-enhanced course modeling and LLM-based interaction-enhanced learner modeling. First, use LLM to generate relevant course knowledge sequences for each course. By combining each course with the generated enhanced knowledge, a course relationship graph is constructed and used as input for the course LLM to obtain enhanced course embeddings. Second, based on the historical sequence of the learner, select learners with similar behavior, where the historical record is represented using enhanced embeddings obtained from course modeling. Then connect the learner course records of similar learners to the end of the target learner sequence. This enhanced sequence is provided as input to the learner LLM to generate course recommendations. Finally, combine the predicted embedding with the target embedding and pass it through a prediction head to obtain the final output. The detailed steps are as follows:

[0055] Further, generating the course sequence comprises:

[0056] S i = LLM P,c i = c i , s i ;

[0057] wherein c i represents the name of the course, s i represents the generated course sequence, i.e. the course relationship description includes prerequisites and advancement paths, and P represents the format prompt template.

[0058] Further, constructing the course relationship graph comprises:

[0059] Each course in the course name is regarded as a center node, and a directed graph is constructed around the center node, and the directed graph is taken as the course relationship graph wherein, denotes a course set related to each course, denotes that a directed edge is established between courses based on the relationship description, forming an edge set.

[0060] Further, in the directed graph, from node v i to node v j The edge is represented as: wherein, node v i is a prerequisite course of node v j , and node v j is a follow-up course of node v i .

[0061] Further, obtaining the knowledge-enhanced embedding vector comprises:

[0062]

[0063] wherein, M c is a specific merging strategy on the final hidden state of the course LLM to generate an enhanced representation of the course, and T i is a simplified text representation of the course relationship graph.

[0064] Specifically, based on the LLM course modeling:

[0065] First, use the LLM-based course modeling unit to generate an embedding vector for each course.

[0066] (1) Relationship establishment: In order to directly solve the problem of course isolation, the embodiment uses an LLM with extensive world knowledge as an extractor of course structure relationships, enriching the connections between courses from a natural language perspective. This method not only provides valuable and detailed information, but also reveals potential inter-course relationships, helping learners to plan learning paths more effectively. Specifically, the embodiment first defines a generation task and a required output format prompt template P. Then, using the prompt and the given course name c i call LLM to generate course sequence S i , i.e. S i = LLM P, c i = c i , s i . Wherein, c i denotes the name of the course, and s iThe generated course relationship description, including prerequisites and subsequent courses. Finally, the generated results will be reviewed by domain experts, and their feedback will be used to iterate and refine the prompt template P and model parameters, improving the accuracy of course sequences and ensuring consistency with established teaching logic.

[0067] The definition includes the generation task and the required output format:

[0068] The prompt template P for the generation task and the required output format:

[0069] Rule:

[0070] 1. Based on the given course name, accurately select the prerequisite courses from the following list, i.e., the courses that need to be completed before studying this course. Also, correctly select the subsequent courses, i.e., the courses that can be studied after completing this course.

[0071] 2. The returned result must be in standard JSON format and must include the fields "prerequisite" and "subsequent". If a course has no prerequisites, the value of "prerequisite" is an empty JSONArray. If a course has no subsequent courses, the value of "subsequent" is also an empty JSONArray.

[0072] 3. The values of the prerequisite and subsequent courses must be chosen from the following list, and absolutely cannot generate course names, otherwise severe punishment will be imposed.

[0073] ChoosenList:

[0074] {course_names};

[0075] OutputFormat:

[0076] {{

[0077] "prerequisite":["coursename1","coursename2",.....],

[0078] "subsequent":["coursename1","coursename2",.....]

[0079] }}

[0080] ThinkStepbyStep:

[0081] 1. Based on the target course and ChoiceList provided by the user, select the prerequisite courses for the target course from ChoiceList and assemble them into a list;

[0082] 2. Based on the target course and ChoiceList provided by the user, select the follow-up courses of the target course from ChoiceList and assemble them into a list;

[0083] 3. Correctly assemble it into a JSON file and return it.

[0084] (2) Relationship diagram: From the structured course sequence S generated in the previous step i In this embodiment, the course name c is extracted. i and its corresponding relation description s i The relationship is described as being related to the course name c. i A description of other related courses and the relationships between them. These elements form the basis for constructing a formal course relationship diagram. Each course c i Each node is considered a central node, and a directed graph is constructed around that node. in, Indicates c i The relevant course set, including |V c |≤5 is a constraint designed to accommodate the computational complexity of LLM. Based on the extracted relational description, this embodiment establishes directed edges between courses, thereby forming an edge set. From node v i Pointing to node v j The edges can be represented as This means v i It is v j prerequisite courses, v j It is v i Advanced courses. (Figure G) c The adjacency matrix is ​​represented as The adjacency matrix is ​​graph G c A digital mapping that describes the relationships between the vertices (i.e., courses) in a graph, in the form of graph G. c Simplified text representation of T i It encapsulates node, edge, and adjacency matrix information. Specifically, if node v... i and node v j If a relationship exists, then Otherwise, the value is 0. Through the above steps, a comprehensive course relationship graph is constructed to capture the structural interconnections between courses, providing a solid foundation for subsequent recommendation tasks.

[0085] (3) Course encoding: Through the above steps, the embodiment derives the course c i in the form of graph G c in the simplified text representation T i , which encapsulates node, edge and adjacency matrix information. The embodiment uses an LLM with excellent text understanding capability as a feature extractor. Input T i into the parameterized LLM to generate a knowledge-enhanced embedding vector e i for course c c :

[0086]

[0087] where M c refers to a specific merging strategy on the final hidden state of the course LLM to generate an enhanced representation of the course.

[0088] Further, obtaining the set of top-scoring learners includes:

[0089]

[0090] where j = 1, 2, …, |U|, and K is a hyperparameter.

[0091] Further, obtaining the future course embedding vector includes:

[0092]

[0093] where, denotes the historical feature sequence of the K learners similar to the target learner u i , and e i,n+1 denotes the embedding vector of the course predicted by the LLM for the future interaction of u i .

[0094] Specifically: LLM-based learner modeling:

[0095] Using the knowledge extraction function of the LLM, each course is added an extended course sequence that contains rich relationship information. The enhanced embedding vector generated by the knowledge-enhanced course modeling unit lays a solid foundation for subsequent learner preference prediction. The following details the interaction-enhanced learner modeling unit.

[0096] (1) Filter similar learners: To solve the problem of sparse learner and course interaction data in the course recommendation system, the embodiment implements a filtering mechanism to identify similar learners for the target user, thereby capturing groups with similar cognitive features and behavior patterns to the target learner. Specifically, first, through LLM-based course modeling, the historical record sequence R = R1, R2, …, R |U|transformed into a history feature sequence H = H1, H2, …, H |U| The history record sequence of a learner is a course name sequence, which represents the courses the learner has ever learned. Each learner u i is associated with a history record R i = r i,1 , r i,2 , …, r i,n , which consists of an ordered sequence of previously participated courses in chronological order, where n represents the number of previously learned courses, r i,j ∈ C represents the jth course participation event of the learner u i , and C represents a set containing all courses. H i = e i,1 , e i,2 , …, e i,n , e i,j represents the enhanced embedding vector of course c i . For the target learner u i , this embodiment calculates the dot product between the history feature embedding of u i and the history embedding of other learners u j to determine the set U i of the top K highest scoring learners:

[0097]

[0098] where j = 1, 2, …, |U|, and K is a hyperparameter. This indicates that these learners exhibit closely aligned preferences and behavior patterns.

[0099] (2) Preference prediction: Integrate the history feature sequence of similar learners obtained above with the interaction sequence of the target learner, and use it as the input of the LLM. Then, the LLM uses the history features of the target learner and information from similar learners to predict the future course embedding:

[0100]

[0101] where H is the history feature sequence of the K learners similar to the target learner u i . e i,n+1 is the embedding vector of the course predicted by the LLM for the future interaction of u i , specifically corresponding to the (n+1)th interaction. Combining the filtered history records of similar learners as supplementary information to enrich the interaction data of the target learner can more comprehensively represent the interest features of the learner, ultimately helping to more accurately recommend courses.

[0102] Further, the method further comprises:

[0103] The embedding of the next recommended course is generated based on the historical interaction of the target learner as the training target, and the predicted interest score of each course is obtained:

[0104]

[0105] where W and b are trainable parameters;

[0106] According to the predicted interest score, the minimum predicted response is determined:

[0107]

[0108] The learning process of the model is optimized by the cross-entropy loss between the minimum predicted response and the true response:

[0109]

[0110] where y i is the true response.

[0111] Specifically: training and optimization:

[0112] Considering the complexity of LLM, a parameter-efficient fine-tuning method based on LoRA is adopted in the training process to reduce the overall cost. The training target of the course recommendation model is to generate the embedding of the next recommended course based on the historical interaction embedding of the target learner. Specifically, based on the historical interaction embedding of u i , the predicted interest score of course c i is obtained by the following formula:

[0113]

[0114] where W and b are trainable parameters. is the corresponding probability of predicting course c i .

[0115] Finally, the learning process is optimized by minimizing the cross-entropy loss between the predicted response and the true response y i , which is defined as follows:

[0116]

[0117] By iteratively updating the loss at each epoch, the optimal weighting parameter can be obtained, thus making an accurate prediction for the next course.

[0118] The embodiment also provides a course recommendation system based on large language model decoupling modeling, comprising: a course modeling unit configured to define a generation task and a required output format prompt template, generate a course sequence from the format prompt template and a course name by using an LLM model; extract a course name and a corresponding relationship description in the course sequence, construct a course relationship graph, input the course relationship graph into the LLM model, and obtain a knowledge-enhanced embedding vector; a learner modeling unit configured to obtain a historical record sequence of a learner, convert the historical record sequence into a historical feature sequence, calculate a dot product between a historical feature embedding of a target learner and historical embeddings of other learners based on the historical feature sequence, and obtain a set of highest-scored learners; and a preference prediction unit configured to obtain a historical feature sequence of a similar learner based on the set of highest-scored learners, integrate the historical feature sequence with an interaction sequence of the target learner, input the integrated result into the LLM model, and obtain a future course embedding vector.

[0119] The above-described embodiments are merely intended to describe the preferred modes of the present application, and are not intended to limit the scope of the present application. Various modifications and improvements to the technical solutions of the present application made by those of ordinary skill in the art without departing from the design spirit of the present application shall fall within the scope of protection of the present application as defined by the claims.

Claims

1. A course recommendation method based on large language model decoupling modeling, characterized in that, The method comprises the following steps: defining a generation task and a required output format prompt template, and generating a course sequence from the format prompt template and a course name by using a course LLM model; extracting the course name and the corresponding relationship description in the course sequence, constructing a course relationship graph, inputting the course relationship graph into the course LLM model, and obtaining a knowledge-enhanced embedding vector; obtaining a historical record sequence of a learner, converting the historical record sequence into a historical feature sequence, calculating the dot product between the historical feature embedding of a target learner and the historical embedding of other learners based on the historical feature sequence, and obtaining a set of highest-scored learners; integrating the historical feature sequence of the similar learners with the interaction sequence of the target learner, inputting the integration result into a learner LLM model, and obtaining a future course embedding vector; The method further comprises the following steps: wherein, represents the historical feature sequence of K learners similar to the target learner u i i,n+1 represents the embedding vector of the course predicted by the learner LLM for future interaction of u i u represents the learner LLM.​​ 2. The course recommendation method based on large language model decoupling modeling according to claim 1, characterized in that, taking the embedding of the next recommended course generated based on the historical interaction embedding of the target learner as a training target, and obtaining a predicted interest score of each course: S i = course LLM(P, c i ) = {(c i , s i )}; where c i represents the name of the course, s i represents the generated course sequence, i.e. the course relationship description includes prerequisites and progression paths, and P represents the format prompt template.

3. The course recommendation method based on large language model decoupling modeling according to claim 1, characterized in that, wherein W and b are trainable parameters; Each course in the course name is regarded as a center node, and a directed graph is constructed around the center node, and the directed graph is taken as the course relationship graph G c = (V c ,ε c ); wherein, denotes a course set related to each course, denotes that a directed edge is established between courses based on the relationship description, forming an edge set.

4. The course recommendation method based on large language model decoupling modeling according to claim 3, characterized in that, An edge from node v i to node u j is denoted as: e ij = (v i , u j ) ∈ ε c ; wherein node v i is a prerequisite for node v j and node v j is an advanced course for node v i .

5. The course recommendation method based on large language model decoupling modeling according to claim 1, characterized in that, determining a minimum predicted response according to the predicted interest score: where M c is a specific merge policy on the final hidden state of the course LLM to generate an augmented representation of the course, T i is a simplified textual representation of the course graph, is the real number field, and d is the dimension of the hidden layer.

6. The course recommendation method based on large language model decoupling modeling according to claim 1, characterized in that, optimizing the learning process of the model through the cross-entropy loss between the minimum predicted response and the true response: where j = {1, 2,..., |U|}, is a hyperparameter, T is the transpose, H i denotes the historical feature sequence of target learner u i denotes the historical feature sequence of target learner u j denotes the historical feature sequence of target learner u j denotes the historical feature sequence of target learner u 7. The course recommendation method based on large language model decoupling modeling according to claim 1, characterized in that, The method comprises the following steps: a course modeling unit configured to define a generation task and a required output format prompt template, and generate a course sequence from the format prompt template and a course name by using a course LLM model; extracting the course name and the corresponding relationship description in the course sequence, constructing a course relationship graph, inputting the course relationship graph into the course LLM model, and obtaining a knowledge-enhanced embedding vector; a learner modeling unit configured to obtain a historical record sequence of a learner, convert the historical record sequence into a historical feature sequence, calculate the dot product between the historical feature embedding of a target learner and the historical embedding of other learners based on the historical feature sequence, and obtain a set of highest-scored learners; a preference prediction unit configured to obtain a historical feature sequence of similar learners based on the set of highest-scored learners, integrate the historical feature sequence of the similar learners with the interaction sequence of the target learner, input the integration result into a learner LLM model, and obtain a future course embedding vector. where y i is the true response.

8. The course recommendation system based on large language model decoupling modeling realized by the method according to any one of claims 1-7, characterized in that, ​ ​ ​ ​ ​

Images