Knowledge graph enhancement-based dynamic RAG learning resource recommendation method, system, device and medium
By constructing a domain knowledge graph and combining multimodal retrieval with a user interest model, the shortcomings of existing learning resource recommendation methods in terms of accuracy and personalization are addressed, achieving more accurate and personalized learning resource recommendations.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INSPUR SOFTWARE CO LTD
- Filing Date
- 2026-01-26
- Publication Date
- 2026-06-05
AI Technical Summary
Existing learning resource recommendation methods perform poorly in cold start scenarios, struggle to capture deep connections between knowledge points, lack knowledge structure awareness and personalized adaptation, resulting in insufficient recommendation accuracy and low personalization.
We construct a domain knowledge graph, acquire and integrate candidate resources through multimodal retrieval strategies, generate personalized recommendation results by combining user interest models, and optimize and adjust strategies through feedback to improve recommendation accuracy and applicability.
It improves the accuracy and applicability of learning resource recommendations, conforms to the knowledge system structure, adapts to different learning preferences and ability levels, and enhances the transparency and credibility of the recommendations.
Smart Images
Figure CN122153150A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of artificial intelligence, natural language processing, and education technology, specifically to a method, system, device, and medium for recommending dynamic RAG learning resources based on knowledge graph enhancement. Background Technology
[0002] Existing learning resource recommendation methods mainly rely on collaborative filtering or simple content matching. The application of RAG technology in the field of education also has obvious limitations, specifically as follows: ① Collaborative filtering methods rely on user behavior data, resulting in poor recommendation performance in cold start scenarios; ② Simple content matching is difficult to capture the deep connections between knowledge, and the recommended resources are limited to superficial similarity. Traditional simple matching only matches whether they are similar or not, but cannot solve whether they should be learned or not. ③ Existing RAG technology in educational scenarios suffers from insufficient retrieval accuracy, lack of knowledge structure awareness, and inadequate personalized adaptation.
[0003] Therefore, how to overcome the shortcomings of traditional recommendation systems, such as insufficient accuracy, weak knowledge connections, and low personalization, and improve the accuracy and applicability of learning recommendation resources is a technical problem that urgently needs to be solved. Summary of the Invention
[0004] The technical objective of this invention is to provide a method, system, device, and medium for recommending dynamic RAG learning resources based on knowledge graph enhancement, in order to overcome the shortcomings of traditional recommendation systems, such as insufficient accuracy, weak knowledge association, and low personalization, and to improve the accuracy and applicability of learning recommendation resources.
[0005] The technical objective of this invention is achieved as follows: a dynamic RAG learning resource recommendation method based on knowledge graph enhancement, the specific method of which is as follows: Construct a domain knowledge graph and link learning resources with knowledge concepts: Construct a domain knowledge graph that includes learning resource entities, concept entities, and the relationships between them; Parse user queries and extract query intent and key entities: perform structured parsing of learning resources, extract resource features and establish mappings with knowledge graph entities; Candidate learning resources are obtained and deduplicated through a multimodal retrieval strategy: combining semantic vector retrieval, keyword retrieval, and knowledge graph path retrieval to obtain candidate resources relevant to the user's query; Personalized recommendation results are generated through a user interest model: a user interest model is constructed based on the user's learning trajectory and interaction feedback, and candidate resources are sorted and filtered in combination with the user interest model to generate personalized recommendation results; among them, the sorting adopts a gradient boosting tree model that integrates multiple features, including the semantic similarity between resources and queries, knowledge graph path length, user interest matching degree, resource quality score and timeliness features. Feedback optimization: Dynamically adjust the retrieval strategy and user interest model based on user feedback on the recommendation results, and optimize system parameters and user interest model.
[0006] As a preferred option, the construction of a domain knowledge graph is as follows: Entity extraction: The BERT-BiLSTM-CRF model is used to extract knowledge points, concepts, and skill entities from learning resources; Relation extraction: Identifying hierarchical relationships, prerequisite relationships, and association relationships between entities using a deep learning model based on an attention mechanism; Graph Update: Supports incremental graph updates based on new learning resources to maintain the timeliness of knowledge.
[0007] As a preferred approach, the multimodal retrieval strategy is as follows: Semantic vector retrieval: Text vectors are generated using a domain-fine-tuned BERT model, and similarity matching is performed through a vector database to obtain the first candidate set based on semantic similarity; Keyword retrieval: Weighted keyword matching is performed by combining domain thesaurus, and a second candidate set is obtained based on keyword matching to ensure accurate retrieval of professional terms; Knowledge graph path retrieval: Based on the user query entity, multi-hop path retrieval is performed in the knowledge graph to mine indirectly related resources, and a third candidate set is obtained based on knowledge graph path reasoning; Search fusion: A dynamic weight fusion strategy is adopted, which adjusts the weight ratio of each search result according to the query type and resource type, and performs deduplication and weighted fusion on the first candidate set, the second candidate set and the third candidate set to form the final candidate resource set.
[0008] As a preferred approach, a user interest model is constructed based on user learning trajectories and interaction feedback, as follows: Extracting interest features: Extracting interest features from users' learning time, resource type preferences, and knowledge mastery levels; Detecting interest drift: Detecting trends in user interest changes through a sliding window mechanism; Interest decay: Set a time decay factor for users' historical interest characteristics to enhance the influence weight of recent interests.
[0009] As a preferred approach, a three-layer filtering mechanism is used to generate personalized recommendation results, as follows: The first layer is based on knowledge graph correlation filtering, retaining resources that are directly or indirectly related to the user's current learning topic; The second layer filters resources based on user interest models, selecting those that match users' learning preferences and ability levels. The third layer filters based on resource quality scores, eliminating low-quality or outdated learning resources.
[0010] Even better, the feedback optimization is as follows: Explicit feedback collection: Collecting user ratings and tags for recommended resources; Implicit feedback analysis: Analyzing user behavior data such as clicks, dwell time, and completion rate; Strategy adjustment: Based on feedback data, reinforcement learning algorithms are used to dynamically adjust retrieval weights and recommendation strategies.
[0011] A dynamic RAG learning resource recommendation system based on knowledge graph enhancement, the system comprising: The knowledge graph construction module is used to build a domain knowledge graph that includes learning resource entities, concept entities, and the relationships between them. The resource preprocessing module is used to perform structured parsing of learning resources, extract resource features, and establish mappings with knowledge graph entities; The multimodal retrieval module combines semantic vector retrieval, keyword retrieval, and knowledge graph path retrieval to obtain candidate resources relevant to the user's query. The personalized interest modeling module is used to build user interest models based on user learning trajectories and interaction feedback; The recommendation generation module is used to integrate candidate resource features, knowledge graph correlation, and user interest models to generate personalized recommendation results. The feedback optimization module is used to dynamically adjust the retrieval strategy and interest model parameters based on user feedback on the recommendation results.
[0012] As a preferred option, the knowledge graph construction module includes: The entity extraction submodule is used to extract knowledge points, concepts, and skill entities from learning resources using the BERT-BiLSTM-CRF model; The relation extraction submodule is used to identify hierarchical relationships, prerequisite relationships, and association relationships between entities using a deep learning model based on an attention mechanism. The graph update submodule is used to support incremental graph updates based on new learning resources to maintain the timeliness of knowledge. The multimodal retrieval module includes, The semantic vector retrieval submodule is used to generate text vectors using a domain-fine-tuned BERT model and perform similarity matching through a vector database. The keyword retrieval submodule is used to perform weighted keyword matching by combining the domain thesaurus to ensure accurate retrieval of professional terms; The knowledge graph path retrieval submodule is used to perform multi-hop path retrieval in the knowledge graph based on user query entities and to mine indirectly related resources. The retrieval fusion submodule is used to adopt a dynamic weight fusion strategy to adjust the weight ratio of each retrieval result according to the query type and resource type; The personalized interest modeling module includes: The interest feature extraction submodule is used to extract interest features from users' learning time, resource type preferences, and knowledge mastery. The interest drift detection submodule is used to detect trends in user interest changes through a sliding window mechanism. The interest decay submodule is used to set a time decay factor for the user's historical interest features, thereby increasing the influence weight of recent interests; The recommendation generation module employs a three-layer filtering mechanism, as follows: The first layer is based on knowledge graph correlation filtering, retaining resources that are directly or indirectly related to the user's current learning topic; The second layer filters resources based on user interest models, selecting those that match users' learning preferences and ability levels. The third layer filters based on resource quality scores, eliminating low-quality or outdated learning resources. The feedback optimization module includes, The explicit feedback collection submodule is used to collect user ratings and tags for recommended resources; The implicit feedback analysis submodule is used to analyze user behavior data such as clicks, dwell time, and completion rate. The strategy adjustment submodule is used to dynamically adjust retrieval weights and recommendation strategies based on feedback data using reinforcement learning algorithms.
[0013] An electronic device includes: a memory and at least one processor; The memory contains computer programs; The at least one processor executes the computer program stored in the memory, causing the at least one processor to perform the knowledge graph-based dynamic RAG learning resource recommendation method as described above.
[0014] A computer-readable storage medium storing a computer program that can be executed by a processor to implement the knowledge graph-based dynamic RAG learning resource recommendation method described above.
[0015] The knowledge graph-based dynamic RAG learning resource recommendation method, system, device, and medium of the present invention have the following advantages: (i) This invention solves the problems of insufficient accuracy, weak knowledge association, and low personalization in traditional recommendation systems by modeling knowledge association through knowledge graphs and combining dynamic retrieval strategies and personalized interest modeling. It improves the accuracy and applicability of learning resource recommendations, focuses on learning resource recommendation scenarios, and the core is to integrate knowledge graphs with retrieval augmentation generation (RAG) technology and apply it to the development of resource recommendation systems in scenarios such as online education and intelligent learning platforms. (II) This invention addresses the problem of knowledge fragmentation through knowledge graph modeling, resulting in recommendation results that better align with the knowledge system structure and improve the accuracy of learning resource recommendations: (iii) This invention accurately captures changes in user needs and interests through a user interest model (dynamic interest model), adapts to different learning preferences and ability levels, and enhances personalization; (iv) This invention reduces the limitations of a single retrieval method by using a multimodal retrieval fusion strategy, adapts to different query and resource types, and improves robustness; (v) The knowledge graph-based recommendation of this invention can display resource association paths, improve the transparency and credibility of recommendations, and enhance interpretability. Attached Figure Description
[0016] The invention will be further described below with reference to the accompanying drawings.
[0017] Appendix Figure 1 A schematic diagram of the system architecture for dynamic RAG learning resource recommendation based on knowledge graph enhancement; Appendix Figure 2 This is a schematic diagram of the multimodal retrieval module; Appendix Figure 3 This is a diagram illustrating the recommended module generation process. Detailed Implementation
[0018] The method, system, device, and medium for recommending dynamic RAG learning resources based on knowledge graph enhancement according to the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this disclosure. In the following detailed description, numerous specific details are set forth to provide a comprehensive understanding of the embodiments of this disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of this disclosure.
[0019] In the technical solution disclosed herein, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation entry points are provided for users to choose to authorize or refuse. Example 1
[0020] This embodiment provides a dynamic RAG learning resource recommendation method based on knowledge graph enhancement, as detailed below: S1. Construct a domain knowledge graph and link learning resources with knowledge concepts: Construct a domain knowledge graph that includes learning resource entities, concept entities, and the relationships between them; S2. Parse user queries and extract query intent and key entities: Perform structured parsing of learning resources, extract resource features and establish mappings with knowledge graph entities; S3. Obtain candidate learning resources through multimodal retrieval strategies and integrate them for deduplication: Combine semantic vector retrieval, keyword retrieval and knowledge graph path retrieval to obtain candidate resources related to user queries; S4. Generate personalized recommendation results through user interest model: Construct user interest model based on user learning trajectory and interaction feedback, and combine user interest model to sort and filter candidate resources to generate personalized recommendation results; Among them, the sorting adopts gradient boosting tree model that integrates multiple features, including semantic similarity between resources and queries, knowledge graph path length, user interest matching degree, resource quality score and timeliness features. S5. Feedback Optimization: Dynamically adjust the retrieval strategy and user interest model based on user feedback on the recommendation results, and optimize system parameters and user interest model.
[0021] The specific steps for constructing the domain knowledge graph in step S1 of this embodiment are as follows: S101, Entity Extraction: The BERT-BiLSTM-CRF model is used to extract knowledge points, concepts, and skill entities from learning resources; S102. Relationship Extraction: Identifying hierarchical relationships, prerequisite relationships, and association relationships between entities using a deep learning model based on attention mechanisms; S103, Graph Update: Supports incremental graph updates based on new learning resources to maintain the timeliness of knowledge.
[0022] The multimodal retrieval strategy in step S3 of this embodiment is as follows: S301, Semantic Vector Retrieval: Text vectors are generated using a domain-fine-tuned BERT model, and similarity matching is performed through a vector database to obtain the first candidate set based on semantic similarity; S302. Keyword retrieval: Combine domain thesaurus for weighted keyword matching, and obtain a second candidate set based on keyword matching to ensure accurate retrieval of professional terms; S303, Knowledge Graph Path Retrieval: Based on the user query entity, perform multi-hop path retrieval in the knowledge graph, mine indirect related resources, and obtain the third candidate set based on knowledge graph path reasoning; S304. Retrieval Fusion: A dynamic weight fusion strategy is adopted, which adjusts the weight ratio of each retrieval result according to the query type and resource type, and performs deduplication and weighted fusion on the first candidate set, the second candidate set and the third candidate set to form the final candidate resource set.
[0023] In this embodiment, step S4, which involves constructing a user interest model based on user learning trajectory and interaction feedback, is as follows: S401. Extracting interest features: Extracting interest features from user learning time, resource type preference, and knowledge mastery level; S402. Detecting interest drift: Detecting the trend of changes in user interests through a sliding window mechanism; S403, Interest Decay: Set a time decay factor for the user's historical interest characteristics to enhance the influence weight of recent interests.
[0024] In this embodiment, step S4, generating personalized recommendation results, employs a three-layer filtering mechanism, as detailed below: The first layer is based on knowledge graph correlation filtering, retaining resources that are directly or indirectly related to the user's current learning topic; The second layer filters resources based on user interest models, selecting those that match users' learning preferences and ability levels. The third layer filters based on resource quality scores, eliminating low-quality or outdated learning resources.
[0025] The feedback optimization in step S5 of this embodiment is as follows: S501, Explicit Feedback Collection: Collect user ratings and tags for recommended resources; S502, Implicit Feedback Analysis: Analyze user behavior data such as clicks, dwell time, and completion rate; S503. Strategy Adjustment: Based on feedback data, a reinforcement learning algorithm is used to dynamically adjust the retrieval weights and recommendation strategies. Example 2
[0026] As attached Figure 1As shown, this embodiment provides a dynamic RAG learning resource recommendation system based on knowledge graph enhancement. The system includes: The knowledge graph construction module is used to build a domain knowledge graph that includes learning resource entities, concept entities, and the relationships between them. The resource preprocessing module is used to perform structured parsing of learning resources, extract resource features, and establish mappings with knowledge graph entities; The multimodal retrieval module combines semantic vector retrieval, keyword retrieval, and knowledge graph path retrieval to obtain candidate resources relevant to the user's query. The personalized interest modeling module is used to build user interest models based on user learning trajectories and interaction feedback; The recommendation generation module is used to integrate candidate resource features, knowledge graph correlation, and user interest models to generate personalized recommendation results. The feedback optimization module is used to dynamically adjust the retrieval strategy and interest model parameters based on user feedback on the recommendation results.
[0027] The knowledge graph construction module in this embodiment includes: The entity extraction submodule is used to extract knowledge points, concepts, and skill entities from learning resources using the BERT-BiLSTM-CRF model; The relation extraction submodule is used to identify hierarchical relationships, prerequisite relationships, and association relationships between entities using a deep learning model based on an attention mechanism. The graph update submodule is used to support incremental graph updates based on new learning resources, thus maintaining the timeliness of knowledge.
[0028] As attached Figure 2 As shown, the multimodal retrieval module in this embodiment includes, The semantic vector retrieval submodule is used to generate text vectors using a domain-fine-tuned BERT model and perform similarity matching through a vector database. The keyword retrieval submodule is used to perform weighted keyword matching by combining the domain thesaurus to ensure accurate retrieval of professional terms; The knowledge graph path retrieval submodule is used to perform multi-hop path retrieval in the knowledge graph based on user query entities and to mine indirectly related resources. The retrieval fusion submodule is used to employ a dynamic weight fusion strategy to adjust the weight ratio of each retrieval result based on the query type and resource type.
[0029] The personalized interest modeling module in this embodiment includes: The interest feature extraction submodule is used to extract interest features from users' learning time, resource type preferences, and knowledge mastery. The interest drift detection submodule is used to detect trends in user interest changes through a sliding window mechanism. The interest decay submodule is used to set a time decay factor for users' historical interest features, thereby increasing the influence weight of recent interests.
[0030] As attached Figure 3 As shown, the recommendation generation module in this embodiment adopts a three-layer filtering mechanism, as detailed below: The first layer is based on knowledge graph correlation filtering, retaining resources that are directly or indirectly related to the user's current learning topic; The second layer filters resources based on user interest models, selecting those that match users' learning preferences and ability levels. The third layer filters based on resource quality scores, eliminating low-quality or outdated learning resources.
[0031] The feedback optimization module in this embodiment includes, The explicit feedback collection submodule is used to collect user ratings and tags for recommended resources; The implicit feedback analysis submodule is used to analyze user behavior data such as clicks, dwell time, and completion rate. The strategy adjustment submodule is used to dynamically adjust retrieval weights and recommendation strategies based on feedback data using reinforcement learning algorithms. Example 3
[0032] Taking the query scenario of "how to learn a binary tree" in the field of computer science as an example, the specific implementation process is as follows: (1) Knowledge graph construction: Extract entities such as “data structure”, “binary tree” and “sorting algorithm” from resources such as textbooks and courseware, identify the associations such as “binary tree is a type of data structure” (hierarchical relationship) and “to learn sorting algorithm, you need to master array first” (prerequisite relationship), and regularly incorporate new resources to achieve incremental updates.
[0033] (2) Resource preprocessing: Extract metadata such as title, abstract, and difficulty level of learning resources, associate the teaching video of "binary tree traversal" with entities such as "binary tree" and "tree traversal", and generate vector representation and keyword index.
[0034] (3) Multimodal retrieval is performed as follows: ① Semantic Vector Retrieval: Converts queries into vectors to match semantically similar teaching videos and courseware; ② Keyword search: Extract keywords such as "learning" and "binary tree" to search for resources containing these terms; ③ Knowledge graph path retrieval: Starting with "binary tree", explore prerequisite related resources such as "tree structure" and "recursive algorithm"; ④ Fusion Strategy: Adjust the weights according to query type (50% for semantic retrieval, 30% for graph retrieval, and 20% for keyword retrieval) to form a candidate resource set.
[0035] (4) Personalized interest modeling, as detailed below: ① Extracting interest features: Based on the user's recent learning of "arrays" and "recursion" and the long time spent on them, it is determined that they have a high interest in data structures and algorithm basics; ② Detecting interest trends: Identifying users' transition from basic data structures to complex data structures; ③ Application interest decay: Reduce the interest weight of "Python syntax" from one month ago.
[0036] (5) Recommended generation, as follows: ① Knowledge association filtering: Retain resources that are directly related to or have prerequisite relationships with the "binary tree"; ② Interest matching filter: Prioritize binary tree resources containing "recursive implementation"; ③ Quality filtering: Remove resources with scores below the threshold; ④ Sorting Output: The final recommendation list is generated by sorting the gradient boosting tree model using multi-feature fusion.
[0037] (6) Feedback optimization: After the user clicks and completes the learning of the "Binary Tree Recursive Traversal" resource, the system records the implicit feedback, increases the association weight between "recursion" and "binary tree", and updates the user's interest weight in "recursive application".
[0038] The specific technical implementation support for this embodiment is as follows: ① Knowledge graph storage: Utilizes graph databases such as Neo4j; ② Vector retrieval: Using vector databases such as Milvus; ③ Interest model construction: based on TensorFlow or PyTorch framework; ④ Deployment scenario: server or cloud platform. Example 4
[0039] This embodiment also provides an electronic device, including: a memory and a processor; The memory stores the instructions executed by the computer. The processor executes computer execution instructions stored in the memory, causing the processor to execute the knowledge graph-enhanced dynamic RAG learning resource recommendation method in any embodiment of the present invention.
[0040] The processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor can be a microprocessor or any conventional processor.
[0041] Memory can be used to store computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory. Memory can mainly include a program storage area and a data storage area. The program storage area can store the operating system, at least one application program required for a function, etc.; the data storage area can store data created based on the use of the terminal, etc. In addition, memory can also include high-speed random access memory, and can also include non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart memory cards (SMC), secure digital cards (SD cards), flash memory cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices. Example 5
[0042] This embodiment also provides a computer-readable storage medium storing multiple instructions, which are loaded by a processor to cause the processor to execute the knowledge graph-enhanced dynamic RAG learning resource recommendation method according to any embodiment of the present invention. Specifically, a system or apparatus equipped with a storage medium may be provided, on which software program code implementing the functions of any of the above embodiments is stored, and the computer (or CPU or MPU) of the system or apparatus may read and execute the program code stored in the storage medium.
[0043] In this case, the program code read from the storage medium can itself implement the function of any of the above embodiments, and therefore the program code and the storage medium storing the program code constitute part of the present invention.
[0044] Storage media embodiments for providing program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RYM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer via a communication network.
[0045] Furthermore, it should be clear that not only can the program code read by the computer be executed, but also the operating system or other components operating on the computer can be instructed based on the program code to perform some or all of the actual operations, thereby realizing the function of any of the embodiments described above.
[0046] Furthermore, it is understood that the program code read from the storage medium is written to the memory set in the expansion board inserted into the computer or to the memory set in the expansion unit connected to the computer. Then, based on the instructions of the program code, the CPU or other components installed on the expansion board or expansion unit execute some and all of the actual operations, thereby realizing the function of any of the embodiments described above.
[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A dynamic RAG learning resource recommendation method based on knowledge graph enhancement, characterized in that, The method is as follows: Construct a domain knowledge graph and link learning resources with knowledge concepts: Construct a domain knowledge graph that includes learning resource entities, concept entities, and the relationships between them; Parse user queries and extract query intent and key entities: perform structured parsing of learning resources, extract resource features and establish mappings with knowledge graph entities; Candidate learning resources are obtained and deduplicated through a multimodal retrieval strategy: combining semantic vector retrieval, keyword retrieval, and knowledge graph path retrieval to obtain candidate resources relevant to the user's query; Personalized recommendation results are generated through a user interest model: a user interest model is constructed based on the user's learning trajectory and interaction feedback, and candidate resources are sorted and filtered in combination with the user interest model to generate personalized recommendation results; among them, the sorting adopts a gradient boosting tree model that integrates multiple features, including the semantic similarity between resources and queries, knowledge graph path length, user interest matching degree, resource quality score and timeliness features. Feedback optimization: Dynamically adjust the retrieval strategy and user interest model based on user feedback on the recommendation results, and optimize system parameters and user interest model.
2. The knowledge graph enhanced dynamic RAG learning resource recommendation method according to claim 1, characterized in that, The construction of the domain knowledge graph is as follows: Entity extraction: The BERT-BiLSTM-CRF model is used to extract knowledge points, concepts, and skill entities from learning resources; Relationship extraction: Identifying hierarchical relationships, prerequisite relationships, and association relationships between entities using a deep learning model based on attention mechanisms; Graph Update: Supports incremental graph updates based on new learning resources to maintain the timeliness of knowledge. 3.The knowledge graph enhanced dynamic RAG learning resource recommendation method according to claim 1, characterized in that, The multimodal retrieval strategy is as follows: Semantic vector retrieval: Text vectors are generated using a domain-fine-tuned BERT model, and similarity matching is performed through a vector database to obtain the first candidate set based on semantic similarity; Keyword retrieval: Weighted keyword matching is performed by combining domain thesaurus, and a second candidate set is obtained based on keyword matching to ensure accurate retrieval of professional terms; Knowledge graph path retrieval: Based on the user query entity, multi-hop path retrieval is performed in the knowledge graph to mine indirectly related resources, and a third candidate set is obtained based on knowledge graph path reasoning; Search fusion: A dynamic weight fusion strategy is adopted, which adjusts the weight ratio of each search result according to the query type and resource type, and performs deduplication and weighted fusion on the first candidate set, the second candidate set and the third candidate set to form the final candidate resource set.
4. The knowledge graph enhanced dynamic RAG learning resource recommendation method according to claim 1, characterized in that, The user interest model is constructed based on user learning trajectory and interaction feedback as follows: Extracting interest features: Extracting interest features from users' learning time, resource type preferences, and knowledge mastery levels; Detecting interest drift: Detecting trends in user interest changes through a sliding window mechanism; Interest decay: Set a time decay factor for users' historical interest characteristics to enhance the influence weight of recent interests.
5. The knowledge graph enhanced dynamic RAG learning resource recommendation method according to claim 1, characterized in that, Personalized recommendation results are generated using a three-layer filtering mechanism, as follows: The first layer is based on knowledge graph correlation filtering, retaining resources that are directly or indirectly related to the user's current learning topic; The second layer filters resources based on user interest models, selecting those that match users' learning preferences and ability levels. The third layer filters based on resource quality scores, eliminating low-quality or outdated learning resources.
6. The dynamic RAG learning resource recommendation method based on knowledge graph enhancement according to any one of claims 1 to 5, characterized in that, The specific feedback and optimizations are as follows: Explicit feedback collection: Collecting user ratings and tags for recommended resources; Implicit feedback analysis: Analyzing user behavior data such as clicks, dwell time, and completion rate; Strategy adjustment: Based on feedback data, reinforcement learning algorithms are used to dynamically adjust retrieval weights and recommendation strategies.
7. A dynamic RAG learning resource recommendation system based on knowledge graph enhancement, characterized in that, The system includes: The knowledge graph construction module is used to build a domain knowledge graph that includes learning resource entities, concept entities, and the relationships between them. The resource preprocessing module is used to perform structured parsing of learning resources, extract resource features, and establish mappings with knowledge graph entities; The multimodal retrieval module combines semantic vector retrieval, keyword retrieval, and knowledge graph path retrieval to obtain candidate resources relevant to the user's query. The personalized interest modeling module is used to build user interest models based on user learning trajectories and interaction feedback; The recommendation generation module is used to integrate candidate resource features, knowledge graph correlation, and user interest models to generate personalized recommendation results. The feedback optimization module is used to dynamically adjust the retrieval strategy and interest model parameters based on user feedback on the recommendation results.
8. The dynamic RAG learning resource recommendation system based on knowledge graph enhancement according to claim 7, characterized in that, The knowledge graph construction module includes: The entity extraction submodule is used to extract knowledge points, concepts, and skill entities from learning resources using the BERT-BiLSTM-CRF model; The relation extraction submodule is used to identify hierarchical relationships, prerequisite relationships, and association relationships between entities using a deep learning model based on an attention mechanism. The graph update submodule is used to support incremental graph updates based on new learning resources to maintain the timeliness of knowledge. The multimodal retrieval module includes, The semantic vector retrieval submodule is used to generate text vectors using a domain-fine-tuned BERT model and perform similarity matching through a vector database. The keyword retrieval submodule is used to perform weighted keyword matching by combining the domain thesaurus to ensure accurate retrieval of professional terms; The knowledge graph path retrieval submodule is used to perform multi-hop path retrieval in the knowledge graph based on user query entities and to mine indirectly related resources. The retrieval fusion submodule is used to adopt a dynamic weight fusion strategy to adjust the weight ratio of each retrieval result according to the query type and resource type; The personalized interest modeling module includes: The interest feature extraction submodule is used to extract interest features from users' learning time, resource type preferences, and knowledge mastery. The interest drift detection submodule is used to detect trends in user interest changes through a sliding window mechanism. The interest decay submodule is used to set a time decay factor for the user's historical interest features, thereby increasing the influence weight of recent interests; The recommendation generation module employs a three-layer filtering mechanism, as follows: The first layer is based on knowledge graph correlation filtering, retaining resources that are directly or indirectly related to the user's current learning topic; The second layer filters resources based on user interest models, selecting those that match users' learning preferences and ability levels. The third layer filters based on resource quality scores, eliminating low-quality or outdated learning resources. The feedback optimization module includes, The explicit feedback collection submodule is used to collect user ratings and tags for recommended resources; The implicit feedback analysis submodule is used to analyze user behavior data such as clicks, dwell time, and completion rate. The strategy adjustment submodule is used to dynamically adjust retrieval weights and recommendation strategies based on feedback data using reinforcement learning algorithms.
9. An electronic device, characterized in that, include: Memory and at least one processor; The memory contains computer programs; The at least one processor executes the computer program stored in the memory, causing the at least one processor to perform the knowledge graph-based dynamic RAG learning resource recommendation method as described in any one of claims 1 to 6.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that can be executed by a processor to implement the knowledge graph-based dynamic RAG learning resource recommendation method as described in any one of claims 1 to 6.