Intelligent personalized topic recommendation method based on knowledge graph

Abstract

The invention discloses an intelligent personalized topic recommendation method based on a knowledge graph, and relates to the technical field of information retrieval, and the method comprises the steps: constructing a multi-dimensional subject knowledge graph which defines knowledge point entities and topic entities through the knowledge graph, and establishes structural relationships and capability dimension attributes between the entities; forming a structured knowledge basis for recommendation; collecting learning behavior data of the user, and generating a user knowledge state model for dynamically evaluating the knowledge state of the user in combination with the multi-dimensional subject knowledge graph; selecting a corresponding recommendation strategy according to an output result of the user knowledge state model, and performing multi-dimensional question matching based on capability dimension matching based on the multi-dimensional subject knowledge graph to generate a personalized question recommendation scheme; and dynamically optimizing the multi-dimensional subject knowledge graph, the recommendation strategy and the matching rule based on the feedback of the user to the recommendation scheme. The problems that a traditional recommendation system is shallow in knowledge association, rough in diagnosis and rigid in strategy are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information retrieval, in particular to an intelligent personalized question recommendation method based on a knowledge graph. BACKGROUND

[0002] At present, personalized learning is increasingly valued, and how to accurately recommend suitable exercises for students has become a core challenge in the field of educational technology. However, the existing question recommendation system does not perform well in practice, and the fundamental reason is that the system design is not deep and dynamic enough. Traditional systems mostly rely on isolated "knowledge point tags" for question matching. This simplified method fails to build and utilize the internal and structured association network between knowledge points, such as prerequisite dependency, ability level progression, and comprehensive application logic across knowledge points. This leads to a situation where the recommended results often do not match the students' real knowledge system, and the fundamental knowledge gaps that lead to learning difficulties cannot be diagnosed and filled. At the same time, the judgment of the user's knowledge state is also too rough, usually only based on the single indicator of correct answer rate, ignoring the deep information contained in multi-dimensional data such as answer time, specific types of errors, and review behavior, resulting in the system being unable to distinguish between students who have poor concept memory, calculation errors, or weak logical application, thus producing ineffective recommendations for "thousands of people". In addition, the existing recommendation strategy is often static and rigid, relying on pre-set fixed rules, which cannot dynamically adjust the difficulty and type of questions according to the students' learning progress, nor can it incorporate cognitive laws such as the Ebbinghaus forgetting curve to arrange reviews in a timely manner, making it difficult for the recommended content to keep up with the students' dynamically changing learning state. Finally, the matching dimension of the question and the user's demand is too narrow, often ignoring the core ability dimension of the question, the adaptability of the question type and the learning scene, and the rationality of the question amount, resulting in a serious mismatch between the recommended questions and the students' actual ability improvement goals and current learning tasks. These systematic defects collectively limit the effectiveness of existing recommendation systems, and cannot provide truly intelligent technical support for "teaching students in accordance with their aptitude". SUMMARY

[0003] In order to match the recommended questions with the students' actual ability improvement goals and current learning tasks in the process of personalized learning, the present application proposes an intelligent personalized question recommendation method based on a knowledge graph, which includes the following steps: S1: Constructing a multi-dimensional subject knowledge graph, the multi-dimensional subject knowledge graph is constructed by defining knowledge point entities and question entities, and according to the structural relationships and ability dimension attributes between the entities; S2: Collecting the user's learning behavior data and combining the multi-dimensional subject knowledge graph to generate a user knowledge state model for dynamically evaluating the user's knowledge state; S3: Select the corresponding recommendation strategy according to the output result of the user knowledge state model, and perform multi-dimensional question matching based on the ability dimension matching based on the multi-dimensional subject knowledge graph, and generate a personalized question recommendation scheme; S4: Based on the feedback of the user to the personalized question recommendation scheme, dynamically optimize the multi-dimensional subject knowledge graph, the recommendation strategy and the matching rule.

[0004] The present application fundamentally solves the problems of traditional recommendation system, such as shallow knowledge association, rough diagnosis and rigid strategy, by constructing a multi-dimensional knowledge graph, realizing dynamic user knowledge state diagnosis, generating personalized recommendation strategy and establishing a closed-loop optimization mechanism, and realizes the breakthrough from 'one face for thousands of people' to 'different for different people'.

[0005] Further, in the S1 step, the structural relationship includes at least the dependency relationship between knowledge points, and the mapping relationship between knowledge points and questions, and the ability dimension attribute is simultaneously associated to the knowledge point entity and the question entity, for representing the ability type of mastering the knowledge point / question.

[0006] Further, in the S1 step, the multi-dimensional subject knowledge graph is constructed by the following way: by integrating the multi-source data of textbooks, curriculum standards and question banks, and performing entity alignment and relationship extraction on the current multi-dimensional subject knowledge graph, the multi-dimensional subject knowledge graph is constructed and completed.

[0007] Further, in the S2 step, the dynamic evaluation is specifically: using a weighted model to fuse and calculate the data of the user's answer accuracy, answer time and error type, and output the mastery score of the user to each knowledge point.

[0008] Further, in the S2 step, the dynamic evaluation further includes: when evaluating the mastery score of any knowledge point, the mastery state of the upstream dependent knowledge point of the knowledge point is synchronously evaluated in combination with the dependency relationship in the multi-dimensional subject knowledge graph.

[0009] Further, in the S3 step, the multi-dimensional question matching is specifically knowledge point matching and question type matching, and the matching is performed using a graph database query language, and the query conditions include the target knowledge point, the upstream dependent knowledge point and the required ability dimension.

[0010] Further, in the S3 step, the recommendation strategy includes three types of recommendation strategies: weak supplement type, consolidation type and expansion type. When the weak supplement type strategy is selected, the basic questions of the target knowledge point and the questions of the upstream dependent knowledge points are preferentially recommended; When the consolidation type strategy is selected, the medium questions and easy-to-mistake questions of the target knowledge point are preferentially recommended; When choosing an expansion strategy, comprehensive questions targeting specific knowledge points and questions that connect different knowledge points are preferred.

[0011] Furthermore, in step S4, the feedback includes explicit evaluations of the recommended questions by the user, as well as implicit feedback obtained by statistically analyzing changes in the completion rate and accuracy of the recommended questions.

[0012] Furthermore, in step S4, the dynamic optimization includes: judging the recommendation effect based on implicit feedback, and if the effect does not meet the expected goal, automatically adjusting the difficulty or number of recommended questions in the recommendation strategy.

[0013] Compared with the prior art, the present invention has at least the following beneficial effects: (1) The present invention proposes an intelligent personalized question recommendation method based on knowledge graph. Based on a deep structured knowledge graph, it fully explores and utilizes the inherent connections and ability requirements between knowledge points, so that question recommendation is no longer limited to surface label matching, but is based on the reasoning of a complete knowledge system, thereby significantly improving the accuracy of recommendation and the ability to solve root problems. (2) By integrating multi-source data to construct a dynamic user knowledge status model, the knowledge mastery status and gap types of each user are identified, realizing the transformation from rough judgment to precise diagnosis, and changing the limitations of traditional recommendation; (3) A dynamic evolution recommendation strategy mechanism is adopted, which can adaptively adjust the recommended content and intensity according to the user's real-time learning status and cognitive patterns, ensuring the optimal planning of the learning path; (4) By establishing a complete feedback optimization loop, core components such as knowledge graphs and recommendation strategies can continuously improve themselves and enhance the recommendation effect. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the steps of a knowledge graph-based intelligent personalized question recommendation method. Detailed Implementation

[0015] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings. However, the present invention is not limited to these embodiments.

[0016] With the increasing popularity of personalized learning, existing question recommendation systems still face significant technical bottlenecks in practical applications. Traditional systems mostly rely on single-dimensional knowledge point tag matching mechanisms, lacking a deep understanding of the inherent logical structure of the knowledge system. They fail to identify the prerequisite dependencies and hierarchical progression between knowledge points, resulting in recommended content often being disconnected from students' actual knowledge structure. Furthermore, these systems diagnose users' knowledge status too crudely, typically relying solely on answer accuracy rates while ignoring the deeper information contained in multi-dimensional data such as answer time, error type characteristics, and learning behavior patterns. This prevents the system from accurately pinpointing the root causes of knowledge gaps, hindering truly personalized recommendations. In addition, existing systems generally employ static recommendation strategies, failing to dynamically adjust question difficulty and type based on user learning progress or incorporate forgetting curves from cognitive science to optimize review timing. This makes recommended content ill-suited to users' continuously changing learning states. More significantly, the matching dimensions between questions and user needs are extremely limited, often neglecting key factors such as the core competency dimensions tested by the questions, the suitability of question types for learning scenarios, and ultimately leading to unsatisfactory recommendation results. Therefore, this invention proposes an intelligent personalized question recommendation method based on knowledge graphs, such as... Figure 1 As shown, the method mainly includes the following steps: S1: Construct a multi-dimensional subject knowledge graph, which is constructed by defining knowledge point entities and question entities, and based on the structural relationships and ability dimension attributes between entities; S2: Collect user learning behavior data and combine it with multi-dimensional subject knowledge graphs to generate a user knowledge state model for dynamically evaluating user knowledge status; S3: Select the appropriate recommendation strategy based on the output of the user's knowledge state model and perform multi-dimensional question matching based on ability dimension matching based on multi-dimensional subject knowledge graph to generate a personalized question recommendation scheme; S4: Based on user feedback on personalized question recommendation schemes, dynamically optimize multi-dimensional subject knowledge graphs, recommendation strategies, and matching rules.

[0017] The core implementation logic of this invention begins with a newly constructed multi-dimensional subject knowledge graph. This knowledge graph breaks through the limitations of traditional single-knowledge-point labels, forming a deeply structured knowledge network foundation by defining knowledge-point entities, question entities, and the rich structural relationships between them. In specific implementation, the first step is to establish the ontology framework of the knowledge graph. This framework not only includes the dependencies between knowledge points—such as the Pythagorean theorem depending on the definition of a right triangle—but also defines the mapping relationships between knowledge points and questions, and labels each knowledge point and question with its corresponding ability dimension attributes. These ability dimensions include, but are not limited to, computational ability, logical reasoning ability, and spatial imagination ability. To construct such a comprehensive knowledge graph, the system needs to integrate multi-source information, including textbook catalogs, curriculum standards, examination outlines, and question bank data. For structured data, such as textbook chapter divisions, entities and relationships can be directly extracted and converted into the knowledge graph; for unstructured data, such as knowledge point associations in teacher lesson plans, natural language processing technology is required for relationship extraction. During data fusion, entity alignment technology ensures that the same knowledge point from different sources has a unified identifier, while graph-based reasoning algorithms can automatically discover and complete indirect relationships between knowledge points that are not explicitly labeled, thus forming a complete and logically rigorous knowledge system. This knowledge graph is not static but is designed with a dynamic update mechanism. When curriculum standards are revised, textbooks are updated, or the question bank is expanded, the system can automatically identify the subject attributes of the newly added content through preset subject adaptation plugins and integrate them into the corresponding knowledge graph instances, ensuring the timeliness and completeness of the knowledge system.

[0018] Then, by analyzing users' multi-dimensional learning behaviors, a dynamically evolving user knowledge state model is constructed. The recommendation system continuously collects users' answer data, learning behavior data, and feedback data through a data acquisition module integrated into the learning platform. Answer data includes not only traditional accuracy rates but also the time spent answering each question, and intelligent analysis of incorrect answers identifies error types such as conceptual misunderstandings, calculation errors, or logical application errors. Learning behavior data covers information such as the frequency with which users view explanations of knowledge points, the number of times they repeatedly practice the same question, and their current learning context. All this data is transmitted in real-time to the knowledge mastery assessment unit, which uses a weighted probability model to assign different weights to different dimensions of data, comprehensively calculating the user's mastery score for each knowledge point. More importantly, this assessment process is not conducted in isolation but fully considers the dependency network defined in the knowledge graph. When the system detects a user's weakness in a particular knowledge point, it automatically traces its upstream dependent knowledge points and performs correlation assessments, thereby accurately pinpointing the root cause of the knowledge gap. For example, when the system detects that a user performs poorly in "the application of the Pythagorean theorem," it will further assess their grasp of the fundamental knowledge of "the definition of a right triangle" to determine whether the problem stems from insufficient understanding of the theorem itself or a lack of basic concepts. Ultimately, the system outputs a comprehensive user knowledge status model that includes a knowledge point mastery matrix, knowledge gap type classifications, and ability requirement labels.

[0019] Based on the diagnosed user knowledge state, personalized recommendation decisions are then made. The core of this stage lies in intelligently selecting the most suitable recommendation strategy and executing precise question matching according to the user's current knowledge state. In this embodiment, three types of recommendation strategies are preset: a reinforcement strategy that prioritizes recommending basic questions on the target knowledge point and questions on its upstream dependent knowledge points; a consolidation strategy that prioritizes recommending medium-difficulty questions and error-prone questions on the target knowledge point; and an expansion strategy that prioritizes recommending comprehensive questions on the target knowledge point and questions related to cross-knowledge points.

[0020] The selection of strategies is automatically triggered by the user's knowledge state model. For example, when a user is identified to have a clear knowledge gap in a certain knowledge point, a remedial strategy will be activated; while when a user has just completed learning a knowledge point and is in the critical period of the forgetting curve, a reinforcement strategy will be activated. After determining the recommendation strategy, the powerful query capabilities of the graph database are used to match questions from multiple dimensions. This matching process not only considers the target knowledge point itself, but also associates its upstream dependent knowledge points according to the guidance of the knowledge graph, while strictly screening the ability dimensions and question types that meet the requirements of the strategy. For example, for users who need to strengthen their "geometric proof ability," the system will prioritize proof questions rather than calculation questions; for pre-exam review scenarios, it will focus on selecting comprehensive questions rather than single-knowledge-point questions. Through such multi-level screening, the system ensures that the recommended questions not only accurately match the knowledge points, but also comprehensively meet the user's needs in terms of ability dimensions, question type characteristics, and scenario adaptability.

[0021] After generating an initial recommendation scheme, the recommendation system intelligently sorts and interprets the questions. The questions are arranged according to increasing difficulty to ensure the user's learning experience aligns with cognitive patterns and avoids frustration from sudden jumps in difficulty. Each recommended question is accompanied by a clear reason, generated based on a knowledge graph and the user's knowledge state model, such as "This question is recommended to reinforce the application of the Pythagorean theorem that the user has not yet fully mastered" or "This question aims to improve the user's spatial imagination ability." This interpretable recommendation not only enhances the user experience but also increases their acceptance and trust in the recommendations. Furthermore, the number of recommended questions is dynamically adjusted based on the user's historical behavior data; for example, the number of questions recommended per session is appropriately reduced for users with low completion rates to ensure the practical feasibility of the recommendation scheme.

[0022] To ensure the continuous optimization and self-improvement of the recommendation system, this invention also designs a complete feedback and optimization closed loop. The recommendation system collects user feedback through both explicit and implicit channels: explicit feedback includes direct evaluations of recommended questions, such as "the question is too difficult" or "it doesn't match the knowledge points"; implicit feedback is obtained indirectly by analyzing behavioral data such as changes in user completion rate and accuracy. This feedback data, after being processed by natural language processing techniques and statistical analysis methods, is used to guide the dynamic optimization of the system.

[0023] The optimization process encompasses three levels: knowledge graph correction, recommendation strategy adjustment, and matching rule optimization. When the system detects inaccurate mappings between questions and knowledge points through feedback, it automatically corrects the correspondences in the knowledge graph. When a certain recommendation strategy proves ineffective, the system adjusts the question difficulty configuration within the strategy or introduces new question types. When a question type is found to be mismatched with the learning scenario, the scenario-question type matching rules are updated. This optimization process continues at different intervals to ensure that the system evolves over time, gradually improving recommendation quality and user satisfaction.

[0024] Finally, to achieve broad application across all disciplines and multiple learning scenarios, this invention also employs a flexible architectural design. By constructing a common core framework and a pluggable subject adaptation layer, it can quickly adapt to the specific needs of different disciplines. The subject adaptation layer stores subject-specific knowledge graph instances and matching rules. When the recommendation system serves a specific subject, it automatically loads the corresponding components without modifying the core code. In terms of deployment, the recommendation system also provides standardized application programming interfaces (APIs) that can be easily integrated into various existing educational platforms, including intelligent learning applications, online homework systems, and exam preparation platforms. It also supports both private and cloud deployment modes to meet the differentiated needs of different institutions in terms of data security and infrastructure.

[0025] In summary, this invention proposes an intelligent personalized question recommendation method based on knowledge graphs, achieving a fundamental improvement in the accuracy, personalization, and adaptability of question recommendation systems. Based on a deeply structured knowledge graph, it fully explores and utilizes the inherent connections and ability requirements between knowledge points, enabling question recommendation to move beyond surface label matching and instead be built upon complete knowledge system reasoning, thereby significantly improving the accuracy of recommendations and the ability to solve root problems. By integrating multi-source data to construct a dynamic user knowledge state model, it identifies each user's knowledge mastery and gap types, achieving a shift from coarse judgment to precise diagnosis, overcoming the limitations of traditional recommendations. A dynamically evolving recommendation strategy mechanism is adopted, which adaptively adjusts the recommended content and intensity according to the user's real-time learning status and cognitive patterns, ensuring optimal planning of the learning path. By establishing a complete feedback optimization loop, core components such as the knowledge graph and recommendation strategy can continuously improve themselves, constantly enhancing the recommendation effect.

[0026] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0027] Furthermore, in this invention, descriptions involving terms such as "first," "second," and "a" are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0028] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0029] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

Claims

1. A knowledge graph-based intelligent personalized question recommendation method, characterized in that, Including the following steps: S1: Construct a multi-dimensional subject knowledge graph, which is constructed by defining knowledge point entities and question entities, and based on the structural relationships and ability dimension attributes between entities; S2: Collect user learning behavior data and combine it with multi-dimensional subject knowledge graphs to generate a user knowledge state model for dynamically evaluating user knowledge status; S3: Select the appropriate recommendation strategy based on the output of the user's knowledge state model and perform multi-dimensional question matching based on ability dimension matching based on multi-dimensional subject knowledge graph to generate a personalized question recommendation scheme; S4: Based on user feedback on personalized question recommendation schemes, dynamically optimize multi-dimensional subject knowledge graphs, recommendation strategies, and matching rules.

2. The intelligent personalized question recommendation method based on knowledge graph as described in claim 1, characterized in that, In step S1, the structural relationship includes at least the dependency relationship between knowledge points and the mapping relationship between knowledge points and questions. The ability dimension attribute is simultaneously associated with the knowledge point entity and the question entity to represent the ability type of mastering the knowledge point / question.

3. The intelligent personalized question recommendation method based on knowledge graph as described in claim 2, characterized in that, In step S1, the multi-dimensional subject knowledge graph is constructed in the following way: by integrating multi-source data from textbooks, curriculum standards and question banks, and performing entity alignment and relation extraction on the current multi-dimensional subject knowledge graph, the multi-dimensional subject knowledge graph is constructed and completed.

4. The intelligent personalized question recommendation method based on knowledge graph as described in claim 2, characterized in that, In step S2, the dynamic evaluation specifically involves using a weighted model to integrate and calculate the user's answer accuracy, answer time, and error type data, and outputting a score on the user's mastery of each knowledge point.

5. The intelligent personalized question recommendation method based on knowledge graph as described in claim 4, characterized in that, In step S2, the dynamic evaluation further includes: when evaluating the mastery score of any knowledge point, combining the dependency relationships in the multi-dimensional subject knowledge graph, simultaneously evaluating the mastery status of the upstream dependent knowledge points of that knowledge point.

6. The intelligent personalized question recommendation method based on knowledge graph as described in claim 2, characterized in that, In step S3, the multi-dimensional question matching specifically includes knowledge point matching and question type matching. The matching is performed using a graph database query language, and the query conditions include the target knowledge point, upstream dependent knowledge points, and required ability dimensions.

7. The intelligent personalized question recommendation method based on knowledge graph as described in claim 1, characterized in that, In step S3, the recommendation strategy includes three types: reinforcement, consolidation, and expansion. When choosing a strategy to address weaknesses, priority should be given to basic questions on the target knowledge point and questions on knowledge points that depend on it. When choosing a reinforcement strategy, priority should be given to medium-difficulty questions and questions that are prone to errors in the target knowledge points; When choosing an expansion strategy, comprehensive questions targeting specific knowledge points and questions that connect different knowledge points are preferred.

8. The intelligent personalized question recommendation method based on knowledge graph as described in claim 1, characterized in that, In step S4, the feedback includes explicit evaluations of the recommended questions by the user, as well as implicit feedback obtained by statistically analyzing changes in the completion rate and accuracy of the recommended questions.

9. The intelligent personalized question recommendation method based on knowledge graph as described in claim 8, characterized in that, In step S4, the dynamic optimization includes: judging the recommendation effect based on implicit feedback, and if the effect does not meet the expected goal, automatically adjusting the difficulty or number of recommended questions in the recommendation strategy.

Images