An intelligent review reminding method and system fusing a knowledge graph and a forgetting curve

Abstract

The application provides an intelligent review reminding method and system fusing a knowledge graph and a forgetting curve, and belongs to the technical field of intelligent learning. The intelligent review reminding method comprises the following steps: collecting learning data information of a user from a target data source, and preprocessing the learning data information to obtain preprocessed learning data; performing edge weight dynamic updating processing on a knowledge graph that has been completed and constructed by using the preprocessed learning data, and obtaining an updated knowledge graph; determining actual forgetting rates of each knowledge point of the user by using the updated knowledge graph, and forming a forgetting curve corresponding to the knowledge point according to the actual forgetting rates of each knowledge point of the user; setting a knowledge point review reminding priority ranking list and an associated review data package according to the forgetting curve corresponding to the knowledge point of the user, and pushing the knowledge point and the associated review data package to the user according to the knowledge point review reminding priority ranking list, so that review reminding is realized. The system comprises modules corresponding to the method steps.

Description

Technical Field

[0001] This invention proposes an intelligent review reminder method and system that integrates knowledge graphs and forgetting curves, belonging to the field of intelligent learning technology. Background Technology

[0002] Existing intelligent review reminder technologies are mainly divided into two categories: one is based on the forgetting curve model, which determines the review timing by modeling the learning time and memory intensity of individual knowledge points, but ignores the logical dependencies and relationships between knowledge points, resulting in the forgetting transmission of related knowledge points not being perceived and the review lacking systematicity; the other is based on knowledge graph mining of knowledge point relationships and recommending review content, but does not combine the forgetting law to dynamically adjust the reminder timing, which is prone to problems such as repeated reminders or omission of key forgotten knowledge points.

[0003] Meanwhile, the aforementioned methods struggle to match users' actual memory characteristics, resulting in poor accuracy and adaptability of review reminders to users' real-world situations. Therefore, there is an urgent need for an intelligent review reminder method that deeply integrates knowledge graphs and forgetting curves to address these technical shortcomings. Summary of the Invention

[0004] This invention provides an intelligent review reminder method and system that integrates knowledge graphs and forgetting curves to solve the technical problems existing in the prior art. The technical solution adopted is as follows: A smart review reminder method integrating knowledge graphs and forgetting curves, the smart review reminder method comprising: The learning data information of the user is collected from the target data source, and the learning data information is preprocessed to obtain the preprocessed learning data; The edge weights of the constructed knowledge graph are dynamically updated using the preprocessed learning data to obtain the updated knowledge graph. The updated knowledge graph is used to determine the actual forgetting rate of each knowledge point of the user, and a forgetting curve corresponding to each knowledge point is formed based on the actual forgetting rate of each knowledge point of the user. Based on the forgetting curve corresponding to the user's knowledge points, a priority list for knowledge point review reminders and associated review data packages are set, and the knowledge points and associated review data packages are pushed to the user according to the priority list to realize review reminders.

[0005] Furthermore, user learning data information is collected from the target data source, and the learning data information is preprocessed to obtain preprocessed learning data, including: Collect user learning data from the target data source; The user's learning data is subjected to duplicate data removal processing, and the learning data after duplicate data removal is standardized to obtain standardized learning data.

[0006] Furthermore, the target data source includes a knowledge point database, user learning behavior logs, and a review feedback interaction interface; and the collection of user learning data information from the target data source includes: Collect the knowledge points that the user has completed learning, their difficulty coefficients, logical types, and initial importance weights from the knowledge point database; Collect the learning duration and repetition frequency of the knowledge points from the user's learning behavior log; The review feedback interface collects the accuracy rate of answering questions and the confused knowledge points corresponding to the knowledge points in the user's learning process.

[0007] Furthermore, the preprocessed learning data is used to dynamically update the edge weights of the constructed knowledge graph to obtain the updated knowledge graph, including: Configure a node for each knowledge point, and map the knowledge point difficulty coefficient, knowledge point logic type, and initial importance weight of the knowledge point to the node attributes; Retrieve the knowledge point logical type coefficient corresponding to the knowledge point logical type of each node from the database; The single-node attribute factors for each node are set using the normalized knowledge point difficulty coefficient, knowledge point logical type coefficient, and initial importance weight; The node composite attribute factor between each pair of nodes is obtained by using the single-node attribute factors of each pair of nodes. Extract the learning duration and repetition count for each knowledge point from the user's learning behavior logs, and use this as the first parameter data. Extract the accuracy rate of answering questions and the confused knowledge points corresponding to the knowledge points from the review feedback interaction interface, and use them as the second parameter data; The edge weights between each pair of nodes are dynamically updated by combining the first parameter data and the second parameter data with the node comprehensive attribute factor between each pair of nodes. The knowledge graph after the edge weights are updated is the updated knowledge graph.

[0008] Furthermore, the edge weights between each pair of nodes are dynamically updated using the first parameter data and the second parameter data combined with the node comprehensive attribute factor between each pair of nodes, including: The first parameter data is normalized to obtain the normalized first parameter data for each node. Retrieve the second parameter data corresponding to each node, and determine whether the knowledge points corresponding to each pair of nodes are confusing knowledge points based on the second parameter data; Based on the judgment result of whether the knowledge points corresponding to each pair of nodes belong to confused knowledge points, the edge weights between each pair of nodes are dynamically updated using the first parameter data, the second parameter data, and the node comprehensive attribute factor between each pair of nodes.

[0009] Furthermore, the updated knowledge graph is used to determine the actual forgetting rate of each knowledge point for the user, and a forgetting curve corresponding to each knowledge point is formed based on the actual forgetting rate of each knowledge point for the user, including: Retrieve the time interval between the user's last review of the knowledge point corresponding to each node and the current time. The basic forgetting rate for each node is obtained by using the time interval between the user's last review of the knowledge point associated with that node and the current time. The actual forgetting rate of each node is obtained by combining the forgetting conduction coefficient within each node with the actual forgetting rate of the associated knowledge points of each node and the basic forgetting rate of each node. Using the actual forgetting rate of each user's node as the y-axis value and the time when the actual forgetting rate of each node is obtained as the x-axis value, a forgetting curve corresponding to the knowledge point of each node is formed.

[0010] Furthermore, the basic forgetting rate for each node is obtained by using the time interval between the user's last review of the knowledge point associated with that node and the current moment, including: Retrieve the user's first learning time and first test score for each knowledge point of each node to obtain the initial memory retention parameters for each node; Retrieve the time interval between the user's last review of the knowledge point corresponding to each node and the current time, as well as the current learning behavior decay coefficient β; The basic forgetting rate for each node is obtained by using the initial memory solidity parameter and the learning behavior decay coefficient β corresponding to each node.

[0011] Furthermore, by combining the forgetting conduction coefficient within each node and the actual forgetting rate of the associated knowledge points at each node with the baseline forgetting rate at each node, the actual forgetting rate for each node is obtained, including: Retrieve the edge weight of each edge corresponding to the knowledge point within each node; Retrieve the rate of change in the accuracy of answering questions for each node, as well as the rate of change in the accuracy of answering questions for the associated nodes that have an edge relationship with each node; The edge weights of each edge corresponding to the knowledge points within each node and the corresponding rate of change in problem-solving accuracy are weighted and averaged to obtain the sensitivity M of the related knowledge points. The actual forgetting rate F for each node is obtained by combining the sensitivity M of the related knowledge points within each node, the change rate of the accuracy rate for each node, and the baseline forgetting rate for each node. s .

[0012] Furthermore, based on the forgetting curve corresponding to the user's knowledge points, a priority list for knowledge point review reminders and associated review data packages are set. The knowledge points and associated review data packages are then pushed to the user according to the priority list to implement review reminders, including: Retrieve the forgetting curve corresponding to each knowledge point, and obtain the slope of the forgetting curve corresponding to each knowledge point based on the forgetting curve; Retrieve the average weight change rate of all edges corresponding to the node to which each knowledge point belongs; The forgetting intensity parameter of each knowledge point is obtained by taking the slope of the forgetting curve corresponding to each knowledge point and the average weight change rate of all edges corresponding to the node to which each knowledge point belongs. The review priority of each knowledge point is sorted according to the forgetting intensity parameter from high to low, and a priority list of knowledge point review reminders is generated. Retrieve the associated review data package for each knowledge point in the priority sorting list of knowledge point review reminders, and push the knowledge points and associated review data packages to the user according to the priority sorting list of knowledge point review reminders.

[0013] An intelligent review reminder system integrating knowledge graphs and forgetting curves, the intelligent review reminder system comprising: The data acquisition and processing module is used to collect the user's learning data information from the target data source, and to preprocess the learning data information to obtain preprocessed learning data. The knowledge graph update module is used to dynamically update the edge weights of the constructed knowledge graph using preprocessed learning data, and obtain the updated knowledge graph. The forgetting curve acquisition module is used to determine the actual forgetting rate of each knowledge point of the user using the updated knowledge graph, and to form the forgetting curve corresponding to each knowledge point based on the actual forgetting rate of each knowledge point of the user. The review reminder and data push module is used to set a priority sorting list of knowledge point review reminders and associated review data packages according to the forgetting curve of the user's knowledge points, and push the knowledge points and associated review data packages to the user according to the priority sorting list to realize review reminders.

[0014] Beneficial effects of this invention: This invention proposes an intelligent review reminder method and system that integrates knowledge graphs and forgetting curves. It achieves deep fusion of knowledge graphs and forgetting curves, overcoming the limitations of simple superposition in existing technologies by dynamically updating edge weights and calculating the actual forgetting rate through associative perception. This makes review reminders both systematically relevant and adaptable to forgetting patterns. Based on user learning data, the edge weights of the knowledge graph are dynamically optimized, and the forgetting curve is corrected by combining the relationships between knowledge points, effectively adapting to the memory characteristics of different users, thereby improving the accuracy of identifying review knowledge points and their adaptability to users. Simultaneously, through the forgetting transmission mechanism of associated knowledge points, it avoids knowledge system gaps caused by isolated reviews, enhances the relevance of reviews, and effectively reduces ineffective repetitive reminders and omissions of key knowledge points, improving review efficiency and user experience. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method described in this invention; Figure 2 This is a system block diagram of the system described in this invention. Detailed Implementation

[0016] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0017] This invention proposes an intelligent review reminder method that integrates knowledge graphs and forgetting curves, such as... Figure 1 As shown, the intelligent review reminder method includes: The learning data information of the user is collected from the target data source, and the learning data information is preprocessed to obtain the preprocessed learning data; The edge weights of the constructed knowledge graph are dynamically updated using the preprocessed learning data to obtain the updated knowledge graph. The updated knowledge graph is used to determine the actual forgetting rate of each knowledge point of the user, and a forgetting curve corresponding to each knowledge point is formed based on the actual forgetting rate of each knowledge point of the user. Based on the forgetting curve corresponding to the user's knowledge points, a priority list for knowledge point review reminders and associated review data packages are set, and the knowledge points and associated review data packages are pushed to the user according to the priority list to realize review reminders.

[0018] This includes collecting user learning data from a target data source, preprocessing the learning data to obtain preprocessed learning data, including: Collect user learning data from the target data source; The user's learning data is subjected to duplicate data removal processing, and the learning data after duplicate data removal is standardized to obtain standardized learning data.

[0019] The target data source includes a knowledge point database, user learning behavior logs, and a review feedback interaction interface; and the collection of user learning data information from the target data source includes: Collect the knowledge points that the user has completed learning, their difficulty coefficients, logical types, and initial importance weights from the knowledge point database; Collect the learning duration and repetition frequency of the knowledge points from the user's learning behavior log; The review feedback interface collects the accuracy rate of answering questions and the confused knowledge points corresponding to the knowledge points in the user's learning process.

[0020] The working principle of the above technical solution is as follows: Based on a multi-dimensional target data source, by collecting basic attributes of knowledge points, user learning behavior, and review feedback data, and through deduplication and standardization preprocessing, high-quality, uniformly formatted learning data is formed to provide reliable data support for subsequent processes; using the preprocessed multi-dimensional learning data, the edge weights of the knowledge graph are dynamically updated, enabling the knowledge graph to adapt to user learning behavior and memory feedback characteristics in real time, accurately depicting the changes in the strength of associations between knowledge points; relying on the updated knowledge graph containing personalized association information, the actual forgetting rate of each knowledge point is calculated by integrating the association relationships of knowledge points, and a forgetting curve that reflects the influence of individual user memory characteristics and knowledge point associations is constructed; based on the accurate forgetting curve, high-priority review knowledge points are selected and associated review data packages are generated, which are then pushed to users according to priority.

[0021] The effects of the above technical solution are as follows: The solution collects comprehensive learning data from multi-dimensional target data sources, and combines deduplication and standardization preprocessing to ensure the accuracy, completeness, and consistency of the data, providing high-quality data support for subsequent knowledge graph updates and forgetting rate calculations. Based on the preprocessed multi-dimensional learning data, the solution dynamically updates the edge weights of the knowledge graph, enabling the knowledge graph to reflect user learning behavior and memory feedback characteristics in real time, thus improving the personalized adaptability of the knowledge graph. Based on the updated knowledge graph, the solution determines the actual forgetting rate of knowledge points and forms a forgetting curve, ensuring that forgetting state modeling takes into account both knowledge point relationships and individual user characteristics, improving the accuracy of the forgetting curve. Based on the accurate forgetting curve, the solution generates a priority list of review reminders and associated review data packages, achieving personalized, targeted, and systematic review reminders, reducing invalid reminders, lowering the risk of forgetting key knowledge points, and improving review efficiency and user experience.

[0022] In one embodiment of the present invention, the edge weights of a constructed knowledge graph are dynamically updated using preprocessed learning data to obtain an updated knowledge graph, including: Configure a node for each knowledge point, and map the knowledge point difficulty coefficient, knowledge point logic type, and initial importance weight of the knowledge point to the node attributes; Extract the learning duration and repetition count for each knowledge point from the user's learning behavior logs, and use this as the first parameter data. Extract the accuracy rate of answering questions and the confused knowledge points corresponding to the knowledge points from the review feedback interaction interface, and use them as the second parameter data; The edge weights between each pair of nodes are dynamically updated by combining the first parameter data and the second parameter data with the node comprehensive attribute factor between each pair of nodes. The knowledge graph after the edge weights are updated is the updated knowledge graph.

[0023] The method of dynamically updating the edge weights between every two nodes using the first and second parameter data includes: The first parameter data is normalized to obtain the normalized first parameter data for each node. Retrieve the second parameter data corresponding to each node, and determine whether the knowledge points corresponding to each pair of nodes are confusing knowledge points based on the second parameter data; The edge weights between each pair of nodes are dynamically updated based on the judgment result of whether the knowledge points corresponding to each pair of nodes belong to confused knowledge points.

[0024] Specifically, when the knowledge points corresponding to any two nodes are confused knowledge points, the edge weight S between any two nodes is obtained using the following formula: S=K ij ×[1+exp(-|T c -N c )] -1 ×[1-0.5×(P 01 +P 02 )] Where S represents the edge weight between any two nodes; T c N represents the absolute difference in learning time after normalization between two nodes; c P represents the absolute difference in the number of repetitions after normalization for the two nodes. 01 and P 02 K represents the accuracy rate of answering questions for the knowledge points corresponding to the two nodes; ij This represents the node composite attribute factor between every two nodes; where, [1+exp(-|T c -N c)] -1 By using an exponential function to nonlinearly reinforce the rule that the smaller the input difference, the stronger the confusion association, the weights change smoothly with the difference value, which conforms to the gradual characteristics of cognitive ambiguity; [1-0.5×(P 01 +P 02 The calculation incorporates the average accuracy of two knowledge points; the lower the accuracy, the larger this value, further amplifying the confusion-related weighting. ij As a basic factor, the influence of the inherent attributes of knowledge points on confusion and association is corrected to ensure that the weights take into account both cognitive interference and the characteristics of the knowledge points themselves. If the knowledge points corresponding to any two nodes are not confusing knowledge points, then the edge weight S between any two nodes is obtained by the following formula: S=K ij ×(1-β)×[1-0.5×(P 01 ×T g01 / N 01 +P 02 ×T g02 / N 01 )] Where S represents the edge weight between any two nodes; β represents the learning behavior decay coefficient, initially set to 0.2, with a maximum value of 0.3. If the user's learning interval exceeds 7 days, β increases, and the increment gradient is 0.02; T g01 and T g02 These represent the normalized learning time for the knowledge points corresponding to the two nodes, respectively; N 01 and N 02 K represents the number of times the knowledge point corresponding to the two nodes is repeated. ij This represents the node composite attribute factor between every two nodes. [1-0.5×(P)] 01 ×T g01 / N 01 +P 02 ×T g02 / N 01 The effect of effective learning efficiency on association strength is quantified; the higher the efficiency, the larger the value, and the higher the weight of conventional associations. As the learning interval increases, β increases, and (1-β) decreases, indicating that the weight dynamically decreases over time, consistent with the objective law of memory decay. ij As a fundamental factor, it corrects the impact of the inherent logical attributes of knowledge points on conventional associations, ensuring that the weights align with the strength of the logical associations of the knowledge points themselves.

[0025] The node comprehensive attribute factor between every two nodes is a unified quantitative representation of the inherent characteristics of knowledge points. It is used to integrate three heterogeneous attributes—knowledge point difficulty coefficient, knowledge point logical type, and initial importance weight—into a single continuous value. This enables dynamic correction of the inherent attributes of knowledge points on the strength of node associations in the knowledge graph. A larger node comprehensive attribute factor between every two nodes indicates a stronger association between the knowledge points corresponding to the two nodes, and vice versa. Specifically, the node comprehensive attribute factor between every two nodes is obtained in the following way: Retrieve the knowledge point logic type coefficient corresponding to the knowledge point logic type of each node from the database. For example, concept type = 0.2, derivation type = 0.5, comprehensive application type = 0.8, etc. The larger the knowledge point logic type coefficient, the more complex the internal logical structure of the knowledge point. The difficulty coefficient, logical type coefficient, and initial importance weight of each node are normalized. Then, the normalized difficulty coefficient, logical type coefficient, and initial importance weight are used to set the single-node attribute factor for each node. This single-node attribute factor is used for non-linear fusion to represent the difficulty, logical type, and initial importance of the knowledge point. A larger value indicates stronger comprehensive inherent attributes, higher coreity, and greater complexity of the knowledge point; a smaller value indicates a more basic and simpler knowledge point with weaker inherent influence. Furthermore, the single-node attribute factor K = 1 + exp[-(D×L] 0.5 +w×D 0.5 ]; where D represents the normalized knowledge point difficulty coefficient of each node; L represents the normalized knowledge point logic type coefficient of each node; w represents the normalized initial importance weight of each node; The node composite attribute factor K between any two nodes is obtained by using the single-node attribute factors of each pair of nodes. ij =(K i ×K j ) / (1+K i +K j ); where K i and K j These represent the single-node attribute factors corresponding to the i-th and j-th nodes of the two nodes, respectively.

[0026] The working principle of the above technical solution is as follows: A dedicated node is configured for each knowledge point, mapping the knowledge point's difficulty coefficient, logical type, and initial importance weight to the node's inherent attributes; learning time, repetition count (first parameter data), and accuracy rate, as well as confused knowledge points (second parameter data), are extracted from user learning behavior logs and review feedback interaction interfaces; the first parameter data is normalized to eliminate dimensional differences and ensure parameter comparability; based on the confused knowledge point information in the second parameter data, it is determined whether the knowledge points corresponding to any two nodes are confused. For both confused and non-confused scenarios, differentiated formulas are used to calculate edge weights: in confused scenarios, the normalized difference in learning time, repetition count, and accuracy rate are integrated; in non-confused scenarios, the normalized learning time, repetition count, and dynamically adjusted learning behavior decay coefficient are combined, ultimately outputting an updated knowledge graph.

[0027] The core characteristic of confusing knowledge points is that users are prone to memory confusion due to differences in cognitive similarity and mastery levels, requiring quantification of the triggering conditions for cognitive confusion. The difference in learning time and the difference in repetition frequency represent the degree of learning investment in two knowledge points; the closer they are (|T... c -N c The smaller the value of the value, the more blurred the user's cognitive boundaries, and the higher the probability of confusion; conversely, the greater the difference in investment, the lower the possibility of confusion. The lower the accuracy, the weaker the user's grasp of the knowledge point, and the easier it is to confuse it with similar knowledge points; the higher the accuracy, the clearer the cognition, and the weaker the confusion association. The association between non-confusing knowledge points is a logical association (such as basic knowledge points - advanced knowledge points), and its strength is mainly affected by the efficiency of learning investment and the timeliness of memory. It is necessary to quantify the dynamic stability of the regular association. When it is a non-confusing knowledge point, the focus is on quantifying the effective mastery efficiency of a single round of learning, reflecting the user's actual mastery quality of the knowledge point. The higher the efficiency, the more stable the regular association. The learning behavior decay coefficient follows the Ebbinghaus forgetting curve. The longer the learning interval, the more obvious the memory decay, and the lower the association strength between knowledge points. Increasing β (maximum 0.3) can dynamically correct this decay.

[0028] The effects of the above technical solution are as follows: It employs a scenario-based weight calculation logic, adapting to both confused and non-confusing knowledge point association types, making edge weight updates more closely aligned with the actual association characteristics of knowledge points; it integrates learning behavior data and review feedback data, avoiding the limitations of single-parameter driven approaches and improving the accuracy of edge weights in depicting user learning states; it unifies parameter scales through normalization processing, ensuring the effective integration of multi-dimensional data and improving the accuracy and reliability of edge weight calculation; it introduces a dynamically adjusted learning behavior decay coefficient to adapt to changes in user learning intervals, enabling edge weights to reflect the timeliness of knowledge point associations in real time; and its dynamic edge weight updates allow the knowledge graph to accurately capture the dynamic changes in the strength of knowledge point associations, effectively improving the matching between edge weights and users' actual knowledge learning situations. Simultaneously, through node comprehensive attribute factors, it normalizes and non-linearly integrates three heterogeneous attributes—knowledge point difficulty coefficient, logic type coefficient, and initial importance weight—to uniformly quantify and represent the inherent characteristics of knowledge points, achieving effective integration of heterogeneous attributes and dynamic correction of association strength. This solution can improve the representation accuracy of knowledge graphs, strengthen the forgetting transmission and influence relationship between related knowledge points, provide reliable data support for subsequent knowledge point forgetting rate calculation and personalized review strategy generation, improve the rationality and pertinence of learning recommendations, and enhance the adaptive ability and learning effect optimization efficiency of the learning system.

[0029] One embodiment of the present invention utilizes an updated knowledge graph to determine the actual forgetting rate corresponding to each knowledge point of a user, and forms a forgetting curve corresponding to each knowledge point based on the actual forgetting rate of each knowledge point of the user, including: Retrieve the time interval between the user's last review of the knowledge point corresponding to each node and the current time. The basic forgetting rate for each node is obtained by using the time interval between the user's last review of the knowledge point associated with that node and the current time. The actual forgetting rate of each node is obtained by combining the forgetting conduction coefficient within each node with the actual forgetting rate of the associated knowledge points of each node and the basic forgetting rate of each node. Using the actual forgetting rate of each user's node as the y-axis value and the time when the actual forgetting rate of each node is obtained as the x-axis value, a forgetting curve corresponding to the knowledge point of each node is formed.

[0030] Specifically, the basic forgetting rate for each node is obtained by using the time interval between the user's last review of the knowledge point associated with that node and the current moment, including: Retrieve the user's initial learning time and initial test score for each knowledge point at each node to obtain the initial memory retention parameter F=K×T for each node. gc ×X c Among them, T gcThis represents the initial learning time of the knowledge point associated with each node after normalization; X c This represents the initial test score after normalization; K represents the corresponding single-node attribute factor in each node. Retrieve the time interval between the user's last review of the knowledge point corresponding to each node and the current time, as well as the current learning behavior decay coefficient β; The basic forgetting rate F0 = C + F × e is obtained by using the initial memory retention parameter and the learning behavior decay coefficient β corresponding to each node. -β×Tk / Tz Where C represents the preset minimum forgetting rate, which is fixed at 0.1. It is a pre-set constant that does not change with the user, knowledge point, time, or learning behavior. It is a fixed value in the entire calculation logic to reflect that there is an inherent 0.1 forgetting rate for memorized knowledge points, because humans cannot remember all knowledge points forever. This makes the model more in line with the laws of real memory and avoids calculation anomalies. Tk represents the time interval between the user's last review of the knowledge point corresponding to each node and the current time. Tz represents the preset memory decay period, which ranges from 10 to 14 days.

[0031] The working principle of the above technical solution is as follows: The time interval between the last review of each knowledge point and the current moment is retrieved as the temporal basis for calculating the forgetting rate; combined with the first learning time and first test score after normalization of the knowledge point, the initial memory solidity parameter is calculated; then, the dynamic learning behavior decay coefficient, time interval, preset minimum forgetting rate, and memory decay cycle are incorporated, and the basic forgetting rate is calculated through an exponential model; based on the forgetting conduction coefficient of nodes in the updated knowledge graph, the actual forgetting rate of the corresponding knowledge point is correlated, and the basic forgetting rate is corrected to obtain the actual forgetting rate that integrates the correlation effects; with the actual forgetting rate on the vertical axis and the acquisition time on the horizontal axis, a unique forgetting curve for each knowledge point is constructed to intuitively present the memory decay pattern.

[0032] The effects of the above technical solution are as follows: The solution integrates initial learning data and a dynamic decay coefficient into the basal forgetting rate, taking into account both initial memory quality and the impact of learning intervals, thus improving the accuracy of depicting the basal forgetting state. It introduces the forgetting conduction coefficient and the forgetting rate of related knowledge points from the knowledge graph, breaking through the limitations of single-knowledge-point forgetting modeling and realizing the dynamic conduction and superposition of related forgetting, making the actual forgetting rate more consistent with the memory logic of the knowledge system. The personalized forgetting curve is constructed based on the user's actual learning sequence and related characteristics, accurately reflecting individual differences in memory decay and providing a scientific basis for subsequent review priority setting. The forgetting rate calculation process integrates multi-dimensional data and knowledge graph related information, avoiding the limitations of static models and improving the dynamic adaptability and reliability of forgetting state modeling. Simultaneously, incorporating single-node attribute factors into the calculation of the initial memory solidity parameter F enables deep quantification of the inherent attributes of knowledge points and user learning behavior. The single-node attribute factors have integrated the inherent characteristics of knowledge point difficulty, logical type, and initial importance, objectively reflecting the ease and core nature of memorizing the knowledge point itself. Using this method to determine the initial memory retention parameter allows the parameter to no longer rely solely on learning time and test scores, but to simultaneously reflect the impact of the knowledge point's own attributes on the initial memory effect, making the parameter more aligned with the objective memory patterns of different knowledge points. Therefore, it significantly improves the accuracy and rationality of initial memory retention assessment, avoids biases caused by using a uniform calculation standard for knowledge points with different attributes, and provides a more realistic and robust initial benchmark for subsequent construction of basal forgetting rate, actual forgetting rate, and forgetting curve, thereby enhancing the scientific rigor and adaptability of the entire forgetting assessment model.

[0033] One embodiment of the present invention uses the forgetting conduction coefficient corresponding to each node and the actual forgetting rate of the associated knowledge points corresponding to each node, combined with the basic forgetting rate corresponding to each node, to obtain the actual forgetting rate of each node, including: Retrieve the edge weight of each edge corresponding to the knowledge point within each node; The algorithm retrieves the rate of change in the accuracy of answering questions for each node, as well as the rate of change in the accuracy of answering questions for related nodes that have an edge relationship with each node. The rate of change in accuracy is calculated by dividing the difference between the average accuracy of the nodes answering questions for the same knowledge point in the current time period and the previous time period by the absolute difference between the accuracy of the previous time period and the zero-prevention constant. The zero-prevention constant is specifically a constant greater than 1. Furthermore, the knowledge point to which the related nodes that have an edge relationship with each node belong is the related knowledge point corresponding to each node. The edge weights of each edge corresponding to the knowledge points within each node and the corresponding rate of change in problem-solving accuracy are weighted and averaged to obtain the sensitivity M of the related knowledge points. The actual forgetting rate F for each node is obtained by combining the sensitivity M of the related knowledge points within each node, the change rate of the accuracy rate for each node, and the baseline forgetting rate for each node. s Furthermore, the actual forgetting rate F corresponding to each node. s =min[F0×(1-MB),1.0]; where M represents the sensitivity of the related knowledge points corresponding to the knowledge points within each node; B represents the rate of change of the accuracy of answering questions corresponding to each node; and F0 represents the basic forgetting rate.

[0034] The working principle of the above technical solution is as follows: retrieve the edge weight (knowledge graph association strength) corresponding to each node, and the rate of change of the accuracy of answering questions between the node and the associated nodes (dynamic feedback of memory state); perform a weighted average of the edge weight and the rate of change of the accuracy of answering questions of the corresponding associated nodes to quantify the influence of associated knowledge points on the current node and obtain the sensitivity M of associated knowledge points; take the basic forgetting rate F0 as the benchmark, incorporate the sensitivity M of associated knowledge points and the rate of change of the accuracy of answering questions of the current node B, calculate the actual forgetting rate through a multiplication correction formula, and use the min function to limit the maximum value to 1.0 to ensure the reasonableness of the result.

[0035] The effects of the above technical solution are as follows: It integrates the edge weights of the knowledge graph with the rate of change in question-answering accuracy, combining the strength of knowledge point associations with dynamic feedback on memory status. This allows the actual forgetting rate to reflect the impact of associated knowledge points, overcoming the limitations of single forgetting modeling. By calculating the sensitivity M of associated knowledge points through weighted averaging, it accurately quantifies the degree to which associations affect the current forgetting state of knowledge points, improving the targeting of forgetting rate calculations. The introduction of the rate of change in question-answering accuracy B adapts to the dynamic fluctuations of the user's memory status in real time, making the actual forgetting rate more closely aligned with the user's latest learning outcomes. The use of a min function constrains the upper limit of the actual forgetting rate, preventing calculation results from exceeding a reasonable range and ensuring the reliability of subsequent review priority setting. It achieves deep integration of basic forgetting rate, associated influence, and dynamic memory feedback, improving the accuracy and dynamic adaptability of forgetting state modeling and providing a scientific basis for personalized review reminders.

[0036] In one embodiment of the present invention, a priority list for knowledge point review reminders and associated review data packages are set according to the forgetting curve corresponding to the user's knowledge points, and the knowledge points and associated review data packages are pushed to the user according to the priority list to realize review reminders, including: Retrieve the forgetting curve corresponding to each knowledge point, and obtain the slope of the forgetting curve corresponding to each knowledge point based on the forgetting curve; Retrieve the average weight change rate of all edges corresponding to the node to which each knowledge point belongs; The forgetting intensity parameter E = (K × RS) for each knowledge point is obtained by using the slope of the forgetting curve corresponding to each knowledge point and the average rate of change of weight of all edges corresponding to the node to which each knowledge point belongs. 0.5 Where K represents the slope of the forgetting curve for each knowledge point; RS represents the average rate of change of weight of all edges corresponding to the node to which each knowledge point belongs; The review priority of each knowledge point is sorted according to the forgetting intensity parameter from high to low, and a priority list of knowledge point review reminders is generated. Retrieve the associated review data package for each knowledge point in the priority sorting list of knowledge point review reminders, and push the knowledge points and associated review data packages to the user according to the priority sorting list of knowledge point review reminders.

[0037] The working principle of the above technical solution is as follows: The slope of the forgetting curve for each knowledge point is extracted to quantify the instantaneous rate of change in forgetting of the knowledge point itself; the average weight change rate of all edges belonging to the node of each knowledge point is calculated to characterize the overall trend of the change in the association strength between the knowledge point and related knowledge points; a square root fusion formula is used to combine the slope of the forgetting curve with the average edge weight change rate to calculate the forgetting intensity parameter E, which comprehensively reflects the forgetting speed combined with the influence of association changes; the forgetting intensity parameter E is sorted from high to low to generate a priority list, ensuring that knowledge points with high forgetting urgency and significant association influence are recommended first; the associated review data packages of each knowledge point in the priority list are retrieved and pushed to the user according to the sorting results to complete the review reminder operation.

[0038] The above technical solution achieves the following effects: It integrates the dynamic slope of the forgetting curve with the average rate of change of edge weights, overcoming the limitations of judging review priority from a single dimension, and making priority assessment more comprehensively reflect the forgetting status and related characteristics of knowledge points; it uses a square root fusion method to balance the influence of the two core indicators, avoiding misjudgment of priority caused by extreme values ​​of a single indicator, and improving the rationality of the ranking results; the forgetting intensity parameter is directly related to the forgetting speed and the change in the strength of the association of knowledge points, so that the review priority accurately matches the urgency of forgetting and the importance of the association of knowledge points; it pushes related review data packages according to priority, ensuring that high-priority knowledge points are reviewed first, reducing invalid reminders and omissions of key knowledge points, and improving the targeting and efficiency of review; it achieves deep linkage between the dynamic characteristics of forgetting and the changes in the association of knowledge graphs, so that review reminders not only conform to the user's memory decay pattern, but also adapt to the dynamic changes in the association relationship of knowledge points, improving personalized adaptability.

[0039] This invention proposes an intelligent review reminder system that integrates knowledge graphs and forgetting curves, such as... Figure 2 As shown, the intelligent review reminder system includes: The data acquisition and processing module is used to collect the user's learning data information from the target data source, and to preprocess the learning data information to obtain preprocessed learning data. The knowledge graph update module is used to dynamically update the edge weights of the constructed knowledge graph using preprocessed learning data, and obtain the updated knowledge graph. The forgetting curve acquisition module is used to determine the actual forgetting rate of each knowledge point of the user using the updated knowledge graph, and to form the forgetting curve corresponding to each knowledge point based on the actual forgetting rate of each knowledge point of the user. The review reminder and data push module is used to set a priority sorting list of knowledge point review reminders and associated review data packages according to the forgetting curve of the user's knowledge points, and push the knowledge points and associated review data packages to the user according to the priority sorting list to realize review reminders.

[0040] The working principle of the above technical solution is as follows: Based on a multi-dimensional target data source, by collecting basic attributes of knowledge points, user learning behavior, and review feedback data, and through deduplication and standardization preprocessing, high-quality, uniformly formatted learning data is formed to provide reliable data support for subsequent processes; using the preprocessed multi-dimensional learning data, the edge weights of the knowledge graph are dynamically updated, enabling the knowledge graph to adapt to user learning behavior and memory feedback characteristics in real time, accurately depicting the changes in the strength of associations between knowledge points; relying on the updated knowledge graph containing personalized association information, the actual forgetting rate of each knowledge point is calculated by integrating the association relationships of knowledge points, and a forgetting curve that reflects the influence of individual user memory characteristics and knowledge point associations is constructed; based on the accurate forgetting curve, high-priority review knowledge points are selected and associated review data packages are generated, which are then pushed to users according to priority.

[0041] The effects of the above technical solution are as follows: The solution collects comprehensive learning data from multi-dimensional target data sources, and combines deduplication and standardization preprocessing to ensure the accuracy, completeness, and consistency of the data, providing high-quality data support for subsequent knowledge graph updates and forgetting rate calculations. Based on the preprocessed multi-dimensional learning data, the solution dynamically updates the edge weights of the knowledge graph, enabling the knowledge graph to reflect user learning behavior and memory feedback characteristics in real time, thus improving the personalized adaptability of the knowledge graph. Based on the updated knowledge graph, the solution determines the actual forgetting rate of knowledge points and forms a forgetting curve, ensuring that forgetting state modeling takes into account both knowledge point relationships and individual user characteristics, improving the accuracy of the forgetting curve. Based on the accurate forgetting curve, the solution generates a priority list of review reminders and associated review data packages, achieving personalized, targeted, and systematic review reminders, reducing invalid reminders, lowering the risk of forgetting key knowledge points, and improving review efficiency and user experience.

[0042] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An intelligent review reminder method integrating knowledge graphs and forgetting curves, characterized in that, The intelligent review reminder method includes: The learning data information of the user is collected from the target data source, and the learning data information is preprocessed to obtain the preprocessed learning data; The edge weights of the constructed knowledge graph are dynamically updated using the preprocessed learning data to obtain the updated knowledge graph. The updated knowledge graph is used to determine the actual forgetting rate of each knowledge point of the user, and a forgetting curve corresponding to each knowledge point is formed based on the actual forgetting rate of each knowledge point of the user. Based on the forgetting curve corresponding to the user's knowledge points, a priority list for knowledge point review reminders and associated review data packages are set, and the knowledge points and associated review data packages are pushed to the user according to the priority list to realize review reminders.

2. The intelligent review reminder method according to claim 1, characterized in that, The learning data information of the user is collected from the target data source, and the learning data information is preprocessed to obtain the preprocessed learning data, including: Collect user learning data from the target data source; The user's learning data is subjected to duplicate data removal processing, and the learning data after duplicate data removal is standardized to obtain standardized learning data.

3. The intelligent review reminder method according to claim 2, characterized in that, The target data source includes a knowledge point database, user learning behavior logs, and a review feedback interaction interface; and the user's learning data information collected from the target data source includes: Collect the knowledge points that the user has completed learning, their difficulty coefficients, logical types, and initial importance weights from the knowledge point database; Collect the learning duration and repetition frequency of the knowledge points from the user's learning behavior log; The review feedback interface collects the accuracy rate of answering questions and the confused knowledge points corresponding to the knowledge points in the user's learning process.

4. The intelligent review reminder method according to claim 1, characterized in that, The preprocessed learning data is used to dynamically update the edge weights of the constructed knowledge graph, resulting in an updated knowledge graph, including: Configure a node for each knowledge point, and map the knowledge point difficulty coefficient, knowledge point logic type, and initial importance weight of the knowledge point to the node attributes; Retrieve the knowledge point logical type coefficient corresponding to the knowledge point logical type of each node from the database; The single-node attribute factors for each node are set using the normalized knowledge point difficulty coefficient, knowledge point logical type coefficient, and initial importance weight; The node composite attribute factor between each pair of nodes is obtained by using the single-node attribute factors of each pair of nodes. Extract the learning duration and repetition count for each knowledge point from the user's learning behavior logs, and use this as the first parameter data. Extract the accuracy rate of answering questions and the confused knowledge points corresponding to the knowledge points from the review feedback interaction interface, and use them as the second parameter data; The edge weights between each pair of nodes are dynamically updated by combining the first parameter data and the second parameter data with the node comprehensive attribute factor between each pair of nodes. The knowledge graph after the edge weights are updated is the updated knowledge graph.

5. The intelligent review reminder method according to claim 4, characterized in that, The edge weights between each pair of nodes are dynamically updated using the first parameter data, the second parameter data, and the node comprehensive attribute factor between each pair of nodes, including: The first parameter data is normalized to obtain the normalized first parameter data for each node. Retrieve the second parameter data corresponding to each node, and determine whether the knowledge points corresponding to each pair of nodes are confusing knowledge points based on the second parameter data; Based on the judgment result of whether the knowledge points corresponding to each pair of nodes belong to confused knowledge points, the edge weight between each pair of nodes is dynamically updated by combining the first parameter data and the second parameter data with the node comprehensive attribute factor between each pair of nodes. Wherein, when the knowledge points corresponding to every two nodes are confused knowledge points, the edge weight between every two nodes is S=K. ij ×[1+exp(-|T c -N c )] -1 ×[1-0.5×(P 01 +P 02 )]; where S represents the edge weight between any two nodes; T c N represents the absolute difference in learning time after normalization between two nodes; c P represents the absolute difference in the number of repetitions after normalization for the two nodes. 01 and P 02 K represents the accuracy rate of answering questions for the knowledge points corresponding to the two nodes; ij This represents the node composite attribute factor between any two nodes; If the knowledge points corresponding to any two nodes are not confusing knowledge points, then the edge weight between any two nodes is S=K. ij ×(1-β)×[1-0.5×(P 01 ×T g01 / N 01 +P 02 ×T g02 / N 01 ]; where S represents the edge weight between any two nodes; β represents the learning behavior decay coefficient, initially set to 0.2, with a maximum value of 0.

3. If the user's learning interval exceeds 7 days, β increases, and the increment gradient is 0.02; T g01 and T g02 These represent the normalized learning time for the knowledge points corresponding to the two nodes, respectively; N 01 and N 02 K represents the number of times the knowledge point corresponding to the two nodes is repeated. ij This represents the node composite attribute factor between any two nodes.

6. The intelligent review reminder method according to claim 1, characterized in that, The updated knowledge graph is used to determine the actual forgetting rate of each knowledge point for a user, and a forgetting curve is generated for each knowledge point based on the actual forgetting rate of each knowledge point, including: Retrieve the time interval between the user's last review of the knowledge point corresponding to each node and the current time. The basic forgetting rate for each node is obtained by using the time interval between the user's last review of the knowledge point associated with that node and the current time. The actual forgetting rate of each node is obtained by combining the forgetting conduction coefficient within each node with the actual forgetting rate of the associated knowledge points of each node and the basic forgetting rate of each node. Using the actual forgetting rate of each user's node as the y-axis value and the time when the actual forgetting rate of each node is obtained as the x-axis value, a forgetting curve corresponding to the knowledge point of each node is formed.

7. The intelligent review reminder method according to claim 6, characterized in that, The base forgetting rate for each node is obtained by using the time interval between the user's last review of the knowledge points associated with that node and the current time, including: Retrieve the user's first learning time and first test score for each knowledge point of each node to obtain the initial memory retention parameters for each node; Retrieve the time interval between the user's last review of the knowledge point corresponding to each node and the current time, as well as the current learning behavior decay coefficient β; The basic forgetting rate for each node is obtained by using the initial memory solidity parameter and the learning behavior decay coefficient β corresponding to each node.

8. The intelligent review reminder method according to claim 6, characterized in that, The actual forgetting rate for each node is obtained by combining the forgetting conduction coefficient within each node, the actual forgetting rate of the associated knowledge points at each node, and the base forgetting rate for each node. This includes: Retrieve the edge weight of each edge corresponding to the knowledge point within each node; Retrieve the rate of change in the accuracy of answering questions for each node, as well as the rate of change in the accuracy of answering questions for the associated nodes that have an edge relationship with each node; The edge weights of each edge corresponding to the knowledge points within each node and the corresponding rate of change in problem-solving accuracy are weighted and averaged to obtain the sensitivity M of the related knowledge points. The actual forgetting rate F for each node is obtained by combining the sensitivity M of the related knowledge points within each node, the change rate of the accuracy rate for each node, and the baseline forgetting rate for each node. s .

9. The intelligent review reminder method according to claim 1, characterized in that, Based on the forgetting curve of the user's knowledge points, a priority list for knowledge point review reminders and associated review data packages are set. These knowledge points and associated review data packages are then pushed to the user according to the priority list to implement review reminders, including: Retrieve the forgetting curve corresponding to each knowledge point, and obtain the slope of the forgetting curve corresponding to each knowledge point based on the forgetting curve; Retrieve the average weight change rate of all edges corresponding to the node to which each knowledge point belongs; The forgetting intensity parameter of each knowledge point is obtained by taking the slope of the forgetting curve corresponding to each knowledge point and the average weight change rate of all edges corresponding to the node to which each knowledge point belongs. The review priority of each knowledge point is sorted according to the forgetting intensity parameter from high to low, and a priority list of knowledge point review reminders is generated. Retrieve the associated review data package for each knowledge point in the priority sorting list of knowledge point review reminders, and push the knowledge points and associated review data packages to the user according to the priority sorting list of knowledge point review reminders.

10. An intelligent review reminder system integrating knowledge graphs and forgetting curves, characterized in that, The intelligent review reminder system includes: The data acquisition and processing module is used to collect the user's learning data information from the target data source, and to preprocess the learning data information to obtain preprocessed learning data. The knowledge graph update module is used to dynamically update the edge weights of the constructed knowledge graph using preprocessed learning data, and obtain the updated knowledge graph. The forgetting curve acquisition module is used to determine the actual forgetting rate of each knowledge point of the user using the updated knowledge graph, and to form the forgetting curve corresponding to each knowledge point based on the actual forgetting rate of each knowledge point of the user. The review reminder and data push module is used to set a priority sorting list of knowledge point review reminders and associated review data packages according to the forgetting curve of the user's knowledge points, and push the knowledge points and associated review data packages to the user according to the priority sorting list to realize review reminders.

Images