Personalized learning path recommendation system based on knowledge graph

Abstract

This invention relates to the field of personalized learning technology and discloses a personalized learning path recommendation system based on knowledge graphs. The system includes a user profiling module, a knowledge graph construction module, a path generation module, and an effectiveness evaluation module. The user profiling module acquires learners' user learning parameters; the knowledge graph construction module receives learning domain information, constructs a domain knowledge graph based on user learning parameters, sets the granularity of knowledge nodes, and performs association analysis on knowledge units to obtain the association weight values ​​of each knowledge node; the path generation module dynamically plans the learning path according to the association weight values ​​and adjusts it in real time during learning; the effectiveness evaluation module monitors learners' knowledge point mastery, learning progress deviation, and path completion rate, and provides early warnings for anomalies. This system can provide learners with learning paths tailored to individual differences, improving the relevance and flexibility of learning.

Description

Technical Field

[0001] This invention relates to the field of personalized learning technology, specifically a personalized learning path recommendation system based on knowledge graphs. Background Technology

[0002] With the rapid development of online education, learners are faced with a vast amount of learning resources. Finding suitable learning content from these resources has become a major challenge in improving learning efficiency. In traditional learning models, learners often have to follow a fixed curriculum, a "one-size-fits-all" approach that ignores individual differences in knowledge reserves, learning abilities, and learning pace. Some learners may struggle with advanced content due to weak initial knowledge, while others may waste time repeatedly learning previously mastered concepts.

[0003] In current educational settings, most learning platforms employ relatively simple recommendation mechanisms, typically based on learners' browsing history or lists of popular courses. This approach lacks in-depth analysis of the inherent logic of the knowledge system, easily leading to fragmented recommended learning content and an inability to form a coherent learning path. For example, when a learner studies a specific mathematical concept, the platform may only recommend a single course directly related to it, ignoring the connections between this concept and prerequisite concepts and subsequent applications, making it difficult for learners to construct a complete knowledge framework.

[0004] Traditional learning paths lack dynamic adjustment mechanisms. Once a learning plan is set, learners must complete it step by step. Even if they encounter difficulties in mastering knowledge points or deviate from the expected progress, the system struggles to respond promptly. Teachers or learning platforms cannot understand the learner's true learning status in real time; problems are often only discovered after periodic tests. By then, learners may have already accumulated significant knowledge gaps on incorrect paths, requiring more effort to fill in the gaps later.

[0005] Existing learning assessment methods also have limitations. Most assessments focus only on the final test scores, making it difficult to comprehensively reflect learners' actual mastery of various knowledge points. There is a lack of effective monitoring methods for abnormal situations that arise during the learning process, such as significant delays in learning progress or repeated errors on key knowledge points, preventing early warnings. This means learners may only realize their learning difficulties after problems have accumulated to a certain extent, impacting their overall learning experience.

[0006] In this era of knowledge explosion, academic knowledge systems are becoming increasingly complex, and the connections between knowledge points are becoming ever closer. Taking computer science as an example, from programming language fundamentals to algorithm design and artificial intelligence applications, each knowledge point is interconnected, and a gap in any link can hinder subsequent learning. In traditional learning models, learners need to independently organize these relationships, which places extremely high demands on their self-learning abilities. For those with weaker self-planning skills, it's easy to get lost in the complex knowledge network.

[0007] Existing learning outcome assessment mechanisms also have significant shortcomings. In most cases, assessments rely solely on periodic test scores. This method only reflects a learner's learning progress at a specific point in time and cannot track their mastery of individual knowledge points in real time. When learners deviate significantly from their learning progress or misunderstand knowledge points, the system fails to issue timely warnings, allowing problems to escalate. For example, in language learning, if learners are forced to engage in reading practice before their grammar foundation is solid, it will not only affect their reading comprehension but may also damage their confidence. This lack of dynamic monitoring and early warning makes it difficult to identify and correct problems in the learning process in a timely manner, hindering the improvement of learning efficiency. Summary of the Invention

[0008] The purpose of this invention is to provide a personalized learning path recommendation system based on knowledge graphs to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, this invention provides a personalized learning path recommendation system based on knowledge graphs, the system comprising:

[0010] The user profile module is used to obtain learners' user learning parameters.

[0011] The knowledge graph construction module is used to receive learning domain information, construct a domain knowledge graph based on user learning parameters, set the granularity of knowledge nodes, perform association analysis on knowledge units, and obtain the association weight value of each knowledge node.

[0012] The path generation module is used to dynamically plan the learning path based on the association weight values ​​of each knowledge node, and to adjust the recommended path in real time during the learning process.

[0013] The effectiveness evaluation module is used to monitor learners' mastery of knowledge points, deviations in learning progress, and completion rates of learning paths, and to issue early warnings for abnormalities.

[0014] Preferably, the learner's user learning parameters include: learning objectives, knowledge base level, learning style preferences, and historical learning duration.

[0015] Preferably, the received learning domain information is used to construct a domain knowledge graph based on user learning parameters, set the granularity of knowledge nodes, and perform correlation analysis on knowledge units, as detailed below:

[0016] It receives user learning parameters from learners, extracts knowledge units within the learning domain, sets the granularity of knowledge nodes, and collects the prerequisite dependencies, difficulty coefficients, and association frequencies of knowledge units at each knowledge node granularity.

[0017] By calculating the ratios of the pre-dependencies and post-extensions of knowledge units at each knowledge node granularity, the maximum and minimum values ​​of the difficulty coefficient, and the mean and variance of the association frequency, the knowledge association degree, difficulty difference degree, and frequency stability at each knowledge node granularity are obtained.

[0018] Preferably, the association analysis of the knowledge units is performed as follows:

[0019] The pre-dependencies of knowledge units within each knowledge node granularity, the user's knowledge base level, and the knowledge correlation degree of each knowledge node granularity are analyzed to obtain the pre-adaptation evaluation value of each knowledge node granularity.

[0020] The difficulty coefficients of knowledge units within each knowledge node granularity, user learning style preferences, and the degree of difficulty difference between each knowledge node granularity are analyzed to obtain the difficulty matching evaluation value for each knowledge node granularity.

[0021] The association frequency of knowledge units within each knowledge node granularity, the user's historical learning time, and the frequency stability of each knowledge node granularity are analyzed to obtain the learning efficiency evaluation value of each knowledge node granularity.

[0022] Preferably, the specific process for obtaining the association weight values ​​of each knowledge node is as follows:

[0023] Extract the pre-adaptation evaluation value, difficulty matching evaluation value, and learning efficiency evaluation value at the granularity of each knowledge node, and process them to obtain the association weight value of each knowledge node.

[0024] Preferably, the learning path is dynamically planned based on the association weight values ​​of each knowledge node, as detailed below:

[0025] The association weight value of each knowledge node is compared with the knowledge node association weight threshold stored in the database;

[0026] If the association weight value of a knowledge node is higher than the knowledge node association weight threshold stored in the database, the prior adaptation evaluation value of the knowledge node is compared with the first benchmark value of the prior adaptation evaluation value. If the prior adaptation evaluation value of the knowledge node is higher than the first benchmark value of the prior adaptation evaluation value, the learning order of the knowledge node is adjusted. If it is lower than or equal to the first benchmark value of the prior adaptation evaluation value, the learning order of the knowledge node does not need to be adjusted.

[0027] The difficulty matching assessment value of the knowledge node is compared with the second benchmark value of the difficulty matching assessment value. If the difficulty matching assessment value of the knowledge node is higher than the second benchmark value of the difficulty matching assessment value, the learning resource type of the knowledge node is adapted and adjusted. If it is lower than or equal to the second benchmark value of the difficulty matching assessment value, the learning resource type of the knowledge node does not need to be adjusted.

[0028] The learning efficiency evaluation value of the knowledge node is compared with the third benchmark value of the learning efficiency evaluation value. If the learning efficiency evaluation value of the knowledge node is higher than the third benchmark value of the learning efficiency evaluation value, the allocation time of the knowledge node is dynamically adjusted. If it is lower than or equal to the third benchmark value of the learning efficiency evaluation value, there is no need to adjust the allocation time of the knowledge node.

[0029] If the association weight value of a knowledge node is lower than or equal to the knowledge node association weight threshold stored in the database, then dynamic programming is not required.

[0030] After dynamic programming is completed, the association weight value of the knowledge node is recalculated and compared with the knowledge node association weight threshold stored in the database to determine whether dynamic programming is complete.

[0031] The learning order, learning resource type, and number of times the allocated time for knowledge nodes are adjusted are recorded. When the number of adjustments reaches the specified number, an early warning is issued.

[0032] Preferably, the recommended path is adjusted in real time during the learning process, specifically as follows:

[0033] During the learning process, learners' real-time learning data is tracked, and the mastery of knowledge nodes is continuously monitored. If the mastery of a knowledge node is detected to be lower than the preset mastery threshold, the path branch adjustment will be automatically triggered.

[0034] Preferably, the process of monitoring learners' mastery of knowledge points, deviations in learning progress, and completion rates of learning paths, and issuing early warnings for anomalies, is as follows:

[0035] When the mastery of a knowledge point is detected to be lower than the safe threshold, additional learning resources are recommended and a prompt is issued. When the deviation from the learning progress is detected to exceed the safe threshold, the time allocated to subsequent nodes is adjusted and a prompt is issued. When the path completion rate is detected to be lower than the safe threshold, the learning plan is reordered and a prompt is issued.

[0036] Preferably, the association weight values ​​of each knowledge node are analyzed under the following specific conditions:

[0037] The knowledge relevance, difficulty difference, and frequency stability of each knowledge node are weighted and summed. The weight coefficient of knowledge relevance is determined according to the domain characteristics of the learning objective, the weight coefficient of difficulty difference is dynamically adjusted according to the user's knowledge level, and the weight coefficient of frequency stability is set according to the cumulative effect of historical learning time. The comprehensive relevance weight value is obtained through multi-level weighted calculation.

[0038] Preferably, the extraction of knowledge units within the learning domain is specifically analyzed as follows:

[0039] Initial knowledge units are extracted from a knowledge ontology annotated by domain experts to determine core concepts and basic terms.

[0040] By combining natural language processing technology, entity recognition and relation extraction are performed on unlabeled learning resources to supplement and expand knowledge units;

[0041] The initial extracted and expanded knowledge units are merged and deduplicated to obtain a structured set of knowledge units.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] By accurately capturing learners' learning parameters through the user profiling module, including their knowledge base, learning pace, and cognitive characteristics, the design of learning paths no longer relies on a uniform standard but is fully tailored to individual differences. When different learners face the same knowledge domain, they can obtain a starting point and progression method that matches their own situation, avoiding the phenomenon in traditional models where a uniform pace leads to some learners falling behind or having extra capacity.

[0044] The knowledge graph construction module performs in-depth processing of information in the learning domain. By setting the granularity of knowledge nodes and performing correlation analysis, it weaves scattered knowledge points into a network with inherent logic. This structured knowledge presentation method allows learners to clearly grasp the connections between knowledge points. For example, in physics learning, it helps learners build a complete knowledge framework, rather than memorizing individual knowledge points in isolation.

[0045] The dynamic planning function of the path generation module allows the learning path to be flexibly adjusted based on the learner's real-time performance. During the learning process, if a learner grasps a certain knowledge point quickly, the system will automatically accelerate the progress of that section or introduce higher-level related content; if difficulties arise in understanding, the pace will be slowed down in a timely manner, supplementing relevant basic content or changing the explanation method. This real-time adjustment mechanism ensures that the learning process is always at a pace most suitable for the learner's current state, avoiding passive learning caused by a fixed path.

[0046] The effectiveness evaluation module monitors multiple dimensions, including knowledge point mastery, learning progress deviation, and path completion rate, providing a comprehensive reflection of learners' learning status. When repeated errors are detected on a particular knowledge point, an early warning is issued, alerting the learner or mentor to pay attention and intervene. Similarly, when learning progress deviates significantly from expectations, warnings prompt learners to adjust their learning plans. This multi-dimensional monitoring and early warning system ensures that problems are identified promptly, preventing small issues from escalating into major obstacles. Furthermore, tracking path completion rate reflects learners' persistence and the rationality of their learning plans, providing a reference for optimizing learning strategies.

[0047] The entire system integrates personalization, dynamism, and comprehensive assessment, making the learning process more targeted and flexible. Learners are no longer bound to fixed course schedules but can explore knowledge along the path best suited to them under the system's guidance. They can steadily improve based on their existing foundation and receive timely adjustments and support when encountering difficulties, thereby enhancing the smoothness and effectiveness of the learning process and allowing each learner to improve their learning experience on a track that suits their individual characteristics. Attached Figure Description

[0048] Figure 1 This is a schematic diagram illustrating the working principle of the knowledge graph-based personalized learning path recommendation system described in this invention.

[0049] Figure 2 A flowchart for knowledge graph construction and association analysis;

[0050] Figure 3 A flowchart for knowledge unit association analysis;

[0051] Figure 4 A flowchart for dynamic programming of learning paths;

[0052] Figure 5 A flowchart for weighted calculation of associated weight values. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Please see Figure 1 This invention provides a personalized learning path recommendation system based on knowledge graphs, the system comprising:

[0055] The user profiling module obtains learners' learning parameters, which form the basis for personalized processing by the system. The knowledge graph construction module receives learning domain information and combines it with user learning parameters to construct a domain knowledge graph. During this process, it sets the granularity of knowledge nodes, performs correlation analysis on knowledge units, and obtains the correlation weight value of each knowledge node. The path generation module dynamically plans the learning path based on the correlation weight value of each knowledge node and can adjust the recommended path in real time during the learning process. The effect evaluation module is responsible for monitoring learners' mastery of knowledge points, deviations in learning progress, and path completion rate, and provides early warnings when abnormal situations occur.

[0056] Example 1:

[0057] Figure 2 A flowchart for knowledge graph construction and association analysis. (e.g.) Figure 2 The steps of receiving learning domain information, constructing a domain knowledge graph based on user learning parameters, setting knowledge node granularity, and performing association analysis on knowledge units in this embodiment are as follows:

[0058] The learner's user learning parameters encompass learning objectives, prior knowledge level, learning style preferences, and historical learning time. Learning objectives represent the learner's desired outcome in a specific learning domain, such as mastering calculus in advanced mathematics or learning basic Python syntax. This provides clear guidance for the system to determine the scope and direction of the learning content. Prior knowledge level reflects the learner's current knowledge base related to the learning domain, including the level of proficiency in concepts, principles, and related skills. For example, before learning linear algebra, the learner's grasp of basic matrix concepts in high school mathematics. Learning style preferences reflect the learner's preferred learning methods. Some learners prefer video lectures, while others prefer reading text materials and independent practice, and still others find group discussions more effective. Historical learning time records the total time spent on related learning activities in the past, as well as the time allocation across different knowledge units. For example, the total time spent on learning basic probability theory in the past month, and the average daily study time.

[0059] The specific process of receiving learning domain information, constructing a domain knowledge graph based on user learning parameters, setting the granularity of knowledge nodes, and performing correlation analysis on knowledge units is as follows:

[0060] First, the learner's user learning parameters are received, and based on these, knowledge units within the learning domain are extracted. A knowledge unit is the basic unit that constitutes the knowledge system of the learning domain; it may be a concept, a principle, or a skill operation step. After extracting the knowledge units, the granularity of the knowledge nodes is set. The setting of the granularity needs to comprehensively consider the characteristics of the learning domain and the user learning parameters. The granularity can be coarse, for example, using "function" as a single knowledge node, or fine, for example, using "definition of a linear function," "graphing a linear function," and "properties of a linear function" as different knowledge nodes.

[0061] At each knowledge node granularity, the prerequisite dependencies, difficulty coefficient, and association frequency of each knowledge unit are collected. Prerequisite dependencies refer to the prerequisites for learning a particular knowledge unit, such as the need to master other knowledge units before learning the "quadratic formula." For example, learning the knowledge unit "quadratic formula" requires the learner to master the "definition of a quadratic equation" and "square root calculation." The difficulty coefficient indicates the level of difficulty of a knowledge unit, typically assessed based on its abstractness, complexity, and the time learners usually need to understand it. It can be set between 0 and 1, with higher values ​​indicating greater difficulty. Association frequency refers to the number of times a knowledge unit appears simultaneously or sequentially with other knowledge units during the learning process. For example, in learning mathematics, the knowledge units "trigonometric functions" and "properties of triangles" often appear together in many learning scenarios, resulting in a relatively high association frequency between them.

[0062] After completing the above data collection, the relevant data at each knowledge node granularity are calculated and processed:

[0063] The knowledge relevance at each knowledge node granularity is calculated by ratioing the pre-dependencies to post-extensions of a knowledge unit. Pre-dependencies are the knowledge units that must be mastered before learning the current knowledge unit, while post-extensions are the knowledge units that can be further learned after mastering the current knowledge unit. The ratio reflects the tightness of the connections between knowledge units within the overall knowledge system. If the ratio of pre-dependencies to post-extensions is large at a certain knowledge node granularity, it indicates a relatively strong connection between the knowledge unit at that granularity and its preceding and following knowledge units.

[0064] The difficulty difference is calculated by taking the ratio of the maximum to the minimum difficulty coefficient of a knowledge unit. This ratio reflects the difference in difficulty between different knowledge units at the same knowledge node granularity. If the ratio of the maximum to the minimum difficulty coefficient of a knowledge unit at a certain knowledge node granularity is large, it indicates that the difficulty of the knowledge unit at that granularity fluctuates greatly, with some knowledge units being very difficult and others relatively easy.

[0065] Frequency stability is calculated by ratioing the mean to the variance of the association frequencies of knowledge units. The mean of the association frequency reflects the average number of associations between knowledge units, while the variance reflects the dispersion of the association frequency. The ratio of the two can reflect the stability of the association frequency of knowledge units. If the ratio of the mean to the variance is large, it indicates that the association frequency of knowledge units at that granularity is relatively stable with small fluctuations; conversely, it indicates that the association frequency fluctuates greatly and has poor stability.

[0066] Example 2:

[0067] Figure 3 This is a flowchart for knowledge unit association analysis. (Example) Figure 3 In this embodiment, when performing association analysis on knowledge units, it is necessary to proceed from multiple dimensions and combine the characteristics of user learning parameters and knowledge node granularity to obtain different evaluation values. The specific implementation steps are as follows:

[0068] This analysis examines the prerequisite dependencies of knowledge units within each knowledge node granularity, combining the user's prior knowledge level with the knowledge relevance of each knowledge node granularity to obtain a prerequisite fit assessment value. Prerequisite dependencies clearly define the other knowledge units that must be mastered before learning a particular knowledge unit, while the user's prior knowledge level reflects the learner's current relevant knowledge reserves. Knowledge relevance reflects the tightness of connection between knowledge units. During the analysis, if the user's prior knowledge level highly matches the prerequisite dependencies of a knowledge unit within a certain knowledge node granularity—meaning the learner has a good grasp of the prerequisite knowledge required for that unit—and the knowledge relevance of that knowledge node granularity is also high, it indicates that the knowledge units at that granularity are closely related to other knowledge units, making it easier for the learner to learn that granularity of knowledge units based on their current foundation. Therefore, the resulting prerequisite fit assessment value will be high. Conversely, if the user's prior knowledge level does not meet the prerequisite dependency requirements, or the knowledge relevance is low, the prerequisite fit assessment value will be low.

[0069] For each knowledge node granularity, the difficulty coefficient of the knowledge unit is analyzed in conjunction with the user's learning style preferences and the difficulty difference between each knowledge node granularity to derive a difficulty matching assessment value. The difficulty coefficient measures the ease or difficulty of the knowledge unit. User learning style preferences include learners' inclinations regarding knowledge presentation methods, learning pace, etc., while the difficulty difference reflects the difficulty fluctuation of different knowledge units within the same knowledge node granularity. When the difficulty coefficient of a knowledge unit matches the user's learning style preferences—for example, when a learner who prefers independent inquiry faces a moderately difficult knowledge unit that requires thinking—and the difficulty difference within that knowledge node granularity is within a reasonable range (i.e., the difficulty of the knowledge unit does not fluctuate too much and can adapt to the learner's style), the difficulty matching assessment value will be high. If the difficulty of the knowledge unit does not match the learning style, or the difficulty difference is too large, making it difficult for the learner to adapt, the difficulty matching assessment value will be low.

[0070] When analyzing the association frequency of knowledge units within each knowledge node granularity, a learning efficiency evaluation value is obtained by combining the user's historical learning time with the frequency stability of each knowledge node granularity. Association frequency refers to the number of times a knowledge unit appears simultaneously or sequentially with other knowledge units. Historical learning time reflects the learner's past learning effort, and frequency stability reflects the stability of the association frequency. If the association frequency of a knowledge unit is compatible with the learning rhythm reflected by the user's historical learning time—for example, if the learner's historical learning time is short, indicating a preference for rapid knowledge acquisition, and the association frequency of the knowledge unit is moderate and does not increase the learning burden due to excessive frequency—and the frequency stability of that knowledge node granularity is good, meaning the association frequency does not fluctuate frequently and maintains a relatively stable rhythm, then the learner will be more efficient in learning knowledge units at that granularity, and the learning efficiency evaluation value will be correspondingly higher. Conversely, if the association frequency does not match the learning rhythm, or if the frequency stability is poor, the learning efficiency evaluation value will decrease.

[0071] To obtain the association weight values ​​for each knowledge node, it is necessary to first extract the pre-adaptation evaluation value, difficulty matching evaluation value, and learning efficiency evaluation value within the granularity of each knowledge node. These evaluation values ​​reflect the fit between the granularity of the knowledge node and the learner, as well as the characteristics of the knowledge unit itself, from different perspectives. After extraction, these three evaluation values ​​are comprehensively processed, taking into account the degree of influence of each evaluation value in the association weight calculation, rather than simply adding them together. For example, for scenarios emphasizing basic knowledge learning, the pre-adaptation evaluation value may have a greater impact; while for scenarios emphasizing learning efficiency, the learning efficiency evaluation value may have a higher weight. Through this comprehensive processing, the association weight value of each knowledge node is finally obtained, which can comprehensively reflect the importance and fit of the knowledge node in personalized learning path recommendation.

[0072] Example 3:

[0073] Figure 4 A flowchart for dynamic programming of the learning path. For example... Figure 4 In this embodiment, the learning path is dynamically planned based on the association weight values ​​of each knowledge node. The specific implementation steps are as follows:

[0074] The association weight value of each knowledge node is compared with the knowledge node association weight threshold pre-stored in the database. This threshold is set based on a large amount of learning data and domain characteristics and is used to determine whether the knowledge node needs to be dynamically adjusted.

[0075] If the association weight value of a knowledge node is higher than the threshold, it indicates that the knowledge node has a high need for adjustment in the current learning scenario and requires further analysis. In this case, the prior adaptation evaluation value of the knowledge node is compared with the first benchmark value of the prior adaptation evaluation value. The first benchmark value of the prior adaptation evaluation value is determined comprehensively based on the prior dependencies of the knowledge node and the user's knowledge base level, and is used to measure whether the learner's current foundation is suitable for prioritizing the learning of this knowledge node. If the prior adaptation evaluation value is higher than the first benchmark value, it indicates that the learner's current knowledge base can well support the learning of this knowledge node, and the association adaptation degree of this knowledge node with other prior knowledge is high. The system will adjust the priority of the learning order of this knowledge node, placing it in a more prominent position. If the prior adaptation evaluation value is lower than or equal to the first benchmark value, it indicates that this knowledge node is not the most urgently needed content to be prioritized at the current stage, and there is no need to adjust its learning order.

[0076] The difficulty matching assessment value of the knowledge node is compared with the second benchmark value for difficulty matching assessment. The second benchmark value for difficulty matching assessment is set based on the difficulty coefficient of the knowledge unit and the user's learning style preferences, and is used to determine whether the current learning resource type matches the learner's style. If the difficulty matching assessment value is higher than the second benchmark value, it means that the existing learning resource type may not be able to fully adapt to the learner's preferences. For example, if a learner who prefers visual learning is faced with text-based resources, the system will adapt the learning resource type by adding visual resources such as charts and videos. If the difficulty matching assessment value is lower than or equal to the second benchmark value, it indicates that the current resource type matches the learner's style and no adjustment is required.

[0077] The learning efficiency assessment value of this knowledge node is compared with the third benchmark value for learning efficiency assessment. The third benchmark value for learning efficiency assessment is determined by combining the association frequency of the knowledge unit and the user's historical learning time, and is used to evaluate whether the currently allocated learning time is reasonable. If the learning efficiency assessment value is higher than the third benchmark value, it indicates that learning this knowledge node with the current time may be inefficient. For example, if the learner's historical learning time is short, but the time allocated to this node is too long, the system will dynamically shorten the allocated time. If the learner's historical learning time is long and the association frequency of this node is high, it may be necessary to extend the time to ensure learning effectiveness, and the time will also be adjusted in this case. If the learning efficiency assessment value is lower than or equal to the third benchmark value, it indicates that the current allocated time matches the learner's efficiency rhythm, and no adjustment is required.

[0078] If the association weight value of a knowledge node is lower than or equal to the knowledge node association weight threshold stored in the database, it means that the knowledge node is not well-suited to the learner in the current situation, or its association importance in the knowledge system is not high. In this case, no dynamic planning adjustment is needed, and the original learning arrangement can be maintained.

[0079] After dynamic programming is completed, the system recalculates the association weight value of the knowledge node. This is because the parameters of the knowledge node may change after adjustment, requiring a reassessment of its association weight. The recalculated association weight value is then compared with the threshold in the database. If the difference is within the allowable range and all evaluation values ​​meet the expected state, the dynamic programming is considered complete. If the difference exceeds the range or the evaluation values ​​do not meet the expectations, the dynamic programming process needs to be repeated until the conditions are met.

[0080] Meanwhile, the system records in real time the number of adjustments to the learning order, learning resource type, and allocated time for knowledge nodes, and each adjustment is counted in the statistics. When the number of adjustments reaches the prescribed limit, it indicates that there may be an anomaly with that knowledge node. For example, if frequent adjustments still fail to achieve the desired fit, the system will issue a warning to attract the attention of learners or relevant administrators, facilitating timely troubleshooting.

[0081] Throughout the dynamic programming process, the baseline values ​​involved are not fixed but dynamically updated based on the learner's real-time learning data and changes in the learning domain to ensure that the dynamic programming results always align with the learner's actual situation. The criteria for determining whether dynamic programming is complete can be expressed as follows:

[0082]

[0083] In the formula, The difference between the recalculated association weight value and the original threshold. For the recalculated association weight values, Associate weight thresholds with knowledge nodes stored in the database. When... When the difference is less than the preset allowable range, dynamic programming is considered complete.

[0084] Example 4:

[0085] The specific process of real-time adjustment of the recommended path during the learning process is as follows: The system continuously tracks various real-time data related to learning activities. For example, when a learner is studying "quadratic equations" on an online learning platform, the system records the duration of watching instructional videos, the accuracy rate of completing homework exercises, and the number of times they participate in discussion forums. Through real-time analysis of this data, the system can dynamically assess the learner's mastery of the knowledge node "solutions of quadratic equations." If the system detects that the learner's accuracy rate on exercises for this knowledge node is lower than a preset mastery threshold—for example, answering multiple practice questions incorrectly consecutively, with errors concentrated in the "discriminative application" sub-section—it indicates that the learner's mastery of this knowledge node is insufficient. In this case, the system will automatically trigger path branch adjustment, pushing specialized explanation videos and exercises for "discriminative application" to the learner, forming a new learning path branch to help them address their weaknesses.

[0086] The specific process for monitoring learners' mastery of knowledge points, deviations in learning progress, and completion rates of learning paths, and for issuing early warnings of anomalies, is as follows:

[0087] When the system detects that the mastery of a knowledge point is below the safe threshold, taking the learning of "subjunctive mood in English grammar" as an example, if the learner's accuracy rate in subjunctive mood tense collocation exercises is consistently below the safe threshold, the system will increase the recommendation of auxiliary learning resources, such as pushing subjunctive mood tense comparison tables, typical wrong question analysis videos, etc. At the same time, the system will send a prompt to the learner through platform messages, informing them that their current mastery of subjunctive mood knowledge points is not solid enough, and suggesting that they prioritize learning the recommended auxiliary resources.

[0088] When the system detects that the learning progress deviation exceeds the safety threshold, assuming the learner originally planned to complete the "TCP / IP Protocol" course within "Computer Network Fundamentals" within a week, including theoretical learning and simulated experiments, but by Wednesday the system finds that the learner has only completed 30% of the theoretical learning, far below the planned 50%, the system will adjust the allocated time for subsequent nodes. For example, it might appropriately shorten the theoretical learning time for the subsequent node "IP Address Classification" and extend its experimental operation time to balance the overall progress. Simultaneously, it will send a prompt to the learner, reminding them of the current delay and suggesting a more reasonable allocation of subsequent learning time.

[0089] When the system detects that the completion rate of a learning path falls below the safety threshold—for example, in a learning path for "Advanced Mathematics Calculus" where the original plan included 10 knowledge nodes, but the learner stopped progressing for two consecutive weeks after reaching the 5th node, resulting in a completion rate of only 40%, which is below the safety threshold—the system will trigger a rescheduling of the learning plan. The rescheduled plan will break down the remaining nodes into smaller learning units based on the learner's recent learning progress, allocating 1-2 units per day and setting daily learning reminders. Simultaneously, the system will send a notification to the learner explaining that the current path completion is unsatisfactory and that the learning plan has been rescheduled, suggesting that they gradually complete the learning according to the new plan.

[0090] Throughout the monitoring and early warning process, the system flexibly sets various safety thresholds based on the characteristics of different learning domains and individual learner differences. For example, for highly practical skills learning, such as "Python programming," the safety threshold for knowledge mastery may be set higher to ensure learners have sufficient practical skills; while for introductory-level theoretical knowledge learning, the safety threshold may be relatively lower, allowing learners a gradual understanding process. Simultaneously, the method of early warning notifications will be adjusted according to learner preferences; some learners prefer pop-up notifications, while others prefer email notifications, and the system will automatically select the appropriate notification method based on user settings.

[0091] Example 5:

[0092] Figure 5 A flowchart for weighted calculation of associated weight values. (e.g.) Figure 5 The specific analysis process of the association weight values ​​of each knowledge node and the extraction of knowledge units within the learning domain in this embodiment are as follows:

[0093] In the specific analysis of the association weight values ​​of each knowledge node, it is necessary to perform a weighted summation of the knowledge association degree, difficulty difference degree, and frequency stability at each knowledge node granularity. The weight coefficient for knowledge association degree is determined based on the domain characteristics of the learning objective. Different learning domains differ in their knowledge structures. For example, in the field of language learning, the association between vocabulary and grammar is particularly close, so the weight coefficient for knowledge association degree will be set relatively high; while in the field of art and design, the association between basic drawing skills and color theory is relatively flexible, and the weight coefficient for knowledge association degree may be lower.

[0094] The weighting coefficient for difficulty variation is dynamically adjusted based on the user's level of knowledge. If the user has a higher level of knowledge and can adapt to a wider range of knowledge difficulty, the weighting coefficient for difficulty variation will be appropriately reduced to allow for greater difficulty fluctuations within the knowledge node granularity. If the user has a lower level of knowledge and cannot cope with excessive difficulty changes, the weighting coefficient for difficulty variation will be increased to limit the difficulty variation within the knowledge node granularity and make it more in line with the user's acceptance level.

[0095] The weighting coefficient for frequency stability is set based on the cumulative effect of historical learning duration. When a user's historical learning duration is short, the learning rhythm is not yet stable, and the requirement for the stability of the frequency of association between knowledge units is high; therefore, the weighting coefficient for frequency stability is set higher. When a user's historical learning duration is long, a relatively stable learning rhythm has been formed, and the dependence on the stability of association frequency decreases, so the weighting coefficient for frequency stability is correspondingly lower. Through this multi-layered weighted calculation, the impact of knowledge relevance, difficulty difference, and frequency stability in different contexts is comprehensively considered to finally obtain the association weight value for each knowledge node.

[0096] When extracting knowledge units within a learning domain, the initial extraction is performed using a knowledge ontology annotated by domain experts. Based on their professional knowledge and teaching experience, domain experts annotate the core concepts and fundamental terms within the domain in the ontology. For example, in physics, concepts such as "definition of force," "Newton's three laws of motion," and "conservation of energy" are identified as core concepts, while terms like "acceleration" and "momentum" serve as fundamental terms. These initially extracted knowledge units constitute the basic framework of the domain's knowledge system.

[0097] This study combines natural language processing (NLP) techniques to perform entity recognition and relation extraction on unlabeled learning resources. These unlabeled resources include supplementary content for textbook chapters, academic papers, and teaching cases. Entity recognition algorithms within NLP are used to identify new entities relevant to the field, such as specific experimental methods or the latest research findings. Relation extraction algorithms are then used to analyze the connections between these new entities and already extracted knowledge units, such as the relationship between a particular experimental method and the application of Newton's second law, thereby supplementing and expanding the scope of knowledge units.

[0098] The system merges and deduplicates the initially extracted and expanded knowledge units. During the fusion process, the system compares the knowledge units from the two sources to check for content that is essentially the same but has different expressions. For example, "law of inertia" and "Newton's first law" refer to the same knowledge unit; in this case, the more standard expression is retained. Through this process, duplicate knowledge units are removed, resulting in a structured set of knowledge units. This set includes both the core foundational content of the domain and the new content added through expansion, comprehensively and systematically reflecting the knowledge system of this learning domain.

[0099] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0100] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A personalized learning path recommendation system based on knowledge graphs, characterized in that, include: The user profile module is used to obtain learners' user learning parameters. The knowledge graph construction module is used to receive learning domain information, construct a domain knowledge graph based on user learning parameters, set the granularity of knowledge nodes, perform association analysis on knowledge units, and obtain the association weight value of each knowledge node. The path generation module is used to dynamically plan the learning path based on the association weight values ​​of each knowledge node, and to adjust the recommended path in real time during the learning process. The effectiveness evaluation module is used to monitor learners' mastery of knowledge points, deviations in learning progress, and completion rates of learning paths, and to issue early warnings for abnormalities. The received learning domain information is used to construct a domain knowledge graph based on user learning parameters, set the granularity of knowledge nodes, and perform correlation analysis on knowledge units. The specific analysis is as follows: It receives user learning parameters from learners, extracts knowledge units within the learning domain, sets the granularity of knowledge nodes, and collects the prerequisite dependencies, difficulty coefficients, and association frequencies of knowledge units at each knowledge node granularity. By calculating the ratios of the pre-dependency and post-extension relationships of knowledge units at each knowledge node granularity, the maximum and minimum values ​​of the difficulty coefficient, and the mean and variance of the association frequency, the knowledge association degree, difficulty difference degree, and frequency stability at each knowledge node granularity are obtained. The association analysis of knowledge units is as follows: The pre-dependencies of knowledge units within each knowledge node granularity, the user's knowledge base level, and the knowledge correlation degree of each knowledge node granularity are analyzed to obtain the pre-adaptation evaluation value of each knowledge node granularity. The difficulty coefficients of knowledge units within each knowledge node granularity, user learning style preferences, and the degree of difficulty difference between each knowledge node granularity are analyzed to obtain the difficulty matching evaluation value for each knowledge node granularity. The association frequency of knowledge units within each knowledge node granularity, the user's historical learning time, and the frequency stability of each knowledge node granularity are analyzed to obtain the learning efficiency evaluation value of each knowledge node granularity. The specific process for obtaining the association weight values ​​of each knowledge node is as follows: Extract the pre-adaptation evaluation value, difficulty matching evaluation value, and learning efficiency evaluation value at the granularity of each knowledge node, and process them to obtain the association weight value of each knowledge node.

2. The knowledge graph-based personalized learning path recommendation system according to claim 1, characterized in that: The learner's user learning parameters include: learning objectives, knowledge base level, learning style preferences, and historical learning duration.

3. The personalized learning path recommendation system based on knowledge graphs according to claim 2, characterized in that: The learning path is dynamically planned based on the association weight values ​​of each knowledge node, and the specific analysis is as follows: The association weight value of each knowledge node is compared with the knowledge node association weight threshold stored in the database; If the association weight value of a knowledge node is higher than the knowledge node association weight threshold stored in the database, the prior adaptation evaluation value of the knowledge node is compared with the first benchmark value of the prior adaptation evaluation value. If the prior adaptation evaluation value of the knowledge node is higher than the first benchmark value of the prior adaptation evaluation value, the learning order of the knowledge node is adjusted. If it is lower than or equal to the first benchmark value of the prior adaptation evaluation value, the learning order of the knowledge node does not need to be adjusted. The difficulty matching assessment value of the knowledge node is compared with the second benchmark value of the difficulty matching assessment value. If the difficulty matching assessment value of the knowledge node is higher than the second benchmark value of the difficulty matching assessment value, the learning resource type of the knowledge node is adapted and adjusted. If it is lower than or equal to the second benchmark value of the difficulty matching assessment value, the learning resource type of the knowledge node does not need to be adjusted. The learning efficiency evaluation value of the knowledge node is compared with the third benchmark value of the learning efficiency evaluation value. If the learning efficiency evaluation value of the knowledge node is higher than the third benchmark value of the learning efficiency evaluation value, the allocation time of the knowledge node is dynamically adjusted. If it is lower than or equal to the third benchmark value of the learning efficiency evaluation value, there is no need to adjust the allocation time of the knowledge node. If the association weight value of a knowledge node is lower than or equal to the knowledge node association weight threshold stored in the database, then dynamic programming is not required. After the dynamic programming is completed, the association weight value of the knowledge node is recalculated and compared with the knowledge node association weight threshold stored in the database to determine whether the dynamic programming is complete. The learning order, learning resource type, and number of times the allocated time for knowledge nodes are adjusted are recorded. When the number of adjustments reaches the specified number, an early warning is issued.

4. The personalized learning path recommendation system based on knowledge graphs according to claim 1, characterized in that: The process of adjusting the recommended path in real time during the learning process is as follows: During the learning process, learners' real-time learning data is tracked, and the mastery of knowledge nodes is continuously monitored. If the mastery of a knowledge node is detected to be lower than the preset mastery threshold, the path branch adjustment will be automatically triggered.

5. The personalized learning path recommendation system based on knowledge graphs according to claim 1, characterized in that: The process of monitoring learners' mastery of knowledge points, deviations in learning progress, and completion rates of learning paths, and issuing early warnings for anomalies, is as follows: When the mastery of a knowledge point is detected to be lower than the safe threshold, additional learning resources are recommended and a prompt is issued. When the deviation from the learning progress is detected to exceed the safe threshold, the time allocated to subsequent nodes is adjusted and a prompt is issued. When the path completion rate is detected to be lower than the safe threshold, the learning plan is reordered and a prompt is issued.

6. The personalized learning path recommendation system based on knowledge graphs according to claim 1, characterized in that: The specific analysis conditions for the association weight values ​​of each knowledge node are as follows: The knowledge relevance, difficulty difference, and frequency stability of each knowledge node are weighted and summed. The weight coefficient of knowledge relevance is determined according to the domain characteristics of the learning objective, the weight coefficient of difficulty difference is dynamically adjusted according to the user's knowledge level, and the weight coefficient of frequency stability is set according to the cumulative effect of historical learning time. The comprehensive relevance weight value is obtained through multi-level weighted calculation.

7. The knowledge graph-based personalized learning path recommendation system according to claim 1, characterized in that: The extraction of knowledge units within the learning domain is analyzed in detail below: Initial knowledge units are extracted from a knowledge ontology annotated by domain experts to determine core concepts and basic terms. By combining natural language processing technology, entity recognition and relation extraction are performed on unlabeled learning resources to supplement and expand knowledge units; The initial extracted and expanded knowledge units are merged and deduplicated to obtain a structured set of knowledge units.

Images