Method and system for constructing pre-post sequence relationship of knowledge points based on user's exercise record

Abstract

The application provides a knowledge point pre-post sequence relationship construction method and system based on user exercise record, and belongs to the technical field of educational data mining. The application obtains the exercise record data set of each user; divides the record into learning sessions according to time, creates a knowledge point pair based on the session, and assigns a time sequence confidence by statistical time sequence; determines a list of logically associated knowledge point pairs by using a textbook directory and the exercise record. The knowledge mastery condition probability of each user is calculated according to the answer result; for each knowledge point pair, the pre-post sequence score is calculated by taking the two knowledge points as pre-sequences in combination with the condition probability and the time sequence confidence, and the pre-post sequence relationship is determined by comparing with a preset threshold; finally, the verification instruction is constructed to verify and correct the relationship by using a large language model to generate the target knowledge point pre-post sequence relationship. The application can improve the lack of timeliness and accuracy of the relationship construction between knowledge points, and provide a basis for personalized learning path planning.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of educational data mining, and particularly relates to a knowledge point pre-post sequence relationship construction method and system based on user exercise records. BACKGROUND

[0002] Knowledge point relationship construction is one of the core technologies in the field of intelligent education, which can identify the learning dependency relationship between knowledge topics, and provide basic support for personalized learning path planning, adaptive test question recommendation and other educational applications. The establishment of such relationship is of great significance for constructing knowledge graph and optimizing teaching sequence, especially in large-scale online education platforms, automated mining of logical progressive relationship between knowledge points can significantly improve learning efficiency and teaching quality.

[0003] The existing knowledge point relationship construction methods mainly rely on manual annotation, such as manually establishing knowledge system based on course outline or teacher experience, and some research attempts to construct knowledge point relationship through association rule mining, such as discovering the relationship between knowledge points through Apriori algorithm or collaborative filtering-based recommendation mechanism. These technologies can reveal the co-occurrence of knowledge points or the similarity of questions to some extent, but often stay in static or surface structure.

[0004] However, the existing methods have the following shortcomings: manual annotation is time-consuming and laborious, and is difficult to adapt to dynamic updates and maintain effectiveness in large-scale learning data; Apriori algorithm can only identify the co-occurrence of knowledge points, and cannot distinguish their logical pre- and post- sequence relationship, resulting in lack of directionality in the constructed association relationship; and the collaborative filtering method pays too much attention to the similarity of questions at the topic level, ignoring the inherent logical progressive relationship between knowledge points; the above shortcomings result in lack of timeliness and accuracy in the construction of relationship between knowledge points, and the learning path planning does not match the actual learning process, which cannot provide basis for personalized learning path planning, thereby affecting the effect of personalized learning. Therefore, there is an urgent need for a method that can automatically and accurately mine the causal dependency relationship between knowledge points from real exercise behavior data to construct the pre- and post- sequence relationship of knowledge points. SUMMARY

[0005] The application aims to provide a knowledge point pre-post sequence relationship construction method and system based on user exercise records, to solve the problem of lack of timeliness and accuracy in the construction of relationship between knowledge points in the prior art, and the inability to provide basis for personalized learning path planning.

[0006] To solve the above technical problems, in a first aspect, the application provides a knowledge point pre-post sequence relationship construction method based on user exercise records, comprising:

[0007] obtain a question record dataset of each user, the question record dataset comprising a plurality of question records sorted in time sequence, each question record comprising a user ID, a question ID, a question-associated knowledge point, an answer result, and a timestamp;

[0008] divide the question record dataset of each user into a plurality of learning sessions according to time intervals, create a plurality of knowledge point pairs with a front-back sequence based on the associated knowledge points in each learning session, and assign a time sequence confidence to each knowledge point pair by counting the time sequence of question records associated with different knowledge points in all learning sessions;

[0009] determine a knowledge point pair with a logical association in all knowledge points using the preset textbook directory information and the question record dataset, and obtain a knowledge point pair list;

[0010] calculate the mastery state of each knowledge point of each user based on the answer result in the question record dataset, and determine the knowledge mastery conditional probability of each knowledge point pair according to the mastery state;

[0011] For each knowledge point pair in the knowledge point pair list, based on the knowledge mastery conditional probability and the time sequence confidence, a set of front-back sequence scores is calculated respectively with the two knowledge points in the knowledge point pair as the front knowledge point, and the knowledge point front-back sequence relationship of each knowledge point pair is determined by comparing with a preset scoring threshold.

[0012] Construct a verification instruction for the knowledge point front-back sequence relationship, process the verification instruction using a large language model to obtain a verification result, and correct the knowledge point front-back sequence relationship using the verification result to generate a target knowledge point front-back sequence relationship.

[0013] In an implementable embodiment, the knowledge point pair list is obtained by determining a knowledge point pair with a logical association in all knowledge points using the preset textbook directory information and the question record dataset, comprising:

[0014] Filter out the knowledge point pairs with a logical association relationship in all knowledge points using the chapter attribution information of the knowledge points in the textbook directory information, and obtain a first knowledge point pair list;

[0015] Construct a knowledge point prior relationship matrix using the chapter attribution information of the knowledge points, each element in the knowledge point prior relationship matrix representing a strength value of the logical association strength of the knowledge point pair;

[0016] Determine the knowledge point pairs with a user proportion greater than a preset proportion in learning a certain knowledge point pair in the question record dataset using an association rule mining algorithm, and obtain a candidate knowledge point pair list;

[0017] Take the strength value as the prior weight of the knowledge point pair in the candidate knowledge point pair list, and use the preset weight threshold to screen the candidate knowledge point pair list to obtain a second knowledge point pair list;

[0018] Based on the intersection of the first knowledge point pair list and the second knowledge point pair list, the knowledge point pair list is determined.

[0019] In an implementable embodiment, the problem solving record data set of each user is divided into multiple learning sessions according to time intervals, multiple groups of knowledge point pairs with front-back sequence are created based on the associated knowledge points in each learning session, and the time sequence confidence of each group of knowledge point pairs is assigned by counting the time sequence of the problem solving records associated with different knowledge points in all learning sessions, including:

[0020] The problem solving record data set of each user is divided into multiple learning sessions according to time intervals, each learning session includes a group of problem solving records arranged in time sequence;

[0021] Based on the appearance order of the question associated knowledge points in each learning session, multiple groups of knowledge point pairs with front-back sequence are created, each knowledge point pair includes a first knowledge point and a second knowledge point, and the first knowledge point is before the second knowledge point;

[0022] Based on the time stamp corresponding to the question ID associated with the knowledge points in each learning session, the number of times that the first knowledge point appears before the second knowledge point in any group of knowledge point pairs in the same learning session and the number of times that the first knowledge point and the second knowledge point appear together are calculated respectively;

[0023] In all learning sessions, the ratio of the sum of the number of times that the first knowledge point appears before the second knowledge point in each group of knowledge point pairs to the sum of the number of times that the first knowledge point and the second knowledge point appear together is calculated to obtain the time sequence confidence of each group of knowledge point pairs.

[0024] In an implementable embodiment, the problem solving record further includes a resource learning identifier indicating whether there is an associated resource learning behavior when answering the question for the question ID, the resource learning identifier includes a resource associated knowledge point and a learning time stamp; The method further comprises:

[0025] When the resource learning identifier indicates that there is an associated resource learning behavior when answering the question for the question ID, it is determined whether the knowledge points associated with the resource associated knowledge points and the question associated knowledge points have the same common knowledge points, and when the resource associated knowledge points and the question associated knowledge points have the same common knowledge points, the learning time stamp is associated with the common knowledge points;

[0026] In a case where the first knowledge point and the second knowledge point of the knowledge point pair both have associated learning time stamps, and the learning time stamp associated with the first knowledge point is prior to the learning time stamp associated with the second knowledge point, the preset weighting coefficient is used to weight the time sequence confidence corresponding to the knowledge point pair, to obtain an updated time sequence confidence.

[0027] In an implementable embodiment, the method further includes:

[0028] In a case where the first knowledge point and the second knowledge point of the knowledge point pair both have associated learning time stamps, and the learning time stamp associated with the first knowledge point is prior to the learning time stamp associated with the second knowledge point, the decay weight is determined in a preset exponential decay function according to the time interval between the learning time stamp associated with the first knowledge point and the learning time stamp associated with the second knowledge point.

[0029] The preset weighting coefficient and the decay weight are used to weight the time sequence confidence corresponding to the knowledge point pair, to obtain an updated time sequence confidence.

[0030] In an implementable embodiment, the knowledge mastery conditional probability includes a knowledge point pair mastery conditional probability and a knowledge point pair non-mastery conditional probability.

[0031] For each knowledge point pair in the knowledge point pair list, a set of pre-posterior scores is calculated based on the knowledge mastery conditional probability and the time sequence confidence, with the two knowledge points in the knowledge point pair as the pre-knowledge point and the posterior knowledge point, respectively.

[0032] For the first knowledge point in each knowledge point pair and the second knowledge point , a first pre-posterior score and a second pre-posterior score are calculated:

[0033]

[0034]

[0035] wherein, is the time sequence confidence, is the knowledge point pair mastery conditional probability, is the knowledge point pair non-mastery conditional probability.

[0036] In an implementable embodiment, the knowledge point pre-posterior relationship of each knowledge point pair is determined by comparing with a preset score threshold, including:

[0037] In a case where the first pre-posterior score is greater than the second pre-posterior score, and the first pre-posterior score is greater than the preset score threshold, it is determined that the first knowledge point is the pre-knowledge point of the second knowledge point, to obtain the knowledge point pre-posterior relationship of the knowledge point pair.

[0038] In a case that the second pre-post sequence score is greater than the first pre-post sequence score and the second pre-post sequence score is greater than a preset score threshold, it is determined that the second knowledge point is a pre-knowledge point of the first knowledge point, and a knowledge point pre-post sequence relationship of the knowledge point pair is obtained.

[0039] In a second aspect, the present application provides a knowledge point pre-post sequence relationship construction system based on user test record, comprising:

[0040] An acquisition module is configured to acquire a test record dataset of each user, the test record dataset comprising a plurality of test records sorted in time sequence, each test record comprising a user ID, a question ID, a question-associated knowledge point, an answer result and a timestamp;

[0041] An allocation module is configured to divide the test record dataset of each user into a plurality of learning sessions according to time intervals, create a plurality of groups of knowledge point pairs with pre-post sequence based on the associated knowledge points in each learning session, and assign a time sequence confidence to each group of knowledge point pairs by counting the time sequence of the test records associated with different knowledge points in all learning sessions;

[0042] A determination module is configured to determine, using preset textbook directory information and the test record dataset, a knowledge point pair with logical association from among all knowledge points, and obtain a knowledge point pair list;

[0043] The determination module is further configured to calculate a mastery state of each knowledge point of each user based on the answer result in the test record dataset, and determine a knowledge mastery conditional probability of each knowledge point pair according to the mastery state;

[0044] The determination module is further configured to, for each knowledge point pair in the knowledge point pair list, calculate a group of pre-post sequence scores by taking the two knowledge points in the knowledge point pair as pre-knowledge points based on the knowledge mastery conditional probability and the time sequence confidence, and determine a knowledge point pre-post sequence relationship of each knowledge point pair by comparing with a preset score threshold;

[0045] A correction module is configured to generate a verification instruction for the knowledge point pre-post sequence relationship construction, process the verification instruction using a large language model to obtain a verification result, and correct the knowledge point pre-post sequence relationship using the verification result to generate a target knowledge point pre-post sequence relationship.

[0046] In a third aspect, the present application provides an electronic device, comprising:

[0047] A memory is configured to store a computer program;

[0048] A processor is configured to execute the computer program to implement the steps of the knowledge point pre-post sequence relationship construction method based on user test record of the first aspect.

[0049] In a fourth aspect, the present application provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the steps of the knowledge point presequence and postsequence relationship construction method based on user test record according to the first aspect can be implemented.

[0050] The knowledge point presequence and postsequence relationship construction method based on user test record provided by the present application automatically constructs the knowledge point presequence and postsequence relationship by using real test behavior data, effectively overcoming the problems of low artificial labeling efficiency and lack of directionality of static association rules in the prior art. By dividing the test record into learning sessions and calculating the timing confidence, the order of knowledge points can be dynamically captured from the learning process, ensuring that the relationship has logical progression. In combination with the textbook directory information, the logically associated knowledge point pairs are screened, and the learning dependency is evaluated based on the knowledge mastery conditional probability, enhancing the accuracy and timeliness of relationship construction. Finally, the relationship is verified and corrected by a large language model, further improving the matching degree of the knowledge point presequence and postsequence relationship with the real learning process, thereby providing a reliable basis for personalized learning path planning and significantly improving learning efficiency and teaching quality.

[0051] Further, by introducing the resource learning identifier and its timestamp, the user's active learning behavior in the learning process can be more finely captured, thereby making up for the accidental bias that may exist only by relying on the timing of the test. When both sides of the knowledge point pair have associated resource learning behaviors and the time sequence is consistent, the timing confidence is improved by weighting, enhancing the reliability of the presequence and postsequence relationship determination. Further, according to the resource learning time interval, an exponential decay weight is introduced for weighting, so that the contribution of recent continuous learning behavior to relationship construction is greater, which is more in line with the natural law of knowledge accumulation and forgetting, thereby significantly improving the accuracy and timeliness of the knowledge point presequence and postsequence relationship inference. Ultimately, this method can construct a knowledge point logical relationship that is more in line with the actual learning process, providing a more solid and reliable basis for personalized learning path planning. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0053] Figure 1 A flowchart of a knowledge point presequence and postsequence relationship construction method based on user test record provided by the present application;

[0054] Figure 2A flowchart of a method for generating a knowledge point pair list according to an embodiment of the present application is shown in FIG. 1.

[0055] Figure 3 A structural diagram of a knowledge point pre-post sequence relationship construction system based on user test records according to an embodiment of the present application is shown in FIG. 2.

[0056] Figure 4 A structural diagram of another knowledge point pre-post sequence relationship construction system based on user test records according to an embodiment of the present application is shown in FIG. 3.

[0057] Figure 5 FIG. 4 is a hardware structural diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0058] To make personnel in the technical field better understand the present application, the present application is further described in detail below in combination with the drawings and specific embodiments. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0059] The core of the present application is to provide a knowledge point pre-post sequence relationship construction method based on user test records, and a flowchart of one specific embodiment of the method is shown in FIG. 1, which includes steps S110 to S160. Figure 1

[0060] S110: Obtain the test record dataset of each user, which includes multiple test records sorted in time sequence, each test record including user ID, question ID, question associated knowledge point, answer result and time stamp.

[0061] The test record dataset refers to a data set containing user answer behavior information obtained from an online learning platform, including multiple test records arranged in time sequence. The question associated knowledge point refers to the mapping relationship between the question and the knowledge point annotated according to the subject textbook chapter and the proposition outline, supporting one question with multiple knowledge point association, for example, a geometry question in mathematics can be associated with both triangle congruence and Pythagorean theorem knowledge points. The answer result refers to the user's answer state identifier for the question, including correct and incorrect. The time stamp refers to the time mark of the user's answer, accurate to millisecond level.

[0062] ​The data set of each user's test record is obtained, including extracting the mapping relationship between the questions and knowledge points from the database, which supports the association of one question with multiple knowledge points. The obtained test record is data cleaned, and invalid records with a test duration of less than 10 seconds are filtered to avoid the influence of invalid answering data on the analysis result. The multiple answering data of the same user for the same question are aggregated, and the last answering result is used as the final answering record of the user for the question. The cleaned test record is sorted by timestamp to ensure that the records are arranged in chronological order. Based on the mapping relationship between the questions and the knowledge points, the question-associated knowledge points are added to each test record to form a complete test record data set containing user ID, question ID, question-associated knowledge points, answering result and timestamp.

[0063] Exemplarily, taking junior high school mathematics as an example, the test records of 1000 junior high school mathematics users are obtained from an online learning platform. The user set is wherein represents the nth student. The test record of the user is represented as wherein represents the question, represents correct or incorrect, represents the answering time, represents the test duration, and the unit is millisecond. The question set in the test record is obtained and the corresponding relationship with the knowledge point label is, for example, and the like. The data preprocessing of the test record includes filtering invalid records with a test duration of less than 10 seconds, and aggregating multiple answers of the same user for the same question to take the last answering result as the standard. The cleaned test records of the 1000 users are sorted by timestamp to form a test record data set arranged in chronological order, each record containing user ID, question ID, question-associated knowledge points, answering result and timestamp, and optionally including resource learning identification, which is used to indicate whether there is associated resource learning behavior such as watching explanation video during answering, the identification containing resource-associated knowledge points and corresponding learning timestamp. High-quality basic data support is provided for subsequent knowledge point sequence relationship mining.

[0064] In one embodiment, in the data acquisition and preprocessing of step S110, the test record further includes an error type label in addition to user ID, question ID, question-associated knowledge points, answering result and timestamp. The error type label is used to identify the specific reason for the user's answering error, which can be classified as concept misunderstanding, calculation error or question bias, etc.

[0065] In subsequent calculation of the mastery state of the knowledge point, the error type label can be used to optimize the calculation of the mastery degree. The specific implementation is as follows: when the error type label of an error record is concept misunderstanding, it indicates that the user has a fundamental understanding deviation for the knowledge point associated with the question. At this time, a higher weight can be given to the error record when calculating the mastery degree of the knowledge point, or the result of this answering can be directly regarded as effective evidence of not mastering. On the contrary, if the error type label is calculation error or question review deviation, it indicates that the error may be caused by accidental operation error or habit problem, rather than the lack of knowledge point. When processing such records, the weight of the record in the calculation of the mastery degree can be reduced, or the record can be regarded as noise data and properly cleaned up in the data preprocessing stage. By introducing the error type label, the nature of the user's error can be more finely distinguished, so that the real mastery state of the user for the knowledge point can be more accurately evaluated, and a more reliable data basis can be provided for subsequent inference of the precedence and sequence relationship of the knowledge points.

[0066] S120: dividing the question-answering record data set of each user into multiple learning sessions according to time intervals, creating multiple groups of knowledge point pairs with precedence and sequence based on the associated knowledge points in each learning session, and assigning a time sequence confidence to each group of knowledge point pairs by statistically analyzing the time sequence of the question-answering records associated with different knowledge points in all learning sessions.

[0067] A learning session refers to a set of continuous answering behaviors of a user within a fixed time interval, including a sequence of question-answering records sorted by timestamps. A knowledge point pair refers to two knowledge points that appear in a learning session in a time sequence, where the first appearing knowledge point is the precedent knowledge point and the second appearing knowledge point is the sequence knowledge point. The time sequence confidence refers to the frequency of the appearance of the precedent knowledge point before the sequence knowledge point in a knowledge point pair.

[0068] The question-answering record data set of each user is divided into multiple learning sessions according to a preset time interval, and the time interval is set to 30 minutes. Continuous question-answering records are classified into the same learning session according to timestamps. Based on the appearance sequence of the knowledge points associated with the questions in each learning session, multiple groups of knowledge point pairs are created, for example, if the knowledge point appearance sequence in a learning session is , the knowledge point pair , , is generated. The appearance of each group of knowledge point pairs in all learning sessions is counted, and the number of times that the precedent knowledge point in a knowledge point pair appears before the sequence knowledge point in the same learning session and the total number of times that the knowledge point pair appears together are calculated. The time sequence confidence is calculated according to the statistical result, for example, the knowledge point pair appears together 100 times in all learning sessions, of which appears before 85 times, and the time sequence confidence is 0.85.

[0069] S130: Determine the knowledge point pairs with logical association in all knowledge points by using the preset textbook directory information and the question record dataset, and obtain a knowledge point pair list.

[0070] The knowledge point pair list refers to a set of knowledge point pairs with logical association in all knowledge points. The first knowledge point pair list refers to a set of knowledge point pairs filtered based on the chapter attribution information of the knowledge points in the textbook directory. The knowledge point prior relationship matrix refers to a matrix constructed by using the chapter attribution information of the knowledge points, wherein each element represents the logical association strength of a knowledge point pair. The candidate knowledge point pair list refers to a set of knowledge point pairs with a user proportion learning a certain knowledge point pair greater than a preset proportion in the question record dataset determined by an association rule mining algorithm. The second knowledge point pair list refers to a set of knowledge point pairs filtered by using a preset weight threshold with the strength value in the knowledge point prior relationship matrix as a prior weight.

[0071] The chapter attribution information of the knowledge points in the textbook directory information is used to filter the knowledge point pairs with logical association, and the first knowledge point pair list CandSet1 is obtained. For example, the knowledge points belonging to the same chapter or adjacent chapters are considered to have logical association. The chapter attribution information of the knowledge points is used to construct the knowledge point prior relationship matrix, wherein the matrix element value is calculated based on the chapter distance, such as 1.0 for the same chapter, 0.8 for adjacent chapters, and 0 for non-adjacent chapters. The association rule mining algorithm is used to calculate the user proportion learning each knowledge point pair in the question record dataset, and if the proportion is greater than a preset proportion, the knowledge point pair is added to the candidate knowledge point pair list. The strength value in the knowledge point prior relationship matrix is used as a prior weight, and the candidate list is filtered by using a preset weight threshold to obtain the second knowledge point pair list CandSet2. The intersection of the first knowledge point pair list and the second knowledge point pair list is used to determine the final knowledge point pair list CandSet.

[0072] In one embodiment, in step S130, when the knowledge point prior relationship matrix is constructed by using the textbook directory information, in addition to calculating the logical association strength based on the chapter attribution information of the knowledge points, a natural language processing technique can also be introduced to perform deep analysis on the textbook content to obtain more accurate strength values.

[0073] In implementation, official teaching syllabus, chapter text of teaching materials, definition and description of knowledge points, and other text materials can be analyzed by natural language processing. For example, the semantic similarity between different knowledge point names or description texts can be calculated using word embedding models (such as Word2Vec, BERT), or the topic distribution correlation of knowledge points in the content of teaching materials can be analyzed using topic models (such as LDA). Based on the NLP analysis results, semantic correlation strength can be assigned to the knowledge point pairs. For example, if two knowledge points are highly related in semantics or frequently appear in the same teaching context, even if they do not belong to the same or adjacent chapters, a higher strength value can be assigned. Finally, the NLP-based semantic correlation strength and the chapter distance-based correlation strength can be weighted and fused to form the element values in the knowledge point prior relationship matrix. This approach can capture more subtle logical connections between knowledge points from the semantic level of the content of the teaching materials, making up for possible shortcomings of relying solely on chapter structure, and making the prior relationship matrix better reflect the internal correlation of knowledge.

[0074] S140: Based on the answer results in the question record data set, the mastery status of each knowledge point of each user is calculated, and the knowledge mastery conditional probability of each knowledge point pair is determined according to the mastery status.

[0075] The mastery status refers to the mastery degree of the knowledge point, with 1 indicating mastery and 0 indicating non-mastery. The mastery degree is defined as Accuracy(u, k) = (the number of k questions answered correctly by user u) / (the total number of k questions answered by user u). When Accuracy(u, k) ≥ 80%, the mastery status is recorded as 1, otherwise as 0. The knowledge point pair mastery conditional probability refers to the probability that knowledge point is mastered under the condition that knowledge point is mastered, i.e. . The knowledge point pair non-mastery conditional probability refers to the probability that knowledge point is not mastered under the condition that knowledge point is not mastered, i.e. . Based on the answer results in the question record data set, the mastery degree of each knowledge point of each user is calculated, and the mastery status is determined according to the 80% threshold.

[0076] For each knowledge point pair , the knowledge point pair mastery conditional probability is calculated: wherein, denotes the set of all users; denotes a single user; denotes the mastery status of user on knowledge point (1 for mastery, 0 for non-mastery); the numerator denotes the number of users who simultaneously master knowledge points and ; the denominator denotes the number of users who master knowledge point The number of users.

[0077] Calculate the conditional probability of not mastering a knowledge point: in, Indicates user Knowledge points The state of not mastering; the molecule represents the knowledge points that have not been mastered at the same time. and The number of users, where the denominator represents the number of knowledge points not yet mastered. The number of users.

[0078] For example, among 10,000 students, the knowledge points The number of questions answered correctly is 6,800, and the number of questions completed is 8,000, indicating a high level of mastery. =0.85≥0.8, Mastery status is 1; Knowledge point The number of questions answered correctly is 6,000, and the number of questions completed is 8,000, indicating a high level of mastery. =0.75 < 0.8, mastery level is 0. For knowledge points... At the same time, master and The number of users is 5,200, and they have mastered... The number of users is 7,500, therefore At the same time, it was not mastered. and The number of users is 1,500, and the exact number is unknown. The number of users is 2,500, therefore .

[0079] S150: For each knowledge point pair in the knowledge point pair list, based on the conditional probability of knowledge mastery and the temporal confidence, calculate a set of preceding and following scores using the two knowledge points in the knowledge point pair as preceding knowledge points, and determine the preceding and following relationship of knowledge points in each knowledge point pair by comparing them with the preset scoring threshold.

[0080] The order of scores refers to the score for knowledge points. middle for The scoring criteria for prerequisite knowledge points are calculated using the following formula: .in For time series confidence, For mastering the knowledge points, conditional probability. This represents the conditional probability of not mastering a knowledge point. The preset scoring threshold is 0.5, used to determine the sequential relationship between knowledge points.

[0081] For each knowledge point pair in the knowledge point pair list Based on time series confidence and the knowledge mastery conditional probability, respectively, to calculate the first pre-post sequence score and the second pre-post sequence score . Compare the two scores, if and , determine that is the pre-knowledge point of ; if and , determine that is the pre-knowledge point of .

[0082] S160: Construct a verification instruction for the knowledge point pre-post sequence relationship, process the verification instruction using a large language model to obtain a verification result, and use the verification result to correct the knowledge point pre-post sequence relationship to generate the target knowledge point pre-post sequence relationship.

[0083] The verification instruction refers to a Prompt constructed based on the subject field + judgment target + reasoning basis, for example, based on junior high school mathematics, analyzing the pre-post sequence of the sum of the interior angles of a triangle and the congruence of a triangle. The verification result refers to the pre-post sequence relationship judgment returned by the large language model after processing the verification instruction. The target knowledge point pre-post sequence relationship refers to the final knowledge point pre-post sequence relationship obtained after correction by the large language model, which has higher educational logical rationality.

[0084] Based on the knowledge point pre-post sequence relationship obtained in S150, construct a verification instruction containing the subject field, judgment target, and reasoning basis. For example, for the knowledge point pair , generate a verification instruction based on junior high school mathematics to analyze the pre-post sequence of and . Call the large language model to process the instruction and obtain the verification result. According to the verification result, if the large language model judges that the pre-post sequence relationship is inconsistent with the S150 result, correct the knowledge point pre-post sequence relationship. For example, if S150 determines that is the pre-knowledge point of , but the large language model returns that is the pre-knowledge point of , then correct it to is the pre-knowledge point of .

[0085] Illustratively, S150 determines that the pre-post sequence relationship of the knowledge point pair is that is the pre-knowledge point of , where is a linear equation of the first degree, and is a quadratic equation of the first degree. Construct a verification instruction based on junior high school mathematics to analyze the pre-post sequence of the linear equation of the first degree and the quadratic equation of the first degree. Call the large language model to process the instruction and return the verification result that the linear equation of the first degree should precede the quadratic equation of the first degree, i.e. is a prerequisite of the target knowledge point. Therefore, without modifying the sequence relationship of knowledge points, if it is inconsistent, it needs to be modified, and the sequence relationship between and is exchanged and adjusted. The sequence relationship of the target knowledge point is verified by the large language model twice, which modifies the possible data deviation and teaching law deviation of S150, ensuring that the knowledge point dependency relationship conforms to the educational logic, and providing accurate basis for personalized learning path planning.

[0086] The application provides a method for generating a knowledge point pair list, and a flowchart of one specific embodiment is shown in Figure 2 The method comprises steps S210 to S250.

[0087] In an implementable embodiment, step S130: using the preset textbook directory information and the question record dataset, determining the knowledge point pairs with logical association in all knowledge points, obtaining the knowledge point pair list, comprising:

[0088] S210: using the chapter attribution information of the knowledge points in the textbook directory information, screening the knowledge point pairs with logical association relationship in all knowledge points, obtaining the first knowledge point pair list.

[0089] The first knowledge point pair list refers to the set of knowledge point pairs with logical association relationship screened based on the chapter attribution information of the knowledge points in the textbook directory, including the knowledge point pairs in the same chapter and the knowledge point pairs in adjacent chapters, for example, knowledge points and belong to the same chapter or adjacent chapters, and are considered to have logical association.

[0090] Using the chapter attribution order of the knowledge points in the textbook directory, the knowledge point pairs are screened according to the chapter sequence relationship. For example, knowledge point belongs to chapter A, and knowledge point belongs to chapter B, if chapter B is after chapter A, then there is logical association. The knowledge point pairs in the same chapter have the highest strength, the knowledge point pairs in adjacent chapters have the second highest strength, and the knowledge point pairs in non-adjacent chapters are not considered to have logical association, thereby obtaining the first knowledge point pair list.

[0091] S220: using the chapter attribution information of the knowledge points, constructing a knowledge point prior relationship matrix, and each element in the knowledge point prior relationship matrix represents the strength value of the logical association strength of the knowledge point pair.

[0092] The knowledge point prior relationship matrix refers to the matrix constructed using the chapter attribution information of the knowledge points, wherein each element represents the knowledge point pair The logical association strength value is calculated as follows: 1.0 for the same chapter, 0.8 for adjacent chapters, and 0.0 for non-adjacent chapters.

[0093] Based on the chapter affiliation information of each knowledge point, a pair of knowledge points is created for each knowledge point. Calculate the intensity value. For example, knowledge point. and If they belong to the same chapter, then If they belong to adjacent chapters, then If they are not adjacent chapters, then The rows and columns of the matrix correspond to the knowledge point indices, generating a complete prior relation matrix of knowledge points.

[0094] S230: Using an association rule mining algorithm, identify knowledge point pairs in the question-answering record dataset where the proportion of users learning a certain knowledge point is greater than a preset proportion, and obtain a list of candidate knowledge point pairs.

[0095] The candidate knowledge point pair list refers to the set of knowledge point pairs identified in the question-answering record dataset by the association rule mining algorithm, where the proportion of users who have learned a certain knowledge point pair is greater than a preset proportion, which is usually 50%.

[0096] Use association rule mining algorithms (such as Apriori or FP-Growth) to calculate each knowledge point pair The percentage of users learning in the problem-solving record dataset. For example, among 10,000 students, the percentage of students learning simultaneously. and If the percentage of users is greater than 50% of the total number of users, then... Add to the candidate knowledge point pair list.

[0097] S240: Use the intensity value as the prior weight of the knowledge point pairs in the candidate knowledge point pair list, and use the preset weight threshold to filter the candidate knowledge point pair list to obtain the second knowledge point pair list.

[0098] The second knowledge point pair list refers to the set of knowledge point pairs that uses the strength values ​​in the prior relation matrix of knowledge points as prior weights and is filtered using a preset weight threshold, which is usually 0.5.

[0099] The strength values ​​in the prior relation matrix of knowledge points As a weight, knowledge point pairs with an intensity value greater than a preset weight threshold are selected. For example, intensity value Then keep Intensity value If the answer is no, then remove it. This will give you the second list of knowledge point pairs.

[0100] S250: determining a knowledge point pair list based on the intersection of the first knowledge point pair list and the second knowledge point pair list.

[0101] The knowledge point pair list refers to a final knowledge point pair set determined based on the intersection of the first knowledge point pair list and the second knowledge point pair list, and contains knowledge point pairs that meet both the textbook logical association and the learning behavior co-occurrence condition.

[0102] The intersection of the first knowledge point pair list and the second knowledge point pair list is calculated to obtain the final knowledge point pair list. For example, if the first knowledge point pair list contains , and the second knowledge point pair list contains , , the knowledge point pair list is and .

[0103] Illustratively, the textbook directory contains the first chapter of rational numbers (knowledge points , ), the second chapter of linear equations (knowledge points , ), and the third chapter of quadratic equations (knowledge points , ). Using chapter affiliation information, the same chapter pairs , , and adjacent chapter pairs , are filtered out to obtain the first knowledge point pair list. A knowledge point prior relationship matrix is constructed, with strength values (same chapter), (adjacent chapters). Among 10,000 students, the learning user proportion of knowledge point pair is 70%>50%, is 65%>50%, and the candidate knowledge point pair list contains , . After screening by strength value, strength 1.0>0.5, strength 0.8>0.5, the second knowledge point pair list contains , . Finally, by taking the intersection, the final knowledge point pair list is and .

[0104] In an implementable embodiment, step S120: dividing each user's problem record dataset into multiple learning sessions according to time intervals, creating multiple groups of knowledge point pairs with front-back sequence based on the associated knowledge points in each learning session, assigning a time sequence confidence to each group of knowledge point pairs by statistically ordering the problem record time sequence of different knowledge point associations in all learning sessions, including:

[0105] Divide each user's problem record dataset into multiple learning sessions according to time intervals, and each learning session includes a group of problem records arranged in time sequence.

[0106] Divide each user's problem record dataset into multiple learning sessions according to time intervals. For example, set the time interval to 30 minutes, and divide the user's continuous problem records completed within 30 minutes into a learning session. If the user continues to do problems after 30 minutes, a new learning session will start. For example, the user did 10 problems from 9:00-9:30 and 8 problems from 9:35-10:00, which are divided into two learning sessions. Each learning session contains problem records arranged in time sequence, indicating the user's learning sequence of knowledge points in the session.

[0107] Based on the appearance sequence of the question associated knowledge points in each learning session, create multiple groups of knowledge point pairs with front-back sequence, each knowledge point pair includes a first knowledge point and a second knowledge point, and the first knowledge point is before the second knowledge point.

[0108] Knowledge point pair refers to the combination of front-back knowledge points formed according to the appearance sequence of knowledge points in the same learning session. Based on the appearance sequence of the question associated knowledge points in each learning session, create knowledge point pairs. For example, the appearance sequence of knowledge points in a learning session is , , , , then the knowledge point pairs include , , . Only keep the knowledge point pairs where the first knowledge point appears before the second knowledge point, that is , , .

[0109] Based on the timestamp corresponding to the question ID associated with the knowledge points in each learning session, calculate the number of times the first knowledge point appears before the second knowledge point in any group of knowledge point pairs in the same learning session, and the number of times the first knowledge point and the second knowledge point appear together.

[0110] Based on the timestamp corresponding to the question ID associated with the knowledge points in each learning session, calculate the time sequence statistics of the knowledge point pairs in the same learning session. For example, for the knowledge point pair In the learning session, calculation exist The number of times it appeared before and and The number of times they co-occur. Suppose in a learning session, and They appeared together 5 times, among which exist If it has occurred 4 times previously, then exist The previous occurrence was 4 times, and the number of times they appeared together was 5 times.

[0111] In all learning sessions, the temporal confidence of each knowledge point pair is obtained by calculating the ratio of the total number of times the first knowledge point appears before the second knowledge point in each knowledge point pair to the total number of times the first and second knowledge points appear together.

[0112] The formula for calculating time series confidence is: Time series confidence = ( exist (total number of times it appeared before) / ( and (Total frequency of co-occurrences). Statistics on the number of knowledge points in all learning sessions. middle exist The sum of the previous occurrences and and The ratio of the total number of times they appear together. For example, among 10,000 students, and They appeared together 10,000 times, among which exist If the sequence has occurred 8,500 times previously, the time series confidence level is 0.85. This value indicates... yes Confidence level of prerequisite knowledge points.

[0113] For example, 10,000 students participated in the learning, and their work records were divided into learning sessions at 30-minute intervals. In 100 learning sessions, the knowledge points were... They appeared together 500 times, among which exist It appeared 425 times previously, therefore the time series confidence level is 0.85. (Knowledge point correct) They appeared together 400 times, among which exist It appeared 340 times previously, with a time series confidence level of 0.85.

[0114] In an implementable embodiment, the question record further comprises a resource learning identifier indicating whether there is an associated resource learning behavior when the question for the question ID is answered, the resource learning identifier comprising a resource associated knowledge point and a learning time stamp, and the method further comprises:

[0115] The resource learning identifier is an identifier indicating whether the user has learned the auxiliary learning resource (such as a video, a lecture, etc.) related to the question during the question solving process, comprising a resource associated knowledge point and a learning time stamp, and is used to record the association between the learning resource and the question. The resource associated knowledge point refers to a set of knowledge points involved in the resource learning content, indicating the specific knowledge points the user is exposed to when learning the resource. For example, a learning video explains two knowledge points of triangle interior angles and triangle congruence, and the corresponding resource associated knowledge point set is .

[0116] When the resource learning identifier indicates that there is an associated resource learning behavior when the question for the question ID is answered, it is determined whether the knowledge points associated with the resource associated knowledge points and the question associated knowledge points have the same common knowledge points, and when the knowledge points associated with the resource associated knowledge points and the question associated knowledge points have the same common knowledge points, the learning time stamp is associated with the common knowledge points.

[0117] When the resource learning identifier indicates that there is an associated resource learning behavior, the resource associated knowledge point set and the question associated knowledge point set are extracted, and the intersection of them is calculated to obtain the common knowledge point set. For example, the resource associated knowledge point set is , and the question associated knowledge point set is , then the common knowledge point set is , and the learning time stamp (such as 10:00) in the resource learning identifier is associated with the common knowledge points and .

[0118] When both the first knowledge point and the second knowledge point of a knowledge point pair have an associated learning time stamp, and the learning time stamp associated with the first knowledge point is before the learning time stamp associated with the second knowledge point, the preset weighting coefficient is used to weight the time sequence confidence corresponding to the knowledge point pair to obtain an updated time sequence confidence.

[0119] In the knowledge point pair , if and both have an associated learning time stamp, and the learning time stamp of is before the learning time stamp of , then the preset weighting coefficient is used to weight the time sequence confidence. For example, the time sequence confidence of the knowledge point pair is , the preset weighting coefficient = 1.2, and the updated time sequence confidence If the calculation result exceeds 1.0, it is truncated to 1.0 to ensure that the time sequence confidence is within a reasonable range.

[0120] For example, the student learns the teaching video related to the first-order equation while doing the question, and the resource associated knowledge point is , wherein is the solution of the first-order equation, is the application of the equation, and the learning time is 10:00. The question associated knowledge point is , (wherein is the first-order equation), and the doing time is 10:05. The common knowledge point is , and the learning time stamp 10:00 is associated with . For the knowledge point pair , , the learning time stamp 10:00 of is before the doing time stamp 10:05 of , the time sequence confidence , the weighting coefficient = 1.1, and the updated time sequence confidence = 0.825. The updated time sequence confidence provides more accurate input data for S140 to calculate the knowledge mastery conditional probability and S160 to construct the verification instruction, effectively improving the educational logic rationality of the knowledge point pre-post sequence relationship.

[0121] The above embodiment can solve the problem of sparse doing data: some knowledge points may have few questions, but the student learns through auxiliary learning resources such as lecture notes. This method uses auxiliary behavior data to make up for the deficiency of the time sequence during pure doing, making the calculation of the time sequence confidence more rich and robust, especially in knowledge points with uneven number of questions, reliable relationships can be established between them, ensuring the accuracy and educational logic rationality of the knowledge point pre-post sequence relationship.

[0122] In an implementable embodiment, the method further comprises:

[0123] When the first knowledge point and the second knowledge point of a set of knowledge point pairs both have associated learning time stamps, and the learning time stamp associated with the first knowledge point is before the learning time stamp associated with the second knowledge point, a decay weight is determined in a preset exponential decay function according to the time interval between the learning time stamp associated with the first knowledge point and the learning time stamp associated with the second knowledge point. The corresponding time sequence confidence of the knowledge point pair is weighted using a preset weighting coefficient and the decay weight to obtain an updated time sequence confidence.

[0124] The decay weight is a time decay factor designed based on the Ebbinghaus forgetting curve principle, reflecting the decay degree of the influence of the knowledge point learning interval on the pre-post sequence relationship, and the calculation formula is , wherein is a preset decay coefficient, The time interval (in minutes) is used, and the weights are designed based on the Ebbinghaus forgetting curve, a classic theory in cognitive psychology, which describes the decay of human memory over time. This model has been incorporated into time-series confidence calculations, addressing the problem of mechanical time-series statistics ignoring the laws governing memory decay. (The last sentence appears to be incomplete and possibly refers to a specific knowledge point.) In the middle, if The learning timestamp Previously, and the time interval Larger (e.g., more than a week), then right The supporting effect of this sequence weakens, therefore its contribution to the confidence level of the time series should be reduced.

[0125] When the knowledge points are correct of and They all have related learning timestamps, and The learning timestamp Previously, the time interval was calculated. Based on the Ebbinghaus forgetting curve principle, a preset decay coefficient is used. Substitute into the exponential decay function Calculate the attenuation weight. For example, The learning timestamp is 10:00. The learning timestamp is 10:30, and the time interval is... minutes, then .

[0126] The updated temporal confidence score incorporates the Ebbinghaus forgetting curve principle, representing the confidence level of a knowledge point in enhancing sequential relationships. The calculation formula is as follows: ,in The original time series confidence level. To preset the weighting coefficients, For attenuation weights. Utilizing preset weighting coefficients. and the calculated attenuation weight Confidence of the original time series Weighting is applied. For example, the original time-series confidence score. Decay weight ,but If the calculated result exceeds 1.0, it is truncated to 1.0 to ensure that the confidence level is within a reasonable range.

[0127] For example, knowledge point pairs middle The learning timestamp is 10:00. The learning timestamp is 10:30, and the time interval is... minutes. Based on the principle of Ebbinghaus forgetting curve, the decay weight is calculated . original timing confidence , preset weighting coefficient , updated timing confidence . The updated timing confidence effectively considers the influence of the learning time interval of the knowledge point on cognitive dependence, making the mined pre-post sequence relationship closer to the real cognitive dependence rather than the accidental time sequence, filtering out the time sequence relationship that may be invalid due to too long time interval, providing more accurate input data for S140 to calculate knowledge mastery conditional probability and S160 to construct verification instructions, and making the final knowledge graph more accurately reflect the effective knowledge dependence chain.

[0128] In an implementable embodiment, the knowledge mastery conditional probability includes knowledge point pair mastery conditional probability and knowledge point pair unmastered conditional probability.

[0129] The knowledge point pair mastery conditional probability refers to the probability of the knowledge point being mastered under the condition that the knowledge point is mastered, i.e. . The knowledge point pair unmastered conditional probability refers to the probability of the knowledge point not being mastered under the condition that the knowledge point is not mastered, i.e. . The pre-post sequence score refers to a scoring index for quantifying the pre-post sequence dependence strength between the knowledge point pair and , and the calculation formula is , wherein is the timing confidence, is the knowledge point pair mastery conditional probability, is the knowledge point pair unmastered conditional probability. The pre-post sequence score effectively captures the causal dependence direction between knowledge points by associating the timing sequence and the mastery state, avoiding the directional bias caused by the traditional method of relying only on co-occurrence frequency, and providing a quantitative basis for subsequent determination of the pre-post sequence relationship of knowledge points.

[0130] For each knowledge point pair in the knowledge point pair list, based on the knowledge mastery conditional probability and the timing confidence, a set of pre-post sequence scores are calculated with the two knowledge points in the knowledge point pair as pre-knowledge points, including:

[0131] For the first knowledge point and the second knowledge point in each knowledge point pair, the first pre-post sequence score and the second pre-post sequence score are calculated:

[0132]

[0133]

[0134] in, For time series confidence, For mastering the knowledge points, conditional probability. This refers to the conditional probability of a knowledge point not yet mastered.

[0135] In one feasible implementation, the sequential relationship of knowledge points in each knowledge point pair is determined by comparing it with a preset scoring threshold, including:

[0136] If the score of the first preceding and following knowledge point is greater than the score of the second preceding and following knowledge point, and the score of the first preceding and following knowledge point is greater than the preset scoring threshold, the first knowledge point is determined to be the preceding knowledge point of the second knowledge point, thus obtaining the preceding and following knowledge point relationship of the knowledge point pair.

[0137] If the score of the second preceding and following knowledge point is greater than the score of the first preceding and following knowledge point, and the score of the second preceding and following knowledge point is greater than the preset scoring threshold, the second knowledge point is determined to be the preceding knowledge point of the first knowledge point, and the preceding and following knowledge point relationship of the knowledge point pair is obtained.

[0138] First, based on each knowledge point Time series confidence The conditional probabilities of knowledge point pairs, namely the conditional probability of mastering a knowledge point pair and the conditional probability of not mastering a knowledge point pair, are calculated for each knowledge point pair in the list. Calculate the preceding and following scores for both directions: First preceding and following score measure As Dependency strength of prior knowledge points; Second prior- and post-precedence scores measure As Dependency strength of prerequisite knowledge points.

[0139] Then and In comparison, if and Then determine for The prerequisite knowledge points, among which The preset scoring threshold can be 0.5. and Then determine for The preceding knowledge points. This process ensures that the judgment of the preceding and following order simultaneously satisfies the temporal logic and the correlation with the mastery state, thus improving the accuracy of the results.

[0140] For example, using knowledge points For example, It is a linear equation in one variable. is a quadratic equation. Timing confidence , , , and synchronous calculation , . Calculate the first pre-post sequence score , the second pre-post sequence score . Compare the results and , determine as pre-knowledge points, i.e. linear equations are pre-knowledge points of quadratic equations. The result will enter the large language model verification link in step S160, ensuring consistency with junior high school mathematics teaching logic, i.e. linear equations first and quadratic equations later, providing a reliable basis for personalized learning path planning.

[0141] The knowledge point pre-post sequence relationship construction method based on user test record provided by the application, the embodiment automatically constructs the knowledge point pre-post sequence relationship by using real test behavior data, effectively overcoming the problems of low efficiency of manual annotation and lack of direction of static association rules in the prior art; by dividing the test record into learning sessions and calculating the timing confidence, the pre-post sequence between knowledge points can be dynamically captured from the learning process, ensuring the logical progression of the relationship; combining the textbook directory information to filter the logical association knowledge point pairs, and based on the knowledge mastery conditional probability to evaluate the learning dependence, the accuracy and timeliness of the relationship construction are enhanced; finally, the relationship is corrected by the large language model verification, further improving the matching degree of the knowledge point pre-post sequence relationship and the real learning process, thereby providing a reliable basis for personalized learning path planning, and significantly improving the learning efficiency and teaching quality.

[0142] Figure 3 is a structural schematic diagram of a specific embodiment of a knowledge point pre-post sequence relationship construction system based on user test record provided by the embodiment of the application, referring to Figure 3 , the system can include:

[0143] The data acquisition and preprocessing module 310 is used to obtain the test record data set of each user, and to perform data cleaning and preprocessing on the test record data set. Specifically, the module extracts user historical test records from the online learning platform database, including user ID, question ID, answer result, timestamp and answer duration, etc. information, and obtains the mapping relationship between questions and knowledge points, supporting multiple knowledge point association for one question. The obtained test record is cleaned, including filtering invalid records with test duration less than a preset threshold (such as 10 seconds), and aggregating multiple answer records of the same user to the same question within the same time interval, taking the last answer result as the final record. Finally, a test record data set with uniform format and arranged in time sequence is formed, providing a high-quality data basis for subsequent analysis.

[0144] The antecedent and postcedent relationship analysis engine module 320 is configured to analyze the preprocessed problem-solving record dataset for knowledge point antecedent and postcedent relationship. The module first divides each user's problem-solving record into multiple learning sessions according to time intervals, creates multiple knowledge point pairs based on the order of appearance of knowledge points in each learning session, and counts the time sequence appearance frequency of knowledge point pairs in all learning sessions to calculate the time sequence confidence of each knowledge point pair. Subsequently, the chapter attribution relationship in the textbook directory information is used to construct a knowledge point prior relationship matrix, and knowledge point pairs with logical association are screened out. In combination with an association rule mining algorithm such as the Apriori algorithm, frequent co-occurring knowledge point pairs are mined from the problem-solving records to form a candidate knowledge point pair list. Based on the answer results in the problem-solving records, the mastery status of each user for each knowledge point is calculated, and the conditional probability of the knowledge point pair, including the knowledge point pair mastery conditional probability and the knowledge point pair non-mastery conditional probability, is calculated according to the mastery status. For each knowledge point pair, the antecedent and postcedent scores in two directions are calculated in combination with the time sequence confidence and the conditional probability, and the antecedent and postcedent relationship of the knowledge point pair is determined by comparing with a preset scoring threshold.

[0145] The large model verification module 330 is configured to perform secondary verification on the knowledge point antecedent and postcedent relationship output by the antecedent and postcedent relationship analysis engine module 320. The module constructs a verification instruction (Prompt) based on the subject field, judgment target and reasoning basis, calls a large language model to process the verification instruction, and obtains a verification result. If the antecedent and postcedent relationship returned by the large language model is inconsistent with the output of the engine module, the relationship is modified according to the verification result to generate the target knowledge point antecedent and postcedent relationship, ensuring that the relationship conforms to the educational logic and subject discipline.

[0146] The relationship graph output module 340 is configured to structure and organize the knowledge point antecedent and postcedent relationship modified by the large model verification, and output to a graph database such as Neo4j for storage. The knowledge point relationship graph generated by the module can support personalized learning path planning, adaptive question recommendation and other educational applications, and provide a dynamic and accurate knowledge dependency relationship basis for online learning systems.

[0147] Figure 4 Another specific embodiment of a knowledge point antecedent and postcedent relationship construction system based on user problem-solving records provided by the embodiments of the present application is shown in a structural schematic diagram, which is described with reference to Figure 4 The system can include:

[0148] The acquisition module 410 is configured to acquire a problem-solving record dataset of each user, and the problem-solving record dataset includes multiple problem-solving records sorted in time sequence, each problem-solving record including a user ID, a question ID, a question associated knowledge point, an answer result and a timestamp;

[0149] The distribution module 420 is configured to divide each user's problem record data set into a plurality of learning sessions according to time intervals, create a plurality of knowledge point pairs with a priori and posteriori sequence based on the associated knowledge points in each learning session, and assign a time sequence confidence to each knowledge point pair by statistically ordering the problem record of different knowledge points associated in all learning sessions.

[0150] The determination module 430 is configured to determine a knowledge point pair with a logical association from all knowledge points by using the preset textbook directory information and the problem record data set, and obtain a knowledge point pair list.

[0151] The determination module 430 is further configured to calculate the mastery state of each knowledge point of each user based on the answer results in the problem record data set, and determine the knowledge mastery conditional probability of each knowledge point pair according to the mastery state.

[0152] The determination module 430 is further configured to calculate a set of priori and posteriori sequence scores for each knowledge point pair in the knowledge point pair list based on the knowledge mastery conditional probability and the time sequence confidence, respectively taking the two knowledge points in the knowledge point pair as the priori knowledge points, and determine the knowledge point priori and posteriori sequence relationship of each knowledge point pair by comparing with a preset scoring threshold.

[0153] The correction module 440 is configured to construct a verification instruction for the knowledge point priori and posteriori sequence relationship, obtain a verification result by using a large language model to process the verification instruction, and correct the knowledge point priori and posteriori sequence relationship by using the verification result, and generate a target knowledge point priori and posteriori sequence relationship.

[0154] In one embodiment, the determination module 430 is specifically configured to filter out the knowledge point pairs with a logical association relationship from all knowledge points by using the chapter attribution information of the knowledge points in the textbook directory information, and obtain a first knowledge point pair list; construct a knowledge point priori relationship matrix by using the chapter attribution information of the knowledge points, each element in the knowledge point priori relationship matrix representing a strength value of the logical association strength of the knowledge point pair; determine the knowledge point pairs with a user proportion greater than a preset proportion for learning a certain knowledge point pair in the problem record data set by using an association rule mining algorithm, and obtain a candidate knowledge point pair list; use the strength value as the priori weight of the knowledge point pairs in the candidate knowledge point pair list, and filter the candidate knowledge point pair list by using a preset weight threshold, and obtain a second knowledge point pair list; and determine the knowledge point pair list based on the intersection of the first knowledge point pair list and the second knowledge point pair list.

[0155] In an embodiment, the distribution module 420 is specifically configured to divide each user's problem record dataset into a plurality of learning sessions according to time intervals, each learning session including a set of problem records arranged in chronological order; based on the order of the knowledge points associated with the questions in each learning session, create a plurality of sets of knowledge point pairs with a front-back sequence, each knowledge point pair including a first knowledge point and a second knowledge point, the first knowledge point preceding the second knowledge point; based on the timestamps corresponding to the question IDs associated with the knowledge points in each learning session, respectively calculate the number of times the first knowledge point appears before the second knowledge point in any one set of knowledge point pairs in the same learning session, and the number of times the first knowledge point and the second knowledge point appear together; in all learning sessions, calculate the ratio of the sum of the number of times the first knowledge point appears before the second knowledge point in each set of knowledge point pairs to the sum of the number of times the first knowledge point and the second knowledge point appear together, to obtain the time sequence confidence of each set of knowledge point pairs.

[0156] In an embodiment, the problem record further includes a resource learning identifier representing whether there is an associated resource learning behavior when answering the question for the question ID, the resource learning identifier including a resource-associated knowledge point and a learning timestamp; the distribution module 420 is further configured to, when the resource learning identifier represents that there is an associated resource learning behavior when answering the question for the question ID, determine whether the knowledge points associated with the resource-associated knowledge point and the question-associated knowledge point have the same common knowledge point, and when the knowledge points associated with the resource-associated knowledge point and the question-associated knowledge point have the same common knowledge point, associate the learning timestamp with the common knowledge point; when both the first knowledge point and the second knowledge point of a set of knowledge point pairs have associated learning timestamps, and the learning timestamp associated with the first knowledge point precedes the learning timestamp associated with the second knowledge point, use a preset weighting coefficient to weight the time sequence confidence corresponding to the knowledge point pair to obtain an updated time sequence confidence.

[0157] In an embodiment, the distribution module 420 is further configured to, when both the first knowledge point and the second knowledge point of a set of knowledge point pairs have associated learning timestamps, and the learning timestamp associated with the first knowledge point precedes the learning timestamp associated with the second knowledge point, determine a decay weight in a preset exponential decay function according to the time interval between the learning timestamp associated with the first knowledge point and the learning timestamp associated with the second knowledge point; use the preset weighting coefficient and the decay weight to weight the time sequence confidence corresponding to the knowledge point pair to obtain an updated time sequence confidence.

[0158] In an embodiment, the knowledge mastery conditional probability includes a knowledge point pair mastery conditional probability and a knowledge point pair non-mastery conditional probability; the determination module 430 is specifically configured to, for the first knowledge point and the second knowledge point in each knowledge point pair, calculate a first front-back sequence score and the second pre-posterior score :

[0159]

[0160]

[0161] wherein, is a timing confidence, is a knowledge point pair mastery conditional probability, is a knowledge point pair non-mastery conditional probability.

[0162] In one embodiment, the determining module 430 is specifically configured to determine that the first knowledge point is a pre-knowledge point of the second knowledge point, and obtain the knowledge point pre-posterior relationship of the knowledge point pair, in a case where the first pre-posterior score is greater than the second pre-posterior score, and the first pre-posterior score is greater than a preset score threshold; and determine that the second knowledge point is a pre-knowledge point of the first knowledge point, and obtain the knowledge point pre-posterior relationship of the knowledge point pair, in a case where the second pre-posterior score is greater than the first pre-posterior score, and the second pre-posterior score is greater than the preset score threshold.

[0163] The knowledge point pre-posterior relationship construction system based on user exercise records provided in the embodiments of the present application is used to implement the knowledge point pre-posterior relationship construction method based on user exercise records described above, and therefore the specific embodiments in the knowledge point pre-posterior relationship construction system based on user exercise records can be seen from the embodiment part of the knowledge point pre-posterior relationship construction method based on user exercise records in the foregoing, and the specific embodiments can be referred to the description of the corresponding embodiment part, which will not be described herein again.

[0164] Figure 5 A hardware structure schematic diagram of an electronic device provided in an embodiment of the present application is shown.

[0165] The electronic device can include a processor 510 and a memory 520 having computer program instructions stored therein.

[0166] Specifically, the processor 510 described above can include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or can be configured as one or more integrated circuits that implement the embodiments of the present application.

[0167] The memory 520 can include mass storage for data or instructions. As an example and not by way of limitation, the memory 520 can include a hard disk drive (HDD), a floppy disk drive, flash memory, an optical disc (e.g., a compact disc (CD) or a digital versatile disc (DVD)), a solid-state drive (SSD), a USB drive, or a combination of two or more of these. Where appropriate, the memory 520 can include removable or non-removable (or fixed) media. Where appropriate, the memory 520 can be internal or external to the integrated gateway disaster recovery appliance. In particular embodiments, the memory 520 is non-volatile, solid-state memory.

[0168] The memory can include read-only memory (ROM), random-access memory (RAM), magnetic disk storage mediums, optical storage mediums, flash memory devices, electrical, optical, or other physical / tangible memory storage devices. Thus, in general, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software that, when executed (by one or more processors), is operable to perform operations described with reference to the methods according to the first aspect of the present disclosure.

[0169] The processor 510 implements any one of the above-mentioned knowledge point sequence relationship construction methods based on user's test records by reading and executing the computer program instructions stored in the memory 520.

[0170] In one example, the electronic device can further include a communication interface 530 and a bus 540. As shown, the processor 510, the memory 520, and the communication interface 530 are connected through the bus 540 and complete communication among each other. Figure 5

[0171] The communication interface 530 is mainly used to realize the communication between the modules, devices, units and / or equipment in the embodiments of the present application.

[0172] ​Bus 540 includes hardware, software, or both, to couple components of the online data traffic metering device to each other and to couple components to other components within the online data traffic metering device. While bus 540 is shown for the sake of clarity as a single bus, bus 540 can include one or more buses operating together, serially, in parallel, etc. Bus 540 can include any suitable bus or interconnect, including a memory bus, a peripheral bus, an external bus, a serial bus, a parallel bus, etc. or a combination of one or more of the above. Bus 540 can include any suitable bus or interconnect, including an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand (IB) interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or another suitable bus or interconnect. In some embodiments, bus 540 can include one or more buses, although only one is shown. Although this application describes and illustrates a particular bus, this application contemplates any suitable bus or interconnect.

[0173] The electronic device can perform the knowledge point pre-post sequence relationship construction method based on user test record in the embodiments of the application, so as to realize the knowledge point pre-post sequence relationship construction method based on user test record described above. Figure 1 and Figure 2 The knowledge point pre-post sequence relationship construction method based on user test record described above.

[0174] In addition, in combination with the knowledge point pre-post sequence relationship construction method based on user test record in the above embodiments, the embodiments of the application can provide a computer readable storage medium to realize. The computer readable storage medium has computer program instructions stored thereon; the computer program instructions are executed by the processor to realize any one of the knowledge point pre-post sequence relationship construction methods based on user test record in the above embodiments.

[0175] It should be noted that the present application is not limited to the specific configurations and processes described above and shown in the drawings. For the sake of brevity, detailed descriptions of well-known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between steps, after understanding the spirit of the present application.

[0176] The functions noted in the description of the structural block diagrams above can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, and the like. When implemented in software, the elements of the present application are program or code segments that are used to perform the required tasks. The program or code segments can be stored in a machine-readable medium, or transmitted through a data signal carried in a carrier wave over a transmission medium or communication link. A "machine-readable medium" includes any medium that can store or transport information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, and the like. The code segments can be downloaded via computer networks such as the Internet, intranets, and the like.

[0177] It is also important to note that the examples mentioned in the present application describe some methods or systems based on a series of steps or devices. However, the present application is not limited to the order of the steps mentioned above, that is, the steps can be performed in the order mentioned in the examples, or in an order different from the examples, or several steps can be performed simultaneously.

[0178] The above-described aspects of the present application can be implemented in a computer program product, a method, an apparatus (system), and a computer program product. It should be understood that each block of the flowchart and / or block diagrams and combinations of blocks in the flowchart and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / acts specified in the flowchart and / or block diagram block or blocks. These processors can be, but are not limited to, general-purpose processors, special-purpose processors, special-purpose application-specific processors, or field-programmable logic arrays (FPLAs). It should also be understood that each block of the flowchart and / or block diagrams and combinations of blocks in the flowchart and / or block diagrams can also be implemented by special-purpose hardware-based computer systems that perform some or all of the functions / acts specified in the flowchart and / or block diagram block or blocks. The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable data processing apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable data processing apparatus implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0179] With the above, only for the specific implementation of the present application, those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, module and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be described here. It should be understood that the protection scope of the present application is not limited to this, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed in the present application, and these modifications or replacements should be covered within the protection scope of the present application. The above provides a kind of knowledge point before and after sequence relationship construction method, system, electronic equipment and storage medium based on user doing record provided in the present application are introduced in detail. The principle and implementation of the present application are described in this paper by applying specific examples. The above description of the embodiments is only used to help understand the method and its core idea of the present application. It should be pointed out that, for those skilled in the art, without departing from the principle of the present application, some improvements and modifications can be made to the present application, and these improvements and modifications also fall within the protection scope of the present application.

Claims

1. A method for constructing a knowledge point sequence relationship based on a user's test record, characterized in that, The method comprises: obtaining a problem-solving record data set of each user, the problem-solving record data set comprising a plurality of time-sequenced problem-solving records, each of the problem-solving records comprising a user ID, a question ID, a question-related knowledge point, an answer result and a timestamp; dividing the problem-solving record data set of each user into a plurality of learning sessions according to time intervals, creating a plurality of knowledge point pairs with a pre-post sequence based on the related knowledge points in each of the learning sessions, assigning a time sequence confidence to each knowledge point pair by counting the time sequence of the problem-solving records associated with different knowledge points in all the learning sessions; determining a knowledge point pair list by determining knowledge point pairs with logical associations among all knowledge points using a preset textbook directory information and the problem-solving record data set; calculating a mastery state of each knowledge point for each user based on the answer results in the problem-solving record data set, and determining a knowledge mastery conditional probability of each knowledge point pair according to the mastery state; for each knowledge point pair in the knowledge point pair list, calculating a pre-post sequence score for each knowledge point pair by taking the two knowledge points in the knowledge point pair as pre-sequence knowledge points based on the knowledge mastery conditional probability and the time sequence confidence, and determining a knowledge point pre-post sequence relationship of each knowledge point pair by comparing with a preset scoring threshold; constructing a verification instruction for the knowledge point pre-post sequence relationship, obtaining a verification result by processing the verification instruction using a large language model, and correcting the knowledge point pre-post sequence relationship using the verification result to generate a target knowledge point pre-post sequence relationship.

2. The method of claim 1, wherein, The method comprises: screening knowledge point pairs with logical association relationships from all knowledge points using chapter attribution information of knowledge points in the textbook directory information to obtain a first knowledge point pair list; constructing a knowledge point prior relationship matrix using the chapter attribution information of knowledge points, each element in the knowledge point prior relationship matrix representing a strength value of the logical association strength of a knowledge point pair; determining knowledge point pairs with a user proportion of learning a certain knowledge point pair greater than a preset proportion in the problem-solving record data set by an association rule mining algorithm to obtain a candidate knowledge point pair list; using the strength value as a prior weight of a knowledge point pair in the candidate knowledge point pair list, and screening the candidate knowledge point pair list using a preset weight threshold to obtain a second knowledge point pair list; determining the knowledge point pair list based on the intersection of the first knowledge point pair list and the second knowledge point pair list.

3. The method of claim 1, wherein, The method comprises: dividing the problem record dataset of each user into a plurality of learning sessions according to time intervals, each of the learning sessions comprising a set of the problem records arranged in time sequence; creating a plurality of knowledge point pairs with a front-back sequence based on an appearance sequence of the knowledge points associated with the questions in each of the learning sessions, each of the knowledge point pairs comprising a first knowledge point and a second knowledge point, the first knowledge point appearing before the second knowledge point; calculating, based on time stamps corresponding to question IDs of the knowledge points associated with each of the learning sessions, a number of times that the first knowledge point appears before the second knowledge point in any one of the knowledge point pairs in the same learning session and a number of times that the first knowledge point and the second knowledge point appear together in the same learning session, respectively; calculating, in all of the learning sessions, a ratio of a sum of the number of times that the first knowledge point appears before the second knowledge point in each of the knowledge point pairs to a sum of the number of times that the first knowledge point and the second knowledge point appear together in each of the knowledge point pairs, to obtain the time sequence confidence of each of the knowledge point pairs.

4. The method of claim 3, wherein, The problem record further comprises a resource learning identifier indicating whether there is an associated resource learning behavior when answering the question corresponding to the question ID, the resource learning identifier comprising a resource associated knowledge point and a learning time stamp; The method further comprises: when the resource learning identifier indicates that there is an associated resource learning behavior when answering the question corresponding to the question ID, determining whether there is a same common knowledge point between the knowledge point associated with the resource associated knowledge point and the knowledge point associated with the question, and associating the learning time stamp to the common knowledge point when there is the same common knowledge point between the knowledge point associated with the resource associated knowledge point and the knowledge point associated with the question; when the first knowledge point and the second knowledge point of a knowledge point pair both have the associated learning time stamp, and the learning time stamp associated with the first knowledge point is before the learning time stamp associated with the second knowledge point, weighting the time sequence confidence corresponding to the knowledge point pair by using a preset weighting coefficient to obtain an updated time sequence confidence.

5. The method of claim 4, wherein, The method further comprises: when the first knowledge point and the second knowledge point of a knowledge point pair both have the associated learning time stamp, and the learning time stamp associated with the first knowledge point is before the learning time stamp associated with the second knowledge point, determining a decay weight in a preset exponential decay function according to a time interval between the learning time stamp associated with the first knowledge point and the learning time stamp associated with the second knowledge point; weighting the time sequence confidence corresponding to the knowledge point pair by using a preset weighting coefficient and the decay weight to obtain an updated time sequence confidence.

6. The method of claim 1, wherein, The knowledge mastery conditional probability comprises a knowledge point pair mastery conditional probability and a knowledge point pair non-mastery conditional probability; The method further comprises: calculating, for each of the knowledge point pairs in the knowledge point pair list, a set of front-back sequence scores based on the knowledge mastery conditional probability and the time sequence confidence, with the two knowledge points in the knowledge point pair as the front-back sequence knowledge points, comprising: For each knowledge point pair of a first knowledge point and a second knowledge point , a first pre-post sequence score and a second pre-post sequence score are calculated: ； ； wherein, is the timing confidence, is the knowledge point pair mastery conditional probability, is the knowledge point pair non-mastery conditional probability.

7. The method of claim 6, wherein, The step of determining the sequential relationship of knowledge points in each knowledge point pair by comparing with a preset scoring threshold includes: If the first preceding-follower score is greater than the second preceding-follower score, and the first preceding-follower score is greater than the preset scoring threshold, then the first knowledge point is determined to be the preceding knowledge point of the second knowledge point, and the preceding-follower relationship of the knowledge point pair is obtained. If the second preceding-follower score is greater than the first preceding-follower score, and the second preceding-follower score is greater than the preset scoring threshold, then the second knowledge point is determined to be the preceding knowledge point of the first knowledge point, and the preceding-follower relationship of the knowledge point pair is obtained.

8. A system for constructing a sequence relationship of knowledge points based on a user's test record, characterized by comprising: include: The acquisition module is used to acquire a dataset of each user's question-answering records. The dataset includes multiple question-answering records sorted in chronological order. Each question-answering record includes a user ID, a question ID, a knowledge point associated with the question, an answer result, and a timestamp. The allocation module is used to divide the dataset of each user's question-answering records into multiple learning sessions according to time intervals, create multiple sets of knowledge point pairs with sequential order based on the associated knowledge points in each learning session, and assign time sequence confidence to each set of knowledge point pairs by statistically analyzing the time order of the question-answering records associated with different knowledge points in all learning sessions. The determination module is used to identify logically related knowledge point pairs among all knowledge points using the preset textbook catalog information and the dataset of question-answering records, and to obtain a list of knowledge point pairs; The determination module is also used to calculate the mastery status of each knowledge point for each user based on the answer results in the question-answering record dataset, and to determine the knowledge mastery conditional probability of each knowledge point pair based on the mastery status; The determination module is also used to calculate a set of preceding and following scores for each knowledge point pair in the knowledge point pair list, based on the knowledge mastery conditional probability and the temporal confidence, taking the two knowledge points in the knowledge point pair as preceding knowledge points, and determine the preceding and following relationship of knowledge points in each knowledge point pair by comparing with a preset scoring threshold. The correction module is used to construct verification instructions for the pre- and post-order relationships of the knowledge points, process the verification instructions using a large language model to obtain verification results, and use the verification results to correct the pre- and post-order relationships of the knowledge points to generate the pre- and post-order relationships of the target knowledge points.

9. An electronic device, comprising: include: Memory, used to store computer programs; A processor, configured to implement the steps of the method for constructing the sequential relationship of knowledge points based on user's question-answering records as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, enables the method for constructing the sequential relationship of knowledge points based on user's question-answering records as described in any one of claims 1 to 7.

Images