Personalized semantic evaluation system combined with knowledge graph

Abstract

The present invention discloses a personalized semantic evaluation system combined with a knowledge graph, and relates to the field of educational assessment technology, including: obtaining a pre-constructed mathematical knowledge graph and a preset mathematical knowledge point set, and determining a target knowledge point framework; obtaining a test paper sequence and a test paper standard answer sequence of a target user in a historical time, performing matching analysis, and obtaining a test question score set sequence; performing a test question semantic analysis, and determining a test knowledge point set sequence; mapping the test knowledge point set sequence and the test question score set into a target knowledge point framework, respectively, and obtaining a test knowledge point score framework sequence; performing iterative association interaction analysis, determining a target interactive test knowledge point score framework, and performing analysis to determine a target mastery evaluation result. The present invention solves the technical problem that the prior art cannot accurately evaluate the user's knowledge point mastery, and achieves the technical effect of providing an accurate basis for learning situation evaluation.

Description

Technical Field

[0001] The present invention relates to the field of educational assessment technology, and in particular to a personalized semantic evaluation system combined with a knowledge graph. Background Art

[0002] In the existing learning assessment system, the assessment of the user's knowledge mastery usually relies on simple statistics of test scores, lacking in-depth analysis of the relationship between user learning behavior and knowledge points. This method is difficult to capture the user's weak links in learning or the specific degree of knowledge mastery. The assessment results usually only reflect the overall learning situation, but cannot provide accurate knowledge mastery information. Traditional methods cannot effectively integrate learning data from different time periods, lack analysis of the dynamic changes in the user's knowledge mastery, resulting in the assessment results being difficult to reflect the user's actual learning situation and unable to accurately assess the user's knowledge mastery. Summary of the invention

[0003] This application provides a personalized semantic evaluation system combined with a knowledge graph, which is used to solve the technical problem that the existing technology cannot accurately evaluate the user's mastery of knowledge points.

[0004] In view of the above problems, this application provides a personalized semantic evaluation system combined with a knowledge graph.

[0005] The present application provides a personalized semantic evaluation system combined with a knowledge graph, the system comprising:

[0006] The target knowledge point framework determination module is used to obtain the pre-built mathematical knowledge graph and the preset mathematical knowledge point set for knowledge point indexing and determine the target knowledge point framework; the historical test paper acquisition module is used to obtain the test paper sequence and the test paper standard answer sequence of the target user in the historical time; the matching analysis module is used to match the test paper sequence with the test paper standard answer sequence to obtain the test question score set sequence; the test question semantic analysis module is used to traverse the test paper sequence to perform test question semantic analysis and determine the test knowledge point set sequence; the mapping module is used to map the test knowledge point set sequence and the test question score set into the target knowledge point framework respectively to obtain the test knowledge point score framework sequence; the association interaction analysis module is used to perform iterative association interaction analysis on the test knowledge point score framework sequence to determine the target interactive test knowledge point score framework; the analysis and evaluation module is used to call the personalized evaluation network layer to analyze the target interactive test knowledge point score framework and determine the target mastery evaluation result.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] The present application obtains a pre-constructed mathematical knowledge graph and a preset mathematical knowledge point set for knowledge point indexing, and determines a target knowledge point framework; obtains a test paper sequence and a test paper standard answer sequence for a target user in a historical time; matches a test paper sequence with a test paper standard answer sequence, and obtains a test question score set sequence; traverses the test paper sequence for question semantic analysis, and determines a test knowledge point set sequence; maps the test knowledge point set sequence and the test question score set into a target knowledge point framework, and obtains a test knowledge point score framework sequence; performs iterative association interaction analysis on the test knowledge point score framework sequence, and determines a target interactive test knowledge point score framework; calls a personalized evaluation network layer to analyze the target interactive test knowledge point score framework, and determines a target mastery evaluation result. The present invention solves the technical problem that the prior art cannot accurately evaluate the user's knowledge point mastery, and generates a target user's mastery evaluation result for a preset mathematical knowledge point set by constructing a mathematical knowledge graph, parsing a test paper, mapping a knowledge point score framework, and calling a personalized evaluation network layer for analysis, so as to achieve a technical effect of providing an accurate basis for learning situation evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A schematic diagram of the structure of a personalized semantic evaluation system combined with a knowledge graph provided in an embodiment of the present application.

[0011] Figure 2 A schematic flow chart of a personalized semantic evaluation method combined with a knowledge graph provided in an embodiment of the present application.

[0012] Explanation of the accompanying drawings: target knowledge point framework determination module 11, history test paper acquisition module 12, matching analysis module 13, test question semantic analysis module 14, mapping module 15, association interaction analysis module 16, analysis and evaluation module 17. DETAILED DESCRIPTION

[0013] This application provides a personalized semantic evaluation system combined with a knowledge graph to solve the technical problem that the existing technology cannot accurately evaluate the user's mastery of knowledge points. By constructing a mathematical knowledge graph, parsing test papers, mapping a knowledge point scoring framework, and calling a personalized evaluation network layer analysis, it generates an evaluation result of the target user's mastery of a preset set of mathematical knowledge points, thereby achieving the technical effect of providing an accurate basis for learning status evaluation.

[0014] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0015] It should be noted that any variations of the terms "include" and "have" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0016] Embodiment 1, as Figure 1 As shown, the embodiment of the present application provides a personalized semantic evaluation system combined with a knowledge graph for performing the following Figure 2 The personalized semantic evaluation method combined with the knowledge graph shown in the figure includes:

[0017] The target knowledge point framework determination module 11 is used to obtain a pre-constructed mathematical knowledge graph and a preset mathematical knowledge point set to perform knowledge point indexing and determine the target knowledge point framework.

[0018] In an embodiment of the present application, the target knowledge point framework determination module first determines the entity type set according to the characteristics of mathematical knowledge, that is, the mathematical knowledge points are divided into different categories, including concepts, formulas, methods and theorems. Then, based on the above-mentioned entity type set, a comprehensive search is performed on the mathematical knowledge content to screen out multiple mathematical knowledge entities related to the learning needs of the target user. These entities correspond to different types of mathematical knowledge points, and as the basic components of the knowledge graph, they lay the foundation for further knowledge organization. Then, the retrieved mathematical knowledge entities are enumerated in pairs to form multiple entity combinations; at the same time, relationship identification is performed for each enumerated combination, such as determining whether there is a dependency relationship, a derivation relationship or an application relationship between the knowledge points, so as to obtain the logical connection between the mathematical knowledge points.

[0019] After completing the above steps, the mathematical knowledge graph is constructed based on the retrieved multiple mathematical knowledge entities and their enumeration relationships. The mathematical knowledge graph is a formal structure of a graph, in which nodes represent knowledge points and edges represent the associations between knowledge points. The graph intuitively reflects the hierarchy and dependency of mathematical knowledge points and is the basis for generating the framework of target knowledge points.

[0020] After the graph is constructed, the preset mathematical knowledge point set uploaded by the user is obtained. The preset mathematical knowledge point set, i.e., the knowledge points that the target user needs to master, is used to search the mathematical knowledge graph by indexing, and the partial graph containing these knowledge points and their related knowledge points is screened out, and finally the target knowledge point framework is generated. The target knowledge point framework is presented in the form of multiple nodes, where each node corresponds to a knowledge point.

[0021] Furthermore, in the system provided by the embodiment of the application, the target knowledge point framework determination module 11 is also used for:

[0022] According to the characteristics of mathematical knowledge, a set of entity types is determined; mathematical knowledge entity retrieval is performed based on the entity type set to determine a plurality of mathematical knowledge entities; the plurality of mathematical knowledge entities are traversed to perform pairwise enumeration to obtain a plurality of enumeration combinations; relationship identification is performed on the plurality of enumeration combinations to obtain a plurality of enumeration combination relationships; and the mathematical knowledge graph is constructed based on the plurality of mathematical knowledge entities and the plurality of enumeration combination relationships.

[0023] In the embodiment of the present application, first, a set of entity types is determined. Technical experts divide the knowledge points into different types of entity sets according to the intrinsic properties of the mathematical knowledge points and the teaching application requirements, such as concepts (such as "linear equation"), formulas (such as "Pythagorean theorem formula"), methods (such as "substitution method to solve equations") and theorems (such as "similar triangle theorem").

[0024] Next, based on the determined set of entity types, knowledge entity retrieval is performed on the mathematical knowledge content. By combining rule matching with context semantic analysis, extraction rules (such as formula symbol pattern detection, definitional statement recognition rules, etc.) are designed to process the mathematical corpus data and extract knowledge point entities that meet the classification requirements. Semantic analysis uses an algorithm based on corpus frequency distribution and context association to analyze the semantic consistency between knowledge points and their context. The extracted entity set is deduplicated and manually verified to ensure accuracy, and finally multiple mathematical knowledge entities are generated, each with clear classification attributes.

[0025] After obtaining the knowledge entities, the full permutation algorithm combined with logical screening is used to enumerate the entity sets in pairs to generate multiple enumeration combinations. Subsequently, the enumeration results are screened according to the association rules of mathematical knowledge points (such as knowledge points must belong to the same field or have potential dependencies), and logical entity combinations are retained to provide input for the relationship identification link.

[0026] In the relationship identification stage, the enumeration combination is analyzed through a rule-based and template matching method. First, the classification rules and templates are designed based on common logical relationships in mathematical knowledge (such as "dependence", "deduction" or "application"). For each enumeration combination, these rules are used to determine its relationship type. For example, "Pythagorean theorem formula" and "right triangle definition" conform to the "deduction relationship", while "equation solution" and "substitution method" conform to the "application relationship". Through rule matching, the specific relationship between knowledge point pairs is determined one by one, and multiple enumeration combination relationships are obtained.

[0027] Finally, combining the identified multiple mathematical knowledge entities and multiple enumeration combination relationships, a mathematical knowledge graph is constructed using a graph data structure. The knowledge entities are mapped to nodes of the graph, each node corresponds to a knowledge point and contains type and basic attribute information; the logical relationship between knowledge points is mapped to the edge of the graph, and each edge is labeled with the corresponding relationship type.

[0028] Through the above process, the construction of the mathematical knowledge graph is completed.

[0029] Furthermore, in the system provided by the embodiment of the application, the target knowledge point framework determination module 11 is also used for:

[0030] The entity type set includes concepts, formulas, methods and theorems.

[0031] In the embodiments of the present application, the entity type set includes concepts, formulas, methods, and theorems. A concept refers to a basic definition or characteristic description of a mathematical object, such as "the definition of a function". A formula is a rule or calculation method expressed in mathematical symbols, such as "the Pythagorean theorem formula". A method refers to a step or technique for solving a mathematical problem, such as "the method of substitution". A theorem is a proven mathematical statement, such as "the similar triangle theorem".

[0032] The historical test paper acquisition module 12 is used to acquire the test paper sequence and the test paper standard answer sequence of the target user in the historical time.

[0033] In the embodiment of the present application, the test paper data and standard answers of the target user are first directly extracted from the historical database by the database query method. Based on the user's unique identifier (such as user ID), the SQL query language is used to query the test paper table and the answer table in a linked manner to extract the test paper number, question list, question type, test time and the corresponding standard answer sequence. These test papers and answer data have been stored in the database in a one-to-one correspondence according to the test paper number.

[0034] Subsequently, the extracted test paper contents are matched with knowledge points and verified for coverage. Using the preset set of mathematical knowledge points, the corresponding knowledge points are marked one by one for each test paper through a rule matching algorithm. For example, by comparing the terms in the question field (such as "Pythagorean Theorem" or "Function Graph") with the entries in the preset knowledge point set, the knowledge points involved in each question are marked. For questions in the database that have already been marked with knowledge points, the existing marking results are called to further improve efficiency. After completing the knowledge point marking, the coverage of the knowledge points in each test paper is counted and compared with the preset set of mathematical knowledge points to ensure that each test paper contains at least one-third of the knowledge points. If some test papers do not meet the coverage requirements, they are marked as ineligible and removed from the results.

[0035] Finally, the test paper sequence and the test paper standard answer sequence are sorted out in chronological order. These sequences contain detailed question information and standard answers for each test paper.

[0036] The matching analysis module 13 is used to match the test paper sequence with the test paper standard answer sequence to obtain the test question score set sequence.

[0037] In the embodiment of the present application, the questions in the test paper and the corresponding standard answers are directly matched through the correspondence between the test paper number and the question number. The mapping relationship between the test paper table and the answer table in the database is used to extract the user answer and the standard answer for each question from the database, and a matching table is generated to ensure that all questions have accurate answer data.

[0038] Then, answer matching analysis is performed on each question, and appropriate methods are used according to the question type. For multiple-choice questions, the character matching method is used to directly compare the answer selected by the user with the standard answer. If the two are exactly the same, it is judged as "correct"; otherwise, it is "wrong". For fill-in-the-blank questions, the character matching method is also used, requiring the user's answer to be exactly the same as the standard answer. If there is any difference between the two, it is judged as "wrong". For problem-solving questions, the user's answer is compared with the standard answer sentence by sentence according to the preset simple rules. For example, check whether the user has written the formula correctly, whether the key steps have been completely derived, and whether the correct final answer has been given. If the user's answer completely covers all the core content of the standard answer, it is judged as "correct"; if it only covers part of the content, it is recorded as "partially correct"; if it does not meet the requirements at all, it is judged as "wrong".

[0039] After the answer matching is completed, the score of each question is calculated. For multiple-choice questions and fill-in-the-blank questions, a binary scoring method is used, with full marks for correct answers and zero marks for incorrect answers. For problem-solving questions, they are broken down into multiple steps according to the scoring rules. For example, a certain score is awarded for a correct formula, another certain score is awarded for a complete derivation process, and the remaining score is awarded for a correct final answer. The score for each step is pre-set. If the user's answer only completes part of the content, the score is calculated based on the completion ratio.

[0040] After the score calculation for each question is completed, the score of each test paper is summarized. For each test paper, the scores of all questions are added together to obtain the total score of the test paper.

[0041] Finally, the scores of all test papers are sorted and summarized by time to generate the final test question score collection sequence. This sequence includes the number of each test paper, the test time, the score of each question, and the total score of the test paper.

[0042] The test question semantic analysis module 14 is used to traverse the test paper sequence to perform test question semantic analysis and determine the test knowledge point set sequence.

[0043] In an embodiment of the present application, the test question semantic analysis module uses a pre-trained test question semantic analyzer to perform semantic analysis on the test questions, identify the knowledge points corresponding to each test question and generate a test knowledge point set sequence. The training process of the analyzer includes obtaining multiple sample test papers and corresponding sample test knowledge point sets as training data; dividing them into training sets and verification sets according to a preset ratio; using the training set to train the analysis framework built based on the feedforward neural network, and supervising the training process through the verification set until the model converges. Finally, the analyzer is called to perform test question semantic analysis on the test paper sequence to determine the test knowledge point set sequence.

[0044] Furthermore, in the system provided in the application embodiment, the test question semantic analysis module 14 is also used for:

[0045] Acquire multiple sample test papers and multiple sample test knowledge point sets as training data; divide the training data into a training set and a validation set according to a preset division ratio; use the training set to train a framework built on a feedforward neural network, and use the validation set to supervise during the training process until the training converges, thereby obtaining a trained test question semantic analyzer; call the test question semantic analyzer to perform test question semantic analysis on the test paper sequence to determine the test knowledge point set sequence.

[0046] In the embodiment of the present application, a sample test paper and a sample test knowledge point set are first obtained from historical data as training data. The sample test paper includes question texts of various question types, and the sample test knowledge point set is annotated knowledge points corresponding to each question, which is used to guide the model to learn the relationship between questions and knowledge points.

[0047] Then, the training data is divided into a training set and a validation set according to the preset division ratio. Usually, the ratio is 80:20. The training set is used for model learning, and the validation set is used to monitor model performance in real time to prevent overfitting of the model during training.

[0048] Then, an analysis framework is built based on a feedforward neural network to train the model on the training set. A feedforward neural network is a common artificial neural network structure that can learn deep features of input data through layers of nonlinear transformations. The training process includes converting the test paper text into an input that the model can understand through word vectors or other feature extraction methods. The network predicts the knowledge point labels of the test questions by calculating layer by layer; by comparing the knowledge points predicted by the model with the actual labeled knowledge points, the loss function (such as cross entropy loss) is calculated, and the model parameters are optimized through the back propagation algorithm. During the training process, the validation set is used regularly to evaluate the model performance and adjust the hyperparameters until the prediction accuracy of the model reaches the convergence standard.

[0049] Finally, the trained test question semantic analyzer is called to analyze the test paper sequence question by question. The analyzer automatically identifies the knowledge point corresponding to each test question based on the mapping relationship between the question features and knowledge points learned in the training, and records the results as a test knowledge point set. After traversing all the test papers, a test knowledge point set sequence arranged in the order of the test papers is generated.

[0050] The mapping module 15 is used to map the test knowledge point set sequence and the test question score set into the target knowledge point framework respectively to obtain the test knowledge point score framework sequence.

[0051] In the embodiment of the present application, firstly, through data extraction and preparation, the knowledge point information corresponding to each test question is extracted from the test knowledge point set sequence, including the knowledge point name and its associated test question number, and the score information of each test question is extracted from the test question score set, and a one-to-one correspondence is established with the test question number to form the basic data for mapping. In addition, the structural information of the target knowledge point framework is loaded, and the knowledge point nodes and their relationships in the framework are extracted to ensure that the mapping process is consistent with the framework structure.

[0052] Next, through the process of mapping knowledge points to the framework, the knowledge points in the test knowledge point set are matched one by one to the target knowledge point framework. Using the rule matching algorithm, the knowledge points in the test knowledge points that have the same node names as those in the framework are directly mapped; for knowledge points with slightly different names, the string similarity algorithm (such as Levenshtein distance) is used to calculate the degree of matching, and the closest node is used as the mapping result. If some knowledge points cannot find corresponding nodes in the framework, they are marked as "unmapped".

[0053] After completing the knowledge point mapping, the test question scores are associated with the corresponding knowledge point nodes according to the test question score set. For questions involving only a single knowledge point, the test question scores are directly assigned to the knowledge point node; for questions involving multiple knowledge points, the test question scores are reasonably allocated to multiple knowledge points according to the preset allocation rules (such as the equal distribution rule). Then, the scores of each knowledge point node are accumulated to calculate the total score of the knowledge point, and the list of test question numbers covering the knowledge point is recorded to provide related information for subsequent analysis.

[0054] Finally, all mappings and associated data are sorted and sorted to generate a test knowledge point score framework sequence. This sequence is based on the target knowledge point framework, records the name, total score, number of covered questions and question number of each knowledge point, and arranges them in chronological order to reflect the dynamic score of the knowledge point.

[0055] The correlation interaction analysis module 16 is used to perform iterative correlation interaction analysis on the test knowledge point scoring framework sequence to determine the target interactive test knowledge point scoring framework.

[0056] In the embodiment of the present application, the association interaction analysis module performs iterative association interaction analysis on adjacent frames in the test knowledge point scoring frame sequence step by step in the order from front to back in time, and finally determines the target interactive test knowledge point scoring frame. Specifically, firstly, the adjacent previous time and next time frames in the test knowledge point scoring frame sequence are extracted, and the association interaction analysis is performed to generate a new association frame; then, the generated association frame is continuously analyzed with the scoring frame of the next time, and it is iterated in sequence until all frames in the time sequence are processed, and finally the target interactive test knowledge point scoring frame is determined.

[0057] Furthermore, in the system provided in the embodiment of the application, the correlation interaction analysis module 16 is also used for:

[0058] Extract the first test knowledge point scoring framework and the second test knowledge point scoring framework in the test knowledge point scoring framework sequence in chronological order; perform association interaction analysis on the first test knowledge point scoring framework and the second test knowledge point scoring framework to obtain a first association interaction test knowledge point scoring framework; perform association interaction analysis on the first association interaction test knowledge point scoring framework and the third test knowledge point scoring framework to obtain a second association interaction test knowledge point scoring framework; perform multiple association interaction analyses in sequence to obtain a Kth association interaction test knowledge point scoring framework, wherein K is the total number of test knowledge point scoring frameworks in the test knowledge point scoring framework sequence minus 1, and K is an integer greater than or equal to 1; use the Kth association interaction test knowledge point scoring framework as the target interaction test knowledge point scoring framework.

[0059] In an embodiment of the present application, firstly, a first test knowledge point scoring frame sorted in time is extracted from the test knowledge point scoring frame sequence, and the frame corresponds to the score data of the earliest completed test paper in the historical event. Then, a second test knowledge point scoring frame is extracted, and the frame corresponds to the data of the second test paper that follows immediately. The corresponding knowledge point nodes in the two frames are mapped and inner product calculations are performed, and the similarity of the knowledge point scores in the two frames is calculated to generate a first score similarity set.

[0060] Then, the first score similarity set is normalized, and the calculated similarity values ​​are normalized to a uniform range to ensure that different score data are comparable. The normalized similarity data is filled into a matrix stored as empty to form a first correlation interaction matrix.

[0061] Then, the graph neural network (GNN) is called to perform convolution calculation on the first association interaction matrix and the second test knowledge point scoring framework. Through the convolution operation, the score results of the knowledge points are updated by combining the interaction features in the association matrix and the scoring data of the second framework, and the first association interaction test knowledge point scoring framework is generated to reflect the association dynamics of the knowledge points before and after.

[0062] Next, based on the first correlation interaction test knowledge point scoring framework, the same correlation interaction analysis is performed with the third test knowledge point scoring framework to generate the second correlation interaction test knowledge point scoring framework. According to this process, adjacent test knowledge point scoring frameworks are extracted in sequence, and correlation interaction analysis is performed one by one until all time frames in the sequence are processed and the Kth correlation interaction analysis is completed (K is the total number of frames minus 1).

[0063] Finally, the K-th associated interaction test knowledge point scoring framework is output as the target interaction test knowledge point scoring framework.

[0064] Furthermore, in the system provided in the embodiment of the application, the correlation interaction analysis module 16 is also used for:

[0065] Perform mapping inner product calculation on the corresponding nodes in the first test knowledge point scoring framework and the second test knowledge point scoring framework to determine the first score similarity set; perform standardization on the first score similarity set, and fill the processing result into the empty matrix to obtain the first association interaction matrix; use the graph neural network to perform convolution calculation on the first association interaction matrix and the second test knowledge point scoring framework to determine the first association interaction test knowledge point scoring framework.

[0066] In an embodiment of the present application, the score values ​​of the corresponding nodes in the first test knowledge point scoring framework and the second test knowledge point scoring framework are first compared one by one, and the score correlation degree between each pair of nodes is calculated by mapping the inner product function. The calculation of the mapped inner product reflects the correlation strength of the knowledge point scores at two time points, and the higher the score, the stronger the correlation of the knowledge point at the two time points. By arranging the calculation results of this step into a first score similarity set, each item in the set represents the correlation degree of a certain knowledge point at two time points, which provides basic data for subsequent standardization processing.

[0067] Then, the first score similarity set is standardized, and all score similarities are mapped to a uniform range (such as 0 to 1) to ensure data comparability and numerical stability. The standardization process adjusts the original score similarity to a standardized set by merging the score similarities of all nodes and using the base of the natural logarithm as a processing parameter. The standardized results are then filled into an empty matrix to generate the first correlation interaction matrix, which represents the correlation and strength of all knowledge points in the first and second test knowledge point scoring frameworks.

[0068] Then, the graph neural network is used to perform convolution calculations on the first association interaction matrix and the second test knowledge point scoring framework. Specifically, the first association interaction matrix is ​​first regarded as the adjacency matrix of the graph, each knowledge point node is used as the vertex of the graph, and each value in the matrix is ​​used as the edge weight between the vertices to reflect the strength of the association between the knowledge points. Secondly, the score value in the second test knowledge point scoring framework is used as the initial feature of the node, indicating the basic score of the knowledge point at the second time point. Then, through the convolution operation of the graph neural network, the association information in the adjacency matrix is ​​fused and updated with the initial features of the node. In each convolution, the feature value of the node not only contains its own initial score, but also combines the scores of adjacent nodes and their association strength. After multiple layers of convolution iteration, the score features of each knowledge point node are gradually updated, so that the final score integrates the association information between the first test knowledge point scoring framework and the second test knowledge point scoring framework.

[0069] Finally, the node features after convolution calculation are output as updated knowledge point scores to generate the first associated interaction test knowledge point score framework.

[0070] Furthermore, in the system provided in the embodiment of the application, the correlation interaction analysis module 16 is also used for:

[0071] Obtain a mapping inner product function, wherein the mapping inner product function is:

[0072] ;

[0073] in, The first score similarity obtained by performing mapping inner product calculation on the first test knowledge point score frame and the i-th test knowledge point score in the first test knowledge point score frame, is the score of the i-th test knowledge point in the first test knowledge point scoring framework, Score the i-th test knowledge point in the second test knowledge point scoring framework.

[0074] In the embodiment of the present application, the mapping inner product function is first obtained, and the similarity of the knowledge point scores is calculated by the mapping inner product function. The mapping inner product function is used to measure the correlation between the corresponding knowledge point scores at two time points, and its calculation formula is: ; In this formula, is the score of the i-th test knowledge point in the first test knowledge point scoring framework, The score of the i-th test knowledge point in the second test knowledge point scoring framework. Indicates the score of the first test knowledge point and the second test knowledge point score This part is used to measure the degree of change in the knowledge point score between the two time frames. The smaller the difference, the more stable the change in the knowledge point score, and the larger the difference, the greater the dynamic change in the knowledge point score. The normalization process is done by the ratio of the numerator to the denominator to process the correlation between the change in the knowledge point score and the current score. A larger value indicates that the change in the knowledge point score in the two time frames is stable and the correlation is high; a smaller value indicates that the score changes more or the correlation is low.

[0075] By performing the above calculation on each knowledge point, similarity values ​​of all knowledge points are generated to form a first score similarity set.

[0076] Furthermore, in the system provided in the embodiment of the application, the correlation interaction analysis module 16 is also used for:

[0077] Construct a normalization function, wherein the normalization function is:

[0078] ;in, is the standard value corresponding to the i-th first score similarity in the first score similarity set, e is the base of the natural logarithm, and m is the total number of first score similarities in the first score similarity set; the first score similarity set is standardized using the standardization function, and the processing result is filled into the empty matrix stored to obtain a first correlation interaction matrix.

[0079] In the embodiment of the present application, the first score similarity set is first extracted from the mapping inner product calculation result. Each similarity value is calculated by the mapping inner product function of the previous stage, indicating the similarity of the corresponding knowledge point scores in the first test knowledge point scoring framework and the second test knowledge point scoring framework. Since the range of similarity values ​​may vary greatly, in order to ensure the comparability and stability of the data, a standardization function is introduced for processing.

[0080] The normalization function is ; is the standard value corresponding to the i-th first score similarity in the first score similarity set, e is the base of the natural logarithm, and m is the total number of first score similarities in the first score similarity set.

[0081] In the process of normalizing the first score similarity set, the numerator part Perform an exponential transformation on each score similarity value. Exponential transformation makes the correlation features between knowledge points more significant by amplifying larger similarity values ​​and compressing smaller similarity values. is the exponentially transformed sum of all score similarity values, used for normalization. This denominator ensures that the sum of the standardized values ​​is 1, reflecting the relative weight of the scores of each knowledge point.

[0082] Generate the standardized similarity value of each knowledge point through the processing of the standardization function , which ranges from (0, 1). The closer the value is to 1, the stronger the correlation between the knowledge point in the first and second time frames; the closer the value is to 0, the weaker the correlation or the greater the score change.

[0083] Then, all the normalized similarity values ​​are filled into a matrix stored as empty to generate a first correlation interaction matrix.

[0084] The analysis and evaluation module 17 is used to call the personalized evaluation network layer to analyze the target interactive test knowledge point scoring framework to determine the target mastery evaluation result.

[0085] In an embodiment of the present application, the target interactive test knowledge point score framework is first received as input data. Then, the personalized evaluation network layer is called for analysis. The network layer is constructed based on a feedforward neural network and is specifically used to predict the user's knowledge point mastery. During the model training process, training data is first collected from the test records of historical users, including the knowledge point score sequence, test time, and the mastery level label of each knowledge point. These mastery level labels are generated by preset rules. For example, when the user's test score exceeds a certain set threshold, it is marked as "mastered", otherwise it is marked as "not mastered". The collected data is cleaned and standardized to unify the input format. Then, the processed data is input into the network model for training. The input layer of the model accepts the user's knowledge point score information, the hidden layer extracts the key features of the knowledge point score through a simple activation function, and the output layer uses a linear regression method to perform weighted summation on the features extracted by the hidden layer to predict the mastery level value. During the training process, a supervised learning method is used to optimize the network parameters by minimizing the loss function (such as mean square error) to ensure that the model can accurately learn the mapping relationship between the knowledge point score and the mastery level.

[0086] In the actual analysis process, the target interactive test knowledge point score framework is input into the trained personalized evaluation network layer one by one. The network layer first extracts the features of each knowledge point score, converts the knowledge point score and related information into a feature vector, and then uses the mapping relationship learned in the training to calculate the mastery value of each knowledge point. The mastery value is expressed as a value between 0 and 1. A value close to 1 indicates that the user has a high degree of mastery of the knowledge point, and a value close to 0 indicates a low degree of mastery.

[0087] Through this process, the evaluation results of the target users' mastery of the preset set of mathematical knowledge points are generated.

[0088] In the embodiments of the present application, in summary, the embodiments of the present application have at least the following technical effects:

[0089] The present application obtains a pre-constructed mathematical knowledge graph and a preset mathematical knowledge point set for knowledge point indexing, and determines a target knowledge point framework; obtains a test paper sequence and a test paper standard answer sequence for a target user in a historical time; matches a test paper sequence with a test paper standard answer sequence, and obtains a test question score set sequence; traverses the test paper sequence for question semantic analysis, and determines a test knowledge point set sequence; maps the test knowledge point set sequence and the test question score set into a target knowledge point framework, and obtains a test knowledge point score framework sequence; performs iterative association interaction analysis on the test knowledge point score framework sequence, and determines a target interactive test knowledge point score framework; calls a personalized evaluation network layer to analyze the target interactive test knowledge point score framework, and determines a target mastery evaluation result. The present invention solves the technical problem that the prior art cannot accurately evaluate the user's knowledge point mastery, and generates a target user's mastery evaluation result for a preset mathematical knowledge point set by constructing a mathematical knowledge graph, parsing a test paper, mapping a knowledge point score framework, and calling a personalized evaluation network layer for analysis, so as to achieve a technical effect of providing an accurate basis for learning situation evaluation.

[0090] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0091] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0092] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A personalized semantic evaluation system combined with knowledge graph, characterized in that the system include: A target knowledge point framework determination module is used to obtain a pre-built mathematical knowledge graph and a preset mathematical knowledge point set to perform knowledge point indexing and determine a target knowledge point framework; A historical test paper acquisition module is used to obtain the test paper sequence and test paper standard answer sequence of the target user in the historical time; A matching analysis module is used to match the test paper sequence with the test paper standard answer sequence to obtain the test question score set sequence; The test question semantic analysis module is used to traverse the test paper sequence to perform test question semantic analysis and determine the test knowledge point set sequence; A mapping module, used for mapping the test knowledge point set sequence and the test question score set into the target knowledge point framework respectively, to obtain the test knowledge point score framework sequence; The correlation interaction analysis module is used to perform iterative correlation interaction analysis on the test knowledge point scoring framework sequence to determine the target interactive test knowledge point scoring framework; The analysis and evaluation module is used to call the personalized evaluation network layer to analyze the target interactive test knowledge point scoring framework and determine the target mastery evaluation result; Among them, iterative correlation interaction analysis is performed on the test knowledge point scoring framework sequence to determine the target interactive test knowledge point scoring framework, including: Extracting the first test knowledge point scoring frame and the second test knowledge point scoring frame in the test knowledge point scoring frame sequence in a time-ordered order; Performing correlation and interaction analysis on the first test knowledge point scoring framework and the second test knowledge point scoring framework to obtain a first correlation and interaction test knowledge point scoring framework; Performing an association interaction analysis on the first association interaction test knowledge point scoring framework and the third test knowledge point scoring framework to obtain a second association interaction test knowledge point scoring framework; Perform multiple association interaction analyses in sequence to obtain a Kth association interaction test knowledge point scoring framework, wherein K is the total number of test knowledge point scoring frameworks in the test knowledge point scoring framework sequence minus 1, and K is an integer greater than or equal to 1; Using the Kth associated interactive test knowledge point scoring framework as the target interactive test knowledge point scoring framework; The first test knowledge point scoring framework and the second test knowledge point scoring framework are subjected to correlation and interaction analysis to obtain a first correlation and interaction test knowledge point scoring framework, including: Performing mapping inner product calculation on corresponding nodes in the first test knowledge point scoring framework and the second test knowledge point scoring framework to determine a first score similarity set; Performing standardization processing on the first score similarity set, and filling the processing result into the empty matrix, to obtain a first correlation interaction matrix; Using a graph neural network to perform convolution calculation on the first association interaction matrix and the second test knowledge point scoring framework to determine the first association interaction test knowledge point scoring framework; Obtain a mapping inner product function, wherein the mapping inner product function is: Among them, lim(x i ,y i ) is the first score similarity obtained by mapping the first test knowledge point score frame and the i-th test knowledge point score in the first test knowledge point score frame. i is the score of the i-th test knowledge point in the first test knowledge point scoring framework, y i Score the i-th test knowledge point in the second test knowledge point scoring framework.

2. The personalized semantic evaluation system combined with knowledge graph as claimed in claim 1, characterized in that: include: Construct a normalization function, wherein the normalization function is: Among them, Mix[lim(x i ,y i )] is the standard value corresponding to the i-th first score similarity in the first score similarity set, e is the base of the natural logarithm, and m is the total number of first score similarities in the first score similarity set; The first score similarity set is standardized using the standardization function, and the processing result is filled into the empty matrix to obtain a first correlation interaction matrix.

3. The personalized semantic evaluation system combined with knowledge graph as claimed in claim 1, characterized in that: Traversing the test paper sequence to perform question semantic analysis and determine a test knowledge point set sequence, including: Acquire multiple sample test papers and multiple sample test knowledge point sets as training data; Dividing the training data into a training set and a validation set according to a preset division ratio; The framework constructed based on the feedforward neural network is trained using the training set, and the verification set is used for supervision during the training process until the training reaches convergence, thereby obtaining a trained test question semantic analyzer; The test question semantic analyzer is called to perform test question semantic analysis on the test paper sequence to determine the test knowledge point set sequence.

4. The personalized semantic evaluation system combined with knowledge graph according to claim 1, characterized in that: Obtain the pre-built mathematical knowledge graph and the preset mathematical knowledge point set for knowledge point indexing and determine the target knowledge point framework, including: Determine the entity type set based on the characteristics of mathematical knowledge; Perform mathematical knowledge entity retrieval based on the entity type set to determine multiple mathematical knowledge entities; Traversing the plurality of mathematical knowledge entities and enumerating them in pairs to obtain a plurality of enumeration combinations; Performing relationship identification on the multiple enumeration combinations to obtain multiple enumeration combination relationships; The mathematical knowledge graph is constructed based on the multiple mathematical knowledge entities and the multiple enumeration combination relationships.

5. The personalized semantic evaluation system combined with knowledge graph as claimed in claim 4, characterized in that: The entity type set includes concepts, formulas, methods and theorems.

Images