Knowledge point mastery degree prediction method, device and equipment

Abstract

The invention provides a knowledge point mastery degree prediction method, apparatus and device. The method comprises the steps of obtaining a knowledge point description sequence, a subject sequence of a target object and an auxiliary information sequence; determining a question embedding feature vector corresponding to the question sequence, an auxiliary embedding feature vector corresponding to the auxiliary information sequence and a knowledge point embedding feature vector corresponding to the knowledge point description sequence; performing multi-head attention operation based on the topic embedding feature vector and the auxiliary embedding feature vector to obtain a capability feature vector; determining a knowledge space feature vector based on the knowledge point embedding feature vector; performing multi-head attention operation based on the capability feature vector, the knowledge point embedded feature vector and the knowledge space feature vector to obtain a knowledge point target feature vector; and based on the knowledge point target feature vector and the knowledge point graph, predicting the target knowledge point mastery degree of the target object for each candidate knowledge point name in the knowledge point graph. Through the technical scheme of the invention, the knowledge point mastery degree of the learner can be accurately evaluated.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device and equipment for predicting the mastery of knowledge points. Background Art

[0002] Knowledge mastery refers to a quantitative assessment indicator of a learner's depth of understanding, application ability, and transferability of specific knowledge units (such as concepts, principles, skills, etc.). Its core lies in the observable manifestation of the degree of knowledge internalization, which is the level of understanding, memory, and application of specific knowledge content in a certain knowledge field.

[0003] In the field of intelligent education, accurately assessing a learner's (e.g., student) knowledge mastery is key to achieving personalized learning and precision teaching. However, existing technologies cannot accurately assess knowledge mastery. For example, regular exams are used to assess knowledge mastery. However, this method is time-consuming and labor-intensive, and lacks a precise assessment of knowledge mastery. Summary of the Invention

[0004] This application provides a method for predicting the mastery of knowledge points, the method comprising: Acquire a knowledge point description sequence, a target object question sequence, and an auxiliary information sequence; the knowledge point description sequence includes semantic description information corresponding to each candidate knowledge point name; the auxiliary information sequence includes the knowledge point name, question difficulty, and question response corresponding to each question in the question sequence; Determine a question embedding feature vector corresponding to the question sequence, an auxiliary embedding feature vector corresponding to the auxiliary information sequence, and a knowledge point embedding feature vector corresponding to the knowledge point description sequence; Performing a multi-head attention operation based on the question embedding feature vector and the auxiliary embedding feature vector to obtain an ability feature vector of the target object for the knowledge point name corresponding to the question sequence; Determine a knowledge space feature vector based on the knowledge point embedding feature vector, wherein the knowledge point embedding feature vector includes an initial feature vector of each candidate knowledge point name, and the knowledge space feature vector includes multiple fused feature vectors, and each fused feature vector is obtained by fusing multiple initial feature vectors; Performing a multi-head attention operation based on the capability feature vector, the knowledge point embedding feature vector, and the knowledge space feature vector to obtain a knowledge point target feature vector; Based on the knowledge point target feature vector and the acquired knowledge point map, the target object's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point map is predicted.

[0005] The present application provides a knowledge point mastery prediction device, the device comprising: An acquisition module is used to acquire a knowledge point description sequence, a target object question sequence, and an auxiliary information sequence; the knowledge point description sequence includes semantic description information corresponding to each candidate knowledge point name; the auxiliary information sequence includes the knowledge point name, question difficulty, and question response corresponding to each question in the question sequence; a determination module, configured to determine a question embedding feature vector corresponding to the question sequence, an auxiliary embedding feature vector corresponding to the auxiliary information sequence, and a knowledge point embedding feature vector corresponding to the knowledge point description sequence; and determine a knowledge space feature vector based on the knowledge point embedding feature vector, wherein the knowledge point embedding feature vector includes an initial feature vector for each candidate knowledge point name, and the knowledge space feature vector includes multiple fused feature vectors, each fused feature vector being obtained by fusing multiple initial feature vectors; a processing module configured to perform a multi-head attention operation based on the question embedding feature vector and the auxiliary embedding feature vector to obtain an ability feature vector of the target object for the knowledge point name corresponding to the question sequence; and perform a multi-head attention operation based on the ability feature vector, the knowledge point embedding feature vector, and the knowledge space feature vector to obtain a knowledge point target feature vector; A prediction module is used to predict the target object's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point map based on the knowledge point target feature vector and the acquired knowledge point map.

[0006] The present application provides an electronic device, comprising: a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions that can be executed by the processor; the processor is used to execute the machine-executable instructions to implement the knowledge point mastery prediction method of the above example of the present application.

[0007] The present application provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the knowledge point mastery prediction method of the above example of the present application.

[0008] The present application provides a machine-readable storage medium, which stores machine-executable instructions that can be executed by a processor; wherein the processor is used to execute the machine-executable instructions to implement the knowledge point mastery prediction method of the above example of the present application.

[0009] As can be seen from the above technical solutions, in the embodiments of the present application, an ability feature vector can be determined based on the topic embedding feature vector and the auxiliary embedding feature vector, a knowledge space feature vector can be determined based on the knowledge point embedding feature vector, a knowledge point target feature vector can be determined based on the ability feature vector, the knowledge point embedding feature vector, and the knowledge space feature vector, and the target knowledge point mastery of the target subject (i.e., the learner) for each candidate knowledge point name can be predicted based on the knowledge point target feature vector and the knowledge point graph. In this way, knowledge point mastery can be predicted in conjunction with the knowledge point graph, accurately assessing the learner's (e.g., student's) knowledge point mastery, reflecting knowledge point mastery in real time, and achieving precise assessment of knowledge point mastery. The assessment process is simple and rapid. By organizing knowledge points into a graph structure and fully utilizing the hierarchical relationships and dependencies between knowledge points, the assessment results are more comprehensive and accurate, achieving a precise assessment of the mastery level of each specific knowledge point, and updating the knowledge point mastery assessment results in real time. The assessment results can be interpreted from the perspective of knowledge point associations, reflecting both the student's overall learning status and specific knowledge weaknesses. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a flow chart of a method for predicting the mastery of knowledge points in one embodiment of the present application; Figure 2 This is a flow chart of a method for predicting the mastery of knowledge points in one embodiment of the present application; Figure 3A This is a schematic diagram of the structure of a knowledge point mastery prediction model in one embodiment of the present application; Figure 3B This is a schematic diagram of the structure of a knowledge point mastery prediction model in one embodiment of the present application; Figure 3C This is a schematic diagram of the structure of a knowledge point mastery prediction model in one embodiment of the present application; Figure 4 is a schematic diagram of the statistical effect of the difficulty distribution of questions on a sample data set in an embodiment; Figure 5 This is a flow chart of a method for predicting the mastery of knowledge points in one embodiment of the present application; Figure 6 This is a schematic diagram of the structure of a knowledge point mastery prediction model in one embodiment of the present application; Figure 7 This is a schematic diagram of the structure of a knowledge point mastery prediction device in one embodiment of the present application; Figure 8 It is a hardware structure diagram of an electronic device in one embodiment of the present application. DETAILED DESCRIPTION

[0011] In the embodiment of the present application, a method for predicting the mastery of a knowledge point is proposed. The method can be applied to electronic devices. Figure 1 FIG. 5 is a flow chart of the method, which may include: Step 101: Obtain a knowledge point description sequence, a target object question sequence, and an auxiliary information sequence; the knowledge point description sequence includes semantic description information corresponding to each candidate knowledge point name; the auxiliary information sequence includes the knowledge point name, question difficulty, and question response corresponding to each question in the question sequence.

[0012] Step 102: Determine the topic embedding feature vector corresponding to the topic sequence, the auxiliary embedding feature vector corresponding to the auxiliary information sequence, and the knowledge point embedding feature vector corresponding to the knowledge point description sequence.

[0013] Step 103: Perform a multi-head attention operation based on the question embedding feature vector and the auxiliary embedding feature vector to obtain the target object's ability feature vector for the knowledge point name corresponding to the question sequence.

[0014] Step 104: determine a knowledge space feature vector based on the knowledge point embedding feature vector, wherein the knowledge point embedding feature vector includes an initial feature vector of each candidate knowledge point name, and the knowledge space feature vector includes multiple fused feature vectors, and each fused feature vector is obtained by fusing multiple initial feature vectors.

[0015] Step 105: Perform a multi-head attention operation based on the capability feature vector, the knowledge point embedding feature vector, and the knowledge space feature vector to obtain a knowledge point target feature vector.

[0016] Step 106: Based on the target feature vector of the knowledge point and the acquired knowledge point graph, predict the target subject's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point graph.

[0017] Exemplarily, obtaining a knowledge point description sequence may include, but is not limited to: obtaining a knowledge point sequence, a question text sequence, and a reference matrix; wherein, the knowledge point sequence includes multiple candidate knowledge point names, the question text sequence includes multiple example question texts, and the reference matrix represents the correspondence between the candidate knowledge point names and the example question texts; inputting the knowledge point sequence, the question text sequence, and the reference matrix into a large language model, determining the semantic description information corresponding to each candidate knowledge point name through the large language model, and obtaining a knowledge point description sequence; wherein, for each candidate knowledge point name, determining the example question text corresponding to the candidate knowledge point name through the reference matrix, performing semantic analysis on the candidate knowledge point name based on the example question text, and obtaining the semantic description information corresponding to the candidate knowledge point name.

[0018] Exemplarily, a multi-head attention operation is performed based on the question embedding feature vector and the auxiliary embedding feature vector to obtain the target object's ability feature vector for the knowledge point name corresponding to the question sequence, which may include but is not limited to: determining a first Q vector, a first K vector and a first V vector based on the question embedding feature vector, and performing a multi-head attention operation based on the first Q vector, the first K vector and the first V vector to obtain a first intermediate vector; determining a second Q vector, a second K vector and a second V vector based on the auxiliary embedding feature vector, and performing a multi-head attention operation based on the second Q vector, the second K vector and the second V vector to obtain a second intermediate vector; determining a third Q vector and a third K vector based on the first intermediate vector, determining a third V vector based on the second intermediate vector, and performing a multi-head attention operation based on the third Q vector, the third K vector and the third V vector to obtain a third intermediate vector, and the third intermediate vector is used to represent the mastery of the problem level; determining a fourth K vector based on the first intermediate vector, determining a fourth V vector based on the third intermediate vector, determining a fourth Q vector based on the knowledge space feature vector, and performing a multi-head attention operation based on the fourth Q vector, the fourth K vector and the fourth V vector to obtain a fourth intermediate vector; and determining an ability feature vector based on the fourth intermediate vector.

[0019] Exemplarily, determining the knowledge space feature vector based on the knowledge point embedding feature vector may include but is not limited to: determining the fifth Q vector, the fifth K vector and the fifth V vector based on the knowledge point embedding feature vector, performing a multi-head attention operation based on the fifth Q vector, the fifth K vector and the fifth V vector to obtain a fifth intermediate vector, and inputting the fifth intermediate vector into a multi-layer perceptron; performing a multi-layer perception operation based on the fifth intermediate vector through the multi-layer perceptron to obtain a knowledge space feature vector.

[0020] Exemplarily, based on the knowledge point target feature vector and the acquired knowledge point graph, predicting the target object's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point graph may include but is not limited to: determining a first graph feature and a second graph feature based on the knowledge point graph, the first graph feature including the initial feature vector of each candidate knowledge point name in the knowledge point graph, and the second graph feature including the similarity of the initial feature vectors of two candidate knowledge point names in the knowledge point graph; generating a target graph feature based on the knowledge point target feature vector and the first graph feature, the target graph feature including a fused feature vector of each candidate knowledge point name in the knowledge point graph, the fused feature vector being obtained by concatenating the knowledge point target feature vector and the initial feature vector; predicting the target object's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point graph based on the target graph feature, the first graph feature and the second graph feature.

[0021] Exemplarily, based on the target graph features, the first graph features and the second graph features, predicting the target object's mastery of the target knowledge points for each candidate knowledge point name in the knowledge point graph may include but is not limited to: inputting the target graph features into a multi-layer perceptron, and performing feature mapping operations based on the target graph features through the multi-layer perceptron to obtain mapped graph features; wherein the feature dimensions of the mapped graph features are consistent with the input feature dimensions of the graph neural network; inputting the mapped graph features, the first graph features and the second graph features into the graph neural network, and performing predictions based on the mapped graph features, the first graph features and the second graph features through the graph neural network to obtain the mastery of the target knowledge points.

[0022] Exemplarily, the question sequence also includes questions to be predicted, and the auxiliary information sequence also includes the knowledge point name and question difficulty corresponding to the questions to be predicted; the predicted question response corresponding to the questions to be predicted can also be determined based on the question embedding feature vector, the ability feature vector and the knowledge space feature vector; wherein the predicted question response indicates whether the target object's answer to the questions to be predicted is correct or incorrect.

[0023] Exemplarily, determining the predicted question response corresponding to the question to be predicted based on the question embedding feature vector, the ability feature vector and the knowledge space feature vector may include but is not limited to: determining the sixth Q vector based on the question embedding feature vector, determining the sixth V vector based on the ability feature vector, and determining the sixth K vector based on the knowledge space feature vector; performing a multi-head attention operation based on the sixth Q vector, the sixth K vector and the sixth V vector to obtain a sixth intermediate vector, and inputting the sixth intermediate vector into a multi-layer perceptron; performing a multi-layer perception operation based on the sixth intermediate vector through the multi-layer perceptron to obtain a post-operation feature, and determining the predicted question response corresponding to the question to be predicted based on the post-operation feature.

[0024] As can be seen from the above technical solutions, in the embodiments of the present application, an ability feature vector can be determined based on the topic embedding feature vector and the auxiliary embedding feature vector, a knowledge space feature vector can be determined based on the knowledge point embedding feature vector, a knowledge point target feature vector can be determined based on the ability feature vector, the knowledge point embedding feature vector, and the knowledge space feature vector, and the target knowledge point mastery of the target subject (i.e., the learner) for each candidate knowledge point name can be predicted based on the knowledge point target feature vector and the knowledge point graph. In this way, knowledge point mastery can be predicted in conjunction with the knowledge point graph, accurately assessing the learner's (e.g., student's) knowledge point mastery, reflecting knowledge point mastery in real time, and achieving precise assessment of knowledge point mastery. The assessment process is simple and rapid. By organizing knowledge points into a graph structure and fully utilizing the hierarchical relationships and dependencies between knowledge points, the assessment results are more comprehensive and accurate, achieving a precise assessment of the mastery level of each specific knowledge point, and updating the knowledge point mastery assessment results in real time. The assessment results can be interpreted from the perspective of knowledge point associations, reflecting both the student's overall learning status and specific knowledge weaknesses.

[0025] The above technical solutions of the embodiments of the present application are described below in conjunction with specific application scenarios.

[0026] Knowledge mastery is a quantitative assessment of a learner's depth of understanding, application ability, and transferability of specific knowledge units (such as concepts, principles, and skills). In the field of intelligent education, accurately assessing a learner's (e.g., student) knowledge mastery is key to achieving personalized learning and precision teaching.

[0027] Related technologies cannot accurately assess knowledge mastery. For example, periodic exams are used to assess knowledge mastery. However, this method is time-consuming and labor-intensive, and lacks a precise assessment of knowledge mastery. For example, only considering whether a student's answers are correct or not ignores the inherent relationships between knowledge points, resulting in one-sided assessment results.

[0028] Related technologies for assessing knowledge mastery suffer from the following problems: Ignoring knowledge structure: treating knowledge points as independent entities without fully leveraging the hierarchical relationships and interdependencies between them, resulting in incomplete assessment results. Coarse assessment granularity: providing only an overall assessment of learning levels, unable to granularly analyze the mastery of specific knowledge points, hindering targeted teaching interventions. Lack of dynamism: difficulty capturing the dynamic changes in students' knowledge status, failing to reflect their learning progress in a timely manner.

[0029] In response to the above findings, the embodiment of the present application proposes a knowledge point mastery prediction method, which realizes the prediction of knowledge point mastery based on graph neural network, and realizes the accurate assessment of knowledge point mastery in the fields of education and artificial intelligence, such as intelligent education assessment and personalized learning diagnosis. In order to solve the problem of ignoring the knowledge structure, the association between knowledge points is modeled by introducing graph neural network. By organizing the knowledge points into a graph structure, the hierarchical relationship and dependency relationship between knowledge points are fully utilized, making the evaluation results more comprehensive and accurate. In order to solve the problem of coarse evaluation granularity, a fine-grained knowledge point mastery prediction method based on the student answer sequence is proposed. By combining the temporal modeling capability of Transformer and the graph structure modeling capability of graph neural network, an accurate assessment of the mastery level of each specific knowledge point is achieved. In order to solve the problem of lack of dynamics, a model structure that can capture the dynamic changes of students' knowledge status is proposed. By analyzing the students' answer sequence, the model can update the evaluation results of students' knowledge mastery in real time. In order to solve the problem of poor interpretability of the evaluation results, by integrating the knowledge point graph structure, the model can explain the evaluation results from the perspective of knowledge point association, which can not only reflect the students' overall learning status, but also point out specific knowledge weaknesses.

[0030] In the embodiment of the present application, a method for predicting the mastery of a knowledge point is proposed. The method can be applied to electronic devices such as PCs (Personal Computers), servers, laptops, smart phones, etc. Figure 2 FIG. 5 is a flow chart of the method, which may include: Step 201: Obtain a knowledge point description sequence, where the knowledge point description sequence includes semantic description information corresponding to each candidate knowledge point name (each knowledge point name is recorded as a candidate knowledge point name).

[0031] Exemplarily, a knowledge point sequence can be obtained, which can include multiple candidate knowledge point names. For example, the knowledge point sequence can include candidate knowledge point name a1, candidate knowledge point name a2,..., candidate knowledge point name aN, etc., that is, taking N candidate knowledge point names as an example, the N candidate knowledge point names are all known candidate knowledge point names, such as 1,000 candidate knowledge point names, etc.

[0032] The knowledge point description sequence may include semantic description information b1 corresponding to the candidate knowledge point name a1, semantic description information b2 corresponding to the candidate knowledge point name a2, ..., semantic description information bN corresponding to the candidate knowledge point name aN, that is, there are N semantic description information in total. Semantic description information b1 is used to describe the semantic information of the candidate knowledge point name a1. Semantic description information b1 is semantic information related to the candidate knowledge point name a1, and is used to interpret and describe the candidate knowledge point name a1. It is explanatory information and descriptive information for the candidate knowledge point name a1. For example, the knowledge point representation of the candidate knowledge point name a1 is poor, and only the knowledge point name label is provided. Therefore, the candidate knowledge point name a1 can be expanded into a vector representation rich in teaching semantics. This vector representation can be the semantic description information b1 corresponding to the candidate knowledge point name a1.

[0033] Exemplarily, a question text sequence may be obtained, which may include multiple example question texts. The example question text is text information for the example question. An example question text may be understood as a question. These example question texts may be part of the example question texts selected from the knowledge base.

[0034] Exemplarily, a reference matrix (such as a Q-matrix) can be obtained, which represents the correspondence between the candidate knowledge point names and the example question texts. For example, assuming that there are N candidate knowledge point names and M example question texts, the reference matrix can include N*M elements, the first element of the first row represents the correspondence between the first candidate knowledge point name and the first example question text, such as a value of 1 indicates that there is a correspondence, and a value of 0 indicates that there is no correspondence, the second element of the first row represents the correspondence between the first candidate knowledge point name and the second example question text, and so on, the Mth element of the first row represents the correspondence between the first candidate knowledge point name and the Mth example question text, the first element of the second row represents the correspondence between the second candidate knowledge point name and the first example question text, and so on, there is no limitation on this reference matrix.

[0035] Exemplarily, the knowledge point sequence, question text sequence and reference matrix can be input into a large language model, and the semantic description information corresponding to each candidate knowledge point name can be determined by the large language model to obtain a knowledge point description sequence, which includes the semantic description information corresponding to each candidate knowledge point name.

[0036] For example, for each candidate knowledge point name in the knowledge point sequence, the large language model (LLM) determines the example question text corresponding to the candidate knowledge point name through the reference matrix, that is, finds the example question text that corresponds to the candidate knowledge point name from the question text sequence through the reference matrix.

[0037] Then, based on the example question text (which may include multiple example question texts), semantic analysis is performed on the candidate knowledge point name to obtain the semantic description information corresponding to the candidate knowledge point name. For example, by providing the large language model (LLM) with a prompt pattern of "name + example question -> explanation", the large language model's semantic generation capabilities can be used to obtain the semantic description information corresponding to each candidate knowledge point name. There are no restrictions on this process; the large language model can simply obtain the semantic description information based on the example question text. In other words, the semantic description information is semantic information related to the candidate knowledge point name, and the large language model can obtain the semantic description information corresponding to the candidate knowledge point name by analyzing the example question text corresponding to the candidate knowledge point name.

[0038] Step 202: Obtain a knowledge point graph. The knowledge point graph is a graph for all candidate knowledge point names (such as N candidate knowledge point names). The knowledge point graph may include N candidate knowledge point names and the connection relationships between the candidate knowledge point names. This embodiment does not limit this knowledge point graph.

[0039] For example, a knowledge point graph is a structured knowledge representation method that graphically displays the relationships between knowledge points. For example, knowledge points can be concepts, entities, events, etc., while relationships can be inheritance, association, dependency, etc. Through the knowledge point graph, one can clearly see the connections between knowledge points, thereby better understanding and applying these knowledge points.

[0040] Knowledge point graphs have widespread applications in many fields. For example, in education, knowledge point graphs can help students better understand the relationships between knowledge points, guiding learning and teaching. Teachers can design syllabi and teaching content based on knowledge point graphs, and students can use them to independently learn and consolidate their knowledge. For example, in search engine optimization, knowledge point graphs can help websites provide more accurate and relevant search results. By annotating and associating knowledge points within website content, search engines can better understand the relationships between web pages, improving the quality and accuracy of search results.

[0041] To construct a knowledge point graph, a manual construction method is used. This involves manually creating a knowledge point graph, with experts or practitioners in the knowledge field identifying, classifying, and associating knowledge points to ensure the accuracy and completeness of the relationships between knowledge points. Alternatively, a machine learning-based construction method is used. This involves automatically constructing a knowledge point graph using machine learning algorithms. By analyzing large amounts of corpora and text data, the knowledge point graph can be automatically identified and extracted, and the relationships between knowledge points can be constructed, speeding up the construction process.

[0042] In one possible implementation, a dual-track strategy may be adopted for the construction of a knowledge point map, where the two complement each other to form a knowledge point map with comprehensive coverage and precise relationships.

[0043] First, an initial knowledge point graph is constructed based on the knowledge of educational experts. For example, a complete initial knowledge point graph can be constructed based on dataset metadata or guidance from educational experts. Metadata or manually annotated knowledge point relationships (such as similarity and precedence) are recorded as an adjacency matrix to form a complete initial knowledge point graph. Basic information such as difficulty coefficient and grade attributes, calculated by strategy statistics or manually annotated, is then added to each knowledge point in the initial knowledge point graph to obtain a structured representation of the initial knowledge point graph.

[0044] Then, based on the question data, the initial knowledge point map is enhanced to obtain the enhanced target knowledge point map, and the target knowledge point map is output. The knowledge point maps in the subsequent embodiments all refer to the target knowledge point map. For example, the question and its knowledge point annotation information are collected, the co-occurrence of the knowledge points is counted, and then the node2vec algorithm is used to represent the knowledge points and learn to capture the structural relationship between the knowledge points through random walks. Based on the knowledge point vector representation obtained by learning, the cosine similarity between the knowledge points is calculated, and the threshold is set to construct an adjacency matrix to supplement the initial knowledge point map constructed by the data set or artificial prior knowledge, so that a complete target knowledge point map can be obtained. This target knowledge point map introduces a graph neural network module, so as to realize the inference of the student's complete knowledge mastery level through limited question-solving records.

[0045] After obtaining the target knowledge point graph, the knowledge relationships within the target knowledge point graph can also be enhanced. For example, by adjusting the structure of the target knowledge point graph, such as adding weakly associated edges or adjusting edge weights, the model's ability to learn knowledge point relationships can be enhanced. In practice, this can be achieved through node2vec and rules (such as adding undirected edges / correlation coefficients representing similarity relationships between the last-level knowledge points belonging to the same parent knowledge point). This enhancement process is not limited in this embodiment.

[0046] In summary, a knowledge point map can be obtained. In this embodiment, there is no restriction on the process of obtaining the knowledge point map. Once the knowledge point map is obtained, subsequent processing can be performed based on the knowledge point map.

[0047] Step 203: Obtain a question sequence and an auxiliary information sequence of the target object. The question sequence may include multiple questions (i.e., the answer sequence of the target object). The auxiliary information sequence may include the knowledge point name, question difficulty, and question response, such as a correct response or an incorrect response, corresponding to each question in the question sequence.

[0048] For example, when predicting the mastery of a knowledge point for a user A (e.g., student A), user A is the target object. A sequence of questions (also called an answer sequence) is required. This sequence can include multiple question texts and the answer results for each question text. For ease of description, the question text and the answer results for each question text are collectively referred to as a question.

[0049] For example, each question text can be a text-based question, such as a multiple-choice question, a fill-in-the-blank question, or an application question. User A can answer each question text and obtain the answer for that question text. Based on this, these question texts and the answer results for each question text can be combined to obtain a question sequence for the target object. For example, a question sequence can include question c1 (e.g., a question text and the target object's answer to that question text), question c2, and so on. In other words, the question sequence can include multiple questions.

[0050] For example, for each question in the question sequence, the knowledge point name corresponding to the question can be obtained, such as question c1 corresponds to knowledge point name d1, question c2 corresponds to knowledge point name d2, ..., and so on. In this way, the knowledge point name corresponding to each question in the question sequence can be obtained. Considering that user A will not answer questions with every candidate knowledge point name, the knowledge point names corresponding to the question sequence are only some of the N candidate knowledge point names. For example, if N candidate knowledge point names are 1,000 candidate knowledge point names, the knowledge point names corresponding to the question sequence are only 50 of them. With respect to the knowledge point name, the knowledge point name represents the name of the knowledge point in the question text. For example, the knowledge point name can be addition, addition operation, addition calculation rule, addition calculation, subtraction, subtraction operation, etc.

[0051] For example, for each question in a question sequence, the question difficulty corresponding to the question can be obtained. The question difficulty can indicate the difficulty of the question (question text). For example, three difficulty levels can be divided, such as low difficulty, medium difficulty, and high difficulty. The question difficulty corresponding to the question can be low difficulty, medium difficulty, or high difficulty. Alternatively, four difficulty levels can be divided, such as level 1 difficulty, level 2 difficulty, level 3 difficulty, and level 4 difficulty. The question difficulty corresponding to the question can be level 1 difficulty, level 2 difficulty, level 3 difficulty, or level 4 difficulty. Alternatively, five difficulty levels can be divided, such as level 0 difficulty, level 1 difficulty, level 2 difficulty, level 3 difficulty, and level 4 difficulty. The question difficulty corresponding to the question can be level 0 difficulty, level 1 difficulty, level 2 difficulty, level 3 difficulty, or level 4 difficulty.

[0052] For example, for each question in the question sequence, the corresponding question response can be obtained. The question response can be a correct response or an incorrect response. For example, if user A's question response to the question is a correct response, it means that user A's answer to the question is correct. If user A's question response to the question is an incorrect response, it means that user A's answer to the question is incorrect.

[0053] In summary, for each question in the question sequence, the corresponding knowledge point name, question difficulty, and question response can be obtained, and then an auxiliary information sequence can be obtained. The auxiliary information sequence can include auxiliary information corresponding to each question in the question sequence, and the auxiliary information can include the knowledge point name, question difficulty, and question response. For example, assuming there are 50 questions in the question sequence, there will be 50 auxiliary information in the auxiliary information sequence, and each auxiliary information includes the knowledge point name, question difficulty, and question response.

[0054] In one possible implementation, a student profile can also be constructed for each target subject (e.g., user A). For example, based on the actual needs of the educational scenario, basic attribute information of the target subject can be designed. This basic attribute information can include key features such as grade, knowledge point mastery, and ability value. Specifically, the target subject's knowledge point mastery takes into account factors such as knowledge point difficulty, grade adaptability, knowledge point dependencies, and individual differences to simulate the distribution of student abilities in real-world scenarios. The subsequent process of this embodiment requires predicting the target subject's knowledge point mastery and then constructing a student profile.

[0055] In one possible implementation, to construct a question sequence (i.e., an answer sequence), the answer sequence can be collected based on the target subject's historical answer data. Specifically, the data in the answer sequence is all real-world data. Alternatively, the answer sequence can be generated using a probabilistic model based on the target subject's ability score, question difficulty, and knowledge mastery. This process takes time into account, making the generated data more consistent with the actual learning process. Specifically, the data in the answer sequence is simulated data. Alternatively, some of the data in the answer sequence is real-world data, while some is simulated data.

[0056] For example, after obtaining the answer sequence, it can be expanded. For example, by appropriately perturbing and transforming the answer sequence, an updated answer sequence can be obtained. For example, perturbations and transformations can include randomly deleting some answer records, adjusting the order of answers, and so on.

[0057] For example, a dynamic knowledge status update mechanism could be designed to adjust the target student's mastery of knowledge points in real time based on their answers. This dynamic knowledge status update mechanism takes into account the interdependencies between knowledge points and reflects the continuity of the learning process.

[0058] Step 204: Determine the topic embedding feature vector corresponding to the topic sequence, the auxiliary embedding feature vector corresponding to the auxiliary information sequence, and the knowledge point embedding feature vector corresponding to the knowledge point description sequence.

[0059] For example, a question sequence may include multiple questions, and the question embedding feature vector may include multiple question feature values, where the multiple question feature values ​​correspond one-to-one to the multiple questions in the question sequence. For example, a feature extraction network may be used to extract features from the question sequence to obtain a question embedding feature vector. When extracting features from the question sequence, it is necessary to extract the question feature value corresponding to each question in the question sequence. In this way, the question feature values ​​corresponding to all questions can be combined into a question embedding feature vector.

[0060] For example, the feature extraction network can be a Rasch Embedding module or another type of embedding module. This is not a limitation, as long as it can convert the question sequence into a question embedding feature vector. A Rasch Embedding module is an embedding module that maps high-dimensional data (such as text, images, and videos) into a low-dimensional space. An embedding vector is an N-dimensional real-valued vector that represents the input data as a point in a continuous numerical space.

[0061] Exemplarily, the auxiliary information sequence may include multiple auxiliary information (such as the name of the knowledge point, the difficulty of the question, and the response to the question), and the auxiliary embedded feature vector may include multiple auxiliary feature values, and the multiple auxiliary feature values ​​correspond one-to-one to the multiple auxiliary information in the auxiliary information sequence. For example, the auxiliary information sequence can be feature extracted through a feature extraction network to obtain an auxiliary embedded feature vector. When extracting features from the auxiliary information sequence, it is necessary to extract the auxiliary feature values ​​corresponding to each auxiliary information in the auxiliary information sequence (the auxiliary feature value corresponding to one auxiliary information can be one feature value, such as the knowledge point name, question difficulty, and question response corresponding to the same feature value, or the auxiliary feature value corresponding to one auxiliary information can be three feature values, such as the knowledge point name, question difficulty, and question response corresponding to different feature values). In this way, the auxiliary feature values ​​corresponding to all the auxiliary information can be combined into an auxiliary embedded feature vector.

[0062] For example, the feature extraction network can be a Rasch Embedding module or other types of embedding networks. There is no restriction on this, as long as the auxiliary information sequence can be converted into an auxiliary embedding feature vector.

[0063] Exemplarily, the knowledge point description sequence may include semantic description information corresponding to multiple candidate knowledge point names, and the knowledge point embedding feature vector may include multiple knowledge point feature values, and the multiple knowledge point feature values ​​correspond one-to-one to the multiple semantic description information in the knowledge point description sequence. For example, the knowledge point description sequence can be feature extracted through a feature extraction network to obtain a knowledge point embedding feature vector. When performing feature extraction on the knowledge point description sequence, it is necessary to extract the knowledge point feature values ​​corresponding to each semantic description information in the knowledge point description sequence, so that all knowledge point feature values ​​can be combined into a knowledge point embedding feature vector. The feature extraction network can be a Rasch Embedding module or other types of embedding networks, and there is no limitation on this, as long as it can convert the knowledge point description sequence into a knowledge point embedding feature vector.

[0064] In one possible implementation, see Figure 3A The diagram below is a schematic diagram of the knowledge point mastery prediction model. This is just an example of the knowledge point mastery prediction model. There is no restriction on the network structure of this knowledge point mastery prediction model, and it can be configured according to actual needs. Figure 3A In the above example, the knowledge point mastery prediction model may include a first feature extraction network (such as a Rasch Embedding module) and a second feature extraction network (such as a Rasch Embedding module). Figure 3A As can be seen, the question sequence can be input into the first feature extraction network (such as the Rasch Embedding module), which determines the question embedding feature vector corresponding to the question sequence. The question embedding feature vector is a historical sequence and can be recorded as the question embedding q_emb. The auxiliary information sequence can be input into the first feature extraction network (such as the RaschEmbedding module), which determines the auxiliary embedding feature vector corresponding to the auxiliary information sequence. The auxiliary embedding feature vector is a historical sequence and can be recorded as the answer embedding s_emb.

[0065] The knowledge point sequence, question text sequence, and reference matrix can be input into a large language model (LLM). The large language model determines the semantic description information (also referred to as knowledge point interpretation information or knowledge point description information) corresponding to each candidate knowledge point name, thereby obtaining a knowledge point description sequence. The knowledge point description sequence includes the semantic description information corresponding to each candidate knowledge point name. The knowledge point description sequence can then be input into a second feature extraction network, which determines the knowledge point embedding feature vector corresponding to the knowledge point description sequence, denoted as the knowledge point embedding c_embed. Alternatively, other methods can be used to determine the knowledge point embedding feature vector corresponding to the knowledge point description sequence, without limitation.

[0066] Step 205: Perform a multi-head attention operation based on the question embedding feature vector and the auxiliary embedding feature vector to obtain the target object's ability feature vector for the knowledge point name corresponding to the question sequence.

[0067] In one possible implementation, see Figure 3A As shown, the knowledge point mastery prediction model can include an MHA (Multi Head Attention) network. The Q vector, K vector, and V vector of the MHA network can be determined based on the question embedding feature vector and the auxiliary embedding feature vector. On this basis, the MHA network can process the Q vector, K vector, and V vector to obtain an ability feature vector, which can also be called a dense ability representation z. Since the knowledge point names corresponding to the question sequence are only some of the N candidate knowledge point names, such as 50 knowledge point names out of 1,000 candidate knowledge point names, the ability feature vector is the target object's ability feature vector for the 50 knowledge point names corresponding to the question sequence, rather than the ability feature vector for the N candidate knowledge point names.

[0068] For example, the MHA network is a key component of the Transformer model, which enables the Transformer model to simultaneously focus on different parts of the input sequence in different representation subspaces. In addition, the Transformer model is a neural network architecture that uses the self-attention mechanism to process sequence data in parallel.

[0069] For example, the Q-vector, K-vector, and V-vector can refer to the three input representation vectors used in the self-attention mechanism. The Q-vector can represent the query vector, the K-vector can represent the key vector, and the V-vector can represent the numeric vector. In the self-attention mechanism, based on the Q-vector, the similarity between the query vector Q and all key vectors K is calculated to obtain a weight distribution, which is used to weight the sum of the associated numeric vectors V. In simple terms, the query vector Q and the key vector K are similarly matched, and the result of the matching is the numeric vector V.

[0070] There is no restriction on how the MHA network performs processing based on the Q vector, K vector and V vector in this embodiment. It is related to the function of the MHA network. The processing process of the MHA network is called multi-head attention operation, that is, the MHA network performs multi-head attention operation based on the Q vector, K vector and V vector, and the output feature of the MHA network is called the capability feature vector, that is, the MHA network outputs the capability feature vector.

[0071] In one possible implementation, see Figure 3B As shown in the figure, it is a structural diagram of the knowledge point mastery prediction model. The MHA network of the knowledge point mastery prediction model can include 4 MHA sub-networks, which are recorded as MHA sub-network 1, MHA sub-network 2, MHA sub-network 3 and MHA sub-network 4. The number of MHA sub-networks can be more or less, and there is no limit on the number of MHA sub-networks.

[0072] See also Figure 3B As shown, the first Q vector, first K vector, and first V vector of MHA sub-network 1 can be determined based on the question embedding feature vector. For example, the question embedding feature vector is used as the first Q vector of MHA sub-network 1, the question embedding feature vector is used as the first K vector of MHA sub-network 1, and the question embedding feature vector is used as the first V vector of MHA sub-network 1. A multi-head attention operation can be performed based on the first Q vector, the first K vector, and the first V vector to obtain a first intermediate vector. For example, MHA sub-network 1 can perform a multi-head attention operation based on the first Q vector, the first K vector, and the first V vector, and the output feature of MHA sub-network 1 is called the first intermediate vector.

[0073] See also Figure 3BAs shown, the second Q vector, second K vector, and second V vector of the MHA sub-network 2 can be determined based on the auxiliary embedding feature vector. For example, the auxiliary embedding feature vector is used as the second Q vector of the MHA sub-network 2, the auxiliary embedding feature vector is used as the second K vector of the MHA sub-network 2, and the auxiliary embedding feature vector is used as the second V vector of the MHA sub-network 2. A multi-head attention operation can be performed based on the second Q vector, the second K vector, and the second V vector to obtain a second intermediate vector. For example, the MHA sub-network 2 can perform a multi-head attention operation based on the second Q vector, the second K vector, and the second V vector, and the output feature of the MHA sub-network 2 is called the second intermediate vector.

[0074] See also Figure 3B As shown, the third Q vector and third K vector of the MHA sub-network 3 can be determined based on the first intermediate vector, and the third V vector of the MHA sub-network 3 can be determined based on the second intermediate vector. For example, the first intermediate vector is used as the third Q vector of the MHA sub-network 3, the first intermediate vector is used as the third K vector of the MHA sub-network 3, and the second intermediate vector is used as the third V vector of the MHA sub-network 3. A multi-head attention operation can be performed based on the third Q vector, the third K vector, and the third V vector to obtain the third intermediate vector. For example, the MHA sub-network 3 can perform a multi-head attention operation based on the third Q vector, the third K vector, and the third V vector, and the output feature of the MHA sub-network 3 is referred to as the third intermediate vector. For example, the third intermediate vector is used to represent the problem-level mastery h, that is, the third intermediate vector can also be referred to as the problem-level mastery vector, reflecting the target subject's mastery of the problem level.

[0075] See also Figure 3BAs shown, the fourth K vector of MHA sub-network 4 can be determined based on the first intermediate vector, the fourth V vector of MHA sub-network 4 can be determined based on the third intermediate vector, and the fourth Q vector of MHA sub-network 4 can be determined based on the knowledge space feature vector (know_params; for the method of obtaining the knowledge space feature vector, see the subsequent embodiments). For example, the first intermediate vector can be used as the fourth K vector of MHA sub-network 4, the third intermediate vector can be used as the fourth V vector of MHA sub-network 4, and the knowledge space feature vector can be used as the fourth Q vector of MHA sub-network 4. A multi-head attention operation can be performed based on the fourth Q vector, the fourth K vector, and the fourth V vector to obtain the fourth intermediate vector. For example, the MHA sub-network 4 can perform a multi-head attention operation based on the fourth Q vector, the fourth K vector, and the fourth V vector, and the output feature of MHA sub-network 4 is referred to as the fourth intermediate vector. After obtaining the fourth intermediate vector, the ability feature vector (i.e., the dense ability representation z) can be determined based on the fourth intermediate vector. For example, the fourth intermediate vector can be used as the ability feature vector, or operations can be performed on the fourth intermediate vector to obtain the ability feature vector, i.e., the ability feature vector of the target subject's ability for the knowledge point names corresponding to the question sequence.

[0076] Step 206: Determine a knowledge space feature vector based on the knowledge point embedding feature vector. The knowledge point embedding feature vector includes an initial feature vector of each candidate knowledge point name. The knowledge space feature vector includes multiple fused feature vectors, and each fused feature vector is obtained by fusing multiple initial feature vectors.

[0077] Exemplarily, the knowledge point embedding feature vector (c_embed) may include multiple knowledge point feature values ​​corresponding to multiple candidate knowledge point names. The knowledge point feature values ​​may be referred to as initial feature vectors, that is, the knowledge point embedding feature vector includes N initial feature vectors corresponding to N candidate knowledge point names.

[0078] When determining the knowledge space feature vector (know_params) based on the knowledge point embedding feature vector, a fusion operation can be performed on multiple initial feature vectors to obtain a fused feature vector. The number of fused feature vectors can also be multiple, but the number of fused feature vectors is less than the number of initial feature vectors, that is, the fusion operation is used to perform a dimensionality reduction operation on the knowledge point embedding feature vector. In this way, the knowledge space feature vector can include multiple fused feature vectors. For example, a fusion operation can be performed on the four initial feature vectors in the knowledge point embedding feature vector to obtain one fused feature vector. In this way, N initial feature vectors can be fused to obtain N / 4 fused feature vectors, and the knowledge space feature vector can include N / 4 fused feature vectors. Of course, this is just an example of a fusion operation, and there is no limitation on this fusion operation.

[0079] In one possible implementation, see Figure 3A and Figure 3B As shown, the knowledge point mastery prediction model can include an MHA network (different from the MHA network in step 205, which is another MHA network) and an MLP network. The fifth Q vector, fifth K vector, and fifth V vector of the MHA network can be determined based on the knowledge point embedding feature vector. For example, the knowledge point embedding feature vector is used as the fifth Q vector of the MHA network, the knowledge point embedding feature vector is used as the fifth K vector of the MHA network, and the knowledge point embedding feature vector is used as the fifth V vector of the MHA network. A multi-head attention operation can be performed based on the fifth Q vector, the fifth K vector, and the fifth V vector to obtain a fifth intermediate vector. For example, the MHA network can perform a multi-head attention operation based on the fifth Q vector, the fifth K vector, and the fifth V vector, and the output feature of the MHA network is called the fifth intermediate vector. After obtaining the fifth intermediate vector, the fifth intermediate vector can be input into the MLP network (Multi-Layer Perceptron).

[0080] After obtaining the fifth intermediate vector, the MLP network can perform a multi-layer perception operation based on the fifth intermediate vector through the MLP network to obtain a knowledge space feature vector (know_params). For example, the MLP network can process based on the fifth intermediate vector, and the output feature of the MLP network is called a knowledge space feature vector. As for how the MLP network processes based on the fifth intermediate vector, this embodiment does not impose any restrictions, and it is related to the function of the MLP network. For example, the MLP network is an artificial neural network that can be composed of an input layer, one or more hidden layers, and an output layer, and is used to process complex nonlinear relationships. The MLP network is a fully connected layer that is used to splice (fuse) the feature vectors in the fifth intermediate vector (such as the N initial feature vectors corresponding to the N candidate knowledge point names), and can map the fused feature vectors to the input space of the subsequent MHA network through the fully connected layer.

[0081] Step 207: Perform a multi-head attention operation based on the capability feature vector, the knowledge point embedding feature vector, and the knowledge space feature vector to obtain a knowledge point target feature vector.

[0082] Exemplarily, the seventh Q vector, seventh K vector and seventh V vector of the MHA network can be determined based on the capability feature vector, the knowledge point embedding feature vector and the knowledge space feature vector. The MHA network can perform multi-head attention operation based on the seventh Q vector, the seventh K vector and the seventh V vector to obtain the knowledge point target feature vector, that is, the output feature of the MHA network is used as the knowledge point target feature vector.

[0083] In one possible implementation, see Figure 3A and Figure 3B As shown, the knowledge point mastery prediction model may include an MHA network (different from the MHA network in steps 205 and 206, which is another MHA network). The seventh V vector of the MHA network may be determined based on the capability feature vector. For example, the capability feature vector is used as the seventh V vector of the MHA network. The seventh K vector of the MHA network may be determined based on the knowledge space feature vector. For example, the knowledge space feature vector is used as the seventh K vector of the MHA network. The seventh Q vector of the MHA network may be determined based on the knowledge point embedding feature vector. For example, the knowledge point embedding feature vector is used as the seventh Q vector of the MHA network. A multi-head attention operation may be performed based on the seventh Q vector, the seventh K vector, and the seventh V vector to obtain a knowledge point target feature vector. For example, the MHA network may perform a multi-head attention operation based on the seventh Q vector, the seventh K vector, and the seventh V vector. The output feature of the MHA network is referred to as the knowledge point target feature vector.

[0084] Exemplarily, after the ability feature vector, the knowledge point embedding feature vector, and the knowledge space feature vector are input into the MHA network, the MHA network is used to calculate the association between the knowledge point embedding feature vector and the ability feature vector (i.e., the student ability vector), that is, the knowledge point target feature vector can reflect the association between the knowledge point embedding feature vector and the ability feature vector. For example, the knowledge point embedding feature vector is used as the Q vector (query) of the MHA network, the knowledge space feature vector is used as the K vector (key) of the MHA network, and the ability feature vector is used as the V vector (value) of the MHA network. Through attention calculation, the MHA network can find the most relevant student ability dimension for each knowledge point, obtain the degree of association between each knowledge point and the student ability vector, and thus obtain a more accurate representation of the degree of mastery of the knowledge point.

[0085] Exemplarily, after obtaining the knowledge point target feature vector, the MHA network can input the knowledge point target feature vector to the next network layer (such as the MLP network). Alternatively, the MHA network can also normalize the knowledge point target feature vector and input the normalized knowledge point target feature vector to the next network layer. Alternatively, the MHA network can also perform residual connection processing on the knowledge point target feature vector and input the residual connection processed knowledge point target feature vector to the next network layer. Alternatively, the MHA network can also perform normalization and residual connection processing on the knowledge point target feature vector and input the residual connection processed knowledge point target feature vector to the next network layer.

[0086] Step 208: Generate a target graph feature based on the knowledge point target feature vector and the first graph feature. The target graph feature may include a fused feature vector of each candidate knowledge point name in the knowledge point graph. The fused feature vector is obtained by concatenating the knowledge point target feature vector and the initial feature vector.

[0087] Exemplarily, the knowledge point graph is a graph for all candidate knowledge point names (such as N candidate knowledge point names). The knowledge point graph can include N candidate knowledge point names and the connection relationship between the candidate knowledge point names. The knowledge point graph can reflect the connection relationship between the N candidate knowledge point names and the candidate knowledge point names. On this basis, the first graph feature can be determined based on the knowledge point graph. The first graph feature includes the initial feature vector of each candidate knowledge point name in the knowledge point graph, such as the initial feature vector of the N candidate knowledge point names. For example, for each candidate knowledge point name in the knowledge point graph, the candidate knowledge point name can be feature extracted to obtain the initial feature vector of the candidate knowledge point name, and the initial feature vectors of all candidate knowledge point names can constitute the first graph feature.

[0088] In addition, a second graph feature can be determined based on the knowledge point graph, and the second graph feature includes the similarity of the initial feature vectors of two candidate knowledge point names in the knowledge point graph. For example, the second graph feature can be a feature matrix of N*N dimensions, and the elements in the first row represent the similarity between the initial feature vectors of the first candidate knowledge point name and each candidate knowledge point name. For example, if the first element is 1 (indicating that its own similarity is 1), the second element represents the similarity between the initial feature vectors of the first candidate knowledge point name and the second candidate knowledge point name, and so on. The elements in the second row represent the similarity between the initial feature vectors of the second candidate knowledge point name and each candidate knowledge point name, and so on.

[0089] For example, for two candidate knowledge point names that have a connection relationship (the knowledge point graph includes N candidate knowledge point names and the connection relationships between the candidate knowledge point names. Based on the knowledge point graph, it is possible to determine which candidate knowledge point names have a connection relationship and which candidate knowledge point names do not have a connection relationship), the similarity between the initial feature vectors of the two candidate knowledge point names can be calculated. For example, the similarity can be a value between 0 and 1. For two candidate knowledge point names that do not have a connection relationship, the similarity between the initial feature vectors of the two candidate knowledge point names can be 0.

[0090] Exemplarily, after obtaining the knowledge point target feature vector and the first graph feature, the knowledge point target feature vector can be spliced ​​with each initial feature vector in the first graph feature to obtain a fused feature vector corresponding to the initial feature vector, that is, a total of N fused feature vectors of candidate knowledge point names are obtained. For example, the knowledge point target feature vector is spliced ​​with the initial feature vector of the candidate knowledge point name a1 (such as the knowledge point target feature vector is located in front of the initial feature vector, or the knowledge point target feature vector is located behind the initial feature vector) to obtain the fused feature vector of the candidate knowledge point name a1, and the knowledge point target feature vector is spliced ​​with the initial feature vector of the candidate knowledge point name a2 to obtain the fused feature vector of the candidate knowledge point name a2, and so on, to obtain N fused feature vectors. In this way, the target graph feature can include the fused feature vectors of N candidate knowledge point names.

[0091] Step 209: Based on the target graph feature, the first graph feature, and the second graph feature, predict the target subject's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point graph.

[0092] For example, see Figure 3A and Figure 3B As shown, the knowledge point mastery prediction model includes MLP network and GCN network, see Figure 3C As shown in FIG, it is a schematic diagram of the structure of the knowledge point mastery prediction model, which only shows part of the structure of the knowledge point mastery prediction model. Figure 3C It can be seen that the knowledge point graph (Concept Map) is input into the MLP network and the GCN network. For example, based on the knowledge point graph, the first graph feature and the second graph feature are determined, the first graph feature is input into the MLP network, and the first graph feature and the second graph feature are input into the GCN network. In addition, the MHA network can input the knowledge point target feature vector into the MLP network. In this way, in the MLP network, the target graph feature can be generated based on the knowledge point target feature vector and the first graph feature. This process can be seen in step 208. Alternatively, in the processing before the MLP network, the target graph feature can be generated based on the knowledge point target feature vector and the first graph feature, and then this target graph feature is input into the MLP network.

[0093] Exemplarily, after obtaining the target graph features, a feature mapping operation is performed based on the target graph features through an MLP network (i.e., a multi-layer perceptron) to obtain mapped graph features. The feature dimensions of the mapped graph features are consistent with the input feature dimensions of the graph neural network (i.e., the GCN network). That is, after concatenating and fusing the output features of the MHA network (such as the target feature vector of the knowledge point) with the first graph features (i.e., the initial feature vector of the candidate knowledge point name) to obtain the enhanced knowledge point representation (i.e., the target graph features), the fused target graph features can be mapped to the input space of the graph neural network through the MLP network, that is, the target graph features are mapped to the input dimensions required by the graph neural network through the MLP network. For example, if the input feature dimension of the graph neural network is A*B, the target graph features are mapped to A*B features through the MLP network, and this A*B feature is recorded as the mapped graph features.

[0094] For example, after obtaining the mapped graph features, the mapped graph features can be input into the graph neural network, and the first graph features and the second graph features can be input into the graph neural network. In this way, the graph neural network performs a prediction based on the mapped graph features, the first graph features, and the second graph features to obtain the mastery of the target knowledge point. This embodiment does not limit this prediction process, and it is sufficient that the graph neural network can obtain the mastery of the target knowledge point based on the mapped graph features, the first graph features, and the second graph features.

[0095] For example, the mapped graph feature is determined based on the knowledge point target feature vector, and the knowledge point target feature vector is determined based on the ability feature vector. Therefore, the mapped graph feature can reflect the target object's ability feature vector for the knowledge point name corresponding to the question sequence, that is, the target object's ability for some candidate knowledge point names. Since the first graph feature reflects the initial feature vector of each candidate knowledge point name in the knowledge point graph, and the second graph feature reflects the similarity of the initial feature vectors of two candidate knowledge point names in the knowledge point graph, the mapped graph feature, the first graph feature, and the second graph feature can be combined to obtain the target object's target knowledge point mastery for each candidate knowledge point name in the knowledge point graph based on the target object's ability for some candidate knowledge point names. For example, based on the target object's ability for 50 candidate knowledge point names, the target object's target knowledge point mastery for 1,000 candidate knowledge point names can be obtained, thereby obtaining the target object's target knowledge point mastery for all candidate knowledge point names.

[0096] For example, this embodiment does not limit how the graph neural network makes predictions based on the mapped graph features, the first graph features, and the second graph features. This is related to the capabilities of the graph neural network, as long as the graph neural network can output the target object's mastery of the target knowledge points for all candidate knowledge point names.

[0097] For example, graph neural networks can also be called graph convolutional neural networks. Graph neural networks can use GCN (Graph Convolutional Network) networks. GCN networks are a type of neural network used to process graph-structured data. By transferring information between graph nodes and their neighbors and learning the embedded representation of nodes, GCN networks can be used to process the association between knowledge points.

[0098] For example, a single-layer GCN network can be used to make predictions based on the mapped graph features, the first graph features, and the second graph features. Alternatively, a multi-layer GCN network can be used to make predictions based on the mapped graph features, the first graph features, and the second graph features. This is illustrated using a two-layer GCN network as an example. In practical applications, the number of GCN layers can be adjusted based on the complexity of the knowledge point graph and the sparsity of the dataset, allowing for reasonable utilization of prior knowledge of the knowledge point graph to aid in predicting knowledge point mastery.

[0099] The first-layer GCN network can convert and pass messages between the mapped graph features, the first graph features, and the second graph features. Through the message passing mechanism of the graph structure, the prediction results of each knowledge point can take into account the information of related knowledge points. The first-layer GCN network can convert features and pass messages between knowledge points. In the first-layer GCN network, nonlinear activation functions and dropout operations can also be added to enhance the expressive power of the model. The second-layer GCN network is used to map features to the category space of knowledge point mastery, that is, to obtain the target knowledge point mastery of the target object for the candidate knowledge point name.

[0100] Exemplarily, in the above processing process, the input dimension can also be configured, which represents the dimension of the student's ability vector, the node feature dimension of the knowledge point can be configured, and the node feature dimension is used to represent the feature information of the knowledge point. The number of input channels of the graph convolutional network can be configured, and the number of input channels is used to control the dimension of feature conversion. There is no restriction on the parameters of the knowledge point mastery prediction model.

[0101] In one possible implementation, after obtaining the target subject's mastery of the candidate knowledge point names, data verification can be performed. Data verification can include rationality verification and distribution verification. Rationality verification refers to verifying the rationality and authenticity of the generated data (i.e., mastery of the target knowledge point) through evaluation by educational experts and statistical analysis. Distribution verification refers to ensuring that the generated data (i.e., mastery of the target knowledge point) meets the expected distribution requirements in all dimensions (such as difficulty distribution, knowledge point coverage, etc.). Figure 4 As shown in the figure, it is a diagram showing the statistical effect of the difficulty distribution of questions on the sample data set. Figure 4The distribution verification is achieved by visual statistical means such as histograms and probability density curves shown in the figure. Figure 4 neutron Figure 1 (i.e. the lower subgraph) represents the difficulty histogram of all questions. Figure 2 (i.e., the upper subgraph) represents the difficulty PDF (probability density function) of all problems.

[0102] At this point, the knowledge point mastery prediction is completed, and the target subject's target knowledge point mastery for each candidate knowledge point name in the knowledge point map can be predicted. In the above process, the knowledge point mastery prediction model is used to predict the target knowledge point mastery. The training process of the knowledge point mastery prediction model can include: Collect sample data, which includes a question sequence and an auxiliary information sequence of the sample object. The question sequence may include multiple questions (i.e., the answer sequence of the sample object), and the auxiliary information sequence may include the knowledge point name, question difficulty, and question response corresponding to each question in the question sequence.

[0103] Based on the sample data, steps 204 to 209 are executed to obtain the sample subject's mastery of each candidate knowledge point name in the knowledge point map. In addition, when obtaining the sample data, the sample subject's mastery of each candidate knowledge point name in the knowledge point map can also be annotated.

[0104] On this basis, the loss value can be calculated based on the sample knowledge point mastery and the label knowledge point mastery, and the network parameters of the knowledge point mastery prediction model can be adjusted based on the loss value. The adjustment goal is to make the loss value smaller and smaller, that is, the sample knowledge point mastery and the label knowledge point mastery are closer and closer. In this way, after multiple iterations, the knowledge point mastery prediction model has converged, and the converged knowledge point mastery prediction model can be output to complete the training process of the knowledge point mastery prediction model.

[0105] When calculating the loss value based on the mastery of sample knowledge points and the mastery of labeled knowledge points, a semi-sparse loss function can be used to calculate the loss value, or other loss functions can be used to calculate the loss value. There is no restriction on the loss function. For example, to avoid deviations in model training caused by class imbalance, a semi-sparse loss function can be used to calculate the loss value. See the following formula for an example of a loss function: In the above loss function, It can represent the loss value, Can represent the cross entropy loss value, Represents the mastery of label knowledge points, Represents the mastery of sample knowledge points, that is, the predicted value of the mastery of knowledge points.

[0106] In the embodiment of the present application, a method for predicting the mastery of a knowledge point is proposed. In addition to predicting the mastery of a target knowledge point, the method can also predict the response to the predicted question corresponding to the question to be predicted, that is, predict whether the answer result is correct or wrong. Figure 5 FIG. 5 is a flow chart of the method, which may include: Step 501: Obtain a knowledge point description sequence, where the knowledge point description sequence includes semantic description information corresponding to each candidate knowledge point name (each knowledge point name is recorded as a candidate knowledge point name).

[0107] Step 502: Obtain a knowledge point graph. The knowledge point graph is a graph for all candidate knowledge point names. The knowledge point graph includes N candidate knowledge point names and connection relationships between the candidate knowledge point names.

[0108] Step 503: Obtain a question sequence and an auxiliary information sequence for the target object. The question sequence may include multiple completed questions (i.e., the target object has completed answering the questions) and questions to be predicted (i.e., the target object has not completed answering the questions, and it is necessary to predict whether the answer to this question may be correct or incorrect). The auxiliary information sequence may include the knowledge point name, question difficulty, and question response, such as a correct response or an incorrect response, corresponding to each completed question in the question sequence. The auxiliary information sequence may also include the knowledge point name and question difficulty corresponding to the question to be predicted, but does not include the question response corresponding to the question to be predicted.

[0109] For example, compared with step 203, the question sequence additionally adds a question to be predicted, and the auxiliary information sequence additionally adds the knowledge point name and question difficulty corresponding to the question to be predicted. In this way, the predicted question response corresponding to the question to be predicted can be predicted by the knowledge point mastery prediction model.

[0110] Step 504: Determine the topic embedding feature vector corresponding to the topic sequence, the auxiliary embedding feature vector corresponding to the auxiliary information sequence, and the knowledge point embedding feature vector corresponding to the knowledge point description sequence.

[0111] Step 505: Perform a multi-head attention operation based on the question embedding feature vector and the auxiliary embedding feature vector to obtain the target object's ability feature vector for the knowledge point name corresponding to the question sequence.

[0112] Step 506: Determine a knowledge space feature vector based on the knowledge point embedding feature vector.

[0113] Step 507: Perform a multi-head attention operation based on the capability feature vector, the knowledge point embedding feature vector, and the knowledge space feature vector to obtain a knowledge point target feature vector.

[0114] Step 508: Generate a target graph feature based on the knowledge point target feature vector and the first graph feature. The target graph feature may include a fused feature vector of each candidate knowledge point name in the knowledge point graph. The fused feature vector is obtained by concatenating the knowledge point target feature vector and the initial feature vector.

[0115] Step 509: Based on the target graph feature, the first graph feature, and the second graph feature, predict the target object's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point graph.

[0116] Illustratively, steps 501 to 509 are similar to steps 201 to 209 and are not described in detail here.

[0117] Step 510: Determine a predicted question response corresponding to the question to be predicted based on the question embedding feature vector, the ability feature vector, and the knowledge space feature vector; wherein the predicted question response may indicate whether the target subject's answer to the question to be predicted is correct or incorrect.

[0118] For example, see Figure 6 As shown in the figure, it is a structural diagram of the knowledge point mastery prediction model. Figure 3A and Figure 3B In comparison, the knowledge point mastery prediction model additionally adds MHA network and MLP network.

[0119] After obtaining the question embedding feature vector, the ability feature vector, and the knowledge space feature vector, the sixth Q vector, the sixth V vector, and the sixth K vector of the MHA network are determined based on the question embedding feature vector, the ability feature vector, and the knowledge space feature vector. For example, the sixth Q vector is determined based on the question embedding feature vector, the sixth V vector is determined based on the ability feature vector, and the sixth K vector is determined based on the knowledge space feature vector. For example, the question embedding feature vector is used as the sixth Q vector of the MHA network, the ability feature vector is used as the sixth V vector of the MHA network, and the knowledge space feature vector is used as the sixth K vector of the MHA network. Then, a multi-head attention operation is performed based on the sixth Q vector, the sixth K vector, and the sixth V vector to obtain the sixth intermediate vector. For example, the MHA network can perform a multi-head attention operation based on the sixth Q vector, the sixth K vector, and the sixth V vector, and the output feature of the MHA network is called the sixth intermediate vector. After obtaining the sixth intermediate vector, the sixth intermediate vector can be input into the MLP network.

[0120] After obtaining the sixth intermediate vector, the MLP network can perform a multi-layer perception operation based on the sixth intermediate vector to obtain a post-operation feature. For example, the MLP network can perform processing based on the sixth intermediate vector, and the output features of the MLP network are referred to as post-operation features. The embodiments of this application do not limit how the MLP network performs processing based on the sixth intermediate vector.

[0121] After obtaining the post-operation feature, the predicted question response corresponding to the question to be predicted can be determined based on the post-operation feature. For example, the post-operation feature can be a predicted question response. For example, the post-operation feature (i.e., the predicted question response) can have a first value (e.g., 1), indicating that the answer is correct, and the post-operation feature (i.e., the predicted question response) can have a second value (e.g., 0), indicating that the answer is incorrect. Alternatively, the post-operation feature can be input into other network layers, which can then make predictions based on the post-operation feature to obtain a predicted question response. For example, the predicted question response can have a first value, indicating that the answer is correct, and the predicted question response can have a second value, indicating that the answer is incorrect.

[0122] At this point, the knowledge point mastery prediction and the prediction of the question to be predicted are completed. The target subject's target knowledge point mastery for each candidate knowledge point name in the knowledge point map can be predicted, and the target subject's answer result for the question to be predicted can be predicted. In the above process, the knowledge point mastery prediction model is used to predict the target knowledge point mastery. The training process of the knowledge point mastery prediction model can include: Collect sample data. The sample data includes a question sequence and an auxiliary information sequence of the sample object. The question sequence may include multiple questions (e.g., 10 questions). The auxiliary information sequence may include the knowledge point name, question difficulty, and question response corresponding to each question in the question sequence. For example, for 10 questions, when the first question is considered a completed question, the second question is considered a question to be predicted. The question response of the first question is a normal value, and the question responses of the remaining questions are set to zero. When the first and second questions are considered completed questions, the third question is considered a question to be predicted. The question responses of the first and second questions are normal values, and the question responses of the remaining questions are set to zero. And so on, thereby obtaining multiple sets of sample data.

[0123] Based on the sample data, steps 504-510 are executed to obtain the sample subject's sample knowledge point mastery for each candidate knowledge point name in the knowledge point graph and the sample subject's answer result for the question to be predicted (i.e., sample question response). When obtaining the sample data, the sample subject's labeled knowledge point mastery for each candidate knowledge point name in the knowledge point graph can be annotated, as well as the sample subject's labeled question response for the question to be predicted (if the third question is the question to be predicted, the question response for the third question is the labeled question response).

[0124] On this basis, a first loss value can be calculated based on the sample knowledge point mastery and the labeled knowledge point mastery, a second loss value can be calculated based on the sample question response and the labeled question response, and a target loss value can be calculated based on the first loss value and the second loss value, such as by performing a weighted operation on the first loss value and the second loss value to obtain the target loss value. Based on the target loss value, the network parameters of the knowledge point mastery prediction model are adjusted. The adjustment goal is to make the target loss value smaller and smaller, that is, the sample knowledge point mastery and the labeled knowledge point mastery are getting closer and closer, and the sample question response and the labeled question response are getting closer and closer. In this way, after multiple iterations, the knowledge point mastery prediction model is converged, and the converged knowledge point mastery prediction model can be output to complete the training process of the knowledge point mastery prediction model.

[0125] As can be seen from the above technical solutions, in the embodiments of the present application, the knowledge point mastery degree is predicted in combination with the knowledge point graph, which can accurately assess the learner's (such as a student's) knowledge point mastery degree, reflect the knowledge point mastery degree in real time, and achieve accurate assessment of the knowledge point mastery degree. The assessment process is simple and fast. By organizing the knowledge points into a graph structure and making full use of the hierarchical relationship and dependency relationship between the knowledge points, the assessment results are made more comprehensive and accurate, achieving an accurate assessment of the mastery level of each specific knowledge point, and updating the assessment results of the knowledge point mastery degree in real time. The assessment results can be interpreted from the perspective of the relationship between knowledge points, which can not only reflect the student's overall learning status, but also point out specific knowledge weaknesses. In the process of predicting the mastery degree of knowledge points based on graph neural networks, data construction and processing are key links. Since educational data has the characteristics of time series, hierarchy, and correlation, it takes into account multiple aspects such as individual differences among students, the correlation between knowledge points, and the dynamic changes in the answering process. By modeling the relationship between knowledge points through graph neural networks, the prediction results not only consider individual knowledge points, but also utilize information from related knowledge points. For example, in mathematics, the mastery level prediction for "solving linear equations with two variables" takes into account the mastery of related knowledge points such as "linear equations with one variable," making the prediction more accurate and reasonable. The multi-head attention mechanism clearly demonstrates the correlation between each knowledge point and each dimension of the student's ability vector. This interpretable prediction helps teachers better understand students' learning status and inform teaching decisions. Graph neural networks possess strong generalization capabilities. They can make reasonable predictions based on the relationships between knowledge points, even for combinations of knowledge points not found in the training data. They are adaptable to knowledge systems across disciplines and exhibit excellent transferability. The model has a simple structure, a small number of parameters, and low resource consumption for both training and inference. It also requires minimal training data size and can achieve good results on smaller datasets. For intelligent question recommendation scenarios, the most appropriate practice questions can be recommended based on the predicted mastery level of knowledge points, achieving personalized learning. For learning diagnostic reporting scenarios, detailed learning diagnostic reports can be generated for teachers and parents, visually demonstrating a student's mastery of each knowledge point. For learning path planning scenarios, the optimal learning path is planned for students based on the dependencies between knowledge points and the student's mastery level.

[0126] Based on the same application concept as the above method, a knowledge point mastery prediction device is proposed in the embodiment of the present application. Figure 7 FIG. 1 is a schematic diagram of the structure of the device, which may include: An acquisition module 71 is configured to acquire a knowledge point description sequence, a target object question sequence, and an auxiliary information sequence; the knowledge point description sequence includes semantic description information corresponding to each candidate knowledge point name; and the auxiliary information sequence includes the knowledge point name, question difficulty, and question response corresponding to each question in the question sequence. Determination module 72, configured to determine a title embedding feature vector corresponding to the title sequence, an auxiliary embedding feature vector corresponding to the auxiliary information sequence, and a knowledge point embedding feature vector corresponding to the knowledge point description sequence; determine a knowledge space feature vector based on the knowledge point embedding feature vectors, wherein the knowledge point embedding feature vector includes an initial feature vector for each candidate knowledge point name, and the knowledge space feature vector includes multiple fused feature vectors, each fused feature vector being obtained by fusing multiple initial feature vectors; Processing module 73 is configured to perform a multi-head attention operation based on the question embedding feature vector and the auxiliary embedding feature vector to obtain an ability feature vector of the target subject for the knowledge point name corresponding to the question sequence; and perform a multi-head attention operation based on the ability feature vector, the knowledge point embedding feature vector, and the knowledge space feature vector to obtain a knowledge point target feature vector; The prediction module 74 is used to predict the target subject's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point map based on the knowledge point target feature vector and the acquired knowledge point map.

[0127] Exemplarily, when the acquisition module 71 acquires the knowledge point description sequence, it is specifically used to: Obtain a knowledge point sequence, a question text sequence, and a reference matrix; wherein the knowledge point sequence includes a plurality of candidate knowledge point names, the question text sequence includes a plurality of example question texts, and the reference matrix represents the correspondence between the candidate knowledge point names and the example question texts; The knowledge point sequence, the question text sequence and the reference matrix are input into a large language model, and the semantic description information corresponding to each candidate knowledge point name is determined by the large language model to obtain the knowledge point description sequence; wherein, for each candidate knowledge point name, the example question text corresponding to the candidate knowledge point name is determined by the reference matrix, and the candidate knowledge point name is semantically analyzed based on the example question text to obtain the semantic description information corresponding to the candidate knowledge point name.

[0128] Exemplarily, the processing module 73 performs a multi-head attention operation based on the question embedding feature vector and the auxiliary embedding feature vector to obtain the ability feature vector of the target object for the knowledge point name corresponding to the question sequence, and is specifically used for: determining a first Q vector, a first K vector and a first V vector based on the question embedding feature vector, and performing a multi-head attention operation based on the first Q vector, the first K vector and the first V vector to obtain a first intermediate vector; determining a second Q vector, a second K vector and a second V vector based on the auxiliary embedding feature vector, and performing a multi-head attention operation based on the second Q vector, the second K vector and the second V vector to obtain a second intermediate vector; determining a third Q vector and a third K vector based on the first intermediate vector, determining a third V vector based on the second intermediate vector, and performing a multi-head attention operation based on the third Q vector, the third K vector and the third V vector to obtain a third intermediate vector, and the third intermediate vector is used to represent the mastery of the problem level; determining a fourth K vector based on the first intermediate vector, determining a fourth V vector based on the third intermediate vector, determining a fourth Q vector based on the knowledge space feature vector, and performing a multi-head attention operation based on the fourth Q vector, the fourth K vector and the fourth V vector to obtain a fourth intermediate vector; and determining the ability feature vector based on the fourth intermediate vector.

[0129] Exemplarily, when the determination module 72 determines the knowledge space feature vector based on the knowledge point embedding feature vector, it is specifically used to: determine the fifth Q vector, the fifth K vector and the fifth V vector based on the knowledge point embedding feature vector, perform a multi-head attention operation based on the fifth Q vector, the fifth K vector and the fifth V vector to obtain a fifth intermediate vector, and input the fifth intermediate vector to a multi-layer perceptron; perform a multi-layer perception operation based on the fifth intermediate vector through the multi-layer perceptron to obtain the knowledge space feature vector.

[0130] Exemplarily, the prediction module 74 predicts the target knowledge point mastery of the target object for each candidate knowledge point name in the knowledge point graph based on the knowledge point target feature vector and the acquired knowledge point graph, and is specifically used to: determine a first graph feature and a second graph feature based on the knowledge point graph, the first graph feature including the initial feature vector of each candidate knowledge point name in the knowledge point graph, and the second graph feature including the similarity of the initial feature vectors of two candidate knowledge point names in the knowledge point graph; generate a target graph feature based on the knowledge point target feature vector and the first graph feature, the target graph feature including a fused feature vector of each candidate knowledge point name in the knowledge point graph, the fused feature vector being obtained by splicing the knowledge point target feature vector and the initial feature vector; predict the target knowledge point mastery of the target object for each candidate knowledge point name in the knowledge point graph based on the target graph feature, the first graph feature and the second graph feature.

[0131] Exemplarily, when the prediction module 74 predicts the target object's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point graph based on the target graph feature, the first graph feature and the second graph feature, it is specifically used to: input the target graph feature to a multilayer perceptron, and perform a feature mapping operation based on the target graph feature through the multilayer perceptron to obtain a mapped graph feature; wherein the feature dimension of the mapped graph feature is consistent with the input feature dimension of the graph neural network; input the mapped graph feature, the first graph feature and the second graph feature to the graph neural network, and perform a prediction based on the mapped graph feature, the first graph feature and the second graph feature through the graph neural network to obtain the mastery of the target knowledge point.

[0132] Exemplarily, the question sequence further includes questions to be predicted, and the auxiliary information sequence further includes the names of knowledge points and question difficulty corresponding to the questions to be predicted; the prediction module 74 is further configured to: Based on the question embedding feature vector, the ability feature vector and the knowledge space feature vector, a predicted question response corresponding to the question to be predicted is determined; wherein the predicted question response indicates whether the target object's answer result to the question to be predicted is correct or incorrect.

[0133] The prediction module 74 determines the predicted question response corresponding to the to-be-predicted question based on the question embedding feature vector, the ability feature vector, and the knowledge space feature vector, specifically for: Determine a sixth Q vector based on the topic embedding feature vector, determine a sixth V vector based on the ability feature vector, and determine a sixth K vector based on the knowledge space feature vector; Performing a multi-head attention operation based on the sixth Q vector, the sixth K vector, and the sixth V vector to obtain a sixth intermediate vector, and inputting the sixth intermediate vector into a multilayer perceptron; A multi-layer perceptron is used to perform a multi-layer perceptron operation based on the sixth intermediate vector to obtain a post-operation feature, and a predicted question response corresponding to the to-be-predicted question is determined based on the post-operation feature.

[0134] Based on the same application concept as the above method, an electronic device is proposed in the embodiment of the present application, see Figure 8 As shown, it includes: a processor 81 and a machine-readable storage medium 82, wherein the machine-readable storage medium 82 stores machine-executable instructions that can be executed by the processor 81; the processor 81 is used to execute the machine-executable instructions to implement the knowledge point mastery prediction method disclosed in the above example of this application.

[0135] Based on the same application concept as the above method, an embodiment of the present application also provides a machine-readable storage medium, on which a number of computer instructions are stored. When the computer instructions are executed by a processor, the knowledge point mastery prediction method disclosed in the above example of the present application can be implemented.

[0136] The machine-readable storage medium may be any electronic, magnetic, optical, or other physical storage device that may contain or store information, such as executable instructions, data, and the like. For example, the machine-readable storage medium may be: RAM (Random Access Memory), volatile memory, non-volatile memory, flash memory, a storage drive (such as a hard disk drive), a solid-state drive, any type of storage disk (such as a CD, DVD, etc.), or similar storage media, or a combination thereof.

[0137] Based on the same application concept as the above method, an embodiment of the present application further provides a computer program product, which may include a computer program. When the computer program is executed by a processor, it implements the knowledge point mastery prediction method disclosed in the above example of the present application.

[0138] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a fully hardware embodiment, a fully software embodiment, or an embodiment combining software and hardware. Furthermore, the embodiments of the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0139] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for predicting the mastery of knowledge points, characterized in that: The method comprises: Acquire a knowledge point description sequence, a target object question sequence, and an auxiliary information sequence; the knowledge point description sequence includes semantic description information corresponding to each candidate knowledge point name; the auxiliary information sequence includes the knowledge point name, question difficulty, and question response corresponding to each question in the question sequence; Determine a question embedding feature vector corresponding to the question sequence, an auxiliary embedding feature vector corresponding to the auxiliary information sequence, and a knowledge point embedding feature vector corresponding to the knowledge point description sequence; Performing a multi-head attention operation based on the question embedding feature vector and the auxiliary embedding feature vector to obtain an ability feature vector of the target object for the knowledge point name corresponding to the question sequence; Determine a knowledge space feature vector based on the knowledge point embedding feature vector, wherein the knowledge point embedding feature vector includes an initial feature vector of each candidate knowledge point name, and the knowledge space feature vector includes multiple fused feature vectors, and each fused feature vector is obtained by fusing multiple initial feature vectors; Perform a multi-head attention operation based on the capability feature vector, the knowledge point embedding feature vector, and the knowledge space feature vector to obtain a knowledge point target feature vector; Based on the knowledge point target feature vector and the acquired knowledge point map, the target object's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point map is predicted.

2. The method according to claim 1, characterized in that Get the knowledge point description sequence, including: Obtain a knowledge point sequence, a question text sequence, and a reference matrix; wherein the knowledge point sequence includes a plurality of candidate knowledge point names, the question text sequence includes a plurality of example question texts, and the reference matrix represents the correspondence between the candidate knowledge point names and the example question texts; The knowledge point sequence, the question text sequence and the reference matrix are input into a large language model, and the semantic description information corresponding to each candidate knowledge point name is determined by the large language model to obtain the knowledge point description sequence; wherein, for each candidate knowledge point name, the example question text corresponding to the candidate knowledge point name is determined by the reference matrix, and the candidate knowledge point name is semantically analyzed based on the example question text to obtain the semantic description information corresponding to the candidate knowledge point name.

3. The method according to claim 1, characterized in that The multi-head attention operation is performed based on the question embedding feature vector and the auxiliary embedding feature vector to obtain the target object's ability feature vector for the knowledge point name corresponding to the question sequence, including: Determine a first Q vector, a first K vector, and a first V vector based on the topic embedding feature vector, and perform a multi-head attention operation on the first Q vector, the first K vector, and the first V vector to obtain a first intermediate vector; Determine a second Q vector, a second K vector, and a second V vector based on the auxiliary embedding feature vector, and perform a multi-head attention operation based on the second Q vector, the second K vector, and the second V vector to obtain a second intermediate vector; Determining a third Q vector and a third K vector based on the first intermediate vector, determining a third V vector based on the second intermediate vector, performing a multi-head attention operation based on the third Q vector, the third K vector, and the third V vector to obtain a third intermediate vector, wherein the third intermediate vector is used to represent the mastery of the problem level; determining a fourth K vector based on the first intermediate vector, determining a fourth V vector based on the third intermediate vector, determining a fourth Q vector based on the knowledge space feature vector, and performing a multi-head attention operation based on the fourth Q vector, the fourth K vector, and the fourth V vector to obtain a fourth intermediate vector; The capability feature vector is determined based on the fourth intermediate vector.

4. The method according to claim 1, wherein The determining of the knowledge space feature vector based on the knowledge point embedding feature vector includes: Based on the knowledge point embedding feature vector, a fifth Q vector, a fifth K vector and a fifth V vector are determined, a multi-head attention operation is performed based on the fifth Q vector, the fifth K vector and the fifth V vector to obtain a fifth intermediate vector, and the fifth intermediate vector is input to a multilayer perceptron; a multilayer perception operation is performed by the multilayer perceptron based on the fifth intermediate vector to obtain the knowledge space feature vector.

5. The method according to claim 1, wherein The step of predicting the target subject's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point graph based on the knowledge point target feature vector and the acquired knowledge point graph includes: Determining a first graph feature and a second graph feature based on the knowledge point graph, wherein the first graph feature includes an initial feature vector of each candidate knowledge point name in the knowledge point graph, and the second graph feature includes a similarity between the initial feature vectors of two candidate knowledge point names in the knowledge point graph; Generate a target graph feature based on the knowledge point target feature vector and the first graph feature, wherein the target graph feature includes a fused feature vector of each candidate knowledge point name in the knowledge point graph, and the fused feature vector is obtained by concatenating the knowledge point target feature vector and the initial feature vector; Based on the target graph feature, the first graph feature and the second graph feature, the target object's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point graph is predicted.

6. The method according to claim 5, characterized in that The predicting, based on the target graph feature, the first graph feature, and the second graph feature, the target subject's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point graph includes: Input the target graph feature into a multilayer perceptron, and perform a feature mapping operation based on the target graph feature through the multilayer perceptron to obtain a mapped graph feature; wherein the feature dimension of the mapped graph feature is consistent with the input feature dimension of the graph neural network; The mapped graph features, the first graph features and the second graph features are input into the graph neural network, and the graph neural network performs prediction based on the mapped graph features, the first graph features and the second graph features to obtain the mastery degree of the target knowledge point.

7. The method according to any one of claims 1 to 6, characterized in that The question sequence also includes questions to be predicted, and the auxiliary information sequence also includes the names of knowledge points and question difficulty corresponding to the questions to be predicted; the method further includes: Based on the question embedding feature vector, the ability feature vector and the knowledge space feature vector, a predicted question response corresponding to the question to be predicted is determined; wherein the predicted question response indicates whether the target object's answer result to the question to be predicted is correct or incorrect.

8. The method according to claim 7, characterized in that The step of determining a predicted question response corresponding to the to-be-predicted question based on the question embedding feature vector, the capability feature vector, and the knowledge space feature vector includes: Determine a sixth Q vector based on the topic embedding feature vector, determine a sixth V vector based on the ability feature vector, and determine a sixth K vector based on the knowledge space feature vector; Performing a multi-head attention operation based on the sixth Q vector, the sixth K vector, and the sixth V vector to obtain a sixth intermediate vector, and inputting the sixth intermediate vector into a multilayer perceptron; A multi-layer perceptron is used to perform a multi-layer perceptron operation based on the sixth intermediate vector to obtain a post-operation feature, and a predicted question response corresponding to the to-be-predicted question is determined based on the post-operation feature.

9. A knowledge point mastery prediction device, characterized in that: The device comprises: An acquisition module is used to acquire a knowledge point description sequence, a target object question sequence, and an auxiliary information sequence; the knowledge point description sequence includes semantic description information corresponding to each candidate knowledge point name; the auxiliary information sequence includes the knowledge point name, question difficulty, and question response corresponding to each question in the question sequence; a determination module, configured to determine a question embedding feature vector corresponding to the question sequence, an auxiliary embedding feature vector corresponding to the auxiliary information sequence, and a knowledge point embedding feature vector corresponding to the knowledge point description sequence; and determine a knowledge space feature vector based on the knowledge point embedding feature vector, wherein the knowledge point embedding feature vector includes an initial feature vector for each candidate knowledge point name, and the knowledge space feature vector includes multiple fused feature vectors, each fused feature vector being obtained by fusing multiple initial feature vectors; a processing module configured to perform a multi-head attention operation based on the question embedding feature vector and the auxiliary embedding feature vector to obtain an ability feature vector of the target object for the knowledge point name corresponding to the question sequence; and perform a multi-head attention operation based on the ability feature vector, the knowledge point embedding feature vector, and the knowledge space feature vector to obtain a knowledge point target feature vector; A prediction module is used to predict the target object's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point map based on the knowledge point target feature vector and the acquired knowledge point map.

10. The device according to claim 9, characterized in that The processing module performs a multi-head attention operation based on the question embedding feature vector and the auxiliary embedding feature vector to obtain the target object's ability feature vector for the knowledge point name corresponding to the question sequence, specifically for: Determine a first Q vector, a first K vector, and a first V vector based on the topic embedding feature vector, and perform a multi-head attention operation on the first Q vector, the first K vector, and the first V vector to obtain a first intermediate vector; Determine a second Q vector, a second K vector, and a second V vector based on the auxiliary embedding feature vector, and perform a multi-head attention operation based on the second Q vector, the second K vector, and the second V vector to obtain a second intermediate vector; Determining a third Q vector and a third K vector based on the first intermediate vector, determining a third V vector based on the second intermediate vector, performing a multi-head attention operation based on the third Q vector, the third K vector, and the third V vector to obtain a third intermediate vector, wherein the third intermediate vector is used to represent the mastery of the problem level; determining a fourth K vector based on the first intermediate vector, determining a fourth V vector based on the third intermediate vector, determining a fourth Q vector based on the knowledge space feature vector, and performing a multi-head attention operation based on the fourth Q vector, the fourth K vector, and the fourth V vector to obtain a fourth intermediate vector; The capability feature vector is determined based on the fourth intermediate vector.

11. The device according to claim 9, characterized in that The prediction module is specifically used to predict the target knowledge point mastery of the target subject for each candidate knowledge point name in the knowledge point graph based on the knowledge point target feature vector and the acquired knowledge point graph: Determining a first graph feature and a second graph feature based on the knowledge point graph, wherein the first graph feature includes an initial feature vector of each candidate knowledge point name in the knowledge point graph, and the second graph feature includes a similarity between the initial feature vectors of two candidate knowledge point names in the knowledge point graph; Generate a target graph feature based on the knowledge point target feature vector and the first graph feature, wherein the target graph feature includes a fused feature vector of each candidate knowledge point name in the knowledge point graph, and the fused feature vector is obtained by concatenating the knowledge point target feature vector and the initial feature vector; Based on the target graph feature, the first graph feature and the second graph feature, the target object's mastery of the target knowledge point for each candidate knowledge point name in the knowledge point graph is predicted.

12. The device according to any one of claims 9 to 11, characterized in that: The question sequence also includes questions to be predicted, and the auxiliary information sequence also includes the knowledge point name and question difficulty corresponding to the question to be predicted; the prediction module is further used to: Based on the question embedding feature vector, the ability feature vector and the knowledge space feature vector, a predicted question response corresponding to the question to be predicted is determined; wherein the predicted question response indicates whether the target object's answer result to the question to be predicted is correct or incorrect.

13. An electronic device, characterized in that: include: a processor and a machine-readable storage medium storing machine-executable instructions capable of being executed by the processor; The processor is configured to execute machine-executable instructions to implement the method according to any one of claims 1 to 8.

Images