Student ability profiling method, learning resource recommendation method and device

Abstract

The present invention provides a student ability profiling method, a learning resource recommendation method and a device. By introducing a target subject knowledge graph, the encoded target knowledge point representation vector can represent the knowledge point information and knowledge relationship structure information of the question to be answered. The relationship between each knowledge point is considered when predicting the ability, which can greatly improve the accuracy of the prediction result. Moreover, by combining the target knowledge point representation vector with the student's historical answer information corresponding to the target subject, the student's knowledge ability is predicted, and the student's memory rules can be introduced, further improving the accuracy of the prediction result.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a student ability profiling method, a learning resource recommendation method and a device. Background Art

[0002] With the rapid development of online education and the gradual popularization of intelligent learning terminals, personalized learning resource recommendations based on big data have become an indispensable part of teaching plans and personal practice. For students with different learning situations, adaptive planning of learning paths and selection of appropriate learning resources can achieve the effect of improving efficiency and reducing burden.

[0003] Compared with mechanical fixed exercises, adaptive methods can effectively capture the changes in students' abilities in various learning scenarios and time periods, and avoid ineffective practice. Therefore, the real-time and automatic characterization of each student's knowledge and ability has become an important research issue. At present, simpler cognitive diagnostic models are usually used to model students' abilities. For example, the learning situation portrait method based on item response theory is limited by the small number of parameters and can only use a single value or vector to model students' abilities. It is difficult to obtain high accuracy in a complex knowledge system. At the same time, this type of method also ignores the potential correlation between knowledge points, reducing the accuracy of the portrait.

[0004] Therefore, there is an urgent need to provide a method for profiling students' abilities. Summary of the invention

[0005] The present invention provides a student ability profiling method, a learning resource recommendation method and a device, so as to solve the defects existing in the prior art.

[0006] The present invention provides a student ability profiling method, comprising:

[0007] Obtaining unanswered questions of a target subject, and extracting initial representations of target knowledge points of the unanswered questions;

[0008] Based on the target subject knowledge graph, the initial representation of the target knowledge point is encoded to determine the target knowledge point representation vector corresponding to the question to be answered, and the target knowledge point representation vector is used to represent the knowledge point information and knowledge point relationship structure information of the question to be answered;

[0009] Based on the target knowledge point representation vector and the student's historical answer information corresponding to the target subject, the student's knowledge ability is predicted.

[0010] According to a student ability profiling method provided by the present invention, encoding the initial representation of the target knowledge point based on the target subject knowledge graph and determining the target knowledge point representation vector corresponding to the question to be answered includes:

[0011] Based on the initial representation of the knowledge points corresponding to each graph node in the target subject knowledge graph, the initial representation of the target knowledge point is encoded to determine the target knowledge point representation vector corresponding to the question to be answered.

[0012] According to a student ability profiling method provided by the present invention, the initial representation of the target knowledge point is encoded based on the initial representation of the knowledge point corresponding to each graph node in the target subject knowledge graph, and the target knowledge point representation vector corresponding to the question to be answered is determined, including:

[0013] Based on the initial representation of the knowledge point corresponding to each graph node in the target subject knowledge graph, multi-layer cascaded graph convolution is performed on the initial representation of the target knowledge point to obtain features obtained by each layer of graph convolution;

[0014] The initial representation of the target knowledge point and the features obtained by the graph convolution of each layer are fused to obtain the representation vector of the target knowledge point.

[0015] According to a student ability profiling method provided by the present invention, the step of extracting an initial representation of a target knowledge point of the question to be answered includes:

[0016] Based on the sample of questions of the target subject carrying the knowledge point representation label, feature extraction is performed on the questions to be answered to obtain the initial representation of the target knowledge point.

[0017] According to a student ability profiling method provided by the present invention, the feature extraction of the to-be-answered questions based on the question samples of the target subject carrying the knowledge point representation labels is performed to obtain the initial representation of the target knowledge point, including:

[0018] Based on the sample of the question of the target subject carrying the knowledge point representation label, the phrase-level masking mechanism of the target subject is adopted to extract the features of the question to be answered, so as to obtain the initial representation of the target knowledge point.

[0019] According to a student ability profiling method provided by the present invention, the history question answering information includes a plurality of history questions and the history answering information corresponding to the plurality of history questions;

[0020] Accordingly, the step of predicting the student's knowledge ability based on the target knowledge point representation vector and the student's historical answer information corresponding to the target subject includes:

[0021] Based on the knowledge point representation vectors of the multiple history questions and the historical answer information corresponding to the multiple history questions, the target knowledge point representation vector is parsed to obtain the answer information of the question to be answered; the answer information is used to represent the knowledge ability of the student.

[0022] According to a student ability profiling method provided by the present invention, the plurality of history questions are arranged according to the corresponding answering times to form a time series, and the history answering information also includes the answering time interval between any history question and the previous adjacent question in the time series;

[0023] Accordingly, the target knowledge point representation vector is parsed based on the knowledge point representation vectors of the plurality of history questions and the history answer information corresponding to the plurality of history questions to obtain the answer information of the question to be answered, including:

[0024] The knowledge point representation vector of the current historical question corresponding to the current answering moment, the embedding vector of the corresponding historical answering information, and the embedding vector of the answering time interval are concatenated to obtain a concatenated feature;

[0025] Determine the knowledge state of the student at the current answering moment based on the splicing feature, the knowledge state of the student at the previous answering moment, and the memory retention feature;

[0026] Determine the answer information prediction result corresponding to the subsequent adjacent question based on the student's knowledge state at the current answering moment and the knowledge point representation vector of the subsequent adjacent question of the current history question;

[0027] Based on the historical answer information and answer information prediction result corresponding to the subsequent adjacent question, the target knowledge point representation vector is parsed to obtain the answer information of the to-be-answered question;

[0028] The previous answering time is the answering time corresponding to the previous adjacent question of the current historical question.

[0029] The present invention also provides a learning resource recommendation method, comprising:

[0030] Determine the student's knowledge and ability, where the knowledge and ability is predicted based on the student ability profiling method described above;

[0031] Based on the knowledge and ability, recommend learning resources matching the knowledge and ability to the student.

[0032] The present invention also provides a student ability profiling device, comprising:

[0033] An acquisition module, used to acquire the unanswered questions of the target subject and extract the initial representation of the target knowledge points of the unanswered questions;

[0034] An encoding module, for encoding the initial representation of the target knowledge point based on the target subject knowledge graph, and determining the target knowledge point representation vector corresponding to the question to be answered, wherein the target knowledge point representation vector is used to represent the knowledge point information and knowledge point relationship structure information of the question to be answered;

[0035] A prediction module is used to predict the student's knowledge ability based on the target knowledge point representation vector and the student's historical answer information corresponding to the target subject.

[0036] The present invention also provides a learning resource recommendation device, characterized in that it comprises:

[0037] A capability determination module, used to determine the student's knowledge capability, the knowledge capability is predicted based on the above-mentioned student capability profiling method;

[0038] A recommendation module is used to recommend learning resources matching the knowledge and ability to the student based on the knowledge and ability.

[0039] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-described student ability profiling methods or the steps of any of the above-described learning resource recommendation methods are implemented.

[0040] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program implements the steps of any of the above-mentioned student ability profiling methods or the steps of any of the above-mentioned learning resource recommendation methods.

[0041] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned student ability profiling methods, or the steps of any of the above-mentioned learning resource recommendation methods.

[0042] The student ability profiling method, learning resource recommendation method and device provided by the present invention first obtain the to-be-answered questions of the target subject, and extract the initial representation of the target knowledge point of the to-be-answered questions; then, based on the target subject knowledge graph, encode the initial representation of the target knowledge point, determine the target knowledge point representation vector corresponding to the to-be-answered questions, and the target knowledge point representation vector is used to represent the knowledge point information and knowledge point relationship structure information of the to-be-answered questions; finally, based on the target knowledge point representation vector and the historical answer information of the students corresponding to the target subject, predict the students' knowledge ability. By introducing the target subject knowledge graph, the encoded target knowledge point representation vector can represent the knowledge point information and knowledge relationship structure information of the to-be-answered questions, and the relationship between each knowledge point is considered in the ability prediction, which can greatly improve the accuracy of the prediction result. Moreover, by combining the target knowledge point representation vector with the historical answer information of the students corresponding to the target subject, the students' knowledge ability is predicted, and the students' memory rules can be introduced, which further improves the accuracy of the prediction result. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on the drawings in the following description without creative work.

[0044] Figure 1 It is a flow chart of the student ability profiling method provided by the present invention;

[0045] Figure 2 It is a structural schematic diagram of the knowledge point representation model provided by the present invention;

[0046] Figure 3 It is a schematic diagram of the training process of the prediction device provided by the present invention;

[0047] Figure 4 It is a flowchart of the learning resource recommendation method provided by the present invention;

[0048] Figure 5 It is a structural schematic diagram of the student ability profiling device provided by the present invention;

[0049] Figure 6 It is a structural schematic diagram of the learning resource recommendation method provided by the present invention;

[0050] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] With the rapid development of online education and the gradual popularization of smart learning terminals, personalized learning resource recommendations based on big data have become an indispensable part of school teaching plans and personal practice. For students with different learning situations, by adaptively planning the learning path and selecting appropriate learning resources, the effect of improving efficiency and reducing burden can be achieved. Compared with mechanized fixed exercises, adaptive methods can effectively capture the changes in students' abilities in various learning scenarios and time periods, and avoid ineffective practice. Among them, the real-time and automatic characterization of each student's knowledge and ability has become an important research issue. Accurately judging the students' mastery of various knowledge can provide understandable knowledge and ability analysis for students and teachers; at the same time, by providing appropriate feedback on students' behavior in real time, long-term ability improvement can be converted into short-term visible incentives, thereby increasing students' interest in independent learning.

[0053] In the past, in academia and industry, simpler cognitive diagnostic models were usually used to model student abilities. However, due to the limitation of the number of parameters, it is difficult for such models to achieve high accuracy in complex and diverse learning situations. Traditional learning situation profiling methods are usually based on item response theory or Bayesian theory, which simulates students' answering behavior through a smaller number of parameters.

[0054] Item response theory models the characteristics of student ability, test difficulty, test discrimination, and blind guess accuracy based on the student's answer sequence, and uses these characteristics to predict the probability of students answering specific questions correctly. Since this type of method has multiple educational psychology parameters, it has good interpretability. However, this method is limited by the small number of parameters and can only use a single value or vector to model student ability, which makes it difficult to obtain high accuracy in a complex knowledge system; at the same time, this method also ignores the potential correlation between knowledge points.

[0055] The Bayesian model models the students' knowledge status based on the hidden Markov process. The process starts with the initial mastery probability of the students' knowledge points, and uses the hidden Markov model to track the changing mastery probability by observing the students' answer records. Finally, the probability of answering the test questions correctly is predicted based on the latest student mastery probability. The Bayesian model can track the students' latest mastery level in real time through conditional probability, and its parameter setting takes into account the students' prior mastery level of the knowledge points before answering the questions. However, the Bayesian model only models a single knowledge point and lacks the potential correlation between knowledge points.

[0056] There are also learning ability profiling methods based on deep learning frameworks in the prior art, which usually use recurrent neural networks to iteratively model student answer sequences. Among them, the Deep Knowledge Tracing (DKT) model is the first method to apply recurrent networks to student ability modeling, and is usually used as a benchmark method. The DKT model inputs the current answer question and the correct / incorrect status at each time step, learns a dense vector representation for each question through trainable embedding parameters, and combines the hidden layer representation of the previous step to cyclically generate the latest knowledge ability. With the help of a neural network architecture, the deep knowledge tracing model has greatly improved the encoding ability of test questions and students, but it also lacks modeling of the potential correlation between knowledge points and student forgetting behavior.

[0057] It can be seen that the student ability profiling methods existing in the prior art all have the technical problem of low accuracy. To solve this technical problem, a student ability profiling method is provided in an embodiment of the present invention.

[0058] Figure 1 A schematic diagram of a process of generating a student capability profile according to an embodiment of the present invention is shown in FIG. Figure 1 As shown, the method includes:

[0059] S11, obtaining the unanswered questions of the target subject, and extracting the initial representation of the target knowledge points of the unanswered questions;

[0060] S12, encoding the initial representation of the target knowledge point based on the target subject knowledge graph, and determining the target knowledge point representation vector corresponding to the question to be answered, wherein the target knowledge point representation vector is used to represent the knowledge point information and knowledge point relationship structure information of the question to be answered;

[0061] S13, predicting the student's knowledge ability based on the target knowledge point representation vector and the student's historical answer information corresponding to the target subject.

[0062] Specifically, the student ability profiling method provided in the embodiment of the present invention is executed by a student ability profiling device, which can be configured in a server. The server can be a local server or a cloud server. The local server can specifically be a computer and a tablet computer, etc., which is not specifically limited in the embodiment of the present invention.

[0063] Among them, the student ability involved refers to the student's knowledge ability, that is, the student's mastery of each knowledge point of a certain subject. The student ability profiling method refers to the method of characterizing the student's ability, which can be understood as a method of predicting and evaluating the student's knowledge ability. The student's ability can usually be characterized by the student's mastery of each knowledge point of a certain subject, and the student's mastery of each knowledge point can be predicted by the probability of the student's answering the corresponding questions to be answered for each knowledge point of a certain subject.

[0064] First, step S11 is executed to obtain the unanswered questions of the target subject. The unanswered questions may correspond to the target knowledge points in the target subject, and the unanswered questions are usually in text form. The target subject can be set as needed, such as Chinese, mathematics, physics, chemistry, biology, etc.

[0065] After determining the questions to be answered, the initial representation of the target knowledge points of the questions to be answered can be extracted. The extraction process can be based on the mapping relationship between the question samples of the target subject and its knowledge point representation, and the questions to be answered are semantically mapped. Here, the mapping relationship can be specifically expressed as a knowledge point representation model obtained by training a randomly initialized neural network model, or it can be embodied as a semantic extraction rule obtained by mining association relationships. The embodiments of the present invention do not specifically limit this. For example, the questions to be answered can be input into the knowledge point representation model, and the knowledge point representation model performs feature extraction on the questions to be answered, and obtains the initial representation of the knowledge points output by the knowledge point representation model.

[0066] Then, step S12 is executed to introduce the target subject knowledge graph. Through the target subject knowledge graph, the initial representation of the target knowledge point can be encoded, and then the target knowledge point representation vector corresponding to the question to be answered can be determined. The target subject knowledge graph refers to the knowledge graph corresponding to the target subject, and the target subject knowledge graph contains various graph nodes and the connection relationship between each graph node, wherein each graph node represents each knowledge point in the target subject, and each graph node carries the initial representation of the knowledge point. The connection relationship between each graph node represents the potential correlation between each knowledge point in the target subject.

[0067] Encoding the initial representation of the target knowledge point through the target subject knowledge graph means that in the process of encoding the initial representation of the target knowledge point, the connection relationship between the graph nodes in the target subject knowledge graph is introduced, so that the encoded target knowledge point representation vector is used to represent the knowledge point information and knowledge point relationship structure information of the question to be answered. Among them, the knowledge point information of the question to be answered refers to the relevant information of the knowledge point corresponding to the question to be answered, and the knowledge point relationship structure information of the question to be answered refers to the relevant information of the adjacent knowledge points of the knowledge point corresponding to the question to be answered.

[0068] The process of encoding the initial representation of the target knowledge point through the target subject knowledge graph can be implemented by an encoder, which is not specifically limited here.

[0069] Finally, step S13 is performed to predict the student's knowledge ability through the target knowledge point representation vector and the student's historical answer information corresponding to the target subject. The student's knowledge ability refers to the student's mastery of the knowledge points corresponding to the question to be answered. The student's mastery of the knowledge points corresponding to the question to be answered can be predicted by the answer information obtained by the student when answering the question to be answered, and the student's knowledge ability can be represented by the answer information. The answer information can be the correct probability of the answer result obtained by the student when answering the question to be answered.

[0070] The historical answer information of students corresponding to the target subject refers to the relevant information when students answer historical questions related to the target subject in the past period of time, which may include multiple historical questions and historical answer information corresponding to each historical question, and may also include the answer time corresponding to each similar historical question belonging to the same knowledge point and the answer time interval between two adjacent similar historical questions, etc. Among them, it can be understood that the historical answer information corresponding to each historical question may be whether the answer result obtained by the student when answering the historical question is correct, and the specific content of the answer result is not concerned in the historical answer information.

[0071] In an embodiment of the present invention, step S3 can realize the prediction of the student's knowledge ability through a prediction model, and the prediction model can be constructed based on a long short-term memory network model. The memory forgetting characteristics contained in the long short-term memory network model can be very consistent with the characteristics of students in the learning process, that is, students learn and remember knowledge points by doing questions and forget them as time goes by. Modeling the student's answer sequence with the long short-term memory network model can well simulate the change process of the student's knowledge state over time.

[0072] Here, the initial model can be trained with the students' historical answer information for the target subject to obtain a prediction model, and then the knowledge point representation vector can be input into the prediction model to obtain the answer information obtained by the students for the questions to be answered output by the prediction model. Taking into account the individual differences of students, the prediction model can correspond to students one by one, that is, different models are required for prediction of different students' knowledge and abilities. Accordingly, the initial model needs to be trained with the historical answer information of different students for the target subject, and then prediction models for different students can be obtained.

[0073] The student ability profiling method provided in the embodiment of the present invention first obtains the to-be-answered questions of the target subject and extracts the initial representation of the target knowledge point of the to-be-answered questions; then, based on the target subject knowledge graph, the initial representation of the target knowledge point is encoded, and the target knowledge point representation vector corresponding to the to-be-answered questions is determined, and the target knowledge point representation vector is used to represent the knowledge point information and knowledge point relationship structure information of the to-be-answered questions; finally, based on the target knowledge point representation vector and the historical answer information of the student corresponding to the target subject, the student's knowledge ability is predicted. Through the introduction of the target subject knowledge graph, the encoded target knowledge point representation vector can represent the knowledge point information and knowledge relationship structure information of the to-be-answered questions, and the relationship between each knowledge point is considered in the ability prediction, which can greatly improve the accuracy of the prediction result. Moreover, by combining the target knowledge point representation vector with the historical answer information of the student corresponding to the target subject, the student's knowledge ability is predicted, and the student's memory rules can be introduced, which further improves the accuracy of the prediction result.

[0074] On the basis of the above-mentioned embodiment, the student ability profiling method provided in the embodiment of the present invention, encoding the initial representation of the target knowledge point based on the target subject knowledge graph, and determining the target knowledge point representation vector corresponding to the question to be answered, includes:

[0075] Based on the initial representation of the knowledge points corresponding to each graph node in the target subject knowledge graph, the initial representation of the target knowledge point is encoded to determine the target knowledge point representation vector corresponding to the question to be answered.

[0076] Specifically, in an embodiment of the present invention, when the initial representation of the knowledge point is encoded through the target subject knowledge graph to determine the knowledge point representation vector corresponding to the question to be answered, the encoding process can be implemented through the BERT model, or through the encoding part in the Transformer model. In order to improve the encoding accuracy, a graph convolutional neural network model can be used to input the initial representation of the target knowledge point into the graph convolutional neural network model to obtain the target knowledge point representation vector output by the graph convolutional neural network model. The graph convolutional neural network model can be obtained by training the initial representation of the knowledge point corresponding to each graph node in the target subject knowledge graph.

[0077] It can be understood that when extracting the initial representation of the knowledge points corresponding to each graph node in the target subject knowledge graph, you can first obtain multiple questions about the knowledge points corresponding to each graph node, and then extract the initial representation of the knowledge points of each question respectively, and finally average pool the initial representation of the knowledge points of each question to obtain the initial representation of the knowledge points corresponding to each graph node.

[0078] By initially representing the knowledge points corresponding to each graph node in the target subject knowledge graph, the graph convolution initial model is trained to obtain a graph convolution neural network model. The graph convolution neural network model may include multiple graph convolution network layers, and the activation function of each graph convolution network layer includes the adjacency matrix between each graph node in the target subject knowledge graph, indicating the edge connection relationship between each graph node.

[0079] The graph convolutional neural network model can be used to encode the initial representation of knowledge points and obtain the knowledge point representation vector corresponding to the question to be answered.

[0080] In an embodiment of the present invention, a target knowledge point representation vector corresponding to a question to be answered is determined through a graph convolutional neural network model, so that the target knowledge point representation vector simultaneously represents the knowledge point information and the knowledge relationship structure information, thereby achieving a perfect combination of the knowledge point information and the knowledge relationship structure information.

[0081] On the basis of the above-mentioned embodiment, the student ability profiling method provided in the embodiment of the present invention, based on the initial representation of the knowledge point corresponding to each graph node in the target subject knowledge graph, encodes the initial representation of the target knowledge point, and determines the target knowledge point representation vector corresponding to the question to be answered, including:

[0082] Based on the initial representation of the knowledge point corresponding to each graph node in the target subject knowledge graph, multi-layer cascaded graph convolution is performed on the initial representation of the target knowledge point to obtain features obtained by each layer of graph convolution;

[0083] The initial representation of the target knowledge point and the features obtained by the graph convolution of each layer are fused to obtain the representation vector of the target knowledge point.

[0084] Specifically, in an embodiment of the present invention, when encoding the initial representation of the target knowledge point, the initial representation of the target knowledge point can be firstly subjected to multi-layer cascade graph convolution according to the initial representation of the knowledge point corresponding to each graph node in the target subject knowledge graph, and the features obtained by each layer of graph convolution can be obtained. Here, a graph convolution neural network model can be firstly trained according to the initial representation of the knowledge point corresponding to each graph node in the target subject knowledge graph, and then the graph convolution neural network model can be used to implement the multi-layer cascade graph convolution operation on the initial representation of the target knowledge point.

[0085] That is, the graph convolutional neural network model can include multiple graph convolutional network layers, each of which is connected in sequence, and each of which has a graph convolution operation function to extract abstract features from the initial representation of the knowledge points of the questions to be answered. The initial representation of the target knowledge point can be sequentially input into each graph convolutional network layer of the graph convolutional neural network model to obtain the features obtained by each graph convolutional network layer. Each graph convolutional network layer can use the frequency domain convolution method to perform Fourier transform on the target subject knowledge graph. The activation functions of the two are the same, and can be expressed as:

[0086]

[0087] Among them, A represents the adjacency matrix between each graph node, which represents the edge connection relationship between each graph node; Where I is the identity matrix; express The degree matrix of is a diagonal matrix, and has H (l) is the output of the lth graph convolutional network layer; σ is the activation function, which can be a nonlinear activation function; W (l) is the weight matrix of the lth graph convolutional network layer. It is similar to the symmetric normalized Laplacian matrix, so it is called graph convolution.

[0088] Then, the initial representation of the target knowledge point and the features obtained by each graph convolutional network can be input into the fusion layer of the graph convolutional neural network model to obtain the target knowledge point representation vector output by the fusion layer.

[0089] Taking the graph convolutional neural network model including 2 graph convolutional network layers and 1 fusion layer as an example, the initial representation of the target knowledge point can be input into the first graph convolutional network layer to obtain the first feature output by the first graph convolutional network layer; then the first feature is input into the second graph convolutional network layer to obtain the second feature output by the second graph convolutional network layer; finally, the initial representation of the target knowledge point, the first feature and the second feature are input into the fusion layer to obtain the target knowledge point representation vector output by the fusion layer.

[0090] For example, the first feature can be expressed as H (0) , the second feature can be expressed as H (1) , the target knowledge point of the question to be answered is initially represented as x0, then the target knowledge point representation vector x of the question to be answered can be obtained through H (0) , the second feature can be expressed as H (1) , that is:

[0091]

[0092] In an embodiment of the present invention, two graph convolutional network layers and a fusion layer are introduced into the graph convolutional neural network model to obtain a more fine-grained and higher-precision knowledge point representation vector, thereby improving the accuracy of the prediction results.

[0093] On the basis of the above-mentioned embodiment, the student ability profiling method provided in the embodiment of the present invention, wherein the step of extracting the initial representation of the target knowledge point of the question to be answered comprises:

[0094] Based on the sample of questions of the target subject carrying the knowledge point representation label, feature extraction is performed on the questions to be answered to obtain the initial representation of the target knowledge point.

[0095] Specifically, in an embodiment of the present invention, when extracting the initial representation of the target knowledge point of the question to be answered, the correspondence between the question and the knowledge point representation can be determined through the question samples of the target subject carrying the knowledge point representation label, and then the initial representation of the target knowledge point of the question to be answered can be determined based on the correspondence.

[0096] The above correspondence can be represented by a knowledge point representation model, which can be obtained by training an initial model based on a sample of questions of a target subject carrying a knowledge point representation label. The question to be answered is input into the knowledge point representation model, after which the knowledge point representation model can extract features of the question to be answered, and then obtain and output the initial representation of the target knowledge point. The above initial model can be a neural network model with randomly initialized model parameters, or a pre-trained language model.

[0097] Taking the example that the knowledge point representation model is obtained by training the pre-trained prediction model with the sample of the target subject with the knowledge point representation label, the specific training process may include:

[0098] First, obtain a pre-trained prediction model. The pre-trained language model can be an open-source large-scale language model, such as a language representation model (Bidirectional Encoder Representation from Transformers, BERT), RoBERTa model, etc. The pre-trained language model is pre-trained on massive open corpus data and can learn richer and more accurate word semantic information than ordinary small-scale word vector models. Based on the BERT model, the RoBERTa model increases the training time and batch size to ensure the performance of the model.

[0099] Then, the sample of the question of the target subject can be input into the pre-trained prediction model to obtain the knowledge point prediction representation of the sample of the question output by the pre-trained language model;

[0100] Finally, based on the knowledge point prediction representation and knowledge point representation label of the question sample, the loss function value is calculated, and the model parameters of the pre-trained language model are updated until the loss function converges. At this point, the training is completed and the knowledge point representation model can be obtained.

[0101] It can be understood that there are multiple question samples used in the training process, each question sample can correspond to a different knowledge point, and the number of question samples corresponding to the same knowledge point can be one or more, which is not detailed here.

[0102] In order to ensure that the initial representation of the knowledge point corresponding to each graph node in the target subject knowledge graph is more universal, for each graph node, all the question samples under the knowledge point corresponding to the graph node are obtained, and each question sample is input into the knowledge point representation model to obtain the initial representation of the knowledge point of each question sample corresponding to the graph node. Then, the initial representation of the knowledge point of each question sample corresponding to the graph node is averaged and pooled, and the average pooled vector obtained is used as the initial representation of the knowledge point corresponding to the graph node.

[0103] Figure 2 Schematic diagram of the structure of the knowledge point representation model provided in the embodiment of the present invention, such as Figure 2 As shown, before inputting the question to be answered into the knowledge point representation model, it is necessary to segment the question to be answered to obtain the words of the question to be answered, and each word is represented by a Token. Suppose that after the question to be answered is segmented, N Tokens are obtained, representing Token1, Token2, ..., TokenN respectively. Figure 2 CLS is the sentence identifier of the question to be answered.

[0104] Then, each word of the question to be answered is input into the knowledge point representation model. The knowledge point representation model includes an input layer, a hidden layer, and an output layer. The input layer is used to determine the embedding vector of each word, which can be represented as E[CLS], E1, E2, …, EN respectively. The hidden layer is used to perform deep feature extraction on the embedding vector of each word, and the hidden layer is usually a Transformer layer. The output layer is used to determine the knowledge point representation vector C of the question to be answered. C is [CLS] corresponding to the output of the last Transformer in the hidden layer, and Ti (1≤i≤N) indicates that Token i corresponds to the output of the last Transformer. In addition, the output layer can also output the target knowledge point of the question to be answered, which can be represented by a knowledge point label (conceptlabel).

[0105] In the embodiment of the present invention, the extraction efficiency can be improved by extracting features from the questions to be answered through the question samples. Moreover, the knowledge point representation model representing the corresponding relationship is obtained by training the pre-trained prediction model, which can simplify the training process, shorten the training time, improve the training efficiency, and improve the reliability of the knowledge point representation model.

[0106] On the basis of the above-mentioned embodiment, the student ability profiling method provided in the embodiment of the present invention, based on the sample of the question of the target subject carrying the knowledge point representation label, performs feature extraction on the question to be answered to obtain the initial representation of the target knowledge point, including:

[0107] Based on the sample of the question of the target subject carrying the knowledge point representation label, the phrase-level masking mechanism of the target subject is adopted to extract the features of the question to be answered, so as to obtain the initial representation of the target knowledge point.

[0108] Specifically, in an embodiment of the present invention, when a knowledge point representation model is used to extract features from questions to be answered, a phrase-level masking mechanism of the target subject can be introduced. The mask used by the masking mechanism is a phrase-level masking template of the target subject, which can not only optimize the semantic abstraction ability of the knowledge point representation model and the accuracy of downstream tasks, but also improve the knowledge point representation model's ability to extract the semantics of phrases within the target subject. In addition, in order to improve the feature extraction efficiency of the knowledge point representation model, the redundant subtask of predicting the next sentence in the pre-trained language model can be removed.

[0109] On the basis of the above-mentioned embodiment, in the student ability profiling method provided in the embodiment of the present invention, the history question answering information includes a plurality of history questions and the history answering information corresponding to the plurality of history questions;

[0110] Accordingly, the step of predicting the student's knowledge ability based on the target knowledge point representation vector and the student's historical answer information corresponding to the target subject includes:

[0111] Based on the knowledge point representation vectors of the multiple history questions and the historical answer information corresponding to the multiple history questions, the target knowledge point representation vector is parsed to obtain the answer information of the question to be answered; the answer information is used to represent the knowledge ability of the student.

[0112] Specifically, when predicting students' knowledge and ability, the knowledge point representation vectors of multiple history questions and the historical answer information corresponding to the multiple history questions can be used to predict the students' knowledge status at the current answering moment, and then the target knowledge point representation vector is analyzed in combination with the knowledge status at the current answering moment to obtain the answer information of the question to be answered. Among them, the historical answer information may include multiple history questions and the historical answer information corresponding to each history question. The knowledge point representation vectors of multiple history questions can be determined in the same way as the knowledge point representation vectors of the questions to be answered, which will not be repeated here. The historical answer information is whether the answer results obtained by the students to each history question are correct, which can be represented by 1 if correct and 0 if wrong.

[0113] The above process can be realized by a prediction model, which has the ability to predict the student's knowledge state at the current answering moment. This ability can be obtained in the process of training the initial model to obtain the prediction model through the knowledge point representation vectors of multiple history questions and the historical answer information corresponding to multiple history questions. Then, the target knowledge point representation vector of the question to be answered is input into the prediction model, and the prediction model analyzes the target knowledge point representation vector based on the student's knowledge state at the current answering moment, and obtains the student's answer information for the question to be answered output by the prediction model. The answer information is the student's answer information at the current moment or at a certain moment in the future, that is, the correct probability of the answer result obtained by the student at the current moment or at a certain moment in the future. The answer information can be used to predict the student's knowledge ability at the current moment or at a certain moment in the future.

[0114] In the process of training the prediction model, the knowledge point representation vectors of multiple history questions and the corresponding historical answer information can be input into the initial model of the prediction model. The initial model can be a neural network model with randomly initialized model parameters.

[0115] The initial model obtains the student's knowledge state at the time of answering any history question according to the knowledge point representation vector of the previous history question in the time series and the corresponding historical answer information, and determines the answer information prediction information of any history question by combining the knowledge point representation vector of any history question, the corresponding knowledge state at the time of answering and the historical answer information. Then, based on the historical answer information and the answer information prediction information of any history question, the loss function value is calculated, and the model parameters of the initial model are updated until the loss function converges. At this time, the training is completed and the prediction model can be obtained.

[0116] In the embodiment of the present invention, the knowledge point representation vectors of multiple history questions and the historical answer information corresponding to each history question are combined to predict the knowledge ability of the student, which can improve the prediction efficiency. Moreover, the knowledge point representation vectors of multiple history questions and the historical answer information corresponding to each history question are combined to determine the knowledge state of the student at a certain moment, which can ensure the accuracy of the prediction result.

[0117] On the basis of the above-mentioned embodiment, in the student ability profiling method provided in the embodiment of the present invention, the plurality of history questions are arranged according to the corresponding answering time to form a time series, and the history answering information also includes the answering time interval between any history question and the previous adjacent question in the time series;

[0118] Accordingly, the target knowledge point representation vector is parsed based on the knowledge point representation vectors of the plurality of history questions and the history answer information corresponding to the plurality of history questions to obtain the answer information of the question to be answered, including:

[0119] The knowledge point representation vector of the current historical question corresponding to the current answering moment, the embedding vector of the corresponding historical answering information, and the embedding vector of the answering time interval are concatenated to obtain a concatenated feature;

[0120] Determine the knowledge state of the student at the current answering moment based on the splicing feature, the knowledge state of the student at the previous answering moment, and the memory retention feature;

[0121] Determine the answer information prediction result corresponding to the subsequent adjacent question based on the student's knowledge state at the current answering moment and the knowledge point representation vector of the subsequent adjacent question of the current history question;

[0122] Based on the historical answer information and answer information prediction result corresponding to the subsequent adjacent question, the target knowledge point representation vector is parsed to obtain the answer information of the to-be-answered question;

[0123] The previous answering time is the answering time corresponding to the previous adjacent question of the current historical question.

[0124] Specifically, in the embodiment of the present invention, since the historical answer information includes multiple historical questions and the historical answer information corresponding to the multiple historical questions, the multiple historical questions can be arranged according to the corresponding answering time to form a time series. Therefore, the historical answer information also includes the answering time interval between any historical question and the previous adjacent question in the time series.

[0125] For any of the multiple historical topics, the previous adjacent topic of the historical topic refers to the historical topic that is before and adjacent to the historical topic in the time series. It is understandable that the first historical topic in the time series has no corresponding previous adjacent topic.

[0126] When parsing the target knowledge point representation vector, the knowledge point representation vector of the current historical question corresponding to the current answering time, the embedding vector of the corresponding historical answer information, and the embedding vector of the answering time interval can be spliced ​​to obtain the splicing feature. Among them, the splicing method can be directly connected head to tail, and the connection order can be set as needed, which is not specifically limited here.

[0127] Then, the knowledge state of the student at the current answering moment can be determined based on the splicing features and the knowledge state of the student at the previous answering moment, and by using the memory rules and forgetting rules. The knowledge state of the student at the current answering moment refers to the student's mastery of each knowledge point of the target subject at the current answering moment. In the determination process, the memory retention feature of the student at the current answering moment can be introduced. The memory retention feature of the student at the current answering moment is used to characterize the information that the student can still master about each knowledge point from the moment when the student first contacts and masters each knowledge point to the current answering moment. The memory retention feature of the student at the current answering moment can be determined by the memory retention feature of the student at the previous answering moment and the selective forgetting feature of the student at the current answering moment. The selective forgetting feature of the student at the current answering moment can be determined based on the splicing features and the knowledge state of the student at the previous answering moment. This process can be achieved through the memory cell layer in the long short-term memory network.

[0128] Then, the answer information prediction result corresponding to the subsequent adjacent question can be determined based on the student's knowledge status at the current answering moment and the knowledge point representation vector of the subsequent adjacent question of the current history question. That is, the knowledge point corresponding to the subsequent adjacent question can be determined by using the knowledge point representation vector of the subsequent adjacent question, and then the answer information prediction result corresponding to the subsequent adjacent question can be predicted in combination with the student's knowledge status at the current answering moment.

[0129] Finally, the target knowledge point representation vector can be parsed according to the historical answer information and answer information prediction results corresponding to the subsequent adjacent questions to obtain the answer information of the question to be answered. Here, the corresponding relationship between the knowledge point representation vector and the answer information of the question can be constructed by combining the above process, the historical answer information and answer information prediction results corresponding to the subsequent adjacent questions. Then, through this corresponding relationship, the target knowledge point representation vector is parsed to obtain the answer information of the question to be answered.

[0130] It can be understood that the above correspondence can be represented by a prediction model, which can be obtained by training an initial model through knowledge point representation vectors of multiple history questions and historical answer information corresponding to multiple history questions.

[0131] The flowchart of the prediction model obtained by training the initial model is as follows: Figure 3 As shown, let the current answering time be t, the previous answering time be t-1, and the next answering time be t+1. It is understandable that the current answering time refers to the answering time of the current historical question, the previous answering time refers to the answering time corresponding to the previous adjacent question of the current historical question in the time series, and the next answering time is the answering time corresponding to the subsequent adjacent question of the current historical question in the time series. Here, the current historical question can be every historical question in the time series. The subsequent adjacent question of the current historical question refers to the historical question that comes after the current historical question in the time series and is adjacent to the current historical question. It is understandable that the last historical question in the time series has no corresponding subsequent adjacent question.

[0132] The training process of the initial model is as follows:

[0133] The knowledge point representation vector x of the current historical question corresponding to the current answering time t t , the corresponding embedding vector a of historical answer information t and the embedding vector s of the answer time interval t , input to the input layer of the initial model of the prediction model, through the input layer for feature splicing, the splicing feature e output by the input layer can be obtained t .

[0134] The splicing feature e t , the student’s knowledge status h at the previous answering time t-1 t-1 and the memory retention characteristics c of the students at the previous answering time t-1 t-1 , which is input into the memory cell layer of the initial model. Through the memory cell layer, the student’s knowledge state h at the previous answer time t can be output t .

[0135] Here, the memory cell layer may include an input gate, a forget gate, and an output gate.t 、h t-1 and c t-1 Input to the input gate, through which the input gate can control the t and h t-1 Perform selective memory and obtain the selective memory feature i t . t 、h t-1 and c t-1 Input to the forget gate, through which the c t-1 Perform selective forgetting and obtain the selective forgetting feature f t . t 、h t-1 and c t-1 Input to the output gate, through the output gate can control h t-1 Update and obtain the knowledge state weight feature o t .

[0136] In this process, the student’s knowledge status h at the current answering time t t Based on the knowledge state weight feature o at the current answering time t t and the student's memory retention characteristics c at the current answering time t t Determine, at the current answering time t, the student's memory retention characteristic c t Based on the memory retention characteristics c of the students at the previous answering time t-1 t-1 And the selective forgetting feature f at the current answer time t t OK. Understandably, Figure 3 In the figure, c represents the student's memory retention action at the memory cell layer, and h represents the student's knowledge status determination action.

[0137] The formulas used include:

[0138] i t =σ(Z xi x t +Zh i h t-1 +b i )

[0139] f t =σ(Z xf x t +Z hf h t-1 +b f )

[0140] o t =σ(Z xo x t +Z ho h t x1+b o )

[0141] c t =f t ·c t-1 +i t ·tanh(Z xc x t +Z hc h t-1 +b c )

[0142] h t =o t ·tanh(c t )

[0143] Among them, σ and tanh are both activation functions, and both can be nonlinear activation functions. xi , Z hi The input gate corresponds to x t The coefficient matrix of h t-1 The coefficient matrix of b. i is the bias matrix of the input gate. xf , Z hf The forget gate corresponds to x t The coefficient matrix of h t-1 The coefficient matrix of b. f is the bias matrix of the forget gate. xo , Z ho The output gate corresponds to x t The coefficient matrix of h t-1 The coefficient matrix of b. i is the bias matrix of the output gate.

[0144] Then, h t And the knowledge point representation vector x of the next adjacent question of the current history question t+1 , input to the output layer of the initial model, and obtain the prediction result of the answer information corresponding to the next adjacent question output by the output layer, that is, the prediction result of the answer information corresponding to the next adjacent question at the next answer time t+1 of the current answer time t, which can be expressed as p t+1 , and there are:

[0145]

[0146] Among them, W is the weight parameter of the output layer, and b is the bias parameter of the output layer.

[0147] Finally, the initial model is trained by combining the historical answer information corresponding to the subsequent adjacent questions and the answer information prediction results to obtain the prediction model. t+1And the answer information prediction result p t+1 , calculate the value of the target loss function L, which can be a cross entropy function. That is:

[0148] L=-∑ t＝0 (a t+1 logp t+1 +(1-a t+1 )log(1-p t+1 ))

[0149] The value of the target loss function is calculated in real time. When the target loss function converges, the training ends and the prediction model is obtained.

[0150] In an embodiment of the present invention, the time interval between answering adjacent history questions is introduced, and a specific training process is given, so that the trained prediction model can learn the forgetting behavior of students' knowledge and ability on target knowledge points over time, which can not only improve the prediction accuracy of the prediction model, but also make the prediction results of the trained prediction model more accurate.

[0151] Figure 4 A flow chart of a learning resource recommendation method provided in an embodiment of the present invention is shown in FIG. Figure 4 As shown, the method includes:

[0152] S21, determining the knowledge ability of the student, wherein the knowledge ability is predicted based on the student ability profiling method provided in the above embodiments;

[0153] S22, based on the knowledge and ability, recommending to the student learning resources that match the knowledge and ability.

[0154] Specifically, the learning resource recommendation method provided in the embodiment of the present invention is executed by a learning resource recommendation device, which can be configured in a server. The server can be a local server or a cloud server. The local server can specifically be a computer and a tablet computer, etc., which is not specifically limited in the embodiment of the present invention.

[0155] First, step S21 is executed to predict the student's knowledge ability through the student ability profiling method provided in the above embodiments. The specific prediction method will not be described in detail here.

[0156] Then, step S22 is executed to recommend learning resources that match the knowledge and ability to the student based on the predicted knowledge and ability. Since the knowledge and ability can be characterized by the answer information obtained by the student to answer the question to be answered, if the answer information is the accuracy probability of the answer result obtained by the student to answer the question to be answered is higher than or equal to the preset threshold, it can be considered that the student has mastered the target knowledge point corresponding to the question to be answered, and there is no need to recommend the learning resources related to the target knowledge point to the student. The student's knowledge and ability for other knowledge points can continue to be predicted. If the accuracy of the answer result is lower than the preset threshold, it can be considered that the student has not yet mastered the target knowledge point corresponding to the question to be answered, and the learning resources corresponding to the target knowledge point need to be recommended to the student. At this time, the learning resource corresponding to the target knowledge point is the learning resource that matches the student's knowledge and ability.

[0157] The learning resource recommendation method provided in the embodiment of the present invention first determines the knowledge ability of the student, and then recommends learning resources matching the knowledge ability to the student based on the knowledge ability. Suitable learning resources can be automatically recommended to the student based on the knowledge ability items of the student, thereby improving the accuracy of the recommendation.

[0158] In summary, the student ability profiling method and learning resource recommendation method provided in the embodiments of the present invention are based on the target subject knowledge graph, and the historical answer information of the students is spatially mapped and modeled through the graph convolutional deep network model, which can effectively abstract the knowledge point representation covering the surrounding related knowledge point information, thereby introducing the associated information in the knowledge graph, and obtaining the vectorized ability representation of the students, that is, the knowledge point representation vector, and then obtaining the students' knowledge ability at each knowledge point. The interpretable and accurate modeling of students' abilities can greatly improve students' learning efficiency and provide the most important basis for personalized recommendations for teaching students in accordance with their aptitude. In addition, by introducing the answer time interval when training the initial model to obtain the prediction model, the forgetting behavior of students' abilities in different knowledge over time can be learned, and the prediction accuracy of the prediction model can be improved.

[0159] like Figure 5 As shown, based on the above embodiment, an embodiment of the present invention provides a student ability profiling device, including:

[0160] An acquisition module 51 is used to acquire the unanswered questions of the target subject and extract the initial representation of the target knowledge points of the unanswered questions;

[0161] An encoding module 52 is used to encode the initial representation of the target knowledge point based on the target subject knowledge graph, and determine the knowledge point representation vector corresponding to the question to be answered, wherein the target knowledge point representation vector is used to represent the knowledge point information and knowledge point relationship structure information of the question to be answered;

[0162] The prediction module 53 is used to predict the student's knowledge ability based on the target knowledge point representation vector and the student's historical answer information corresponding to the target subject.

[0163] On the basis of the above-mentioned embodiment, in the student ability profiling device provided in the embodiment of the present invention, the encoding module is specifically used for:

[0164] Based on the initial representation of the knowledge points corresponding to each graph node in the target subject knowledge graph, the initial representation of the target knowledge point is encoded to determine the target knowledge point representation vector corresponding to the question to be answered.

[0165] On the basis of the above-mentioned embodiment, in the student ability profiling device provided in the embodiment of the present invention, the encoding module is specifically used for:

[0166] Based on the initial representation of the knowledge point corresponding to each graph node in the target subject knowledge graph, multi-layer cascaded graph convolution is performed on the initial representation of the target knowledge point to obtain features obtained by each layer of graph convolution;

[0167] The initial representation of the target knowledge point and the features obtained by the graph convolution of each layer are fused to obtain the representation vector of the target knowledge point.

[0168] On the basis of the above-mentioned embodiment, in the student ability profiling device provided in the embodiment of the present invention, the acquisition module is used for:

[0169] Based on the sample of questions of the target subject carrying the knowledge point representation label, feature extraction is performed on the questions to be answered to obtain the initial representation of the target knowledge point.

[0170] On the basis of the above-mentioned embodiment, in the student ability profiling device provided in the embodiment of the present invention, the acquisition module is specifically used for:

[0171] Based on the sample of the question of the target subject carrying the knowledge point representation label, the phrase-level masking mechanism of the target subject is adopted to extract the features of the question to be answered, so as to obtain the initial representation of the target knowledge point.

[0172] On the basis of the above-mentioned embodiment, in the student ability profiling device provided in the embodiment of the present invention, the history question answering information includes a plurality of history questions and the history answering information corresponding to the plurality of history questions;

[0173] Accordingly, the prediction module is used to:

[0174] Based on the knowledge point representation vectors of the multiple history questions and the historical answer information corresponding to the multiple history questions, the target knowledge point representation vector is parsed to obtain the answer information of the question to be answered; the answer information is used to represent the knowledge ability of the student.

[0175] On the basis of the above-mentioned embodiment, in the student ability profiling device provided in the embodiment of the present invention, the plurality of history questions are arranged according to the corresponding answering time to form a time series, and the history answering information also includes the answering time interval between any history question in the time series and the previous adjacent question of any history question in the time series;

[0176] Accordingly, the prediction module is also used for:

[0177] The knowledge point representation vector of the current historical question corresponding to the current answering moment, the embedding vector of the corresponding historical answering information, and the embedding vector of the answering time interval are concatenated to obtain a concatenated feature;

[0178] Determine the knowledge state of the student at the current answering moment based on the splicing feature and the knowledge state of the student at the previous answering moment;

[0179] Determine the answer information prediction result corresponding to the subsequent adjacent question based on the student's knowledge state at the current answering moment and the knowledge point representation vector of the subsequent adjacent question of the current history question;

[0180] Based on the historical answer information and answer information prediction result corresponding to the subsequent adjacent question, the target knowledge point representation vector is parsed to obtain the answer information of the to-be-answered question;

[0181] The previous answering time is the answering time corresponding to the previous adjacent question of the current historical question.

[0182] Specifically, the functions of each module in the student ability portrait device provided in the embodiment of the present invention correspond one-to-one to the operating procedures of each step in the above-mentioned method embodiment, and the effects achieved are also consistent. Please refer to the above-mentioned embodiment for details, and no further details will be given in the embodiment of the present invention.

[0183] like Figure 6 As shown, based on the above embodiment, an embodiment of the present invention provides a learning resource recommendation device, including:

[0184] Ability determination module 61, used to determine the knowledge ability of the student, the knowledge ability is predicted based on the student ability profiling method provided in the above embodiments;

[0185] The recommendation module 62 is used to recommend learning resources matching the knowledge and ability to the student based on the knowledge and ability.

[0186] Specifically, the functions of each module in the learning resource recommendation device provided in the embodiment of the present invention correspond one-to-one to the operation flow of each step in the above method embodiment, and the effects achieved are also consistent. Please refer to the above embodiment for details, which will not be repeated in the embodiment of the present invention.

[0187] Figure 7 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 7 As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730 and a communication bus 740, wherein the processor 710, the communication interface 720 and the memory 730 communicate with each other through the communication bus 740. The processor 710 may call the logic instructions in the memory 730 to execute the student ability profiling method provided in the above embodiments, the method comprising: obtaining the unanswered questions of the target subject, and extracting the initial representation of the target knowledge points of the unanswered questions; encoding the initial representation of the target knowledge points based on the target subject knowledge graph, determining the target knowledge point representation vector corresponding to the unanswered questions, the target knowledge point representation vector is used to represent the knowledge point information and knowledge point relationship structure information of the unanswered questions; based on the target knowledge point representation vector and the student's historical answer information corresponding to the target subject, predicting the student's knowledge ability. Or execute the learning resource recommendation method provided in the above embodiments, the method comprising: determining the student's knowledge ability, the knowledge ability is predicted based on the student ability portrait method provided in the above embodiments; based on the knowledge ability, recommending learning resources matching the knowledge ability to the student.

[0188] In addition, the logic instructions in the above-mentioned memory 730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0189] On the other hand, the present invention also provides a computer program product, the computer program product includes a computer program, the computer program can be stored on a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the student ability profiling method provided in the above embodiments, the method includes: obtaining the to-be-answered questions of the target subject, and extracting the target knowledge point initial representation of the to-be-answered questions; based on the target subject knowledge graph, encoding the target knowledge point initial representation, determining the target knowledge point representation vector corresponding to the to-be-answered questions, the target knowledge point representation vector is used to represent the knowledge point information and knowledge point relationship structure information of the to-be-answered questions; based on the target knowledge point representation vector and the student's historical answer information corresponding to the target subject, predicting the student's knowledge ability. Or executing the learning resource recommendation method provided in the above embodiments, the method includes: determining the student's knowledge ability, the knowledge ability is predicted based on the student ability profiling method provided in the above embodiments; based on the knowledge ability, recommending learning resources matching the knowledge ability to the student.

[0190] On the other hand, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it is implemented to execute the student ability profiling method provided in the above embodiments, the method comprising: obtaining the unanswered questions of the target subject, and extracting the initial representation of the target knowledge point of the unanswered questions; based on the target subject knowledge graph, encoding the initial representation of the target knowledge point, determining the target knowledge point representation vector corresponding to the unanswered questions, the target knowledge point representation vector is used to represent the knowledge point information and knowledge point relationship structure information of the unanswered questions; based on the target knowledge point representation vector and the student's historical answer information corresponding to the target subject, predicting the student's knowledge ability. Or executing the learning resource recommendation method provided in the above embodiments, the method comprises: determining the student's knowledge ability, the knowledge ability is predicted based on the student ability profiling method provided in the above embodiments; based on the knowledge ability, recommending learning resources matching the knowledge ability to the student.

[0191] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0192] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0193] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A student ability profiling method, characterized in that: include: Obtaining unanswered questions of a target subject, and extracting initial representations of target knowledge points of the unanswered questions; Based on the target subject knowledge graph, the initial representation of the target knowledge point is encoded to determine the target knowledge point representation vector corresponding to the question to be answered, and the target knowledge point representation vector is used to represent the knowledge point information and knowledge point relationship structure information of the question to be answered; Predicting the student's knowledge ability based on the target knowledge point representation vector and the student's historical answer information corresponding to the target subject; The historical question answering information includes a plurality of historical questions and the historical answering information corresponding to the plurality of historical questions; the plurality of historical questions are arranged according to the corresponding answering time to form a time series, and the historical question answering information also includes the answering time interval between any historical question and the previous adjacent question in the time series; The knowledge point representation vector of the current historical question corresponding to the current answering moment, the embedding vector of the corresponding historical answering information, and the embedding vector of the answering time interval are concatenated to obtain a concatenated feature; Determine the knowledge state of the student at the current answering moment based on the splicing feature and the knowledge state of the student at the previous answering moment; Determine the answer information prediction result corresponding to the subsequent adjacent question based on the student's knowledge state at the current answering moment and the knowledge point representation vector of the subsequent adjacent question of the current history question; Based on the historical answer information and answer information prediction result corresponding to the subsequent adjacent question, the target knowledge point representation vector is parsed to obtain the answer information of the to-be-answered question; The previous answering time is the answering time corresponding to the previous adjacent question of the current historical question.

2. The student ability profiling method according to claim 1 is characterized in that: The encoding of the target knowledge point initial representation based on the target subject knowledge graph and determining the target knowledge point representation vector corresponding to the question to be answered includes: Based on the initial representation of the knowledge points corresponding to each graph node in the target subject knowledge graph, the initial representation of the target knowledge point is encoded to determine the target knowledge point representation vector corresponding to the question to be answered.

3. The student ability profiling method according to claim 2 is characterized in that: The encoding of the target knowledge point initial representation based on the knowledge point initial representation corresponding to each graph node in the target subject knowledge graph to determine the target knowledge point representation vector corresponding to the question to be answered includes: Based on the initial representation of the knowledge point corresponding to each graph node in the target subject knowledge graph, multi-layer cascaded graph convolution is performed on the initial representation of the target knowledge point to obtain features obtained by each layer of graph convolution; The initial representation of the target knowledge point and the features obtained by the graph convolution of each layer are fused to obtain the representation vector of the target knowledge point.

4. The student ability profiling method according to claim 1, characterized in that: The step of extracting an initial representation of a target knowledge point of the question to be answered includes: Based on the sample of questions of the target subject carrying the knowledge point representation label, feature extraction is performed on the questions to be answered to obtain the initial representation of the target knowledge point.

5. The student ability profiling method according to claim 4 is characterized in that: The step of extracting features of the to-be-answered questions based on the question samples of the target subject carrying the knowledge point representation tags to obtain the initial representation of the target knowledge point includes: Based on the sample of the question of the target subject carrying the knowledge point representation label, the phrase-level masking mechanism of the target subject is adopted to extract the features of the question to be answered, so as to obtain the initial representation of the target knowledge point.

6. A learning resource recommendation method, characterized in that: include: Determining the student's knowledge ability, wherein the knowledge ability is predicted based on the student ability profiling method according to any one of claims 1 to 5; Based on the knowledge and ability, recommend learning resources matching the knowledge and ability to the student.

7. A student ability profiling device, characterized in that: include: An acquisition module, used to acquire the unanswered questions of the target subject and extract the initial representation of the target knowledge points of the unanswered questions; An encoding module, for encoding the initial representation of the target knowledge point based on the target subject knowledge graph, and determining the target knowledge point representation vector corresponding to the question to be answered, wherein the target knowledge point representation vector is used to represent the knowledge point information and knowledge point relationship structure information of the question to be answered; A prediction module, used to predict the student's knowledge ability based on the target knowledge point representation vector and the student's historical answer information corresponding to the target subject; The historical question answering information includes a plurality of historical questions and the historical answering information corresponding to the plurality of historical questions; the plurality of historical questions are arranged according to the corresponding answering time to form a time series, and the historical question answering information also includes the answering time interval between any historical question and the previous adjacent question in the time series; The knowledge point representation vector of the current historical question corresponding to the current answering moment, the embedding vector of the corresponding historical answering information, and the embedding vector of the answering time interval are concatenated to obtain a concatenated feature; Determine the knowledge state of the student at the current answering moment based on the splicing feature and the knowledge state of the student at the previous answering moment; Determine the answer information prediction result corresponding to the subsequent adjacent question based on the student's knowledge state at the current answering moment and the knowledge point representation vector of the subsequent adjacent question of the current history question; Based on the historical answer information and answer information prediction result corresponding to the subsequent adjacent question, the target knowledge point representation vector is parsed to obtain the answer information of the to-be-answered question; The previous answering time is the answering time corresponding to the previous adjacent question of the current historical question.

8. A learning resource recommendation device, characterized in that: include: An ability determination module, used to determine the knowledge ability of the student, wherein the knowledge ability is predicted based on the student ability profiling method according to any one of claims 1 to 5; A recommendation module is used to recommend learning resources matching the knowledge and ability to the student based on the knowledge and ability.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the student ability profiling method as described in any one of claims 1 to 5, or the steps of the learning resource recommendation method as described in claim 6 are implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the student ability profiling method as described in any one of claims 1 to 5, or the steps of the learning resource recommendation method as described in claim 6 are implemented.

Images