A data processing method and device, computer equipment and a storage medium

Abstract

Embodiments of the present application disclose a data processing method and device, computer equipment and a storage medium, wherein the method comprises the following steps: obtaining a question answered by a target user and a target question not answered; generating a question sequence, a knowledge point sequence and an answer record sequence of the target user according to the answered question and the target question; performing feature extraction on the question sequence, the knowledge point sequence and the answer record sequence to generate a question vector corresponding to the question sequence, a knowledge point vector corresponding to the knowledge point sequence and an answer record vector corresponding to the answer record sequence; performing graph feature extraction on the question vector and the knowledge point vector to generate a corresponding graph vector; and generating a correct probability of the target user answering the target question based on the graph vector, the question vector, the knowledge point vector and the answer record vector through a question answering prediction model. The present application can accurately grasp the degree of user's mastery of knowledge points and improve the accuracy of predicting user's answers.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet, and particularly relates to a data processing method and device, computer equipment and storage medium. BACKGROUND

[0002] In a knowledge tracking task, a learning record of a student includes a question sequence and a knowledge point sequence. Since there is a data scarcity problem in the question sequence, a traditional knowledge tracking method usually uses the knowledge point sequence to model the learning record of the student. However, there is a difference between questions with the same knowledge point, and using the knowledge point sequence for modeling affects the modeling accuracy. Meanwhile, some methods attempt to model by integrating the question sequence into the knowledge point sequence, such as establishing a global question-knowledge point network graph and integrating question difficulty. However, the above methods still cannot well utilize the question sequence, cannot accurately predict the accuracy of the user's answering, and cannot accurately grasp the user's mastery of the knowledge point. SUMMARY

[0003] The embodiments of the present application provide a data processing method and device, computer equipment and storage medium, which can accurately grasp the user's mastery of the knowledge point and improve the accuracy of predicting the user's answering.

[0004] The embodiments of the present application provide a data processing method, which can include:

[0005] Obtaining a question answered by a target user and a target question not answered by the target user; the question answered carries knowledge point information and answering record information, and the target question carries knowledge point information;

[0006] Generating a question sequence, a knowledge point sequence and an answering record sequence of the target user according to the question answered and the target question;

[0007] Performing feature extraction on the question sequence, the knowledge point sequence and the answering record sequence to generate a question vector corresponding to the question sequence, a knowledge point vector corresponding to the knowledge point sequence and an answering record vector corresponding to the answering record sequence;

[0008] Performing graph feature extraction on the question vector and the knowledge point vector to generate a corresponding graph vector; the graph vector is used to represent the relationship between the question vector and the knowledge point vector;

[0009] Generating a correct probability of the target user answering the target question through a question answering prediction model based on the graph vector, the question vector, the knowledge point vector and the answering record vector.

[0010] In a feasible implementation, the method further includes:

[0011] Obtain the initial question-answering prediction model, using the questions already answered by the target user as training sample data; the training sample data includes the answer record for each question.

[0012] Based on the training sample data and the initial question answer prediction model, the correct probability corresponding to the training sample data is generated, and the model parameters of the initial question answer prediction model are adjusted according to the correct probability corresponding to the training sample data and the answer records of the training sample data.

[0013] When the adjusted initial question-answering prediction model meets the convergence condition, the initial question-answering prediction model containing the adjusted model parameters is determined as the question-answering prediction model.

[0014] In one feasible implementation, a question sequence, a knowledge point sequence, and an answer record sequence for the target user are generated based on the answered questions and the target question, including:

[0015] Arrange the answered questions and target questions to generate a question sequence, and generate a knowledge point sequence based on the knowledge point information corresponding to each question in the question sequence;

[0016] Generate an answer record sequence based on the answer record information corresponding to each question in the question sequence.

[0017] In one feasible implementation, graph feature extraction is performed on the question vector and knowledge point vector to generate a graph vector, including:

[0018] Based on the multi-head self-attention mechanism, a first relation vector between question vectors and a second relation vector between knowledge point vectors are obtained; the first relation vector is used to represent the relationship between question vectors, and the second relation vector is used to represent the relationship between knowledge point vectors.

[0019] Based on the cross-gating mechanism, a comprehensive relationship vector between the question vector and the knowledge point vector is generated based on the first relationship vector and the second relationship vector. The comprehensive relationship vector is then fused to generate a graph vector.

[0020] In one feasible implementation, based on a multi-head self-attention mechanism, a first relation vector between question vectors and a second relation vector between knowledge point vectors are obtained, including:

[0021] Obtain the first transformation matrix between the problem vectors, and generate the first relation vector corresponding to the problem vectors based on the first transformation matrix;

[0022] Obtain the second transformation matrix between knowledge point vectors, and generate the second relationship vector corresponding to the knowledge point vectors based on the second transformation matrix.

[0023] In a feasible implementation, according to a cross-gating mechanism, a comprehensive relation vector between the question vector and the knowledge point vector is generated based on the first relation vector and the second relation vector, and the comprehensive relation vector is fused to generate a graph vector, including:

[0024] The first weight coefficient and the second weight coefficient between the first relation vector and the second relation vector are obtained through an activation function;

[0025] The first comprehensive relation vector is obtained according to the first relation vector, the second relation vector and the first weight coefficient; the first comprehensive relation vector represents the question vector fused with the knowledge point vector;

[0026] The second comprehensive relation vector is obtained according to the first relation vector, the second relation vector and the second weight coefficient; the second comprehensive relation vector represents the knowledge point vector fused with the question vector;

[0027] The first comprehensive relation vector and the second comprehensive relation vector are spliced to generate the graph vector.

[0028] In a feasible implementation, the correct probability of the target user answering the target question is generated based on the graph vector, the question vector, the knowledge point vector and the answer record vector through a question answering prediction model, including:

[0029] The knowledge point vector and the graph vector are spliced to generate a first input vector;

[0030] The knowledge point vector and the answer record vector are spliced to generate a second input vector;

[0031] The first input vector is processed by an attenuation function in the question answering prediction model, and a learning state vector of the target user is generated according to the second input vector and the first input vector after the attenuation processing;

[0032] According to the learning state vector and an output result of a feedforward neural network layer in the question answering prediction model, the output result is taken as the correct probability of the target user answering the target question.

[0033] Embodiments of the present application provide a data processing apparatus, which can include:

[0034] A question acquisition unit is configured to acquire a question answered by a target user and a target question not answered by the target user; the question answered by the target user carries knowledge point information and answer record information, and the target question carries the knowledge point information;

[0035] A sequence generation unit is configured to generate a question sequence, a knowledge point sequence and an answer record sequence of the target user according to the question answered by the target user and the target question;

[0036] The feature extraction unit is configured to perform feature extraction on the question sequence, the knowledge point sequence, and the answer record sequence to generate a question vector corresponding to the question sequence, a knowledge point vector corresponding to the knowledge point sequence, and an answer record vector corresponding to the answer record sequence.

[0037] The graph vector generation unit is configured to perform graph feature extraction on the question vector and the knowledge point vector to generate a corresponding graph vector, the graph vector being used to represent a relationship between the question vector and the knowledge point vector.

[0038] The probability prediction unit is configured to generate, by using a question answer prediction model, a correct probability of an answer to a target question by a target user based on the graph vector, the question vector, the knowledge point vector, and the answer record vector.

[0039] In an embodiment, the method further includes:

[0040] The model training unit is configured to obtain an initial question answer prediction model, and take a question that has been answered by the target user as training sample data, the training sample data including an answer record of each question.

[0041] Based on the training sample data and the initial question answer prediction model, a correct probability corresponding to the training sample data is generated, and a model parameter of the initial question answer prediction model is adjusted according to the correct probability corresponding to the training sample data and the answer record of the training sample data.

[0042] When the adjusted initial question answer prediction model meets a convergence condition, the initial question answer prediction model including the adjusted model parameter is determined as the question answer prediction model.

[0043] In an embodiment, the sequence generation unit is specifically configured to:

[0044] arrange the question that has been answered and the target question to generate a question sequence, and generate a knowledge point sequence according to knowledge point information corresponding to each question in the question sequence.

[0045] generate an answer record sequence according to answer record information corresponding to each question in the question sequence.

[0046] In an embodiment, the graph vector generation unit includes:

[0047] The relationship vector generation subunit is configured to obtain, according to a multi-head self-attention mechanism, a first relationship vector between the question vectors and a second relationship vector between the knowledge point vectors, the first relationship vector being used to represent a connection between the question vectors, and the second relationship vector being used to represent a connection between the knowledge point vectors.

[0048] The graph vector generation subunit is configured to generate a comprehensive relation vector between the question vector and the knowledge point vector based on the first relation vector and the second relation vector according to a cross-gating mechanism, and fuse the comprehensive relation vector to generate a graph vector.

[0049] In an implementation, the relation vector generation subunit is specifically configured to:

[0050] obtain a first conversion matrix between the question vectors, and generate the first relation vector corresponding to the question vectors according to the first conversion matrix;

[0051] obtain a second conversion matrix between the knowledge point vectors, and generate the second relation vector corresponding to the knowledge point vectors according to the second conversion matrix.

[0052] In an implementation, the graph vector generation subunit is specifically configured to:

[0053] obtain a first weight coefficient and a second weight coefficient between the first relation vector and the second relation vector through an activation function;

[0054] obtain a first comprehensive relation vector according to the first relation vector, the second relation vector, and the first weight coefficient; the first comprehensive relation vector represents the question vector fused with the knowledge point vector;

[0055] obtain a second comprehensive relation vector according to the first relation vector, the second relation vector, and the second weight coefficient; the second comprehensive relation vector represents the knowledge point vector fused with the question vector;

[0056] perform vector splicing on the first comprehensive relation vector and the second comprehensive relation vector to generate the graph vector.

[0057] In an implementation, the probability prediction unit is specifically configured to:

[0058] perform splicing on the knowledge point vector and the graph vector to generate a first input vector;

[0059] perform splicing on the knowledge point vector and the answer record vector to generate a second input vector;

[0060] perform attenuation processing on the first input vector through an attenuation function in the question answering prediction model, and generate a learning state vector of the target user according to the second input vector and the first input vector after the attenuation processing;

[0061] output a corresponding output result according to the learning state vector and a feedforward neural network layer in the question answering prediction model, and take the output result as a correct probability of the target user answering the target question.

[0062] The embodiment of the present application provides a computer readable storage medium, and the computer readable storage medium stores a computer program.

[0063] The embodiment of the present application provides a computer device, which comprises a processor, a memory and a network interface; the processor is connected with the memory and the network interface; the network interface is used for providing network communication function; the memory is used for storing program code; and the processor is used for calling the program code to execute the method steps.

[0064] The embodiment of the present application provides a computer program product or a computer program, which comprises computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method steps.

[0065] In the embodiment of the present application, by acquiring the questions answered by the target user and the target question not answered, the question sequence, the knowledge point sequence and the answer record sequence of the target user are further generated according to the answered questions and the target question, the question sequence, the knowledge point sequence and the answer record sequence are subjected to feature extraction, the question vector corresponding to the question sequence, the knowledge point vector corresponding to the knowledge point sequence and the answer record vector corresponding to the answer record sequence are generated, then the graph feature extraction is performed on the question vector and the knowledge point vector to generate the corresponding graph vector, the graph vector is used to represent the relationship between the question vector and the knowledge point vector, finally, the correct probability of the target user answering the target question is generated based on the graph vector, the question vector, the knowledge point vector and the answer record vector through the question answering prediction model. By using the above method, the question sequence and the knowledge point sequence in the historical learning record of the student can be fused, the problem that the question sequence cannot be effectively utilized is avoided, the mastering degree of the user to the knowledge point can be accurately mastered, and the accuracy of predicting the user to answer the question is improved. BRIEF DESCRIPTION OF DRAWINGS

[0066] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0067] Figure 1 It is a system architecture diagram for data processing provided by the embodiment of the present application.

[0068] Figure 2is a flowchart of a data processing method provided by an embodiment of the present application;

[0069] Figure 3 is a flowchart of a data processing method provided by an embodiment of the present application;

[0070] Figure 4a is an example schematic diagram of generating a graph vector provided by an embodiment of the present application;

[0071] Figure 4b is an example schematic diagram of a data processing method provided by an embodiment of the present application;

[0072] Figure 5 is a structural schematic diagram of a data processing apparatus provided by an embodiment of the present application;

[0073] Figure 6 is a structural schematic diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0074] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0075] As shown in Figure 1 The network architecture diagram can include a service server 100 and a user terminal cluster, and the user terminal cluster can include a user terminal 10a, a user terminal 10b, …, and a user terminal 10c. There can be a communication connection between the user terminals in the user terminal cluster, for example, there is a communication connection between the user terminal 10a and the user terminal 10b, and there is a communication connection between the user terminal 10b and the user terminal 10c. Any user terminal in the user terminal cluster can have a communication connection with the service server 100, for example, there is a communication connection between the user terminal 10a and the service server 100, and there is a communication connection between the user terminal 10b and the service server 100.

[0076] The aforementioned user terminal cluster (including user terminal 10a, user terminal 10b, and user terminal 10c) can all be integrated with the target application. Optionally, the target application may include an application capable of displaying data information such as text, images, and videos. Database 10d stores a question bank for the target user, which includes answered questions and unanswered target questions. Answered questions carry knowledge point information and answer record information, while target questions carry knowledge point information. In one feasible implementation, the user terminal obtains the questions already answered and the target questions not yet answered by the target user. It then generates a question sequence, a knowledge point sequence, and an answer record sequence based on the answered and target questions. Feature extraction is performed on these sequences to generate question vectors corresponding to the question sequence, knowledge point vectors corresponding to the knowledge point sequence, and answer record vectors corresponding to the answer record sequence. The user terminal then performs graph feature extraction on the question vectors and knowledge point vectors to generate corresponding graph vectors. These graph vectors represent the relationship between the question vectors and knowledge point vectors. Finally, a question-answering prediction model is used to generate the probability that the target user answered the target questions correctly, based on the graph vectors, question vectors, knowledge point vectors, and answer record vectors.

[0077] Optionally, the aforementioned user terminal can be one of the above-mentioned... Figure 1 Any user terminal selected in the user terminal cluster of the corresponding embodiment, for example, the user terminal can be the aforementioned user terminal 10b. It is understood that the method provided in this application embodiment can be executed by a computer device, which includes, but is not limited to, a terminal or a server. The business server 100 in this application embodiment can be a computer device, and the user terminals in the user terminal cluster can also be computer devices; this is not limited here. The aforementioned business server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The aforementioned terminals can include: smartphones, tablets, laptops, desktop computers, smart TVs, smart speakers, desktop computers, smartwatches, and other smart terminals with image recognition functions, but are not limited thereto. The user terminals and the business server can be directly or indirectly connected via wired or wireless communication; this application does not impose any restrictions on this connection.

[0078] Furthermore, for ease of understanding, please refer to Figure 2 , Figure 2 This is a flowchart illustrating the data processing method provided in an embodiment of this application. The method can be implemented by a user terminal (e.g., the one described above). Figure 1The operation can be performed by the user terminal shown, or by the user terminal and the service server (as described above). Figure 1 The service server 100 in the corresponding embodiment performs the operation together. For ease of understanding, this embodiment uses the method executed by the aforementioned user terminal as an example for explanation. The data processing method may include at least the following steps S101-S104:

[0079] S101, Obtain the questions that the target user has answered and the target questions that have not been answered;

[0080] Specifically, the user terminal can obtain the questions that the target user has answered and the target questions that have not yet been answered. This means that the answered and unanswered questions can be stored in a corresponding question bank, which can be created by the target user or from a question bank provided by the learning software. The questions in this question bank can be uploaded to the network or stored on the user terminal. Answered questions carry knowledge point information and answer record information. The knowledge point information indicates the knowledge type of the question, such as addition, subtraction, and square root operations. The answer record information indicates whether the answer was correct or incorrect. Target questions carry knowledge point information, but since the target user has not yet answered them, there is no answer record information for target questions. The following example illustrates the questions, their corresponding knowledge points, and answer records. For instance, the answered question 1 could be "1+2=?", the answered question 2 could be "2*3=?", and the target question could be "4-3=?". The knowledge point for question 1 is addition, the knowledge point for question 2 is multiplication, and the knowledge point for the target question is subtraction. If the target user answers question 1 with 3, then the answer record for question 1 is correct. If the target user answers question 2 with 8, then the answer record for question 2 is incorrect. There is no answer record for question 3.

[0081] S102, Generate a question sequence, knowledge point sequence, and answer record sequence for the target user based on the questions already answered and the target question;

[0082] Specifically, the user terminal generates a question sequence, a knowledge point sequence, and an answer record sequence based on the answered questions and the target question. It's understood that the number of answered questions is usually large. The method in this solution is used to predict the target user's accuracy rate in answering the target question. Therefore, taking a single target question as an example, the user terminal arranges the answered questions and the target question to generate a question sequence. Typically, the answered questions are randomly arranged, or arranged according to the target user's settings, with the target question placed last in the sequence. Further, a knowledge point sequence is generated based on the knowledge point information corresponding to each question in the question sequence, and an answer record sequence is generated based on the answer record information corresponding to the answered questions in the question sequence.

[0083] S103, feature extraction is performed on the question sequence, the knowledge point sequence, and the answer record sequence to generate a question vector corresponding to the question sequence, a knowledge point vector corresponding to the knowledge point sequence, and an answer record vector corresponding to the answer record sequence;

[0084] Specifically, the user terminal extracts features of the question sequence, the knowledge point sequence, and the answer record sequence through an embedding layer in the question answering prediction model to generate a question vector corresponding to the question sequence, a knowledge point vector corresponding to the knowledge point sequence, and an answer record vector corresponding to the answer record sequence. The question answering prediction model is used to predict a correct probability of a target user answering a target question.

[0085] S104, graph feature extraction is performed on the question vector and the knowledge point vector to generate a corresponding graph vector; the graph vector is used to represent a relationship between the question vector and the knowledge point vector;

[0086] Specifically, the user terminal obtains a first conversion matrix between the question vectors, generates a first relationship vector corresponding to the question vectors according to the first conversion matrix, obtains a second conversion matrix between the knowledge point vectors, and generates a second relationship vector corresponding to the knowledge point vectors according to the second conversion matrix.

[0087] Further, the first weight coefficient and the second weight coefficient between the first relationship vector and the second relationship vector are obtained through an activation function, the activation function can be Sigmoid, the first comprehensive relationship vector is obtained according to the first relationship vector, the second relationship vector, and the first weight coefficient, the first comprehensive relationship vector represents the question vector fused with the knowledge point vector, the second comprehensive relationship vector is obtained according to the first relationship vector, the second relationship vector, and the second weight coefficient, the second comprehensive relationship vector represents the knowledge point vector fused with the question vector, and the first comprehensive relationship vector and the second comprehensive relationship vector are spliced to generate the graph vector.

[0088] S105, a correct probability of a target user answering a target question is generated based on the graph vector, the question vector, the knowledge point vector, and the answer record vector through the question answering prediction model.

[0089] Specifically, the user terminal splices the knowledge point vector and the graph vector to generate a first input vector, and splices the knowledge point vector and the answer record vector to generate a second input vector. Further, the first input vector is attenuated by an attenuation function in the question answering prediction model. In a real situation, the target user will gradually forget what has been learned before, so in order to ensure the accuracy of the result prediction, the attenuation function is used to simulate this process. Further, according to the second input vector and the attenuated first input vector, a learning state vector of the target user is generated. The learning state vector is used to represent the current learning state of the target user. Specifically, the learning state vector of the target user can be generated based on a multi-head self-attention mechanism of attention weight calculation based on dot product operation. The input of the attention mechanism includes Query, Key, and Value. The first input vector is used as Query and Key for fusion and attenuation processing, and then the learning state vector is generated through the attention mechanism and value.

[0090] Further, according to the learning state vector and the output of the feedforward neural network layer in the question answering prediction model, a corresponding output result is output. The output result is used as the correct probability of the target user answering the target question. It should be noted that the output result is a number between 0 and 1. The data result can be converted to 0 or 1 by a threshold value. For example, if the output result is greater than 0.5, the output result is converted to 1, and the target user will answer the target question correctly. If the output result is less than or equal to 0.5, the output result is converted to 0, and the target user will answer the target question incorrectly.

[0091] In the embodiments of the present application, the questions answered by the target user and the target questions not answered by the target user are obtained. Further, the question sequence, the knowledge point sequence and the answer record sequence of the target user are generated according to the answered questions and the target questions. The question sequence, the knowledge point sequence and the answer record sequence are feature extracted to generate the question vector corresponding to the question sequence, the knowledge point vector corresponding to the knowledge point sequence and the answer record vector corresponding to the answer record sequence. Then, the graph feature of the question vector and the knowledge point vector is extracted to generate the corresponding graph vector. The graph vector is used to represent the relationship between the question vector and the knowledge point vector. Finally, the correct probability of the target user answering the target question is generated based on the graph vector, the question vector, the knowledge point vector and the answer record vector by the question answering prediction model. By using the above method, the question sequence and the knowledge point sequence in the historical learning records of the student can be fused, the problem of not being able to effectively utilize the question sequence can be avoided, the user's mastery of the knowledge point can be accurately mastered, and the accuracy of predicting the user's answer can be improved.

[0092] Please refer to Figure 3 , Figure 3 is a flowchart of a data processing method provided by the embodiments of the present application. The method can be performed by a user terminal (for example, the user terminal 100 described above).Figure 1 The data processing method can be executed by the user terminal or the service server (e.g., the service server 100 in the embodiment shown in FIG. 1) alone or jointly by the user terminal and the service server (e.g., the service server 100 in the embodiment shown in FIG. 1). Figure 1 For ease of understanding, the embodiment is described by taking the method executed by the user terminal as an example. The data processing method can at least include the following steps S201-S207.

[0093] S201, obtaining an initial problem answer prediction model and training sample data, and generating a problem answer prediction model according to the training sample data and the initial problem answer prediction model.

[0094] Specifically, the user terminal obtains an initial problem answer prediction model, takes the questions answered by the target user as training sample data, and the training sample data includes each question answer record. Further, based on the training sample data and the initial problem answer prediction model, the correct probability corresponding to the training sample data is generated. Specifically, a question sequence, a knowledge point sequence and an answer record sequence are generated according to the training sample data, the question sequence, the knowledge point sequence and the answer record sequence are feature extracted to generate a question vector corresponding to the question sequence, a knowledge point vector corresponding to the knowledge point sequence and an answer record vector corresponding to the answer record sequence, the question vector and the knowledge point vector are graph feature extracted to generate a corresponding graph vector, the correct probability of the training sample data is generated based on the graph vector, the question vector, the knowledge point vector and the answer record vector through the initial problem answer prediction model, the model parameters of the initial problem answer prediction model are adjusted according to the correct probability corresponding to the training sample data and the answer record of the training sample data, and when the adjusted initial problem answer prediction model meets the convergence condition, the initial problem answer prediction model containing the adjusted model parameters is determined as the problem answer prediction model.

[0095] S202, obtaining the questions answered by the target user and the target questions not answered; the answered questions carry knowledge point information and answer record information, and the target questions carry knowledge point information.

[0096] The step S202 of the embodiment of the present application is described in detail with reference to the step S101 of the embodiment shown in FIG. 1, which is not described herein. Figure 2 The step S101 of the embodiment shown in FIG. 1 is described in detail, which is not described herein.

[0097] S203, generating a question sequence, a knowledge point sequence and an answer record sequence of the target user according to the answered questions and the target questions.

[0098] The step S203 of the embodiment of the present application is described in detail with reference to the step S102 of the embodiment shown in FIG. 1, which is not described herein. Figure 2 The step S102 of the embodiment shown in FIG. 1 is described in detail, which is not described herein.

[0099] S204, feature extraction is performed on the question sequence, the knowledge point sequence and the answer record sequence to generate a question vector corresponding to the question sequence, a knowledge point vector corresponding to the knowledge point sequence and an answer record vector corresponding to the answer record sequence;

[0100] In the embodiment of the present application, step S204 is described in detail in the embodiment shown in Figure 2 The specific description of step S103 of the embodiment shown in the embodiment is not described here.

[0101] S205, according to the multi-head self-attention mechanism, a first relationship vector between the question vectors and a second relationship vector between the knowledge point vectors are obtained; the first relationship vector is used to represent the relationship between the question vectors, and the second relationship vector is used to represent the relationship between the knowledge point vectors;

[0102] Specifically, the user terminal can construct a graph structure between the question vector and the knowledge point vector, the question vector and the knowledge point vector are nodes in the graph structure, the relationship between the nodes is the edge between the nodes, and the information of the edge between the nodes can be obtained, that is, all node relationships in the graph structure. Please refer to Figure 4a , Figure 4a is an example of generating a graph vector provided by the embodiment of the present application. The nodes in the figure include two types, namely question vector nodes and knowledge point vector nodes, the question vector nodes correspond to the question vectors, and the knowledge point vector nodes correspond to the knowledge point vectors, the figure includes 5 question vector nodes and 5 knowledge point vector nodes, and the edges between the nodes have three types, namely the question vector nodes and the question vector nodes (the dashed line between the question nodes), the knowledge point vector nodes and the knowledge point vector nodes (the dashed line between the knowledge point nodes), and the question vector nodes and the knowledge point vector nodes (the solid line between the question node vector and the knowledge point node vector). The edges between the question vector nodes and the question vector nodes, the knowledge point vector nodes and the knowledge point vector nodes will be described below, specifically, a first conversion matrix between the question vectors can be obtained, a first relationship vector corresponding to the question vector is generated according to the first conversion matrix, a second conversion matrix between the knowledge point vectors is obtained, and a second relationship vector corresponding to the knowledge point vector is generated according to the second conversion matrix. The first relationship vector is the information of the edge between the question vector nodes and the question vector nodes, and the second relationship vector is the information of the edge between the knowledge point vector nodes and the knowledge point vector nodes, specifically, the first relationship vector and the second relationship vector can be obtained by using a multi-head self-attention mechanism function MultiHead, and the acquisition formula is as follows:

[0103] H x =MultiHead(W q E x ,W k E x ,W v Ex )

[0104] H c = MultiHead(W q E c ,W k E c ,W v E c )

[0105] wherein H x is the first relationship vector, H c is the first relationship vector, W q , W k and W v are weight coefficients, E x is the question vector, and E c is the knowledge point vector.

[0106] The edges between the question vector node and the knowledge point vector node are obtained as follows, please refer to step S206.

[0107] S206, according to the cross-gating mechanism, based on the first relationship vector and the second relationship vector, a comprehensive relationship vector between the question vector and the knowledge point vector is generated, and the comprehensive relationship vector is fused to generate a graph vector.

[0108] Specifically, the user terminal can obtain the first weight coefficient and the second weight coefficient between the first relationship vector and the second relationship vector through an activation function, and the activation function can be Sigmoid, the first weight coefficient A x and the second weight coefficient A c The formula for obtaining is as follows:

[0109] A x = Sigmoid(W x1 H x +W x2 H c )

[0110] A c = Sigmoid(W c1 H c +W c2 H x )

[0111] wherein W x1 , W x2 , W c1 and W c2 are weight coefficients, H x is the first relationship vector, and H c is the first relationship vector.

[0112] Furthermore, through a cross-gating mechanism, a first comprehensive relation vector is obtained based on the first relation vector, the second relation vector, and the first weight coefficient. This first comprehensive relation vector represents the question vector integrated with the knowledge point vector. Through the cross-gating mechanism, a second comprehensive relation vector is obtained based on the first relation vector, the second relation vector, and the second weight coefficient. This second comprehensive relation vector represents the knowledge point vector integrated with the question vector. The first comprehensive relation vector M... x Second comprehensive relation vector M c The formula is as follows:

[0113] M x =H x +A x H c

[0114] M c =H x +A c H x

[0115] Among them, A x As the first weighting coefficient, A c H is the second weighting coefficient. x H is the first relation vector. c This is the first relation vector.

[0116] Furthermore, the first and second comprehensive relation vectors are input into the feedforward neural network layer for vector concatenation to generate a graph vector. The graph vector represents the relationship between problem vector nodes and knowledge point vector nodes in the graph structure.

[0117] S207 uses a question-and-answer prediction model to generate the probability of a target user answering a target question correctly, based on graph vectors, question vectors, knowledge point vectors, and answer record vectors.

[0118] In this embodiment of the invention, step S207 is described below. Figure 2 The specific description of step S105 in the illustrated embodiment will not be repeated here.

[0119] Please see Figure 4b , Figure 4b This is an example schematic diagram of a data processing method provided in an embodiment of this application. For example... Figure 4bAs shown, the figure is a knowledge tracking model framework, which can be a question answering prediction model. The model framework includes a learning model and a prediction model. The learning model includes four sub-modules, namely a question and embedding layer module, a graph module, a fusion module, and a knowledge retrieval module based on a time decay attention mechanism. The prediction model is composed of a multi-layer feedforward neural network and a sigmoid activation function. First, the target user's answered questions and unanswered target questions in the question bank are obtained. The question sequence X, knowledge point sequence C, and answer record sequence Y of the target user are generated through the embedding module. The module is further converted into corresponding question vectors E x , knowledge point vectors E c , and answer record vectors E y . Then, the question vector and the knowledge point vector are sent to the graph module to encode the learning state of the target user. For each user, a graph structure G is constructed according to his learning record. The graph structure G is composed of two parts, nodes V and edges E, wherein the edges include edges between nodes of the same type (question-question or knowledge point-knowledge point) and edges between nodes of different types (question-knowledge point). The same type of nodes are fused through a multi-head self-attention mechanism, and the nodes of different types are fused through a cross-gate mechanism, and finally a graph vector is generated. Further, the graph vector is sent to the knowledge retrieval module based on the time decay attention mechanism through the integration function in the fusion module. In real conditions, students will gradually forget what they have learned before, so in order to ensure the accuracy of the result prediction, the decay function is used for decay processing to simulate this process. The multi-head self-attention mechanism calculates the attention weight based on the dot product operation, takes the concatenated vector of the knowledge point vector and the graph vector as Query and Key, and performs fusion decay processing. Then, the learning state vector is generated through the attention mechanism and the value. Finally, the prediction model is used to predict the probability of the student answering the current question correctly.

[0120] In the embodiment of the present application, by obtaining the answered questions of the target user and the target question not answered, the question sequence, the knowledge point sequence and the answer record sequence of the target user are further generated according to the answered questions and the target question, the question sequence, the knowledge point sequence and the answer record sequence are subjected to feature extraction, the question vector corresponding to the question sequence, the knowledge point vector corresponding to the knowledge point sequence and the answer record vector corresponding to the answer record sequence are generated, then the graph feature extraction is performed on the question vector and the knowledge point vector to generate the corresponding graph vector, the graph vector is used to represent the relationship between the question vector and the knowledge point vector, finally, the correct probability of the target user answering the target question is generated by the question answering prediction model based on the graph vector, the question vector, the knowledge point vector and the answer record vector. By using the above method, the question sequence and the knowledge point sequence in the historical learning record of the student can be fused, the problem that the question sequence cannot be effectively utilized is avoided, the mastering degree of the user to the knowledge point can be accurately mastered, and the accuracy of predicting the user's answer is improved.

[0121] It should be noted that the above scheme can be used for adaptive education problems of multiple disciplines. Knowledge tracking is the key to building an adaptive education system. In multiple scenarios of the adaptive education system, the mastering degree of the students to the knowledge points needs to be accurately predicted. For example, accurate question pushing, student learning path planning, knowledge graph construction and other tasks. Specifically, the historical answered questions of the student and the target question not answered for each knowledge point are obtained, the correct probability of the target question corresponding to each knowledge point of the student is obtained by using the method in the above scheme, the mastering degree of the student to the knowledge point is predicted according to the correct probability, further, the knowledge graph of the student can be constructed according to the mastering degree of each knowledge point, and the mastering degree of each knowledge point is marked in the knowledge graph. Further, the learning path planning of the student can be made according to the knowledge graph, for example, more time is arranged for learning the knowledge point with a lower mastering degree, and the student can also be accurately pushed questions according to the learning path planning, for example, more related questions are pushed for the knowledge point with a lower mastering degree for targeted training.

[0122] Please refer to Figure 5 , Figure 5 is a structural schematic diagram of a data processing apparatus provided by an embodiment of the present application. The data processing apparatus can be a computer program (including program code) running in a computer device, for example, the data processing apparatus is an application software; the apparatus can be used to execute the corresponding steps in the method provided by the embodiment of the present application. As shown in Figure 5 , the data processing apparatus 1 of the embodiment of the present application can include: a question obtaining unit 11, a sequence generating unit 12, a feature extraction unit 13, a graph vector generating unit 14, and a probability predicting unit 15.

[0123] The problem obtaining unit 11 is configured to obtain answered questions of the target user and target questions not answered by the target user; the answered questions carry knowledge point information and answer record information, and the target questions carry knowledge point information;

[0124] The sequence generating unit 12 is configured to generate a question sequence, a knowledge point sequence and an answer record sequence of the target user according to the answered questions and the target questions;

[0125] The feature extracting unit 13 is configured to perform feature extraction on the question sequence, the knowledge point sequence and the answer record sequence to generate a question vector corresponding to the question sequence, a knowledge point vector corresponding to the knowledge point sequence and an answer record vector corresponding to the answer record sequence;

[0126] The graph vector generating unit 14 is configured to perform graph feature extraction on the question vector and the knowledge point vector to generate corresponding graph vectors; the graph vectors are used to represent the relationship between the question vector and the knowledge point vector.

[0127] The probability predicting unit 15 is configured to generate a correct probability of the target user answering the target questions by a question answering prediction model based on the graph vectors, the question vector, the knowledge point vector and the answer record vector.

[0128] Referring to Figure 5 The data processing apparatus 1 can further include a model training unit 16.

[0129] The model training unit 16 is configured to obtain an initial question answering prediction model and take the answered questions of the target user as training sample data; the training sample data includes an answer record of each question;

[0130] Based on the training sample data and the initial question answering prediction model, a correct probability corresponding to the training sample data is generated, and the model parameters of the initial question answering prediction model are adjusted according to the correct probability corresponding to the training sample data and the answer record of the training sample data.

[0131] When the adjusted initial question answering prediction model meets a convergence condition, the initial question answering prediction model containing the adjusted model parameters is determined as the question answering prediction model.

[0132] In a feasible implementation, the sequence generating unit 12 is specifically configured to:

[0133] arrange the answered questions and the target questions to generate a question sequence, and generate a knowledge point sequence according to the knowledge point information corresponding to each question in the question sequence;

[0134] generate an answer record sequence according to the answer record information corresponding to each question in the question sequence.

[0135] Referring to Figure 5The graph vector generation unit 14 in the embodiments of the present application can further include a relationship vector generation sub-unit 141 and a graph vector generation sub-unit 142.

[0136] The relationship vector generation sub-unit 141 is configured to obtain first relationship vectors between the question vectors and second relationship vectors between the knowledge point vectors according to a multi-head self-attention mechanism; the first relationship vectors are used to represent the relationship between the question vectors, and the second relationship vectors are used to represent the relationship between the knowledge point vectors.

[0137] The graph vector generation sub-unit 142 is configured to generate a comprehensive relationship vector between the question vectors and the knowledge point vectors based on the first relationship vectors and the second relationship vectors according to a cross-gated mechanism, and fuse the comprehensive relationship vector to generate a graph vector.

[0138] In a possible implementation, the relationship vector generation sub-unit 141 is specifically configured to:

[0139] obtain a first conversion matrix between the question vectors, and generate the first relationship vectors corresponding to the question vectors according to the first conversion matrix;

[0140] obtain a second conversion matrix between the knowledge point vectors, and generate the second relationship vectors corresponding to the knowledge point vectors according to the second conversion matrix.

[0141] In a possible implementation, the graph vector generation sub-unit 142 is specifically configured to:

[0142] obtain first and second weight coefficients between the first and second relationship vectors through an activation function;

[0143] obtain a first comprehensive relationship vector according to the first relationship vector, the second relationship vector and the first weight coefficient; the first comprehensive relationship vector represents the question vectors fused with the knowledge point vectors;

[0144] obtain a second comprehensive relationship vector according to the first relationship vector, the second relationship vector and the second weight coefficient; the second comprehensive relationship vector represents the knowledge point vectors fused with the question vectors;

[0145] perform vector splicing on the first and second comprehensive relationship vectors to generate a graph vector.

[0146] In a possible implementation, the probability prediction unit 15 is specifically configured to:

[0147] splicing the knowledge point vectors and the graph vector to generate a first input vector;

[0148] splicing the knowledge point vectors and the answer record vector to generate a second input vector;

[0149] The attenuation function in the question answering prediction model is used to perform attenuation processing on the first input vector, and a learning state vector of the target user is generated according to the second input vector and the first input vector after the attenuation processing.

[0150] According to the learning state vector and the output result corresponding to the output result of the feedforward neural network layer in the question answering prediction model, the output result is used as the correct probability of the target user answering the target question.

[0151] In the embodiment of the present application, by obtaining the questions answered by the target user and the target questions not answered, the question sequence, the knowledge point sequence and the answer record sequence of the target user are further generated according to the answered questions and the target questions, the feature extraction is performed on the question sequence, the knowledge point sequence and the answer record sequence, the question vector corresponding to the question sequence, the knowledge point vector corresponding to the knowledge point sequence and the answer record vector corresponding to the answer record sequence are generated, and then the graph feature extraction is performed on the question vector and the knowledge point vector to generate the corresponding graph vector, the graph vector is used to represent the relationship between the question vector and the knowledge point vector, and finally the correct probability of the target user answering the target question is generated by the question answering prediction model based on the graph vector, the question vector, the knowledge point vector and the answer record vector. By using the above method, the question sequence and the knowledge point sequence in the historical learning records of the student can be fused, the problem that the question sequence cannot be effectively utilized is avoided, the mastery degree of the user to the knowledge point can be accurately mastered, and the accuracy of predicting the answer of the user is improved.

[0152] Please refer to Figure 6 , Figure 6 is a structural schematic diagram of a computer device provided by the embodiment of the present application. As shown in Figure 6 , the computer device 1000 can include at least one processor 1001, such as a CPU, at least one network interface 1004, a user interface 1003, a memory 1005, and at least one communication bus 1002. The communication bus 1002 is used to realize the connection and communication between the components. The user interface 1003 can include a display screen (Display), and the optional user interface 1003 can also include a standard wired interface and a wireless interface. The network interface 1004 can optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 can be a random access memory (RAM), and can also be a non-volatile memory (NVM), such as at least one disk memory. The memory 1005 can also be at least one storage device located away from the aforementioned processor 1001. As shown in Figure 6 , the memory 1005, as a computer storage medium, can include an operating system, a network communication module, a user interface module and a data processing application program.

[0153] In Figure 6 In the computer device 1000 shown in the figure, the network interface 1004 can provide network communication functions, the user interface 1003 is mainly used to provide an interface for user input; and the processor 1001 can be used to invoke a data processing application program stored in the memory 1005 to implement the above Figures 2-4b The description of the data processing method in any of the corresponding embodiments will not be repeated here.

[0154] It should be understood that the computer device 1000 described in the embodiments of the present application can execute the foregoing Figures 2-4b The description of the data processing method in any of the corresponding embodiments will not be repeated here. Figure 5 The description of the data processing device in any of the corresponding embodiments will not be repeated here. In addition, the beneficial effects of using the same method will not be repeated.

[0155] In addition, it should be noted that the embodiments of the present application also provide a computer readable storage medium, and the computer readable storage medium stores the computer program executed by the data processing device mentioned above, and the computer program includes program instructions, when the processor executes the program instructions, the foregoing Figures 2-4b The description of the data processing method in any of the corresponding embodiments will not be repeated here. In addition, the beneficial effects of using the same method will not be repeated. For technical details of the computer readable storage medium embodiments involved in the present application, please refer to the description of the method embodiments of the present application. As an example, the program instructions can be deployed to execute on one computing device, or on multiple computing devices located in one place, or on multiple computing devices distributed in multiple places and interconnected through a communication network, and the multiple computing devices distributed in multiple places and interconnected through a communication network can constitute a blockchain system.

[0156] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The above-mentioned program can be stored in a computer readable storage medium. When the program is executed, the program can include the processes of the above-mentioned embodiment methods. The computer readable storage medium can be an internal storage unit of the data processing device provided in any of the above-mentioned embodiments or the above-mentioned device, for example, a hard disk or a memory of an electronic device. The computer readable storage medium can also be an external storage device of the electronic device, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, and the like. The computer readable storage medium can also include a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory, and the like. Further, the computer readable storage medium can include both the internal storage unit of the electronic device and the external storage device. The computer readable storage medium is used to store the computer program and other programs and data required by the electronic device. The computer readable storage medium can also be used to temporarily store data that has been output or will be output.

[0157] The terms "first", "second", and the like in the claims and specification and drawings of the present application are used to distinguish different objects, and are not used to describe a particular order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed or can optionally include other steps or units inherent to the process, method, product, or device. In this document, "embodiment" means that the specific features, structures, or characteristics described in connection with the embodiment can be included in at least one embodiment of the present application. The phrase is shown at various places in the specification does not necessarily refer to the same embodiment, nor is it independent or alternative to other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments. The term "and / or" used in the specification and claims of the present application means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0158] Those skilled in the art can realize the units and algorithm steps of each example described in connection with the embodiments disclosed herein can be realized in electronic hardware, computer software, or a combination of both. The above described embodiments are intended to cover the application and its generic equivalents, and the steps of the described examples are not necessarily limited to the order in which they are presented.

[0159] The functional units in the various embodiments of the present application can be integrated in one processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0160] The above disclosure is merely preferred embodiments of the present application, and of course cannot be used to limit the scope of the present application, and any equivalent changes made according to the claims of the present application are still within the scope of the present application.

Claims

1. A data processing method, characterized in that, include: Obtain the questions that the target user has answered and the target questions that have not been answered; the answered questions carry knowledge point information and answer record information, and the target questions carry knowledge point information; Generate a question sequence, a knowledge point sequence, and an answer record sequence for the target user based on the already answered questions and the target question; Feature extraction is performed on the question sequence, knowledge point sequence, and answer record sequence to generate a question vector corresponding to the question sequence, a knowledge point vector corresponding to the knowledge point sequence, and an answer record vector corresponding to the answer record sequence. Graph features are extracted from the question vector and the knowledge point vector to generate corresponding graph vectors; the graph vectors are used to represent the relationship between the question vector and the knowledge point vector. The question-and-answer prediction model generates the probability that the target user answers the target question correctly, based on the graph vector, question vector, knowledge point vector, and answer record vector.

2. The method according to claim 1, characterized in that, Also includes: Obtain an initial question-answering prediction model, using the questions already answered by the target user as training sample data; The training sample data includes the answer records for each question; Based on the training sample data and the initial question answering prediction model, the correct probability corresponding to the training sample data is generated, and the model parameters of the initial question answering prediction model are adjusted according to the correct probability corresponding to the training sample data and the answer records of the training sample data. When the adjusted initial question-answering prediction model meets the convergence condition, the initial question-answering prediction model containing the adjusted model parameters is determined as the question-answering prediction model.

3. The method according to claim 1, characterized in that, The step of generating a question sequence, a knowledge point sequence, and an answer record sequence for the target user based on the answered questions and the target question includes: Arrange the answered questions and the target questions to generate a question sequence, and generate a knowledge point sequence based on the knowledge point information corresponding to each question in the question sequence; A sequence of answer records is generated based on the answer record information corresponding to each question in the question sequence.

4. The method according to claim 1, characterized in that, The step of extracting graph features from the question vector and knowledge point vector to generate a graph vector includes: Based on the multi-head self-attention mechanism, a first relation vector between the question vectors and a second relation vector between the knowledge point vectors are obtained; the first relation vector is used to represent the relationship between the question vectors, and the second relation vector is used to represent the relationship between the knowledge point vectors. Based on the cross-gating mechanism, a comprehensive relationship vector between the question vector and the knowledge point vector is generated based on the first relationship vector and the second relationship vector, and the comprehensive relationship vector is fused to generate a graph vector.

5. The method according to claim 4, characterized in that, The step of obtaining the first relation vector between the question vectors and the second relation vector between the knowledge point vectors based on the multi-head self-attention mechanism includes: Obtain the first transformation matrix between the problem vectors, and generate the first relation vector corresponding to the problem vectors based on the first transformation matrix; Obtain the second transformation matrix between the knowledge point vectors, and generate the second relationship vector corresponding to the knowledge point vector based on the second transformation matrix.

6. The method according to claim 4, characterized in that, The step of generating a comprehensive relationship vector between the question vector and the knowledge point vector based on the first and second relationship vectors according to the cross-gating mechanism, and fusing the comprehensive relationship vector to generate a graph vector, includes: The first weight coefficient and the second weight coefficient between the first relation vector and the second relation vector are obtained by using an activation function; A first comprehensive relation vector is obtained based on the first relation vector, the second relation vector, and the first weight coefficient; the first comprehensive relation vector represents a question vector that incorporates knowledge point vectors. A second comprehensive relation vector is obtained based on the first relation vector, the second relation vector, and the second weight coefficient; the second comprehensive relation vector represents a knowledge point vector integrated into the question vector; The first comprehensive relation vector and the second comprehensive relation vector are concatenated to generate a graph vector.

7. The method according to claim 1, characterized in that, The question-and-answer prediction model generates the probability that the target user answers the target question correctly based on the graph vector, question vector, knowledge point vector, and answer record vector, including: The knowledge point vector and the graph vector are concatenated to generate the first input vector; The knowledge point vector and the answer record vector are concatenated to generate the second input vector; The first input vector is attenuated using the attenuation function in the question-and-answer prediction model. Based on the second input vector and the attenuated first input vector, the learning state vector of the target user is generated. Based on the learning state vector and the output of the feedforward neural network layer in the question answering prediction model, the output is used as the probability that the target user answers the target question correctly.

8. A data processing apparatus, characterized in that, include: The question acquisition unit is used to acquire questions that the target user has already answered and target questions that have not yet been answered. The answered questions carry knowledge point information and answer record information, and the target question carries knowledge point information; A sequence generation unit is used to generate a question sequence, a knowledge point sequence, and an answer record sequence for the target user based on the answered questions and the target question. The feature extraction unit is used to extract features from the question sequence, the knowledge point sequence, and the answer record sequence, and generate a question vector corresponding to the question sequence, a knowledge point vector corresponding to the knowledge point sequence, and an answer record vector corresponding to the answer record sequence. The graph vector generation unit is used to extract graph features from the question vector and the knowledge point vector to generate corresponding graph vectors; the graph vectors are used to represent the relationship between the question vector and the knowledge point vector. The probability prediction unit is used to generate the probability that the target user answers the target question correctly based on the graph vector, question vector, knowledge point vector, and answer record vector through a question-answer prediction model.

9. A computer device, characterized in that, include: Processor, memory, and network interface; The processor is connected to the memory and the network interface, wherein the network interface is used to provide network communication functions, the memory is used to store program code, and the processor is used to call the program code to execute the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and to execute the method of any one of claims 1-7.

Images