Learning suggestion generation method, device and equipment and computer readable storage medium

Abstract

The invention relates to a learning suggestion generation method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring answer data of a target object; performing feature extraction on the answer data to obtain an answer feature vector; constructing a target graph structure according to the answer feature vector; determining a target error cause type corresponding to the target graph structure through an error cause prediction model; the error cause prediction model is obtained by training based on an answer sample training set; and according to the target error cause type, generating a learning suggestion of the target object. Through the method, the labor cost can be reduced, and personalized learning suggestions can be obtained.

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing, and in particular to a method, apparatus, computer device, computer-readable storage medium, and computer program product for generating learning suggestions. Background Art

[0002] In the field of education, error analysis is an important means to improve students' learning outcomes during the education process. By conducting in-depth analysis of the errors made by students during the test-taking process and finding out the causes of the errors, students can be helped to correct their learning methods, consolidate relevant knowledge points, and improve their academic performance.

[0003] For wrong question analysis, traditional technology mainly relies on teachers' manual correction and explanation, which consumes high labor costs. Summary of the invention

[0004] Based on this, it is necessary to provide a learning suggestion generation method, apparatus, computer device, computer-readable storage medium and computer program product that can reduce labor costs in response to the above technical problems.

[0005] In a first aspect, the present application provides a method for generating learning suggestions, comprising:

[0006] Get the answer data of the target object;

[0007] Extracting features from the answer data to obtain an answer feature vector;

[0008] Constructing a target graph structure according to the answer feature vector;

[0009] Determine the target error cause type corresponding to the target graph structure through an error cause prediction model; the error cause prediction model is trained based on a training set of answer samples;

[0010] Generate learning suggestions for the target object according to the target error cause type.

[0011] In one embodiment, constructing a target graph structure according to the answer feature vector includes:

[0012] The target graph structure is constructed by taking the target data body in the answer data as the graph node, the answer feature vector corresponding to the target data body as the value of the graph node, and the connection lines between the graph nodes as the edges.

[0013] In one embodiment, the target data body includes questions, answers, knowledge points and causes of errors; the edges include the connection between the question node and the knowledge point node, the connection between the knowledge point node and the cause of error node, the connection between the cause of error node and the answer node, and the connection between the answer node and the question node.

[0014] In one embodiment, the error cause prediction model includes an attention network; and determining the target error cause type corresponding to the target graph structure through the error cause prediction model includes:

[0015] Updating the node feature vector corresponding to the graph node in the target graph structure through the attention network to obtain an updated node feature vector;

[0016] Determining an embedded representation of the graph node based on the updated node feature vector;

[0017] The embedded representations of the graph nodes are classified to obtain target error cause types corresponding to the target graph structure.

[0018] In one embodiment, generating a learning suggestion for the target object according to the target error cause type includes:

[0019] generating a prompt word according to the error cause type and the learning information of the target object;

[0020] According to the prompt words, learning suggestions for the target object are generated based on the artificial intelligence model.

[0021] In one embodiment, the training method of the error cause prediction model includes:

[0022] Obtaining a training set of answer samples; the training set of answer samples includes answer sample data of multiple answering subjects, and the answer sample data includes error cause classification labels;

[0023] Performing feature extraction on the answer sample training set to obtain an answer sample feature vector;

[0024] Constructing a sample graph structure according to the answer sample feature vector;

[0025] Determine the prediction error cause type corresponding to the sample graph structure through an initial prediction model;

[0026] The initial prediction model is trained according to the difference between the predicted error cause type and the error cause classification label to obtain the error cause prediction model.

[0027] In a second aspect, the present application also provides a learning suggestion generating device, comprising:

[0028] The answer data acquisition module is used to obtain the answer data of the target object;

[0029] A feature vector determination module is used to extract features from the answer data to obtain an answer feature vector;

[0030] A graph structure construction module, used to construct a target graph structure according to the answer feature vector;

[0031] An error cause type determination module, used to determine the target error cause type corresponding to the target graph structure through an error cause prediction model; the error cause prediction model is trained based on a training set of answer samples;

[0032] The learning suggestion generating module is used to generate the learning suggestion for the target object according to the target error cause type.

[0033] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the learning suggestion generating method provided in the first aspect is implemented.

[0034] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for generating learning suggestions provided in the first aspect is implemented.

[0035] In a fifth aspect, the present application further provides a computer program product, including a computer program, which, when executed by a processor, implements the learning suggestion generating method provided in the first aspect.

[0036] The above-mentioned learning suggestion generation method, device, computer equipment, computer-readable storage medium and computer program product obtain the answer data of the target object, extract features of the answer data, obtain the answer feature vector, construct the target graph structure according to the answer feature vector, determine the target error cause type corresponding to the target graph structure through the error cause prediction model, and generate learning suggestions for the target object according to the target error cause type. The error cause prediction model is trained based on the answer sample training set, and can automatically perform error cause analysis based on the answer data of the target object, predict the corresponding error cause type, and reduce labor costs. In addition, based on the graph structure, it can effectively perceive the complex relationship between different data, and comprehensively consider the context information, which can improve the accuracy of error cause analysis, so as to obtain personalized learning suggestions based on the error cause. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 A diagram of an application environment of a learning suggestion generation method in one embodiment;

[0039] Figure 2 A flowchart of a method for generating learning suggestions in one embodiment is shown;

[0040] Figure 3 A schematic diagram of the structure of the figure in one embodiment;

[0041] Figure 4 A schematic diagram of determining a target error cause type corresponding to a target graph structure based on a graph attention network in one embodiment;

[0042] Figure 5 A schematic diagram of a training process of an error cause prediction model in one embodiment;

[0043] Figure 6 A flowchart of a method for generating student suggestions in another embodiment;

[0044] Figure 7 A structural block diagram of a device for generating student suggestions in one embodiment;

[0045] Figure 8 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0047] The learning suggestion generation method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The server 104 can obtain the answer data of the target object sent by the terminal 102, and the server 104 extracts the features of the answer data to obtain the answer feature vector, constructs the target graph structure according to the answer feature vector, determines the target error type corresponding to the target graph structure through the error cause prediction model, and generates the learning suggestion of the target object according to the target error type. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car devices, projection devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The server 104 may be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. It should be noted that the learning suggestion generation method provided in the embodiment of the present application is not limited to the application scenario in which the server and the terminal interact with each other, but is also applicable to the application scenario of a single server or a single terminal.

[0048] In an exemplary embodiment, Figure 2 As shown, a learning suggestion generation method is provided, which is applied to Figure 1 The server in the example is used to illustrate, including the following steps 202 to 210. Among them:

[0049] Step 202, obtaining the target object's answer data.

[0050] Among them, the target object refers to the object to be analyzed for the cause of error and given learning suggestions. The answer data is the relevant data generated by the target object's answering, and the answer data may include question data, answer data, review and annotation data, knowledge point information, historical wrong question information, object learning data, and wrong question data of objects of the same level, etc. Among them, the question data refers to the data of the questions answered by the target object, the answer data refers to the data obtained by the target object to answer the questions corresponding to the question data, the review and annotation data refers to the data for reviewing and annotating the answer of the target object, the knowledge point information refers to the knowledge points involved in the question data, the historical wrong question information refers to the wrong question information of the target object in the historical answering process, the object learning data refers to the data generated by the target object's learning, such as the target object's study hours, study time, study subjects, grades or learning progress, etc., the wrong question data of objects of the same level refers to the wrong question data of other objects of the same level as the target object, wherein the same level means that the learning ability is at the same level.

[0051] Optionally, the question data and answer data can be obtained through an online question bank or paper test questions. For example, the paper question data and answer data can be identified by manual input or OCR (Optical Character Recognition) technology, and the review annotation data on paper can be identified by online annotation data or manual input or OCR technology. Knowledge point information can be obtained based on the knowledge graph and question bank system.

[0052] Optionally, the initial answer data of the target object is obtained, and the initial answer data is cleaned, such as noise removal, spelling check and correction, and format standardization, etc., to obtain cleaned answer data, and then feature extraction is performed on the cleaned answer data to obtain an answer feature vector. Among them, noise removal can include removing extra spaces, special characters, and unnecessary punctuation marks, etc., and format standardization refers to unifying the expression format of various answer data.

[0053] Step 204, extract features from the answer data to obtain an answer feature vector.

[0054] Optionally, the answer data may be encoded to obtain the answer data encoding, and then feature extraction may be performed on the answer data encoding to obtain the answer feature vector. The answer data may include multiple types of data, such as text type, numerical type, etc. Different types of data may use different encoding methods. The answer data may include multiple types, and the corresponding answer feature vectors may include multiple types.

[0055] Optionally, if the answer data includes text data, it can be encoded by text encoding, such as question data, answer data, review annotation data, etc. If the answer data has discrete classification characteristics, it can be encoded by one-hot encoding, such as knowledge point information, historical error type, class information, course information, etc. If the answer data includes numerical data, it can be encoded by numerical encoding, such as object learning data, such as grades, learning progress, etc.

[0056] Exemplarily, if a type of answer data includes multiple types of data, each type of answer data is encoded by a corresponding decoding method, and then feature extraction is performed on each encoding to obtain a feature vector of the corresponding type of data, and the feature vectors of various types of data are concatenated to obtain the answer feature vector corresponding to the type of answer data. For example, the question data may include text type data and numerical type data, then the text type data in the question data is encoded by a text encoding method to obtain first encoded data, and the numerical type data in the question data is encoded by a numerical encoding method to obtain second encoded data, and feature extraction is performed on the first encoded data to obtain a first feature vector, and feature extraction is performed on the second encoded data to obtain a second feature vector, and then the first feature vector and the second feature vector are concatenated to obtain the answer feature vector corresponding to the question data.

[0057] Step 206, constructing a target graph structure according to the answer feature vector.

[0058] Among them, different types of answer data can obtain corresponding answer feature vectors, for example, question data can obtain corresponding question feature vectors, answer data can obtain corresponding answer feature vectors, etc. It can be understood that the target graph structure can be constructed based on multiple answer feature vectors.

[0059] Optionally, each answer feature vector can be used as the value of the corresponding graph node, that is, one answer feature vector corresponds to one graph node, and the connection between each answer feature vector is used as the edge of the graph structure to construct the target graph structure.

[0060] Step 208, determining the target error cause type corresponding to the target graph structure through an error cause prediction model; the error cause prediction model is trained based on a training set of answer samples.

[0061] The error cause prediction model refers to a model that can predict the error cause type of the graph structure. The error cause prediction model can be trained based on a question answering sample training set, where the question answering sample training set includes question answering sample data of multiple question answering objects, and each question answering sample data includes an error cause type label.

[0062] Optionally, the error cause prediction model includes an attention network, which provides attention weights for edges in the target graph structure, calculates values ​​of graph nodes in the target graph structure based on the attention weights, and determines target error cause types based on the values ​​of each graph node. For example, the values ​​of each graph node can be classified to obtain the target error cause type.

[0063] Alternatively, the target error cause type corresponding to the target graph structure may be predicted by the error cause prediction model. For example, the value of each graph node in the target graph structure may be classified and predicted by the error cause prediction model to obtain the target error cause type.

[0064] Step 210: Generate learning suggestions for the target object according to the target error cause type.

[0065] Among them, learning suggestions refer to suggestions for learning of the target object, so as to avoid the target object from repeating the same mistakes in the future, strengthen the understanding and application of relevant knowledge points, and improve learning efficiency and learning effect. After obtaining the target error cause type, the learning suggestions for the target object can be generated according to the target error cause type. Optionally, a prompt word can be generated based on the target error cause type, and then the prompt word can be input into the artificial intelligence large model to obtain the learning suggestions for the target object.

[0066] Optionally, prompt words can be constructed based on the target error cause type, knowledge point information, historical wrong question information, object learning data, etc., and then the prompt words are input into the artificial intelligence big model to obtain learning suggestions for the target object. In an exemplary scenario, the server generates learning suggestions for the target object, and can send the learning suggestions to the terminal, which pushes the learning suggestions to the target object's account in the form of a message.

[0067] In the above-mentioned learning suggestion generation method, the answer data of the target object is obtained, and features are extracted from the answer data to obtain the answer feature vector. A target graph structure is constructed according to the answer feature vector, and the target error cause type corresponding to the target graph structure is determined by the error cause prediction model. According to the target error cause type, a learning suggestion for the target object is generated. The error cause prediction model is trained based on the answer sample training set, and can automatically perform error cause analysis based on the answer data of the target object and predict the corresponding error cause type, thereby reducing labor costs. In addition, the graph structure can effectively perceive the complex relationship between different data, and comprehensively consider the context information to improve the accuracy of error cause analysis, so that personalized learning suggestions can be obtained based on the error cause.

[0068] In some embodiments, step 206 of constructing a target graph structure according to the answer feature vector includes:

[0069] The target data body in the answer data is taken as the graph node, the answer feature vector corresponding to the target data body is taken as the value of the graph node, and the connection between the graph nodes is taken as the edge to construct the target graph structure.

[0070] The target data subject refers to the data subject that participates in constructing the graph structure. The data subject can be understood as the subject of the data. For example, the subject of the question data is the question, and the corresponding data subject is the question. The subject of the answer data is the answer, and the corresponding data subject is the answer. The target data subject can be any data subject. Optionally, all data subjects can be determined from the answer data, and then the target data subject can be determined from all data subjects. In other words, the target data subject can be part of the data subjects or all of the data subjects. The target data subject includes at least one data subject. For example, the target data subject includes questions, answers, knowledge points, historical error types, learning conditions, etc.

[0071] Optionally, the target data subject is used as a graph node, the answer feature vector corresponding to the target data subject is used as the value of the corresponding graph node, and the connection between the graph nodes is used as an edge, wherein the connection between the graph nodes can be a connection between any two graph nodes, or a connection between some graph nodes. For example, the graph nodes include questions, answers, knowledge points, and error causes, and the corresponding edges can include the connection between the question node and the knowledge point node, the connection between the knowledge point node and the error cause node, the connection between the error cause node and the answer node, and the connection between the answer node and the question node, or, it can also include the connection between the answer node and the knowledge point node, and the connection between the question node and the error cause node. Among them, the answer feature vector corresponding to the error cause node is the characteristic vector corresponding to the historical error cause type.

[0072] The answer feature vector corresponding to the target data subject refers to the feature vector corresponding to the answer data corresponding to the target data subject, which can be obtained by encoding and feature extracting the answer data corresponding to the target data subject.

[0073] In this embodiment, by constructing a target graph structure with the target data body in the answer data as the graph node, the answer feature vector corresponding to the target data body as the value of the graph node, and the lines between the graph nodes as the edges, a graph structure that matches the answer data of the target object can be constructed.

[0074] In some embodiments, the target data body includes questions, answers, knowledge points and causes of errors; the edges include the connection between the question node and the knowledge point node, the connection between the knowledge point node and the cause of error node, the connection between the cause of error node and the answer node, and the connection between the answer node and the question node.

[0075] For example, the graph structure is as follows Figure 3As shown, the target data body includes questions, answers, knowledge points and causes of errors. Each target data body is taken as a graph node. Then, the graph nodes include question nodes 302, answer nodes 304, knowledge point nodes 306 and cause of errors nodes 308. The lines between the question nodes 302 and the knowledge point nodes 306, the lines between the knowledge point nodes 306 and the cause of errors nodes 308, the lines between the cause of errors nodes 308 and the answer nodes 304, and the lines between the answer nodes 304 and the question nodes 302 serve as the edges of the target graph structure.

[0076] In this embodiment, by taking the main data subjects such as questions, answers, knowledge points and error causes as graph nodes, that is, taking the data subjects that have a greater impact on the error cause type as graph nodes, and determining the edges of the graph structure according to the degree of data association between the data subjects, it is possible to establish a graph structure that can better express the relationship between data and improve the accuracy of the graph structure.

[0077] In some embodiments, the error cause prediction model includes an attention network; in step 208, determining the target error cause type corresponding to the target graph structure through the error cause prediction model includes:

[0078] The node feature vectors corresponding to the graph nodes in the target graph structure are updated through the attention network to obtain the updated node feature vectors; based on the updated node feature vectors, the embedded representations of the graph nodes are determined; the embedded representations of the graph nodes are classified to obtain the target error cause types corresponding to the target graph structure.

[0079] Among them, the node feature vector can be the answer feature vector of the data subject corresponding to the node. For example, the data subject corresponding to the question node is the question, and the answer feature vector corresponding to the question can be a feature vector obtained based on the question data, and the answer feature vectors of other data subjects can be deduced by analogy.

[0080] Optionally, the attention network can provide attention weights between the graph node and each adjacent graph node in the target graph structure through the attention mechanism, update the node feature vector of each graph node according to the attention weight, obtain an updated node feature vector, and then determine the embedded representation of the graph node based on the updated node feature vector, for example, use the updated node feature vector as the embedded representation of the graph node, or perform dimensionality reduction processing on the updated node feature vector to obtain the embedded representation of the graph node. Then, the embedded representation of the graph node is classified and predicted to obtain the target error cause type corresponding to the target graph structure.

[0081] Exemplarily, the updating method of the node feature vector of the target node in the target graph structure is as follows: the initialization node feature vector of the target node can be the answer feature vector of the data subject corresponding to the target node, and after providing the attention weight between the target node and its adjacent nodes through the attention network, the initialization node vector corresponding to the target node and its adjacent nodes is weighted summed with the attention weight to obtain the updated node feature vector, and the updated node feature vector is used as the value of the target node. For example, the nodes adjacent to the target node P are P1, P2 and P3, the initialization node feature vector corresponding to the target node P is V1, the initialization node feature vector corresponding to the node P1 is V2, the initialization node feature vector corresponding to the node P2 is V2, the initialization node feature vector corresponding to the node P3 is V3, the attention weight between P and P1 is K1, the attention weight between P and P2 is K2, and the attention weight between P and P3 is K3. Then, the updated node feature vector V of the target node can be calculated by the following formula (1).

[0082] Formula (1)

[0083] It is easy to understand that by using the above formula (1), it is easy to obtain the updated node feature vector corresponding to each graph node in the target graph structure. The nodes adjacent to the target node refer to the nodes that form the edge of the graph structure with the target node, such as Figure 3 In the graph structure of the example, the nodes adjacent to the question node 302 include the answer node 304 and the knowledge point node 306, and the nodes adjacent to the answer node 304 include the question node 302 and the error cause node 308, etc.

[0084] Each graph node corresponds to an embedded representation, and the similarity between nodes can be judged through the embedded representation. The embedded representations of the graph nodes in the target graph structure can be classified by a classifier. For example, the embedded representations of each graph node are input into the classifier, and the classification results are used to characterize the corresponding error cause type, thereby obtaining the target error cause type corresponding to the target graph structure.

[0085] In one example, if Figure 4 As shown, the attention network includes a graph attention network, and the target graph structure includes question nodes, answer nodes, knowledge point nodes and error cause nodes. The values ​​of each node in the target graph structure are input into the graph attention network. The graph attention network can assign attention weights to the edges of the target graph nodes, and update the values ​​of each node in the target graph structure based on the attention weights to obtain an updated node feature vector. The embedded representation of the graph node is determined according to the updated node feature vector, and the embedded representation of the graph node is classified to obtain the target error cause type corresponding to the target graph structure.

[0086] In this embodiment, the node feature vectors corresponding to the graph nodes in the target graph structure are updated by using the attention network, the embedded representations of the graph nodes are determined based on the updated node feature vectors, and the embedded representations of the graph nodes are classified to obtain the target error cause type corresponding to the target graph structure. This enables the attention network in the error cause prediction model to perceive the complex relationship between different nodes in the graph structure, determines the error cause type corresponding to the graph structure according to the similarity between the nodes, and predicts a more accurate error cause type.

[0087] In some embodiments, step 210 of generating a learning suggestion for a target object according to the target error cause type includes:

[0088] Generate prompt words based on the error type and the target object's learning information; based on the prompt words, generate learning suggestions for the target object based on the artificial intelligence large model.

[0089] Among them, prompt words refer to keywords used to prompt the artificial intelligence large model to generate target answers, and prompt words can include words or sentences.

[0090] Optionally, prompt words can be generated according to preset prompt templates based on the error cause type, target object's learning information, knowledge point data, historical wrong question records, learning goals, etc., and the prompt words can be input into the artificial intelligence big model to obtain learning suggestions for the target object.

[0091] Exemplarily, the preset prompt template includes:

[0092] Promt = "Student Information:

[0093] - Name: {student_name}

[0094] - Grade: {grade}

[0095] - Class: {class}

[0096] - History exam scores: {exam_scores}

[0097] - Study progress: {study_progress}

[0098] Current wrong question analysis:

[0099] - Question: {current_question}

[0100] - Student answer: {current_answer}

[0101] - Error reason: {current_error_reason}

[0102] - Related knowledge points: {current_knowledge_points}

[0103] Historical error records:

[0104] {historical_mistakes}

[0105] Learning objectives:

[0106] - Short-term goals: {short_term_goals}

[0107] - Long-term goals: {long_term_goals}

[0108] Based on the above information, please generate personalized learning suggestions for students, focusing on helping students make progress in the following three areas:

[0109] First, the improvement methods for the current wrong questions;

[0110] Second, strengthen the understanding and application of relevant knowledge points;

[0111] Third, improve overall learning efficiency and test scores."

[0112] In this embodiment, prompt words are generated according to the error cause type and the learning information of the target object, and learning suggestions for the target object are generated based on the prompt words through an artificial intelligence large model, so that personalized learning suggestions can be generated that are closer to the students' actual situation and learning needs.

[0113] In some exemplary embodiments, Figure 5 As shown, the training method of the error cause prediction model includes the following steps 502 to 510.

[0114] Step 502, obtaining a training set of answer samples; wherein the training set of answer samples includes answer sample data of multiple answering subjects, and the answer sample data includes error cause classification labels.

[0115] Among them, the answer sample training set includes answer sample data of multiple answer objects, and the answer sample data of each answer object may include question sample data, answer sample data, review annotation sample data, knowledge point sample information, error cause classification label, historical wrong question sample information, object learning sample data, wrong question sample data of objects of the same level, etc. Optionally, the knowledge point samples corresponding to the question sample data can be labeled. It is easy to understand that the answer sample data can refer to the explanation and description of the answer data in the aforementioned embodiment.

[0116] Step 504, extract features from the answer sample training set to obtain the answer sample feature vector.

[0117] Step 506, constructing a sample graph structure according to the answer sample feature vector.

[0118] It is easy to understand that the implementation of the corresponding terms or steps in step 504 and step 506 can refer to the description and explanation of step 204 and step 206 in the above embodiment. It is only necessary to replace the answer sample training set with the answer data.

[0119] Step 508: Determine the prediction error cause type corresponding to the sample graph structure through the initial prediction model.

[0120] The initial prediction model refers to the model to be trained. For example, the initial prediction model may include a graph attention network (GAT) or a graph neural network (GNNs). The graph attention network is used to calculate the values ​​of nodes in the sample graph structure, and then error cause analysis and prediction are performed based on the values ​​of the graph nodes to obtain the predicted error cause type.

[0121] For example, the graph attention network provides the attention weight corresponding to each edge in the sample graph structure, and then calculates and updates the value of each graph node in the sample graph structure based on the attention weight, and predicts the prediction error cause type corresponding to the sample graph structure based on the value of each graph node.

[0122] Step 510: training the initial prediction model according to the difference between the predicted error cause type and the error cause classification label to obtain an error cause prediction model.

[0123] The error cause classification label is used to characterize the correct error cause classification. By comparing the predicted error cause type predicted by the initial prediction model with the error cause classification label, the difference between the predicted error cause type and the error cause classification label is obtained, and the initial prediction model is trained according to the difference to obtain the error cause prediction model.

[0124] Optionally, according to the difference between the predicted error cause type and the error cause classification label, the parameters of the initial prediction model are adjusted until the difference between the predicted error cause type and the error cause classification label is less than a difference threshold, or the number of training times reaches a preset number of training times, thereby obtaining an error cause prediction model.

[0125] Exemplarily, the initial prediction model includes an attention network, which provides attention weights between nodes for the sample graph structure through the attention network, updates the values ​​of the nodes in the sample graph structure through the attention weights, obtains the embedded representation of the nodes, and then classifies the embedded representation of the nodes to obtain the predicted error cause type, and then compares the predicted error cause type with the error cause classification label to obtain the difference between the predicted error cause type and the error cause classification label, and determines whether the difference is less than the difference threshold. If the difference is not less than the difference threshold, the attention weights between the nodes are re-provided for the sample graph structure until the provided attention weight makes the difference between the predicted error cause type and the error cause classification label less than the difference threshold, thereby obtaining the error cause prediction model.

[0126] In this embodiment, feature extraction is performed on the answer sample training set to obtain the answer sample feature vector, and a sample graph structure is constructed based on the answer sample feature vector. The predicted error cause type corresponding to the sample graph structure is determined by the initial prediction model. According to the difference between the predicted error cause type and the error cause classification label, the initial prediction model is trained to obtain an error cause prediction model for predicting the error cause type of the answer data of the answer object, thereby avoiding manual error cause analysis and explanation, reducing the labor cost of error cause analysis, and improving the efficiency of error cause analysis. At the same time, error cause analysis is performed based on the answer data of the answer object, which can personalize the actual situation of the answer object, perceive the complex relationship between the answer data based on the graph structure, and achieve accurate error cause analysis, and predict the accurate error cause type.

[0127] In an exemplary embodiment, Figure 6 As shown, the student suggestion generating method includes the following steps 602 to 616.

[0128] Step 602, obtaining the target object's answer data.

[0129] Step 604: extract features from the answer data to obtain an answer feature vector.

[0130] Step 606, constructing a target graph structure with the target data body in the answer data as the graph node, the answer feature vector corresponding to the target data body as the value of the graph node, and the connection between the graph nodes as the edge.

[0131] Step 608: Update the node feature vector corresponding to the graph node in the target graph structure through the graph attention network to obtain an updated node feature vector.

[0132] Step 610, determining an embedding representation of the graph node based on the updated node feature vector.

[0133] Step 612, classify the embedded representations of the graph nodes to obtain the target error cause type corresponding to the target graph structure.

[0134] Step 614, generating prompt words according to a preset prompt template based on the target error cause type, the target object's learning information, historical wrong question data, and knowledge point information.

[0135] Step 616, based on the prompt words, generate learning suggestions for the target object based on the artificial intelligence model.

[0136] In the above embodiment, a graph structure is constructed according to the answer data of the answering subject, the complex relationship between the nodes in the graph structure is perceived through the graph neural network, the similarity between the nodes is identified, the embedded representation of the node is obtained, the embedded representation of the node is classified, and the accurate error cause type corresponding to the answer data is obtained. Then, according to the error cause type and historical learning data, personalized learning suggestions are generated through a large model, which is more in line with the actual situation and learning needs of the learning subject.

[0137] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0138] Based on the same inventive concept, the embodiment of the present application also provides a learning suggestion generation device for implementing the learning suggestion generation method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more learning suggestion generation device embodiments provided below can refer to the limitations of the learning suggestion generation method above, and will not be repeated here.

[0139] In an exemplary embodiment, Figure 7 As shown, a learning suggestion generating device is provided, including a question answering data acquiring module 702, a feature vector determining module 704, a graph structure building module 706, an error cause type determining module 708 and a learning suggestion generating module 710, wherein:

[0140] The answer data acquisition module 702 is used to acquire the answer data of the target object;

[0141] A feature vector determination module 704 is used to extract features from the answer data to obtain an answer feature vector;

[0142] A graph structure construction module 706, configured to construct a target graph structure according to the answer feature vector;

[0143] The error cause type determination module 708 is used to determine the target error cause type corresponding to the target graph structure through an error cause prediction model; the error cause prediction model is trained based on the answer sample training set;

[0144] The learning suggestion generating module 710 is used to generate a learning suggestion for the target object according to the target error cause type.

[0145] In some embodiments, the graph structure construction module 706 is also used to construct the target graph structure with the target data body in the answer data as the graph node, the answer feature vector corresponding to the target data body as the value of the graph node, and the lines between the graph nodes as the edges.

[0146] In some embodiments, the error cause prediction model includes an attention network; the error cause type determination module 708 is also used to update the node feature vector corresponding to the graph node in the target graph structure through the attention network to obtain an updated node feature vector; based on the updated node feature vector, determine the embedded representation of the graph node; classify the embedded representation of the graph node to obtain the target error cause type corresponding to the target graph structure.

[0147] In some embodiments, the learning suggestion generation module 710 is further used to generate prompt words according to the error cause type and the learning information of the target object; and to generate learning suggestions for the target object based on the artificial intelligence model according to the prompt words.

[0148] In some embodiments, the above-mentioned device also includes a model training module, which is used to implement the training process of the error cause prediction model, including: obtaining a question answer sample training set; the question answer sample training set includes question answer sample data of multiple answer objects, and the question answer sample data includes an error cause classification label; performing feature extraction on the question answer sample training set to obtain an answer sample feature vector; constructing a sample graph structure according to the answer sample feature vector; determining the predicted error cause type corresponding to the sample graph structure through an initial prediction model; and training the initial prediction model according to the difference between the predicted error cause type and the error cause classification label to obtain the error cause prediction model.

[0149] Each module in the above learning suggestion generating device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module.

[0150] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 8 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the answer data of the answering object. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for generating learning suggestions is implemented.

[0151] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0152] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps in the above-mentioned learning suggestion generating method embodiment are implemented.

[0153] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned learning suggestion generating method embodiment are implemented.

[0154] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above-mentioned learning suggestion generating method embodiment when executed by a processor.

[0155] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0156] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0157] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0158] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for generating learning suggestions, characterized in that: The method comprises: Get the answer data of the target object; Extracting features from the answer data to obtain an answer feature vector; Constructing a target graph structure according to the answer feature vector; Determine the target error cause type corresponding to the target graph structure through an error cause prediction model; the error cause prediction model is trained based on a training set of answer samples; Generate learning suggestions for the target object according to the target error cause type.

2. The method according to claim 1, characterized in that The step of constructing a target graph structure according to the question answering feature vector comprises: The target graph structure is constructed by taking the target data body in the answer data as the graph node, the answer feature vector corresponding to the target data body as the value of the graph node, and the connection lines between the graph nodes as the edges.

3. The method according to claim 2, characterized in that The target data body includes questions, answers, knowledge points and causes of errors; the edges include the connection between question nodes and knowledge point nodes, the connection between knowledge point nodes and cause of errors nodes, the connection between cause of errors nodes and answer nodes, and the connection between answer nodes and question nodes.

4. The method according to claim 1, characterized in that: The error cause prediction model includes an attention network; and determining the target error cause type corresponding to the target graph structure through the error cause prediction model includes: Updating the node feature vector corresponding to the graph node in the target graph structure through the attention network to obtain an updated node feature vector; Determining an embedded representation of the graph node based on the updated node feature vector; The embedded representations of the graph nodes are classified to obtain target error cause types corresponding to the target graph structure.

5. The method according to claim 1, characterized in that Generating a learning suggestion for the target object according to the target error cause type includes: generating a prompt word according to the error cause type and the learning information of the target object; According to the prompt words, learning suggestions for the target object are generated based on the artificial intelligence model.

6. The method according to any one of claims 1 to 5, characterized in that The training method of the error cause prediction model includes: Obtaining a training set of answer samples; the training set of answer samples includes answer sample data of multiple answering subjects, and the answer sample data includes error cause classification labels; Performing feature extraction on the answer sample training set to obtain an answer sample feature vector; Constructing a sample graph structure according to the answer sample feature vector; Determine the prediction error cause type corresponding to the sample graph structure through an initial prediction model; The initial prediction model is trained according to the difference between the predicted error cause type and the error cause classification label to obtain the error cause prediction model.

7. A learning suggestion generating device, characterized in that: The device comprises: The answer data acquisition module is used to obtain the answer data of the target object; A feature vector determination module is used to extract features from the answer data to obtain an answer feature vector; A graph structure construction module, used to construct a target graph structure according to the answer feature vector; An error cause type determination module, used to determine the target error cause type corresponding to the target graph structure through an error cause prediction model; the error cause prediction model is trained based on a training set of answer samples; The learning suggestion generating module is used to generate the learning suggestion for the target object according to the target error cause type.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A 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 method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images