Test question search method, device, storage medium and electronic device

Abstract

A method, device, storage medium and electronic device for searching test questions. The method includes obtaining the answer data of the target subject to the reference test questions; evaluating the mastery degree of the target subject to the reference test questions according to the answer data to obtain mastery degree data; obtaining the test question attribute data of the reference test questions; and searching for the target test questions matching the reference test questions from the candidate test questions according to the test question attribute data and the mastery degree data. The application can improve the accuracy of test question search.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a test question searching method, device, storage medium and electronic device. Background Art

[0002] With the advent of the intelligent era and the national education double reduction policy, the demand for intelligent and precise education is becoming more and more urgent. How to enable students to practice more effectively in a limited time while liberating teachers from repetitive work to the greatest extent has become a problem that needs to be solved urgently.

[0003] However, at present, teachers still need to manually search for test questions in a massive test question bank based on students' test results for targeted training after the test. This not only increases the teaching burden of teachers, but also makes the efficiency of test question searching very low. Summary of the invention

[0004] The present application provides a test question search method, device, storage medium and electronic device, which can improve the efficiency of test question search.

[0005] The test question search method provided by this application includes:

[0006] Obtaining the target subject's answer data on the reference test questions;

[0007] Assessing the target subject's mastery of the reference test questions according to the answer data to obtain mastery degree data;

[0008] Acquiring test question attribute data of the reference test question;

[0009] According to the test question attribute data and the mastery level data, a target test question matching the reference test question is searched from candidate test questions.

[0010] The test question search device provided by this application includes:

[0011] The first acquisition module is used to obtain the target subject's answer data to the reference test questions;

[0012] An evaluation module, used to evaluate the mastery degree of the target subject for the reference test questions according to the answer data, and obtain mastery degree data;

[0013] A second acquisition module, used for test question attribute data of the reference test question;

[0014] The search module is used to search for a target test question matching the reference test question from candidate test questions according to the test question attribute data and the mastery level data.

[0015] The storage medium provided in the present application stores a computer program thereon, and when the computer program is loaded by a processor, the steps in the test question searching method provided in the present application are executed.

[0016] The electronic device provided in the present application includes a processor and a memory, wherein the memory stores a computer program, and the processor is used to execute the steps in the test question search method provided in the present application by loading the computer program.

[0017] In this application, the target subject's answer data for the reference test questions are obtained; the target subject's mastery of the reference test questions is evaluated based on the answer data to obtain mastery data; the test attribute data of the reference test questions is obtained; and the target test questions that match the reference test questions are searched from the candidate test questions based on the test attribute data and the mastery data. Compared with the related art, this application searches for test questions based on the target subject's answer data for the reference test questions and the test attribute data of the reference test questions themselves, which fully considers the target subject's mastery of the test questions, thereby greatly improving the accuracy of the test question search. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 A scenario diagram of a test question search system provided in an embodiment of the present application.

[0020] Figure 2 A flowchart of a test question searching method provided in an embodiment of the present application.

[0021] Figure 3 It is a structural diagram of the test question mastery level assessment model provided in the embodiments of the present application.

[0022] Figure 4 It is a structural diagram of the correction result evaluation model provided in the embodiment of the present application.

[0023] Figure 5 It is a schematic diagram of the structure of the first category network in this application.

[0024] Figure 6 It is a structural diagram of the second category network in this application.

[0025] Figure 7 It is a structural diagram of the question data generation model provided in this application.

[0026] Figure 8 It is a structural diagram of the test question similarity evaluation model provided in this application.

[0027] Fig. 9 It is a structural block diagram of the test question search device provided in an embodiment of the present application.

[0028] Fig.10 It is a structural block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0029] It should be noted that the principles of the present application are illustrated by implementing in an appropriate computing environment. The following description is based on the illustrated specific embodiments of the present application, which should not be construed as limiting other specific embodiments not described in detail herein. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0030] The relational terms such as first and second involved in the following embodiments of the present application are only used to distinguish one object or operation from another object or operation, and are not used to limit the actual sequential relationship between these objects or operations. In the description of the embodiments of the present application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0031] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.

[0032] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes machine learning (ML) technology, among which deep learning (DL) is a new research direction in machine learning. It is introduced into machine learning to make it closer to the original goal, namely artificial intelligence. At present, deep learning is mainly used in computer vision, natural language processing and other fields.

[0033] Deep learning is the process of learning the inherent laws and representation levels of sample data. The information obtained in the learning process is very helpful for interpreting data such as text, images, and sound. Using deep learning technology and corresponding training data sets, network models that implement different functions can be trained. For example, a deep learning network for gender classification can be trained based on one training data set, and a deep learning network for image optimization can be trained based on another training data set.

[0034] In order to improve the efficiency of question search, the present application introduces deep learning into question search, and accordingly provides a question search method device, a storage medium and an electronic device. The question search method can be executed by an electronic device.

[0035] Please refer to Figure 1 ,This application also provides a test question search system, such as Figure 1 As shown, it is a schematic diagram of the use scenario of the test question search system provided by the present application. The test question search system includes an electronic device 100. For example, the electronic device can obtain the answer data of the target object to the reference test question; evaluate the mastery degree of the target object for the reference test question based on the answer data to obtain the mastery degree data; obtain the test question attribute data of the reference test question; and search for the target test question that matches the reference test question from the candidate test questions based on the test question attribute data and the mastery degree data.

[0036] The electronic device 100 may be any device equipped with a processor and having processing capabilities, such as a mobile electronic device with a processor, such as a smart phone, a tablet computer, a PDA, a laptop computer, or a fixed electronic device with a processor, such as a desktop computer, a television, a server, etc.

[0037] In addition, if Figure 1 As shown, the test question search system may also include a storage device 200 for storing data, including but not limited to original data, intermediate data, and result data obtained during the test question search process. For example, the electronic device 100 may store the acquired answer data of the target subject to the reference test questions, the target subject's mastery of the reference test questions obtained based on the answer data, the test question attribute data of the reference test questions, and the candidate test questions in the storage device 200.

[0038] It should be noted that Figure 1 The scenario diagram of the test question search system shown is merely an example. The test question search system and scenario described in the embodiment of the present application are intended to more clearly illustrate the technical solution of the embodiment of the present application, and do not constitute a limitation on the technical solution provided in the embodiment of the present application. A person of ordinary skill in the art can appreciate that with the evolution of the test question search system and the emergence of new business scenarios, the technical solution provided in the embodiment of the present application is equally applicable to similar technical problems.

[0039] Please refer to Figure 2 , Figure 2 The flowchart of the test question search method provided in the embodiment of the present application is shown in FIG. Figure 2 As shown, the process of the test question search method provided in the embodiment of the present application can be as follows:

[0040] In S310, the target subject's answer data to the reference test questions is obtained.

[0041] Among them, the reference test questions can be reference test questions used for test question search, that is, the purpose of the test question search can be to search for test questions similar to the reference test questions. For example, in the scenario of consolidation practice after the exam, if the candidate scores low on test question A in the exam, then the candidate needs to conduct consolidation practice after the exam to improve it in a targeted manner, so it is necessary to search for test questions similar to test question A in the candidate test questions in the question bank. In this scenario, test question A is the aforementioned reference test question. The reference test question can be a Chinese test question or a test question in another language.

[0042] Among them, the target object can be an object that answers the above-mentioned reference test questions. The object can be a person, a computer device, or an artificial intelligence virtual character in the computer device.

[0043] The target subject's answer data to the reference test questions may include answer result data and standard answer data of the reference test questions. When the reference test questions are of multiple-choice or true-or-false question type, the answer result data may be a data selected by the target subject from a plurality of preset results; when the reference test questions are of short-answer question type, the answer result data may be the answer text data entered by the target subject in the answer area. Correspondingly, the standard answer data of the reference test questions may also be a certain data or text data among a plurality of preset results. The target subject's answer result data to the reference test questions may be obtained from the answer area, wherein the answer area may be the corresponding area of ​​the photographed paper test questions or the corresponding storage area of ​​the electronic test questions. The standard answer data of the reference test questions may be obtained from the question bank.

[0044] In S320, the mastery degree of the target subject with respect to the reference test questions is evaluated based on the answer data to obtain mastery degree data.

[0045] In the related art, when searching for similar test questions, similar searches are often performed based only on relevant data of the reference test questions, such as knowledge points, question types, difficulty, and fields. However, this search method does not take into account the interaction between the answering subject and the reference test questions, that is, it does not take into account the answering subject's mastery of the reference test questions. For example, whether the answering subject did not score any points for the question due to lack of ability or due to carelessness, etc., can be described by the answering subject's mastery of the reference test questions. Therefore, similar test questions searched by the test question search method in the existing related art may not be suitable for the answering subject to conduct consolidation training.

[0046] In this embodiment, the target subject's mastery of the reference test questions can be evaluated through the target subject's answer data to obtain mastery data. Then, a joint search can be performed based on the target subject's mastery data of the reference test questions combined with the relevant data of the reference test questions themselves, so that a target test question that is more suitable for the target subject to consolidate the connection can be obtained.

[0047] Specifically, the target subject's mastery of the reference test questions can be evaluated based on the target subject's answer result data to the reference test questions and the standard answer data of the reference test questions.

[0048] In some embodiments, the target subject's mastery of the reference test questions is evaluated based on the answer data to obtain mastery data, including:

[0049] Using a word embedding model, word embedding processing is performed on the first answer result data and the first standard answer data to obtain a first answer feature and a first answer feature;

[0050] The first response feature and the first answer feature are input into the test question mastery evaluation model to obtain the target subject's mastery data for the reference test question.

[0051] In this embodiment, the test question mastery level assessment model can be trained in advance, and then the target subject's answer result data for the reference test questions and the features corresponding to the standard answer data of the reference test questions are input into the test question mastery level assessment model to obtain the target subject's mastery level data for the reference test questions.

[0052] Among them, the features corresponding to the target object's answer result data to the reference test questions and the standard answer data of the reference test questions can be obtained by using the word embedding method. Specifically, the target object's answer result data to the reference test questions can be called the first answer result data, and the standard answer data of the reference test questions can be called the first standard answer data. The first answer result data and the first standard answer data can be pre-trained by word embedding using the following word embedding models: BERT model or ELMo model. The full name of the BERT model is Bidirectional Encoder Representation from Transformers, which means a bidirectional encoding representation model based on transformation, and is a pre-trained language representation model. The full name of the ELMo model is Embeddings from Language Models, which means embedding based on a language model, and is also a pre-trained language representation model. There are many introductions to the BERT model and the ELMo model in the prior art, which will not be repeated here.

[0053] Any of the above-mentioned word embedding models is used to perform word embedding processing on the first answer result data and the first standard answer data to implement the pre-training process and obtain the first answer feature and the first answer feature. Then, the first answer feature and the first answer feature are input into the aforementioned test question mastery evaluation model to obtain the target subject's mastery data of the reference test question.

[0054] In some embodiments, the first answer result data and the first standard answer data are respectively subjected to word embedding processing by a word embedding model to obtain a first answer feature and a first answer feature, including:

[0055] Performing text cleaning on the first answer result data to obtain a cleaned answer result text;

[0056] Performing text cleaning on the first standard answer data to obtain a cleaned standard answer text;

[0057] Segment the answer result text and the standard answer text to obtain the answer result word sequence and the standard answer word sequence;

[0058] The word embedding model is used to perform word embedding processing on the answer result word sequence and the standard answer word sequence respectively to obtain the first answer feature and the first answer feature.

[0059] In this embodiment, a word embedding model is used to perform word embedding processing on the first answer result data and the first standard answer data. Specifically, the first answer result data and the first standard answer data can be firstly subjected to text cleaning to obtain the corresponding cleaned texts. Then, the cleaned texts are respectively subjected to word segmentation processing to obtain corresponding word sequences, i.e., the answer result word sequence and the standard answer word sequence. Among them, the cleaned text is subjected to word segmentation, and a preset word segmentation tool can be used for word segmentation. When the reference test questions are Chinese test questions, the word segmentation tool can be an LTP word segmentation tool or a jieba word segmentation tool. Among them, LTP is a set of Chinese language processing systems open sourced by Harbin Institute of Technology, and jieba is a Chinese word segmentation tool. There are many introductions to the two word segmentation tools in the prior art, which will not be repeated here.

[0060] In some embodiments, the test question mastery evaluation model includes a first convolutional neural network, a second convolutional neural network, and an attention network. The first answer feature and the first answer feature are input into the test question mastery evaluation model to obtain the target subject's mastery data of the reference test question, including:

[0061] The first convolutional neural network is used to perform feature sampling on the first answer feature to obtain a second answer feature; the second convolutional neural network is used to perform feature sampling on the first answer feature to obtain a second answer feature;

[0062] An attention network is used to perform attention calculation on the second answer feature and the second answer feature to obtain first attention information;

[0063] The target subject's mastery of the reference test questions is determined based on the first attention information.

[0064] In this embodiment, the aforementioned test question mastery level assessment model may include a first convolutional neural network (CNN), a second convolutional neural network, and an attention network. Figure 3As shown, it is a structural diagram of the test question mastery assessment model in the present application. Among them, the first convolutional neural network and the second convolutional neural network can be a single-layer CNN network or a multi-layer CNN. The first answer feature and the first answer feature obtained by the aforementioned word embedding are input into the test question mastery assessment model to perform test question mastery assessment. The test question mastery assessment model can use the first convolutional neural network to perform feature sampling on the first answer feature to obtain the second answer feature; at the same time, the second convolutional neural network can perform feature sampling on the first answer feature to obtain the second answer feature. Then, the second answer feature and the second answer feature are input into the attention layer for attention calculation, and the first attention information of the second answer feature on the second answer feature is calculated. The first attention information can be an attention matrix, and then the attention matrix is ​​output. According to the attention matrix, the mastery data of the target object for the reference test question can be determined.

[0065] In some embodiments, the first attention information includes a weight matrix, and determining the mastery degree data of the target subject on the reference test questions according to the first attention information includes:

[0066] Performing pooling processing on the weight matrix according to the feature dimension of the second answer feature to obtain a weight vector;

[0067] The weight vector is determined as the target subject's mastery degree data on the reference test questions.

[0068] In this embodiment, the target subject's mastery of the reference test questions can be obtained by compressing the attention matrix corresponding to the aforementioned first attention information. Specifically, the weight matrix output by the test question mastery evaluation model can be pooled according to the feature dimension of the second answer feature, and the pooling process can be maximum pooling to obtain a weight vector. The weight vector can be the target subject's mastery of the reference test questions.

[0069] In some embodiments, before inputting the first answer feature and the first answer feature into the test question mastery evaluation model to obtain the target subject's mastery data of the reference test question, the method further includes:

[0070] Obtaining a first training sample data set of a correction result evaluation model, the first training sample data set comprising a plurality of first training sample data, the first training sample data comprising first text data of a first sample test question, second standard answer data of the first sample test question, second answer result data for the first sample test question, and correction data for the second answer result data, the correction data comprising a first batch correction conclusion and a first batch correction score;

[0071] The features corresponding to the first text data, the second standard answer data and the second answer result data in each first training sample data are used as input, and the correction data of the second answer result data is used as output to train the correction result evaluation model.

[0072] Before using the test question mastery degree assessment model to perform mastery degree assessment, the test question mastery degree assessment model needs to be trained first. In this embodiment, the test question mastery degree assessment model can be trained by training a grading result assessment model that includes the aforementioned test question mastery degree assessment model. That is, in this embodiment, the test question mastery degree assessment model is a sub-model in the grading result assessment model. In the process of training the grading result assessment model, the model parameters of the test question mastery degree assessment model can also be trained.

[0073] To train the correction result evaluation model, it is necessary to first obtain training samples, and the training samples can be a first training sample data set, and the first training sample data set includes multiple first training sample data. Each first training sample data includes the first text data of the first sample test question, the second standard answer data of the first sample test question, the second answer result data for the first sample test question, and the correction data for the second answer result data. Among them, the correction data includes the first batch of correction conclusions and the first batch of correction scores. Specifically, the correction conclusion can be right or wrong, and the correction score is the score data corresponding to the second answer result data. For example, in some short-answer questions, although the final answer result is wrong, it does not mean that the process cannot be scored. Therefore, the correction data contains both the correction conclusion and the correction score.

[0074] After obtaining the training samples, the input data and corresponding label data for model training can be determined, and the label data is the theoretical output data. In this embodiment, the correction result evaluation model is trained by taking the first text data, the second standard answer data, and the features corresponding to the second answer result data in each first training sample data as input, and taking the correction data of the second answer result data as output to train the correction result evaluation model.

[0075] In some embodiments, the grading result evaluation model includes a test question mastery evaluation model, a third convolutional neural network, a first feature fusion network, a first feature calculation network, a first classification network, and a second classification network. The test question mastery evaluation model includes a first convolutional neural network, a second convolutional neural network, and an attention network. The features corresponding to the first text data, the second standard answer data, and the second answer result data in each training sample data are used as inputs, and the grading data of the second answer result data is used as output to train the grading result evaluation model, including:

[0076] The first convolutional neural network is used to sample the word embedding features corresponding to the second answer result data to obtain a third answer feature, the second convolutional neural network is used to sample the word embedding features corresponding to the second standard answer data to obtain a third answer feature, and the third convolutional neural network is used to sample the word embedding features corresponding to the first text data to obtain a first text feature;

[0077] Using the attention network to perform attention calculation on the third answer feature and the third answer feature to obtain second attention information;

[0078] The first feature calculation network is used to calculate the product of the second attention information and the third answer feature to obtain a fourth answer feature;

[0079] The first feature fusion network is used to fuse the first text feature, the third answer feature, and the fourth answer feature to obtain a first fusion feature;

[0080] Inputting the first fusion feature into the first classification network to obtain the output second batch correction conclusion, and inputting the training splicing feature into the second classification network to obtain the output second batch correction score;

[0081] The model parameters of the correction result evaluation model are iteratively updated according to the first batch correction conclusions, the first batch correction scores, the second batch correction conclusions and the second batch correction scores until a first preset iteration stop condition is reached, thereby completing the training of the correction result evaluation model.

[0082] Among them, this application provides a specific structural example of a correction result evaluation model, such as Figure 4 As shown, a structural schematic diagram of the grading result evaluation model provided by the present application is shown. As shown in the figure, in addition to the first convolutional neural network, the second convolutional neural network and the attention network in the aforementioned test question mastery evaluation model, the grading result evaluation model also includes a third convolutional neural network, a first feature calculation network, a first feature fusion network, a first classification network and a second classification network.

[0083] Specifically, when using the features corresponding to the first text data, the second standard answer data and the second answer result data in each first training sample data as input and using the correction data of the second answer result data as output to train the correction result evaluation model of the structure, the word embedding features corresponding to the first text data can be first input into the third convolutional neural network for feature sampling to obtain the first text features; the word embedding features corresponding to the second answer result data can be input into the first convolutional neural network for feature sampling to obtain the third answer features; the word embedding features corresponding to the second standard answer data can be input into the second convolutional neural network for feature sampling to obtain the third answer features.

[0084] Then, the attention network is used to perform an attention process on the third answer feature and the third answer feature to obtain the second attention information. The first feature calculation network is used to calculate the product of the second attention information and the third answer feature to obtain the fourth answer feature.

[0085] Furthermore, the first feature fusion network may be used to fuse the first text feature, the third answer feature, and the fourth answer feature to obtain a first fused feature. The fusion process here may be to add or average the feature vectors corresponding to each feature in the same dimension, or to concatenate the feature vectors corresponding to each feature. In this embodiment, the first text feature, the third answer feature, and the fourth answer feature may be concatenated to obtain a first fused feature.

[0086] Then, the first fusion feature can be input into the first classification network and the second classification network for classification, and the second correction conclusion and the second correction score can be output. The first classification network can be a binary classification model, and the output is the correction conclusion of this question, that is, whether this question is right or wrong; the second classification network can be a regression model, and the output is the correction score of this question, that is, the predicted score of this question.

[0087] Then, according to the second correction conclusion and the second correction score output by the correction result evaluation model, the loss value is calculated in combination with the label data, that is, the first correction conclusion and the first correction score. Among them, the setting of the loss function can be the cross entropy between the calculation output and the label. According to the calculated loss value, gradient back propagation is performed to adjust the model parameters of the correction result evaluation model, specifically, the parameters of the first convolutional neural network, the second convolutional neural network, the third convolutional neural network, the first classification network, and the second classification network. Then, based on all the first training sample data in the first training sample data set, the correction result evaluation model is iteratively trained until the first preset iteration stop condition is reached, the training of the correction result evaluation model is completed, and the trained correction result evaluation model is obtained.

[0088] It can be understood that the model structure of the above-mentioned grading result evaluation model is only an example, and the model structure of the grading result evaluation model may also have other model structures.

[0089] Among them, completing the training of the grading result evaluation model also completes the training of its sub-model, namely the test question mastery degree estimation model.

[0090] In some embodiments, the first classification network includes a first fully connected network and a first classification module, the second classification network includes a second fully connected network and a second classification module, the first fusion feature is input to the first classification network to obtain an output second correction conclusion, and the first fusion feature is input to the second classification network to obtain an output second correction score, including:

[0091] Inputting the first fusion feature into the first fully connected network to obtain the correction conclusion feature;

[0092] The first classification module is used to classify and calculate the features of the correction conclusion to obtain the second correction conclusion;

[0093] Input the first fusion feature into the second fully connected network to obtain the correction score feature;

[0094] The second classification module is used to classify and calculate the correction score features to obtain the second correction score.

[0095] In this embodiment, the first classification network and the second classification network can be composed of a fully connected network and a classification module, wherein the fully connected network can be a multi-layer perceptron or a deep neural network (Deep Neural Networks, DNN), and the classification module can implement a classification function, and the classification function can be a Softmax function. Figure 5 As shown, it is a schematic diagram of the structure of the first classification network in this application; Figure 6 The figure shows a schematic diagram of the structure of the second classification network in this application.

[0096] Specifically, the first classification network may include a first fully connected network and a first classification module, and the second classification network may include a second fully connected network and a second classification module. In specific processing, the first fusion feature may be first input into the first fully connected network to obtain the correction conclusion feature, and then the first classification module may be used to classify and calculate the correction conclusion feature to obtain the second correction conclusion. Also, the first fusion feature may be input into the second fully connected network to obtain the correction score feature, and then the second classification module may be used to classify and calculate the correction score feature to obtain the second correction score.

[0097] In S330, the test question attribute data of the reference test question is obtained.

[0098] Among them, the test question attribute data of the reference test questions is data related to the attributes of the reference test questions themselves, and may specifically include but are not limited to the knowledge point data of the reference test questions, the question type data of the reference test questions, the test question text data of the reference test questions, etc.

[0099] In S340, based on the test question attribute data and the mastery level data, a target test question matching the reference test question is searched from the candidate test questions.

[0100] Among them, in this embodiment, after obtaining the target object's mastery data of the reference test questions and the test question attribute data of the reference test questions themselves, test questions can be further searched among the candidate test questions based on the mastery data and the test question attribute data of the reference test questions themselves, and the target candidate test questions that match the candidate test questions among the candidate test questions are determined, that is, the target test questions that match the reference test questions.

[0101] Specifically, the target subject's mastery data of the reference test questions and the test question attribute data of the reference test questions may be matched with the mastery data and the test question attribute data of each candidate test question, and the target test question may be determined based on the matching result.

[0102] In some embodiments, searching for a target test question matching a reference test question from candidate test questions according to the test question attribute data and the mastery level data includes:

[0103] Generate the question data of the reference test question according to the second text data and the first knowledge point data included in the test question attribute data;

[0104] Generate basic attribute data of the reference test question according to the first knowledge point data and the question type data included in the test question attribute data;

[0105] According to the mastery degree data, question data and basic attribute data, the target test questions that match the reference test questions are searched from the candidate test questions.

[0106] In this embodiment, the question attribute data of the reference question may include the second text data, the first knowledge point data and the question type data of the reference question. In this embodiment, the attribute data of the question may be upgraded to obtain the question surface data and basic attribute data of the question. Specifically, the upgraded question surface data may be generated based on the second text data and the first knowledge point data in the question attribute data. Further, the basic attribute data of the reference question may be generated based on the first knowledge point data and the question type data in the question attribute data. Then, a question search is further performed from the candidate questions based on the mastery degree data, the question surface data and the basic attribute data to obtain a target question that matches the reference question.

[0107] Specifically, the basic attribute data of the reference test question is generated according to the first knowledge point data and the question type data in the test question attribute data, which can be obtained by combining the knowledge point description text and the question type description text of the reference test question, and then performing word embedding processing on the combined text to obtain the basic attribute data of the reference test question. The basic attribute data can be a word feature vector, and the word embedding processing method has been introduced in the above embodiment, which will not be repeated here.

[0108] In some embodiments, generating the question data of the reference test question according to the second text data and the first knowledge point data included in the test question attribute data includes:

[0109] Performing word embedding processing on the second text data included in the test question attribute data by using a word embedding model to obtain a second text feature;

[0110] The second text feature and the first knowledge point data included in the test question attribute data are input into the question data generation model to obtain the question data of the reference test question.

[0111] In this embodiment, a pre-trained question data generation model can be used to generate the question data of the reference test questions. Specifically, the second text data of the reference test questions can be firstly processed by word embedding to obtain the second text feature, and then the second text feature and the first knowledge point data are input into the aforementioned trained question data generation model to generate the question data of the reference test questions. Among them, the word embedding processing here can also be performed by using any one of the BERT model or the ELMo model, which will not be described in detail here.

[0112] In some embodiments, the question data generation model includes a second feature calculation network, a second feature fusion network, a third fully connected network, and a fourth fully connected network. The second text feature and the question knowledge point data included in the question attribute data are input into the question data generation model to obtain the question data of the reference question, including:

[0113] Inputting the first knowledge point data into the third fully connected network to obtain a first knowledge point feature set;

[0114] Using the second feature calculation network to calculate the average value of all knowledge point features in the first knowledge point feature set to obtain a first average knowledge point feature;

[0115] The second feature fusion network is used to fuse the second text feature with the first average knowledge point feature, and the fused feature is input into the fourth fully connected network to obtain the question data of the reference test question.

[0116] In this embodiment, a possible structure of the aforementioned question data generation model is provided. Specifically, Figure 7 As shown, it is a schematic diagram of the structure of the question data generation model provided by the present application. In this structure, the question data generation model can be composed of a third fully connected network, a second feature calculation network, a second feature fusion network and a fourth fully connected network. Among them, the third fully connected network and the fourth fully connected network can be DNN networks or multi-layer perceptrons.

[0117] Specifically, according to the question data generation model of this structure, the question data of the reference test questions can be generated according to the following logic: first, the first knowledge point data of the reference test questions is input into the third fully connected network to obtain the first knowledge point feature set. Specifically, each knowledge point can correspond to a knowledge point feature. When the reference test questions involve multiple knowledge points, multiple knowledge point features can be output to obtain the first knowledge point set.

[0118] Then, the second feature calculation network is used to calculate the average value of all knowledge point features in the first knowledge point feature set, and the first average knowledge point feature can be obtained. Then, the second feature fusion network is used to fuse the second text feature with the first average knowledge point feature obtained previously. Among them, the fusion of the second text feature and the first average knowledge point feature here can be to add or average the vectors corresponding to the features at the same latitude, or to splice the feature vectors corresponding to the second text feature and the first average knowledge point feature. Then, the fused features are further input into the fourth fully connected network, and the surface data of the reference test questions can be obtained.

[0119] In some embodiments, before inputting the second text feature and the first knowledge point data included in the question attribute data into the question surface data generation model to obtain the question surface data of the reference question, the method further includes:

[0120] Obtain a second training sample data set of a test question similarity evaluation model, the second training sample data set including a plurality of second training sample data, the second training sample data including third text data of a second sample test question, second knowledge point data of the second sample test question, fourth text data of a third sample test question, third knowledge point data of the third sample test question, and first similarity data between the second sample test question and the third sample test question;

[0121] A test question similarity assessment model is trained with the third text feature corresponding to the third text data, the second knowledge point data, the fourth text feature corresponding to the fourth text data, and the third knowledge point data as input and the first similarity data as output.

[0122] Before using the question data generation model to generate the question data of the reference test questions, it is necessary to first train the question data generation model, and then use the trained question data generation model to generate the question data of the reference test questions. In the present embodiment, a method of training a question similarity assessment model including the aforementioned question data generation model is specifically adopted to indirectly train the question data generation model. That is, in the present embodiment, the question data generation model is a sub-model of the question similarity assessment model. In the process of training the question similarity assessment model, the model parameters of the question data generation model can be trained at the same time, so that while completing the training of the question similarity assessment model, the training of the question data generation model can also be completed.

[0123] Specifically, before training the test question similarity evaluation model in this embodiment, it is necessary to first obtain sample data for training the test question similarity evaluation model. The sample data may be a second training sample data set including a plurality of second training sample data, wherein the second training sample data includes a pair of sample test questions, specifically a second sample test question and a third sample test question. Specifically, the second training sample data includes the third text data of the second sample test question, the second knowledge point data of the second sample test question, the fourth text data of the third sample test question, the third knowledge point data of the third sample test question, and the first similarity data between the second sample test question and the third sample test question.

[0124] Then, the test question similarity evaluation model is trained according to the above sample data. Specifically, the input data and output data for training the test question similarity evaluation model are first determined in the sample data. Specifically, the third text feature corresponding to the third text data, the second knowledge point data, the fourth text feature corresponding to the fourth text data, and the third knowledge point data can be used as input, and the first similarity data can be used as output to train the test question similarity evaluation model.

[0125] In some embodiments, the test question similarity assessment model includes a test data generation model, a third feature calculation network, a fourth feature calculation network, a third feature fusion network, and a third classification network. The test data generation model includes a second feature calculation network, a second feature fusion network, a third fully connected network, and a fourth fully connected network. The test question similarity assessment model is trained with the third text feature corresponding to the third text data, the second knowledge point data, the fourth text feature corresponding to the fourth text data, and the third knowledge point data as inputs and the first similarity data as output, including:

[0126] Inputting the second knowledge point data and the third knowledge point data into the third fully connected network respectively, and obtaining the output second knowledge point feature set and the third knowledge point feature set;

[0127] Using the second feature calculation network to respectively calculate the average values ​​of the knowledge point features in the second knowledge point feature set and the third knowledge point feature set, to obtain a second average knowledge point feature and a third average knowledge point feature;

[0128] The third text feature and the second average knowledge point feature are fused using a second feature fusion network to obtain a second fused feature;

[0129] The fourth text feature and the third average knowledge point feature are fused using the second feature fusion network to obtain a third fused feature;

[0130] Inputting the second fused feature and the third fused feature into the fourth fully connected network respectively, to obtain the first question feature and the second question feature as output;

[0131] The third feature calculation network is used to calculate the difference between the first question feature and the second question feature to obtain the difference feature;

[0132] The fourth feature calculation network is used to calculate the product of the first question feature and the second question feature to obtain a product feature;

[0133] The third feature fusion network is used to fuse the difference feature and the product feature to obtain the fourth fusion feature;

[0134] Inputting the fourth fusion feature into the third classification network to obtain output second similarity degree data;

[0135] The model parameters of the test question similarity assessment model are iteratively updated according to the first similarity data and the second similarity data until a second preset iteration stop condition is reached, thereby completing the training of the test question similarity assessment model.

[0136] In this embodiment, a possible model structure of the above-mentioned test question similarity evaluation model is provided. In this model structure, in addition to the second feature calculation network, the second feature fusion network, the third fully connected network and the fourth fully connected network included in the above-mentioned test question data generation model, there are also a third feature calculation network, a fourth feature calculation network, a third feature fusion network and a third classification network. Figure 8 As shown, it is a structural diagram of the test question similarity evaluation model provided in this application.

[0137] The specific process of training the test question similarity evaluation model of the structure may include the following steps: first, the second knowledge point data and the third knowledge point data are respectively input into the third fully connected network to obtain the output second knowledge point feature set and the third knowledge point feature set. Then, the second feature calculation network is used to calculate the average values ​​of the knowledge point features in the second knowledge point feature set and the third knowledge point feature set, respectively, to obtain the second average knowledge point feature and the third average knowledge point feature. The second feature fusion network is further used to fuse the third text feature with the second average knowledge point feature to obtain the second fused feature. Here, the feature fusion can be the addition or averaging of the corresponding dimensions of the feature vectors corresponding to the third text feature and the second average knowledge point feature, or it can be the splicing of the feature vectors corresponding to the third text feature and the second average knowledge point feature to obtain the second fused feature. Then, the second feature fusion network is used to fuse the fourth text feature with the third average knowledge point feature to obtain the third fused feature; the method for fusing the fourth text feature with the third average knowledge point feature here can be consistent with the aforementioned method for fusing the third text feature with the second average knowledge point feature. Furthermore, the second fused feature and the third fused feature are respectively input into the fourth fully connected network to obtain the output first question feature and the second question feature. As mentioned above, the fourth fully connected network here can be a DNN. Then, the third feature calculation network is used to calculate the difference between the first question feature and the second question feature to obtain the difference feature; and the fourth feature calculation network is used to calculate the product of the first question feature and the second question feature to obtain the product feature. Then, the third feature fusion network is used to fuse the calculated difference feature and product feature to obtain the fourth fused feature, and the feature fusion here can be feature splicing. Finally, the fourth fused feature is input into the third classification network to obtain the output second similarity data.

[0138] After obtaining the second similarity data output by the question similarity assessment model, the model parameters of the question similarity assessment model can be further iteratively updated according to the first similarity data and the second similarity data until the second preset iteration stop condition is reached, thereby completing the training of the question similarity assessment model. Specifically, the process of iteratively updating the model parameters of the question similarity assessment model includes iteratively updating the parameters of the third fully connected network and the fourth fully connected network, that is, iteratively updating the model parameters of the question surface data generation model until the preset iteration stop condition is met, thereby completing the training of the question surface data generation model.

[0139] In some embodiments, searching for a target test question matching a reference test question from candidate test questions based on the mastery level data, the question data, and the basic attribute data includes:

[0140] Obtaining candidate test question mastery degree data, candidate test question surface data and candidate test question basic attribute data for each candidate test question;

[0141] Calculate the similarity between the reference test question and each candidate test question based on the mastery degree data, the question surface data, the basic attribute data, the candidate test question mastery degree data, the candidate test question surface data, and the candidate test question basic attribute data;

[0142] The target test questions that match the reference test questions are searched from the candidate test questions according to the similarity.

[0143] In this embodiment, matching data extraction can be performed on all candidate test questions in the candidate test question database in advance, and specifically, the mastery degree data, the candidate test question surface data and the candidate test question basic attribute data of each candidate test question can be extracted. Then, the similarity between the reference test question and each candidate test question is calculated based on the mastery degree data, the surface data and the basic attribute data of the reference test question and the mastery degree data, the candidate test question surface data and the candidate test question basic attribute data of each candidate test question.

[0144] Specifically, the feature vector corresponding to the mastery degree data of the target candidate test question, the feature vector corresponding to the question data, and the feature vector corresponding to the basic attribute data can be first obtained, and then the three feature vectors are weighted to obtain the feature vector of the target candidate test question. Similarly, the feature vector corresponding to the mastery degree data, question data, and basic attribute data of the reference test question can be weighted to obtain the feature vector of the reference test question. Among them, the weight coefficient of the weighting can be specifically adjusted according to the actual situation.

[0145] Then, the cosine similarity of the feature vector of the target candidate question and the feature vector of the reference question is calculated to obtain the similarity between the target candidate question and the reference question. Then, each candidate question is traversed to obtain the similarity between each candidate question and the reference question. Finally, the candidate question with the greatest similarity is determined as the target question based on the ranking of the similarity values.

[0146] In some embodiments, obtaining the candidate test question mastery degree data, candidate test question surface data and candidate test question basic attribute data of each candidate test question includes:

[0147] Obtaining candidate test question text data, candidate test question knowledge point data, candidate test question type data, candidate test question standard answer data and answer data for each candidate test question;

[0148] Generate basic attribute data of each candidate test question according to the knowledge point data of each candidate test question and the question type data of the candidate test question;

[0149] Inputting candidate test question text data and candidate test question knowledge point data of each candidate test question into a question surface data generation model to obtain candidate test question surface data of each candidate test question;

[0150] The candidate test question standard answer data of each candidate test question and the answer data of each candidate test question are input into the test question mastery degree evaluation model to obtain the test question mastery degree data of each candidate test question.

[0151] Among them, in this embodiment, to obtain the mastery level data, question data and basic attribute data of each candidate question, the candidate question text data, candidate question knowledge point data, candidate question type data, candidate question standard answer data and answer data for each candidate question can be obtained first, and then the mastery level data and question data of each candidate data can be obtained respectively according to the aforementioned trained test question mastery level estimation model and question data generation model.

[0152] In some embodiments, the candidate test question standard answer data of each candidate test question and the answer data for each candidate test question are input into the test question mastery level assessment model to obtain the test question mastery level data of each candidate test question, including:

[0153] Classifying the answer data for each candidate test question into a plurality of categories of candidate test question answer data according to attribute information of the answer object;

[0154] Inputting candidate test question standard answer data of each candidate test question and multiple categories of candidate test question answer data into a test question mastery level assessment model to obtain multiple test question mastery level data of each candidate test question;

[0155] The similarity between the reference test question and each candidate test question is calculated based on the mastery degree data, the question data, the basic attribute data, the candidate test question mastery degree data, the candidate test question question data and the candidate test question basic attribute data, including:

[0156] Get the target category of the target object;

[0157] determining target test question mastery level data from a plurality of test question mastery level data of each candidate test question according to the target category;

[0158] The similarity between the reference test question and each candidate test question is calculated based on the mastery level data, test question surface data, basic attribute data, target test question mastery level data, candidate test question surface data and candidate test question basic attribute data.

[0159] In this embodiment, there may be multiple answer data for each candidate test question, for example, one candidate test question may have 100 answer data. At this time, these answer data may be divided according to the academic performance of the answering object, for example, the academic performance of the answering object may be divided into good academic performance, medium academic performance and poor academic performance. The specific classification standard for the academic performance of the answering object may be specifically set as needed. Then the multiple answer data of the candidate test question may also be divided into three categories, for example, there are 20 answer data corresponding to the answering object with good academic performance, 50 answer data corresponding to the answering object with medium academic performance, and 30 answer data corresponding to the answering object with poor academic performance.

[0160] Since a mastery degree data can be generated for each answer data according to the test question mastery degree estimation model, then the mastery degree data corresponding to 20 answering objects with good academic performance, the mastery degree data corresponding to 50 answering objects with medium academic performance, and the mastery degree data corresponding to 30 answering objects with poor academic performance can be obtained. For the multiple mastery degree data corresponding to each type of answering object, its average value can be further calculated to obtain an average mastery degree data corresponding to each type of answering object. Since the mastery degree data can be a feature vector, that is, multiple feature vectors are averaged and pooled in the corresponding dimension to obtain a mastery degree data corresponding to each type of answering object.

[0161] Thus, when searching for test questions, the category of the target object can be determined first, for example, the target object can be determined to be a person with good academic performance. Then, according to the category of the target object, the mastery level data corresponding to the category of the target object can be determined from the multiple mastery level data corresponding to the candidate test questions as the mastery level data of the candidate test questions used for test question matching.

[0162] In this way, the accuracy of the test question search results can be improved by stratifying the mastery level data of the candidate test questions according to the categories of the answer objects.

[0163] In some embodiments, the test question search method provided by the present application may also include:

[0164] Push the target test questions to the target objects.

[0165] That is, in this embodiment, after searching for the target question that best matches the reference question from the candidate questions, the target question can be automatically pushed to the target object, for example, displayed in the terminal where the target object logs in, so that the target object can conduct consolidation exercises.

[0166] According to the above description, the test question search method provided by the present application obtains the target object's answer data for the reference test question; evaluates the target object's mastery of the reference test question based on the answer data to obtain mastery data; obtains the test question attribute data of the reference test question; and searches for the target test question that matches the reference test question from the candidate test questions based on the test question attribute data and the mastery data. Compared with the related art, the present application searches for test questions based on the target object's answer data for the reference test question and the test question attribute data of the reference test question itself, which fully considers the target object's mastery of the test question, thereby greatly improving the accuracy of the test question search.

[0167] Please refer to Fig. 9 In order to better implement the test question search method provided by this application, this application further provides a test question search device 400, such as Fig. 9 As shown, the test question searching device 400 comprises:

[0168] The first acquisition module 410 is used to acquire the answer data of the target subject to the reference test questions;

[0169] An evaluation module 420 is used to evaluate the mastery degree of the target subject for the reference test questions according to the answer data to obtain mastery degree data;

[0170] The second acquisition module 430 is used to acquire the test question attribute data of the reference test question;

[0171] The search module 440 is used to search for a target test question that matches the reference test question from the candidate test questions according to the test question attribute data and the mastery level data.

[0172] In an optional embodiment, the answer data includes first answer result data and first standard answer data, and the evaluation module 420 is used to:

[0173] Using a word embedding model, word embedding processing is performed on the first answer result data and the first standard answer data to obtain a first answer feature and a first answer feature;

[0174] The first response feature and the first answer feature are input into the test question mastery evaluation model to obtain the target subject's mastery data for the reference test question.

[0175] In an optional embodiment, the test question mastery level assessment model includes a first convolutional neural network, a second convolutional neural network and an attention network, and the assessment module 420 is further used to:

[0176] The first convolutional neural network is used to perform feature sampling on the first answer feature to obtain a second answer feature; the second convolutional neural network is used to perform feature sampling on the first answer feature to obtain a second answer feature;

[0177] An attention network is used to perform attention calculation on the second answer feature and the second answer feature to obtain first attention information;

[0178] The target subject's mastery of the reference test questions is determined based on the first attention information.

[0179] In an optional embodiment, the first attention information includes a weight matrix, and the evaluation module 420 is further configured to:

[0180] Performing pooling processing on the weight matrix according to the feature dimension of the second answer feature to obtain a weight vector;

[0181] The weight vector is determined as the target subject's mastery degree data on the reference test questions.

[0182] In an optional embodiment, the evaluation module 420 is further configured to:

[0183] Performing text cleaning on the first answer result data to obtain a cleaned answer result text;

[0184] Performing text cleaning on the first standard answer data to obtain a cleaned standard answer text;

[0185] Segment the answer result text and the standard answer text to obtain the answer result word sequence and the standard answer word sequence;

[0186] The word embedding model is used to perform word embedding processing on the answer result word sequence and the standard answer word sequence respectively to obtain the first answer feature and the first answer feature.

[0187] In an optional embodiment, the test question search device provided by the present application further includes a first training module, which is used to:

[0188] Obtaining a first training sample data set of a correction result evaluation model, the first training sample data set comprising a plurality of first training sample data, the first training sample data comprising first text data of a first sample test question, second standard answer data of the first sample test question, second answer result data for the first sample test question, and correction data for the second answer result data, the correction data comprising a first batch correction conclusion and a first batch correction score;

[0189] The features corresponding to the first text data, the second standard answer data and the second answer result data in each first training sample data are used as input, and the correction data of the second answer result data is used as output to train the correction result evaluation model.

[0190] In an optional embodiment, the grading result evaluation model includes a test question mastery evaluation model, a third convolutional neural network, a first feature fusion network, a first feature calculation network, a first classification network, and a second classification network. The test question mastery evaluation model includes a first convolutional neural network, a second convolutional neural network, and an attention network. The first training module is also used for:

[0191] The first convolutional neural network is used to sample the word embedding features corresponding to the second answer result data to obtain a third answer feature, the second convolutional neural network is used to sample the word embedding features corresponding to the second standard answer data to obtain a third answer feature, and the third convolutional neural network is used to sample the word embedding features corresponding to the first text data to obtain a first text feature;

[0192] Using the attention network to perform attention calculation on the third answer feature and the third answer feature to obtain second attention information;

[0193] The first feature calculation network is used to calculate the product of the second attention information and the third answer feature to obtain a fourth answer feature;

[0194] The first feature fusion network is used to fuse the first text feature, the third answer feature, and the fourth answer feature to obtain a first fusion feature;

[0195] Inputting the first fusion feature into the first classification network to obtain the output second batch correction conclusion, and inputting the training splicing feature into the second classification network to obtain the output second batch correction score;

[0196] The model parameters of the correction result evaluation model are iteratively updated according to the first batch correction conclusions, the first batch correction scores, the second batch correction conclusions and the second batch correction scores until a first preset iteration stop condition is reached, thereby completing the training of the correction result evaluation model.

[0197] In an optional embodiment, the first classification network includes a first fully connected network and a first classification module, the second classification network includes a second fully connected network and a second classification module, and the first training module is further used for:

[0198] Inputting the first fusion feature into the first fully connected network to obtain the correction conclusion feature;

[0199] The first classification module is used to classify and calculate the features of the correction conclusion to obtain the second correction conclusion;

[0200] Input the first fusion feature into the second fully connected network to obtain the correction score feature;

[0201] The second classification module is used to classify and calculate the correction score features to obtain the second correction score.

[0202] In an optional embodiment, the search module 440 is further configured to:

[0203] Generate the question data of the reference test question according to the second text data and the first knowledge point data included in the test question attribute data;

[0204] Generate basic attribute data of the reference test question according to the first knowledge point data and the question type data included in the test question attribute data;

[0205] According to the mastery degree data, question data and basic attribute data, the target test questions that match the reference test questions are searched from the candidate test questions.

[0206] In an optional embodiment, the search module 440 is further configured to:

[0207] Performing word embedding processing on the second text data included in the test question attribute data by using a word embedding model to obtain a second text feature;

[0208] The second text feature and the first knowledge point data included in the test question attribute data are input into the question data generation model to obtain the question data of the reference test question.

[0209] In an optional embodiment, the question data generation model includes a second feature calculation network, a second feature fusion network, a third fully connected network, and a fourth fully connected network, and the search module 440 is further used to:

[0210] Inputting the first knowledge point data into the third fully connected network to obtain a first knowledge point feature set;

[0211] Using the second feature calculation network to calculate the average value of all knowledge point features in the first knowledge point feature set to obtain a first average knowledge point feature;

[0212] The second feature fusion network is used to fuse the second text feature with the first average knowledge point feature, and the fused feature is input into the fourth fully connected network to obtain the question data of the reference test question.

[0213] In an optional embodiment, the test question search device provided by the present application further includes a second training module, which is used to:

[0214] Obtain a second training sample data set of a test question similarity evaluation model, the second training sample data set including a plurality of second training sample data, the second training sample data including third text data of a second sample test question, second knowledge point data of the second sample test question, fourth text data of a third sample test question, third knowledge point data of the third sample test question, and first similarity data between the second sample test question and the third sample test question;

[0215] A test question similarity assessment model is trained with the third text feature corresponding to the third text data, the second knowledge point data, the fourth text feature corresponding to the fourth text data, and the third knowledge point data as input and the first similarity data as output.

[0216] In an optional embodiment, the test question similarity evaluation model includes a test data generation model, a third feature calculation network, a fourth feature calculation network, a third feature fusion network and a third classification network, the test data generation model includes a second feature calculation network, a second feature fusion network, a third fully connected network and a fourth fully connected network, and the second training module is further used for:

[0217] Inputting the second knowledge point data and the third knowledge point data into the third fully connected network respectively, and obtaining the output second knowledge point feature set and the third knowledge point feature set;

[0218] Using the second feature calculation network to respectively calculate the average values ​​of the knowledge point features in the second knowledge point feature set and the third knowledge point feature set, to obtain a second average knowledge point feature and a third average knowledge point feature;

[0219] The third text feature and the second average knowledge point feature are fused using a second feature fusion network to obtain a second fused feature;

[0220] The fourth text feature and the third average knowledge point feature are fused using the second feature fusion network to obtain a third fused feature;

[0221] Inputting the second fused feature and the third fused feature into the fourth fully connected network respectively, to obtain the first question feature and the second question feature as output;

[0222] The third feature calculation network is used to calculate the difference between the first question feature and the second question feature to obtain the difference feature;

[0223] The fourth feature calculation network is used to calculate the product of the first question feature and the second question feature to obtain a product feature;

[0224] The third feature fusion network is used to fuse the difference feature and the product feature to obtain the fourth fusion feature;

[0225] Inputting the fourth fusion feature into the third classification network to obtain output second similarity degree data;

[0226] The model parameters of the test question similarity assessment model are iteratively updated according to the first similarity data and the second similarity data until a second preset iteration stop condition is reached, thereby completing the training of the test question similarity assessment model.

[0227] In an optional embodiment, the search module 440 is further configured to:

[0228] Obtaining candidate test question mastery degree data, candidate test question surface data and candidate test question basic attribute data for each candidate test question;

[0229] Calculate the similarity between the reference test question and each candidate test question based on the mastery degree data, the question surface data, the basic attribute data, the candidate test question mastery degree data, the candidate test question surface data, and the candidate test question basic attribute data;

[0230] The target test questions that match the reference test questions are searched from the candidate test questions according to the similarity.

[0231] In an optional embodiment, the search module 440 is further configured to:

[0232] Obtaining candidate test question text data, candidate test question knowledge point data, candidate test question type data, candidate test question standard answer data and answer data for each candidate test question;

[0233] Generate basic attribute data of each candidate test question according to the knowledge point data of each candidate test question and the question type data of the candidate test question;

[0234] Inputting candidate test question text data and candidate test question knowledge point data of each candidate test question into a question surface data generation model to obtain candidate test question surface data of each candidate test question;

[0235] The candidate test question standard answer data of each candidate test question and the answer data of each candidate test question are input into the test question mastery degree evaluation model to obtain the test question mastery degree data of each candidate test question.

[0236] In an optional embodiment, the search module 440 is further configured to:

[0237] Classifying the answer data for each candidate test question into a plurality of categories of candidate test question answer data according to attribute information of the answer object;

[0238] Inputting candidate test question standard answer data of each candidate test question and multiple categories of candidate test question answer data into a test question mastery level assessment model to obtain multiple test question mastery level data of each candidate test question;

[0239] Get the target category of the target object;

[0240] determining target test question mastery level data from a plurality of test question mastery level data of each candidate test question according to the target category;

[0241] The similarity between the reference test question and each candidate test question is calculated based on the mastery level data, test question surface data, basic attribute data, target test question mastery level data, candidate test question surface data and candidate test question basic attribute data.

[0242] In an optional embodiment, the test question search device provided by the present application further includes a push module, which is used to:

[0243] Push the target test questions to the target objects.

[0244] It should be noted that the test question search device 400 provided in the embodiment of the present application belongs to the same concept as the test question search method in the above embodiment, and its specific implementation process is detailed in the above related embodiments and will not be repeated here.

[0245] An embodiment of the present application also provides an electronic device, including a memory and a processor, wherein the processor is used to execute the steps in the test question search method provided in this embodiment by calling a computer program stored in the memory.

[0246] Please refer to Fig.10 , Fig.10 A schematic diagram of the structure of an electronic device 100 provided in an embodiment of the present application.

[0247] The electronic device 100 may include components such as a network interface 110, a memory 120, a processor 130, and a screen assembly. Those skilled in the art will appreciate that Fig.10 The structure of the electronic device 100 shown in the figure does not constitute a limitation on the electronic device 100, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0248] The network interface 110 may be used to establish a network connection between devices.

[0249] The memory 120 may be used to store computer programs and data. The computer programs stored in the memory 120 include executable codes. The computer programs may be divided into various functional modules. The processor 130 executes various functional applications and data processing by running the computer programs stored in the memory 120.

[0250] The processor 130 is the control center of the electronic device 100. It uses various interfaces and lines to connect various parts of the entire electronic device 100, and executes various functions of the electronic device 100 and processes data by running or executing computer programs stored in the memory 120 and calling data stored in the memory 120, thereby controlling the electronic device 100 as a whole.

[0251] In this embodiment, the processor 130 in the electronic device 100 will load the executable code corresponding to one or more computer programs into the memory 120 according to the following instructions, and the processor 130 will execute the steps in the test question search method provided by the present application, such as:

[0252] Obtaining the target subject's answer data on the reference test questions;

[0253] Evaluate the target subject's mastery of the reference test questions based on the answer data to obtain mastery data;

[0254] Obtaining test question attribute data of reference test questions;

[0255] According to the test question attribute data and the mastery degree data, a target test question matching the reference test question is searched from the candidate test questions.

[0256] It should be noted that the electronic device 100 provided in the embodiment of the present application and the test question search method in the above embodiment belong to the same concept, and its specific implementation process is detailed in the above related embodiments and will not be repeated here.

[0257] The present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program stored therein is executed on a processor of an electronic device provided in an embodiment of the present application, the processor of the electronic device executes any of the steps in the above test question search method suitable for an electronic device. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0258] The above is a detailed introduction to a test question search method, device, storage medium and electronic device provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A test question search method, characterized in that: The method comprises: Obtaining the target subject's answer data on the reference test questions; Assessing the target subject's mastery of the reference test questions according to the answer data to obtain mastery degree data; Acquiring test question attribute data of the reference test question; Searching for a target test question matching the reference test question from candidate test questions according to the test question attribute data and the mastery level data; The step of searching for a target test question matching the reference test question from candidate test questions according to the test question attribute data and the mastery level data includes: Generating the question surface data of the reference question according to the second text data and the first knowledge point data included in the question attribute data; said generating the question surface data of the reference question according to the second text data and the first knowledge point data included in the question attribute data comprises: Performing word embedding processing on the second text data included in the test question attribute data by using a word embedding model to obtain a second text feature; Inputting the second text feature and the first knowledge point data included in the test question attribute data into a test question data generation model to obtain the test question data of the reference test question; The question data generation model includes a second feature calculation network, a second feature fusion network, a third fully connected network, and a fourth fully connected network. The second text feature and the question knowledge point data included in the question attribute data are input into the question data generation model to obtain the question data of the reference question, including: Inputting the first knowledge point data into the third fully connected network to obtain a first knowledge point feature set; Using the second feature calculation network to calculate the average value of all knowledge point features in the first knowledge point feature set to obtain a first average knowledge point feature; The second feature fusion network is used to fuse the second text feature with the first average knowledge point feature, and the fused feature is input into the fourth fully connected network to obtain the question data of the reference test question; Generate basic attribute data of the reference test question according to the first knowledge point data and question type data included in the test question attribute data; According to the mastery degree data, the question data and the basic attribute data, a target question matching the reference question is searched from candidate questions.

2. The method according to claim 1, characterized in that: The answer data includes first answer result data and first standard answer data, and the mastery degree of the target object for the reference test question is evaluated according to the answer data to obtain the mastery degree data, including: Perform word embedding processing on the first answer result data and the first standard answer data respectively through a word embedding model to obtain a first answer feature and a first answer feature; The first response feature and the first answer feature are input into a test question mastery level assessment model to obtain the target subject's mastery level data for the reference test question.

3. The method according to claim 2, characterized in that The test question mastery degree assessment model includes a first convolutional neural network, a second convolutional neural network and an attention network. The first answer feature and the first answer feature are input into the test question mastery degree assessment model to obtain the target subject's mastery degree data for the reference test question, including: Using the first convolutional neural network to perform feature sampling on the first answer feature to obtain a second answer feature; Using the second convolutional neural network to perform feature sampling on the first answer feature to obtain a second answer feature; Using the attention network to perform attention calculation on the second response feature and the second answer feature to obtain first attention information; The target object's mastery of the reference test questions is determined based on the first attention information.

4. The method according to claim 3, characterized in that The first attention information includes a weight matrix, and determining the mastery degree data of the target subject on the reference test question according to the first attention information includes: The weight matrix is ​​pooled according to the feature dimension of the second answer feature to obtain a weight vector; and the weight vector is determined as the mastery degree data of the target subject on the reference test question.

5. The method according to claim 2, characterized in that: The word embedding model is used to perform word embedding processing on the first answer result data and the first standard answer data to obtain a first answer feature and a first answer feature, including: Performing text cleaning on the first answer result data to obtain a cleaned answer result text; Performing text cleaning on the first standard answer data to obtain a cleaned standard answer text; Segmenting the answer result text and the standard answer text to obtain an answer result word sequence and a standard answer word sequence; The answer result word sequence and the standard answer word sequence are respectively subjected to word embedding processing by a word embedding model to obtain a first answer feature and a first answer feature.

6. The method according to claim 2, characterized in that Before inputting the first answer feature and the first answer feature into a test question mastery degree evaluation model to obtain the target subject's mastery degree data of the reference test question, the method further includes: Acquire a first training sample data set of a correction result evaluation model, wherein the correction result evaluation model includes the test question mastery degree evaluation model, wherein the first training sample data set includes a plurality of first training sample data, wherein the first training sample data includes first text data of a first sample test question, second standard answer data of the first sample test question, second answer result data for the first sample test question, and correction data for the second answer result data, wherein the correction data includes a first batch correction conclusion and a first batch correction score; The features corresponding to the first text data, the second standard answer data and the second answer result data in each first training sample data are used as input, and the correction data of the second answer result data is used as output to train the correction result evaluation model.

7. The method according to claim 1, characterized in that Before inputting the second text feature and the first knowledge point data included in the test question attribute data into the test question data generation model to obtain the test question data of the reference test question, the method further includes: Obtain a second training sample data set for a test question similarity evaluation model, wherein the second training sample data set includes a plurality of second training sample data, and the second training sample data includes third text data of a second sample test question, second knowledge point data of the second sample test question, fourth text data of a third sample test question, third knowledge point data of the third sample test question, and first similarity data between the second sample test question and the third sample test question; The test question similarity assessment model is trained with the third text feature corresponding to the third text data, the second knowledge point data, the fourth text feature corresponding to the fourth text data, and the third knowledge point data as input and the first similarity data as output.

8. The method according to claim 1, characterized in that The step of searching for a target test question matching the reference test question from candidate test questions according to the mastery degree data, the question surface data and the basic attribute data comprises: Obtaining candidate test question mastery degree data, candidate test question surface data and candidate test question basic attribute data for each candidate test question; Calculating the similarity between the reference test question and each candidate test question based on the mastery degree data, the question surface data, the basic attribute data, the candidate test question mastery degree data, the candidate test question surface data and the candidate test question basic attribute data; A target test question matching the reference test question is searched from the candidate test questions according to the similarity.

9. The method according to claim 8, characterized in that The step of obtaining the candidate test question mastery degree data, candidate test question surface data and candidate test question basic attribute data of each candidate test question includes: Obtaining candidate test question text data, candidate test question knowledge point data, candidate test question type data, candidate test question standard answer data and answer data for each candidate test question; Generate basic attribute data of each candidate test question according to the knowledge point data of each candidate test question and the question type data of the candidate test question; Inputting the candidate test question text data and the candidate test question knowledge point data of each candidate test question into the question surface data generation model to obtain the candidate test question question surface data of each candidate test question; The candidate test question standard answer data of each candidate test question and the answer data of each candidate test question are input into the test question mastery degree evaluation model to obtain the test question mastery degree data of each candidate test question.

10. The method according to claim 9, characterized in that The step of inputting the candidate test question standard answer data of each candidate test question and the answer data of each candidate test question into the test question mastery degree evaluation model to obtain the test question mastery degree data of each candidate test question includes: Classifying the answer data for each candidate question into a plurality of categories of candidate question answer data according to attribute information of the answer object; Inputting the candidate test question standard answer data of each candidate test question and the candidate test question answer data of the multiple categories into the test question mastery level assessment model to obtain multiple test question mastery level data of each candidate test question; The calculating the similarity between the reference test question and each candidate test question according to the mastery degree data, the question surface data, the basic attribute data, the candidate test question mastery degree data, the candidate test question surface data and the candidate test question basic attribute data comprises: Obtaining a target category of the target object; Determining target test question mastery level data from a plurality of test question mastery level data of each candidate test question according to the target category; The similarity between the reference question and each candidate question is calculated based on the mastery level data, the question surface data, the basic attribute data, the target question mastery level data, the candidate question surface data and the candidate question basic attribute data.

11. The method according to any one of claims 1 to 10, characterized in that The method further comprises: The target test question is pushed to the target object.

12. A test question search device, characterized in that: include: The first acquisition module is used to obtain the target subject's answer data to the reference test questions; An evaluation module, used to evaluate the mastery degree of the target subject for the reference test questions according to the answer data, and obtain mastery degree data; A second acquisition module, used for test question attribute data of the reference test question; A search module, for searching for a target test question matching the reference test question from candidate test questions according to the test question attribute data and the mastery level data; The search module is also used to: Generate the question data of the reference test question according to the second text data and the first knowledge point data included in the test question attribute data; Generate basic attribute data of the reference test question according to the first knowledge point data and question type data included in the test question attribute data; Searching for a target test question matching the reference test question from candidate test questions according to the mastery degree data, the question surface data and the basic attribute data; The search module is also used to: Performing word embedding processing on the second text data included in the test question attribute data by using a word embedding model to obtain a second text feature; Inputting the second text feature and the first knowledge point data included in the test question attribute data into a test question data generation model to obtain the test question data of the reference test question; The question data generation model includes a second feature calculation network, a second feature fusion network, a third fully connected network and a fourth fully connected network, and the search module is further used for: Inputting the first knowledge point data into the third fully connected network to obtain a first knowledge point feature set; Using the second feature calculation network to calculate the average value of all knowledge point features in the first knowledge point feature set to obtain a first average knowledge point feature; The second text feature is fused with the first average knowledge point feature using the second feature fusion network, and the fused feature is input into the fourth fully connected network to obtain the question data of the reference test question.

13. A storage medium having a computer program stored thereon, characterized in that: When the computer program is loaded by the processor, the steps in the test question searching method as described in any one of claims 1 to 11 are executed.

14. An electronic device comprising a processor and a memory, wherein the memory stores a computer program, characterized in that: The processor is used to execute the steps in the test question searching method as described in any one of claims 1 to 11 by loading the computer program.

Images