Computer adaptive testing method, system, electronic device and storage medium

Abstract

The present application relates to the technical field of computer application, and provides a computer adaptive test method and system, an electronic device and a storage medium, which first determines the current ability information of a target subject for a target subject; then inputs the historical answering information and the current ability information of the target subject for the target subject into a question selection model to obtain test resources output by the question selection model. When the basic model is trained to obtain the question selection model, a first loss function is introduced to ensure that the knowledge coverage rate of the test resources output by the question selection model is overall for the target subject, and a second loss function is introduced to ensure that the knowledge point correlation of the test resources output by the question selection model is overall small, so that the credibility of the test resources can be improved, the target subject can be comprehensively tested, the test efficiency can be improved, the subsequent test result is more accurate, and accurate recommendation of learning resources can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer application, and particularly relates to a computer adaptive testing method and system, an electronic device and a storage medium. BACKGROUND

[0002] With the gradual popularization of intelligent learning terminals and the rise of the concept of big data precise teaching, improving student learning efficiency through artificial intelligence technology has become an important part of many school teaching and personal improvement scenarios. Because students with different learning conditions and different ability levels have large differences in learning resources and learning modes, personalized learning systems focus on how to more accurately depict student learning conditions and how to recommend more appropriate learning resources for students with different learning conditions, thereby reducing ineffective practice and reducing the burden and increasing efficiency.

[0003] Before learning resource recommendation, an important step is to test the ability of each student through a certain amount of test resources (such as test papers or test question sets, etc.). For online learning systems and intelligent learning terminals, pre-learning ability testing plays an irreplaceable role in accumulating initial student answer information, positioning student ability levels, and providing students with reliable and interpretable diagnostic results.

[0004] In order to improve test efficiency and reduce student burden, computer adaptive testing (CAT) has attracted more and more attention from academia and industry. Computer adaptive testing is a new form of testing that uses item response theory to build a question bank and uses a computer to automatically generate customized test resources for students based on their ability levels, using computer and educational big data technology, to ultimately estimate the ability of the student.

[0005] In the process of computer adaptive testing, the initial prediction of the student's ability is determined before the test begins, and the initial mastery of the student at each knowledge point or test question is understood. The initial prediction is used to maximize the information content during the test, and then the test question with the maximum information content is output as the test resource. Although the information content is more effective when the ability estimate is close to the true ability, the effectiveness is difficult to guarantee when the ability estimate deviates from the true ability. This means that when there are fewer historical answer records, the information content has considerable limitations. In addition, the information content lacks interpretability, and it is difficult to abstract the understandable selection logic for the test process directly facing teachers or students, which reduces the credibility of the test resources recommended to students when put into actual use.

[0006] Therefore, there is an urgent need to provide a computer adaptive testing method. SUMMARY

[0007] The application provides a self-adaptive test method, system, electronic device and storage medium to solve the defects in the prior art.

[0008] The application provides a computer self-adaptive test method, comprising:

[0009] determining current ability information of a target subject for a target subject;

[0010] inputting historical answering information of the target subject for the target subject and the current ability information into a question selection model to obtain test resources output by the question selection model;

[0011] The question selection model is obtained by training a basic model based on a first training sample, and the first training sample comprises a sample of ability information of a first reference subject for the target subject, historical answering information of the first reference subject for the target subject, and a first test resource sample sequence.

[0012] The target loss function used when training the basic model is determined based on at least one of a first loss function and a second loss function, the first loss function is a loss function corresponding to coverage of an output result of the basic model on a knowledge graph of the target subject, and the second loss function is a loss function corresponding to a knowledge point chain distance of the output result in the knowledge graph.

[0013] According to the computer self-adaptive test method provided by the application, the current ability information of the target subject for the target subject is determined, comprising:

[0014] obtaining current answering information of the target subject for the target subject;

[0015] inputting the current answering information into an academic situation diagnosis model to obtain the current ability information output by the academic situation diagnosis model;

[0016] The academic situation diagnosis model is constructed by training a recurrent neural network model based on a sequence of answering information of the target subject for the target subject; the sequence of answering information comprises a plurality of answered test questions and corresponding answering results of each answered test question, or comprises a plurality of answered test questions, corresponding answering results of each answered test question and knowledge points.

[0017] According to the computer self-adaptive test method provided by the application, the basic model is obtained by training based on the following steps:

[0018] The initial model is trained based on a second training sample to obtain a candidate model; the second training sample comprises a capability information sample of a second reference subject for the target subject, historical answering information of the second reference subject for the target subject, and a second test resource sample sequence carrying a score;

[0019] A test sample is determined, which comprises a capability information sample of a third reference subject for the target subject, historical answering information of the third reference subject for the target subject, and a third test resource sample sequence not carrying a score;

[0020] The capability information sample of the third reference subject for the target subject and the historical answering information of the third reference subject for the target subject are input into the candidate model to obtain a confidence of a test resource prediction sequence output by the candidate model, and based on the confidence, an artificial score of a target test resource prediction result in the test resource prediction sequence is collected;

[0021] Based on the target test resource prediction result and the artificial score thereof, the second training sample is expanded, and based on the expanded second training sample, the candidate model is iteratively trained until the candidate model converges, to obtain the basic model.

[0022] According to the computer adaptive test method provided by the application, each test resource prediction result in the test resource prediction sequence corresponds to a confidence; and the target test resource prediction result is determined based on the following steps:

[0023] Based on the confidence corresponding to each test resource prediction result, a test resource prediction result with a confidence within a preset range or closest to a preset threshold is selected as the target test resource prediction result.

[0024] According to the computer adaptive test method provided by the application, the target loss function is determined based on at least one of the first loss function and the second loss function, and at least one of a third loss function and a fourth loss function;

[0025] The third loss function is a loss function corresponding to an information amount of the output result, and the fourth loss function is a loss function corresponding to an information divergence of the output result.

[0026] According to the computer adaptive test method provided by the application, the target loss function is determined based on the following steps:

[0027] The first loss function, the second loss function, the third loss function and the fourth loss function are weighted and summed to obtain the target loss function.

[0028] According to the computer adaptive test method provided by the application, the historical answering information of the target subject and the current ability information of the target subject are input into a question selection model, and a test resource output by the question selection model is obtained.

[0029] The test resource is displayed to the target subject.

[0030] The target answering result of the target subject to the test resource is received.

[0031] Based on the target answering result, a learning resource is recommended to the target subject.

[0032] The application further provides a computer adaptive test system, comprising:

[0033] An ability determination module is configured to determine current ability information of a target subject for a target subject.

[0034] A question selection module is configured to input historical answering information of the target subject for the target subject and the current ability information into a question selection model to obtain a test resource output by the question selection model.

[0035] The question selection model is obtained by training a basic model based on a first training sample, and the first training sample comprises a sample of ability information of a first reference subject for the target subject, historical answering information of the first reference subject for the target subject and a first test resource sample sequence.

[0036] A target loss function used when the basic model is trained is determined based on at least one of a first loss function and a second loss function, the first loss function is a loss function corresponding to coverage of an output result of the basic model on a knowledge graph of the target subject, and the second loss function is a loss function corresponding to a knowledge point chain distance of the output result in the knowledge graph.

[0037] The application further provides an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the computer adaptive test method of any one of the above when executing the program.

[0038] The application further provides a non-transitory computer readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the computer adaptive test method of any one of the above.

[0039] The application further provides a computer program product comprising a computer program which, when executed by a processor, implements the computer adaptive testing method according to any one of the above.

[0040] The computer adaptive testing method, system, electronic device and storage medium provided by the application first determine the current ability information of a target subject for a target subject; then input the historical answering information and the current ability information of the target subject for the target subject into a question selection model to obtain the testing resources output by the question selection model. When the question selection model is trained from the base model, the first loss function is introduced to ensure that the knowledge coverage rate of the testing resources output by the question selection model as a whole is high, and the second loss function is introduced to ensure that the correlation of the knowledge points of the testing resources output by the question selection model as a whole is low. In this way, the credibility of the testing resources can be improved, the target subject can be comprehensively tested, the testing efficiency can be improved, the subsequent testing results can be more accurate, and the accurate recommendation of learning resources can be realized. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0042] Figure 1 is one of the flowcharts of the computer adaptive testing method provided by the application;

[0043] Figure 2 is a structural diagram of a learning emotion diagnosis model in the computer adaptive testing method provided by the application;

[0044] Figure 3 is a flowchart of determining a base model in the computer adaptive testing method provided by the application;

[0045] Figure 4 is another flowchart of the computer adaptive testing method provided by the application;

[0046] Figure 5 is a structural diagram of the computer adaptive testing system provided by the application;

[0047] Figure 6 is a structural diagram of the electronic device provided by the application. DETAILED DESCRIPTION

[0048] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in conjunction with the drawings in the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0049] When the ability of each student is tested, the traditional teaching system uses a fixed test resource arrangement and learning order, which can meet the use requirements of some scenarios, but usually requires a large number of standard test questions to obtain the basic learning ability of students, and the fixed order of test questions will also cause different degrees of invalid test for students with different ability levels, reduce the test efficiency, and occupy the limited learning and practice time of students.

[0050] Moreover, the traditional test method usually mechanically stratifies the students based on the simple student characteristics such as the score rate or the school examination level, needs to consume more manpower to construct artificial characteristics, and the stratification dimension is limited to the overall answer correctness.

[0051] In terms of test selection strategy, the traditional test method usually selects test questions from the test question bank according to the artificial strategy based on the personal experience of the teaching and research teachers. At the student group level, this kind of method is more in line with the overall teaching trend, but cannot conduct personalized test for different student individuals.

[0052] In order to improve the test efficiency and reduce the burden of students, the computer adaptive test technology is introduced. The computer adaptive test technology usually consists of two parts of cognitive diagnosis model and test selection model. Among them, the cognitive diagnosis model is responsible for determining the initial prediction result of the student ability based on the item response theory or neural cognitive diagnosis model (NeuralCD) before the test starts, and understanding the initial mastery of the student at each knowledge point or test question; the test selection model uses the initial prediction result to maximize the information amount by means of the measurement theory, and then outputs the test questions with the maximum information amount as the test resources by means of the fixed strategy or machine learning model.

[0053] Item Response Theory (OPT) represents a series of psychological models that set various characteristic parameters, such as student ability, test difficulty, test discrimination, and blind guessing accuracy, to predict the probability of a student answering a specific question correctly. Due to its interpretable parameter system, OPT possesses relatively good model interpretability. Neurocognitive diagnostic models, on the other hand, use neural network structures to directly output initial student ability models and knowledge representations from the input sequence of student answer records. Predictive results are generated through the interaction between the student and the knowledge representations (such as dot product). Cognitive diagnostic models have a larger parameter space than OPT models and therefore typically possess stronger static modeling capabilities.

[0054] Common theories for maximizing information content include Fisher Information and KL Divergence. Taking Fisher Information as an example, this theory states that when the student's current state value (… Under this section, select the question with the most Fisher information. As a test item, i.e., a test item used for testing, the Fisher information is calculated by the following formula, which is the expectation of the square of the derivative of the likelihood function: .

[0055] Regarding question selection strategies, while the effectiveness of information content is high when the estimated ability value is close to the actual ability value, its effectiveness is difficult to guarantee when the estimated ability value deviates from the actual ability value. This means that when there are few historical answer records, the amount of information content has considerable limitations. Furthermore, the lack of interpretability of the information content makes it difficult to abstract an understandable question selection logic for testing processes directly aimed at teachers or students, which can easily reduce the credibility of testing resources recommended to students when put into practical use. Based on this, this embodiment of the invention provides a computer adaptive testing method.

[0056] Figure 1 This is a flowchart illustrating a computer-adaptive testing method provided in an embodiment of the present invention, such as... Figure 1 As shown, the method includes:

[0057] S1, determine the target subject's current ability information for the target subject;

[0058] S2, input the target subject's historical answer information and current ability information for the target subject into the question selection model to obtain the test resources output by the question selection model;

[0059] The topic selection model is obtained by training a basic model based on a first training sample. The first training sample includes a sample of the ability information of a first reference subject for the target subject, the historical answer information of the first reference subject for the target subject, and a sequence of first test resource samples.

[0060] The target loss function used when training the base model is determined based on at least one of a first loss function and a second loss function, the first loss function being a loss function corresponding to coverage of an output result of the base model on a knowledge graph of the target subject, and the second loss function being a loss function corresponding to a knowledge point chain distance of the output result in the knowledge graph.

[0061] Specifically, the computer adaptive testing method provided in the embodiments of the present application has a computer adaptive testing system as an execution subject, which can be configured in a computer, which can be a local computer or a cloud computer, and the local computer can be a computer, a tablet, etc., which is not specifically limited here. The system can recommend testing resources to a subject to test the ability of the subject, and then recommend appropriate learning resources to the subject.

[0062] First, step S1 is performed to determine current ability information of a target subject for a target subject. The target subject refers to a user who needs to be recommended testing resources, which can be a student and a person who needs testing resources. The testing resources can be test questions used for testing.

[0063] The target subject can be a subject such as mathematics, Chinese, and foreign languages, and the current ability information can refer to the degree of mastery of the target subject by the target subject. The current ability information can be determined by a cognitive diagnosis model in a conventional computer adaptive testing technology, or can be obtained by other methods, which is not specifically limited here.

[0064] Then, step S2 is performed to input historical answering information of the target subject for the target subject and the current ability information into a question selection model to obtain testing resources output by the question selection model. The historical answering information of the target subject for the target subject can include historical test questions answered by the target subject, answering results of the historical test questions by the target subject, and unanswered test questions in a test question bank corresponding to the target subject except the historical test questions answered by the target subject. The answering results can include correct answers or wrong answers.

[0065] The question selection model can obtain the confidence of each unanswered test question by analyzing the historical answering information of the target subject for the target subject and the current ability information, and then output an unanswered test question with the highest confidence as the testing resource to the target subject for answering.

[0066] The selected topic model can be trained by using a first training sample, the first training sample comprising a first reference subject's ability information sample for the target subject, the first reference subject's historical answering information for the target subject, and a first test resource sample sequence. The first reference subject is a subject whose test resources are known, and the test resources are the first test resource sample sequence, which can be obtained by manual selection. The first test resource sample sequence includes a plurality of first test resource samples that need to be sent to the first reference subject in sequence.

[0067] The first reference subject's ability information sample for the target subject refers to the first reference subject's current mastery of knowledge points of the target subject, and the determination method of the current ability information of the target subject for the target subject is the same. The first reference subject's historical answering information for the target subject can include historical test questions answered by the first reference subject, the answering results of the first reference subject to the historical test questions, and unanswered test questions in the test question bank corresponding to the target subject except for the historical test questions answered by the first reference subject.

[0068] When training the base model, the base model can predict each test resource sample in the first test resource sample sequence in sequence, and the base model can obtain one test resource prediction result each time, and a total of test resource prediction results equal to the number of test resource samples in the first test resource sample sequence can be obtained, which is the output result of the base model.

[0069] When training the base model, the target loss function is determined based on at least one of the first loss function and the second loss function. That is, the target loss function can be the first loss function or the second loss function, or can be determined based on the first loss function in combination with other loss functions, or can be determined based on the second loss function in combination with other loss functions, or can be determined based on the first loss function and the second loss function, or can be determined based on the first loss function and the second loss function in combination with other loss functions, which is not limited here. When the target loss function is determined based on the first loss function and the second loss function in combination with other loss functions, the target loss function can be obtained by weighted summation of the first loss function, the second loss function and other loss functions.

[0070] It can be understood that when the target loss function is determined based on at least two loss functions, the process of training the base model is the process of multi-task learning, and each loss function corresponds to a task.

[0071] The first loss function is a loss function corresponding to the coverage rate of the output result of the base model on the knowledge graph of the target subject, and the second loss function is a loss function corresponding to the knowledge point chain distance of the output result in the knowledge graph of the target subject.

[0072] The knowledge graph of the target subject can be constructed with each knowledge point in the target subject as a node and the association relationship between each knowledge point as the connection relationship between each node. Among them, the association relationship between each knowledge point represents the directed logical correlation between each knowledge point. Taking the target subject as a mathematics subject as an example, "one function and one equation" and "one function and moving point problem", "one function and one equation" and "one function and triangle" are all knowledge point pairs with a preceding relationship, and there is a strong logical and difficulty progression relationship, and the subject needs to master the preceding knowledge point before mastering the subsequent knowledge point.

[0073] Therefore, the knowledge graph of the target subject can play a guiding role in two aspects: the knowledge point span of the test questions and the recommended order between related knowledge points. For knowledge points with strong correlation, i.e. a series of knowledge point sets with a connection relationship on the knowledge graph, only a small number of knowledge points need to be selected for test questions, and a high prediction accuracy can be obtained on the entire knowledge chain. Therefore, the knowledge points selected in the test process need to cover more knowledge chains to improve the overall knowledge coverage. In addition, selecting test questions of knowledge points close to each other on the same knowledge chain as test resources in the test process will reduce the test efficiency.

[0074] Therefore, the first loss function can represent one of the tasks of training the base model, which is to make the test resources output by the test question model cover as many different knowledge points of the target subject as possible, thereby ensuring the overall knowledge coverage of the test resources. The second loss function can represent the second task of training the base model, which is to make the test resources output by the test question model correspond to different knowledge points with less correlation in the target subject as possible, thereby ensuring the difficulty variation of the test resources. Due to the existence of the first loss function and the second loss function, the test question model can achieve multi-dimensional test question selection.

[0075] The computer adaptive test method provided in the embodiment of the application first determines the current ability information of the target subject for the target subject; then inputs the historical answer information and the current ability information of the target subject for the target subject into the test question model to obtain the test resources output by the test question model. When training the base model to obtain the test question model, the first loss function is introduced to ensure the overall knowledge coverage of the test resources output by the test question model for the target subject, and the second loss function is introduced to ensure that the test resources output by the test question model have less correlation in the overall knowledge points. In this way, the credibility of the test resources can be improved, the target subject can be tested comprehensively, the test efficiency can be improved, the subsequent test results can be more accurate, and the accurate recommendation of learning resources can be realized.

[0076] Due to the cognitive diagnosis model adopted in the conventional computer adaptive testing technology, a static modeling mechanism is mostly used to generate the subject ability state before testing, and it is assumed that the subject ability is stable during the testing process. Taking the item response theory as an example, the subject state is initialized before the testing starts, and the item attribute representation is learned with the historical data. However, the static and relatively fixed ability state will limit the parameter correction ability during the testing process, and when the initial training of the ability parameter has a large deviation, it will cause errors in the entire testing process. The neural cognitive diagnosis model has a larger parameter space than the item response theory, but it also does not have the ability to dynamically update.

[0077] Therefore, on the basis of the above-mentioned embodiments, the computer adaptive testing method provided in the embodiments of the present application, the current ability information of the target subject for the target subject is determined, comprising:

[0078] Obtaining the current answer information of the target subject for the target subject;

[0079] Inputting the current answer information into the learning model to obtain the current ability information output by the learning model;

[0080] Wherein, the learning model is constructed by training a recurrent neural network model based on the answer information sequence of the target subject for the target subject; the answer information sequence comprises a plurality of answered questions and the corresponding answer results of each answered question, or comprises a plurality of answered questions, the corresponding answer results of each answered question and knowledge points.

[0081] Specifically, when determining the current ability information of the target subject for the target subject, in order to solve the inherent bias problem existing in the testing process (especially in the case that the target subject has a small amount of answers), the dot product mechanism in the cognitive diagnosis model of the conventional computer adaptive testing technology can be replaced by a recurrent neural network architecture. The structure of the recurrent neural network can better realize the state update in time sequence than the vector dot product.

[0082] Firstly, the current answer information of the target subject for the target subject can be obtained, and the current answer information can include the current answered question and the corresponding answer result, or can include the current answered question, the corresponding answer result and the knowledge point. The introduction of the knowledge point corresponding to the current answered question can provide more rich test attributes to the learning model and improve the accuracy of the learning model. The current answered question can be the test resource output by the previous question selection model.

[0083] After that, the current answer information is input into the learning model, and the current ability information of the target subject for the target subject output by the learning model can be obtained.

[0084] In the embodiments of the present application, the answer information sequence of the target subject for the target subject to the target subject can be used to train the recurrent neural network model, and then the learning situation diagnosis model is obtained. The answer information sequence can include multiple answer questions and corresponding answer results, or can include multiple answer questions, corresponding answer results and knowledge points.

[0085] As shown in the figure, the input of the learning situation diagnosis model at time t (1≤t≤T, T is the current time) can include the answer question at time t or include the answer question at time t and the corresponding knowledge point. The learning situation diagnosis model can perform feature extraction on the input to obtain a feature vector Figure 2 . Thereafter, the learning situation diagnosis model splices the feature vector with the answer result corresponding to the question at time t to obtain a splicing result .

[0086] The learning situation diagnosis model can obtain the hidden layer state at time t based on the splicing result and the hidden layer state at time t-1. The hidden layer state of the learning situation diagnosis model at the initial time, i.e. t=0, is . The hidden layer of the learning situation diagnosis model can be updated by the input of the answer question and the corresponding answer result at each time combined with the hidden layer state at the previous time. The input of the learning situation diagnosis model at time T is the current answer information of the target subject for the target subject, and the hidden layer state is the current ability information of the target subject.

[0087] Further, the test question at time T+1 and the corresponding knowledge point are input to the learning situation diagnosis model , and the feature vector is obtained by feature extraction of the learning situation diagnosis model. The learning situation diagnosis model matches the feature vector with the hidden layer state at time T to obtain the answer prediction result of the target subject when answering the test question at time T+1.

[0088] Based on this, in the embodiments of the present application, the ability information of the target subject is considered to be constantly changing. This dynamic feedback mechanism ensures that inaccurate modeling when the number of answers is insufficient at the beginning of the test can be gradually corrected as the test progresses, so as to gradually improve the accuracy of the current ability information of the target subject.

[0089] ​​​​​​​​In the embodiment of the present application, the learning situation diagnosis model can generate real-time updated current ability information, realize dynamic knowledge tracking of the target subject, and effectively solve the systematic deviation problem of static modeling when lacking of answer records.

[0090] On the basis of the above-mentioned embodiment, the computer adaptive test method provided in the embodiment of the present application is trained based on the following steps:

[0091] The initial model is trained based on a second training sample to obtain a candidate model; the second training sample includes a capability information sample of a second reference subject for the target subject, historical answer information of the second reference subject for the target subject, and a second test resource sample sequence carrying a score;

[0092] A test sample is determined, which includes a capability information sample of a third reference subject for the target subject, historical answer information of the third reference subject for the target subject, and a third test resource sample sequence not carrying a score;

[0093] The capability information sample of the third reference subject for the target subject and the historical answer information of the third reference subject for the target subject are input into the candidate model to obtain a confidence of a test resource prediction sequence output by the candidate model, and based on the confidence, an artificial score of a target test resource prediction result in the test resource prediction sequence is collected;

[0094] Based on the target test resource prediction result and the artificial score thereof, the second training sample is expanded, and based on the expanded second training sample, the candidate model is iteratively trained until the candidate model converges, to obtain the base model.

[0095] Specifically, in the embodiment of the present application, when determining the base model, as shown in Figure 3 the second training sample can be used to train the initial model to obtain the candidate model. The second training sample can include a capability information sample of a second reference subject for the target subject, historical answer information of the second reference subject for the target subject, and a second test resource sample sequence.

[0096] The second reference subject is a subject whose test resources are known, and the test resources thereof are the second test resource sample sequence, which can be obtained by manual selection. The second test resource sample sequence includes a plurality of second test resource samples to be sent to the second reference subject in sequence, and each second test resource sample carries a score. Since the second test resource sample sequence is obtained by manual selection, the score carried by each second test resource sample can be configured as 1.

[0097] The ability information sample of the second reference subject for the target subject is the current mastery of the second reference subject for each knowledge point of the target subject, and the determination manner of the current ability information of the target subject for the target subject is the same. The historical answering information of the second reference subject for the target subject can include historical test questions answered by the second reference subject, answering results of the historical test questions by the second reference subject, and unanswered test questions in the test question library corresponding to the target subject except the historical test questions answered by the second reference subject.

[0098] The process of training the initial model can be a conventional supervised training process, the loss function used can be a conventional loss function, and the obtained candidate model can be an end-to-end deep learning model capable of outputting optimal test resources.

[0099] Thereafter, a test sample of the candidate model can be determined, which includes an ability information sample of a third reference subject for a target subject, historical answering information of the third reference subject for the target subject, and a third test resource sample sequence. The third test resource sample sequence includes a plurality of third test resource samples to be sent to the third reference subject in sequence. The third test resource sample sequence does not carry a score, i.e., each third test resource sample does not carry a score.

[0100] The ability information sample of the third reference subject for the target subject is the current mastery of the third reference subject for each knowledge point of the target subject, and the determination manner of the current ability information of the target subject for the target subject is the same. The historical answering information of the third reference subject for the target subject can include historical test questions answered by the third reference subject, answering results of the historical test questions by the third reference subject, and unanswered test questions in the test question library corresponding to the target subject except the historical test questions answered by the third reference subject.

[0101] The ability information sample of the third reference subject for the target subject and the historical answering information of the third reference subject for the target subject are input into the candidate model, and the confidence of the test resource prediction sequence output by the candidate model can be obtained. It can be understood that each test resource prediction result in the test resource prediction sequence corresponds to a confidence.

[0102] According to the confidence corresponding to each test resource prediction result, a target test resource prediction result in the test resource prediction sequence can be screened out. The screening rule can be set according to needs, which is not specifically limited here.

[0103] Further, the artificial score of the target test resource prediction result can be collected, which can be obtained by a human annotator according to teaching experience for multi-dimensional scoring of the target test resource prediction result, and the multi-dimensions can include difficulty, knowledge point coverage rate, and question type, etc.

[0104] Finally, the second training sample can be expanded according to the target test resource prediction result and the artificial score thereof, that is, the target test resource prediction result and the artificial score thereof are backfilled into the training data set constituted by the second training sample, and the candidate model is iteratively trained based on the expanded second training sample, that is, the backfilled training data set, until the candidate model converges, that is, the basic model is obtained.

[0105] Whether the candidate model converges can be determined by introducing a test set, if the difference between the output test resource prediction sequence and the confidence thereof of the candidate model obtained in the current iteration and the candidate model obtained in the previous iteration is within a preset range, it is considered that the candidate model converges.

[0106] In the embodiment of the application, the basic model is trained by using an active learning mechanism. The active learning mechanism is different from the traditional semi-supervised learning, which allows the basic model to autonomously select which data needs to be manually annotated. Compared with the random annotation strategy, the active learning mechanism can reduce the artificial workload, and the added training samples usually have greater gain for the basic model.

[0107] Based on the above-mentioned embodiment, the computer adaptive test method provided in the embodiment of the application, each test resource prediction result in the test resource prediction sequence corresponds to a confidence; and the target test resource prediction result is determined based on the following steps:

[0108] Based on the confidence corresponding to each test resource prediction result, the test resource prediction result with the confidence within a preset range or closest to a preset threshold is selected as the target test resource prediction result.

[0109] Specifically, when selecting the target test resource prediction result, the confidence corresponding to each test resource prediction result can be used to select the test resource prediction result with the confidence within a preset range or closest to a preset threshold as the target test resource prediction result. The preset range can be 45%-55%, and the preset threshold can be 50%. In this way, the target test resource prediction result selected can have the highest confusion degree, and further intervention by artificial scoring can not only expand the second training sample, but also accelerate the convergence speed of the candidate model, so that a question selection model with high accuracy can be generated only by a small number of iterations.

[0110] On the basis of the above-mentioned embodiments, the computer adaptive test method provided in the embodiments of the present application is characterized in that the target loss function is determined based on at least one of the first loss function and the second loss function, and at least one of a third loss function and a fourth loss function.

[0111] The third loss function is a loss function corresponding to the information amount of the output result, and the fourth loss function is a loss function corresponding to the information divergence of the output result.

[0112] Specifically, in the embodiments of the present application, the target loss function can be determined based on at least one of the first loss function and the second loss function, and at least one of a third loss function and a fourth loss function can be introduced on this basis.

[0113] The third loss function is a loss function corresponding to the information amount of the output result of the base model, and the fourth loss function is a loss function corresponding to the information divergence of the output result of the base model.

[0114] In the embodiments of the present application, the presence of at least one of the third loss function and the fourth loss function can further improve the accuracy and reliability of the question selection model.

[0115] On the basis of the above-mentioned embodiments, the computer adaptive test method provided in the embodiments of the present application is characterized in that the target loss function is determined based on the following steps:

[0116] The first loss function, the second loss function, the third loss function, and the fourth loss function are weighted and summed to obtain the target loss function.

[0117] Specifically, when determining the target loss function, the first loss function, the second loss function, the third loss function, and the fourth loss function can be directly weighted and summed to obtain the target loss function. That is,

[0118] ;

[0119] wherein, is the target loss function, is the number of loss functions, which can be 4, is the i-th loss function, which can be the first loss function, the second loss function, the third loss function, and the fourth loss function, is the corresponding weight.

[0120] ​In the embodiment of the present application, the four loss functions are considered when determining the target loss function, which can improve the topic selection accuracy and credibility of the topic selection model, and is more consistent with artificial teaching experience.

[0121] On the basis of the above-mentioned embodiment, the computer adaptive testing method provided in the embodiment of the present application comprises the following steps:

[0122] displaying the testing resource to the target subject;

[0123] receiving a target answer result of the target subject to the testing resource;

[0124] recommending a learning resource to the target subject based on the target answer result.

[0125] Specifically, after obtaining the testing resource, the testing resource can be sent to a display interface to display the testing resource to the target subject.

[0126] Thereafter, the computer adaptive testing system can receive the target answer result of the target subject to the testing resource, and determine the knowledge mastery of the target subject to the target subject based on the target answer result, and then recommend appropriate learning resources to the target subject.

[0127] In the embodiment of the present application, the testing resource can be used to recommend the learning resource, which can improve the recommendation effect and further improve the experience of the target subject.

[0128] In summary, as Figure 4As shown, the embodiment of the present application provides a computer adaptive test method fusing dynamic knowledge tracking and a test question selection strategy based on active learning. For a target subject, the current ability information of the target subject is determined through dynamic knowledge tracking, and then the test resource is determined and displayed based on the test question selection strategy based on active learning and the current ability information. The test question selection strategy based on active learning can train a test question model conforming to multiple explainable dimensions (including test question quality and test question comprehensiveness) through relatively limited manually annotated data, and select target test resource prediction results with high model confusion degree for manual annotation. Only a small amount of iteration is needed to generate a test question model with high accuracy. When training the base model, the evaluation dimensions conforming to the subject knowledge graph are introduced, which greatly improves the test question reliability and is more consistent with the artificial teaching experience. The method can be applied to a personalized learning system on a learning terminal, and is a method for automatically selecting test resources. Based on the instant feedback provided by the learning situation diagnosis model, the method dynamically selects the knowledge points and corresponding test resources for the test on the knowledge graph according to the real-time performance of the target subject in the test process, so as to improve the test efficiency and achieve more accurate diagnosis effect through fewer test questions, thereby providing an important guarantee for the learning system of individualized teaching and load reduction and efficiency improvement.

[0129] As Figure 5 shown, on the basis of the above embodiment, the embodiment of the present application provides a computer adaptive test system, comprising:

[0130] The ability determination module 51 is configured to determine the current ability information of the target subject for the target subject.

[0131] The test question module 52 is configured to input the historical answer information of the target subject for the target subject and the current ability information into a test question model to obtain a test resource output by the test question model.

[0132] The test question model is obtained by training a base model based on a first training sample. The first training sample includes a capability information sample of a first reference subject for the target subject, historical answer information of the first reference subject for the target subject, and a first test resource sample sequence.

[0133] The target loss function used when training the base model is determined based on at least one of a first loss function and a second loss function. The first loss function is a loss function corresponding to the coverage rate of the output result of the base model on the knowledge graph of the target subject. The second loss function is a loss function corresponding to the knowledge point chain distance of the output result in the knowledge graph.

[0134] On the basis of the above-mentioned embodiments, the computer adaptive test system provided in the embodiments of the present application comprises a capability determination module, which is specifically used for:

[0135] obtaining current answering information of the target subject for the target subject;

[0136] inputting the current answering information into a learning situation diagnosis model to obtain the current capability information output by the learning situation diagnosis model;

[0137] The learning situation diagnosis model is constructed based on training of a recurrent neural network model based on a sequence of answering information of the target subject for the target subject. The sequence of answering information comprises a plurality of answered test questions and an answered result corresponding to each answered test question, or comprises a plurality of answered test questions, an answered result corresponding to each answered test question and a knowledge point.

[0138] On the basis of the above-mentioned embodiments, the computer adaptive test system provided in the embodiments of the present application further comprises a model training module, which is used for:

[0139] training an initial model based on a second training sample to obtain a candidate model; the second training sample comprises a capability information sample of a second reference subject for the target subject, historical answering information of the second reference subject for the target subject and a second test resource sample sequence, and the second test resource sample sequence carries a score;

[0140] determining a test sample, which comprises a capability information sample of a third reference subject for the target subject, historical answering information of the third reference subject for the target subject and a third test resource sample sequence, and the third test resource sample sequence does not carry a score;

[0141] inputting the capability information sample of the third reference subject for the target subject and the historical answering information of the third reference subject for the target subject into the candidate model to obtain a confidence of a test resource prediction sequence output by the candidate model, and collecting an artificial score of a target test resource prediction result in the test resource prediction sequence based on the confidence;

[0142] based on the target test resource prediction result and the artificial score thereof, expanding the second training sample, and based on the expanded second training sample, iteratively training the candidate model until the candidate model converges to obtain the basic model.

[0143] On the basis of the above-mentioned embodiments, the computer adaptive test system provided in the embodiments of the present application comprises a capability determination module, which is specifically used for:

[0144] Based on the confidence corresponding to each test resource prediction result, a test resource prediction result with a confidence within a preset range or closest to a preset threshold is selected as a target test resource prediction result.

[0145] On the basis of the above-mentioned embodiments, the computer adaptive test system provided in the embodiments of the present application comprises a target loss function determined based on at least one of the first loss function and the second loss function, and at least one of a third loss function and a fourth loss function;

[0146] The third loss function is a loss function corresponding to the information amount of the output result, and the fourth loss function is a loss function corresponding to the information divergence of the output result.

[0147] On the basis of the above-mentioned embodiments, the computer adaptive test system provided in the embodiments of the present application further comprises a loss function determination module configured to:

[0148] The first loss function, the second loss function, the third loss function and the fourth loss function are weighted and summed to obtain the target loss function.

[0149] On the basis of the above-mentioned embodiments, the computer adaptive test system provided in the embodiments of the present application further comprises a recommendation module configured to:

[0150] The test resource is displayed to the target subject.

[0151] The target subject's target answer result to the test resource is received.

[0152] Based on the target answer result, a learning resource is recommended to the target subject.

[0153] Specifically, the roles of the modules in the computer adaptive test system provided in the embodiments of the present application are one-to-one corresponding to the operation processes of the steps in the above-mentioned method embodiment, and the effects achieved are consistent. For details, refer to the above-mentioned embodiments, which will not be described here again.

[0154] Figure 6 An example of an entity structure diagram of an electronic device is shown in FIG. 1. Figure 6As shown, the electronic device can include a processor 610, a communications interface 620, a memory 630, and a communications bus 640, wherein the processor 610, the communications interface 620, and the memory 630 complete mutual communication through the communications bus 640. The processor 610 can invoke a logical instruction in the memory 630 to execute the computer adaptive test method provided in each of the embodiments described above, which includes determining current ability information of a target subject for a target subject; inputting historical answering information of the target subject for the target subject and the current ability information into a question selection model to obtain a test resource output by the question selection model; wherein the question selection model is obtained by training a base model based on a first training sample, the first training sample including an ability information sample of a first reference subject for the target subject, historical answering information of the first reference subject for the target subject, and a first test resource sample sequence; a target loss function used when the base model is trained is determined based on at least one of a first loss function and a second loss function, the first loss function being a loss function corresponding to coverage of an output result of the base model on a knowledge graph of the target subject, and the second loss function being a loss function corresponding to a knowledge point chain distance of the output result in the knowledge graph.

[0155] In addition, the logical instruction in the memory 630 described above can be implemented in the form of a software functional unit and sold or used as an independent product, which can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, and various program code storage media.

[0156] In another aspect, the present application also provides a computer program product comprising a computer program, which can be stored on a non-transitory computer-readable storage medium, and the computer program, when executed by a processor, enables a computer to perform the computer adaptive testing method provided by any of the above methods, which comprises: determining current ability information of a target subject for a target subject; inputting historical answering information of the target subject for the target subject and the current ability information into a question selection model to obtain a test resource output by the question selection model; wherein the question selection model is obtained by training a base model based on a first training sample, the first training sample comprising an ability information sample of a first reference subject for the target subject, historical answering information of the first reference subject for the target subject, and a first test resource sample sequence; a target loss function used when training the base model is determined based on at least one of a first loss function and a second loss function, the first loss function being a loss function corresponding to coverage of an output result of the base model on a knowledge graph of the target subject, and the second loss function being a loss function corresponding to a knowledge point chain distance of the output result in the knowledge graph.

[0157] In another aspect, the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and the computer program, when executed by a processor, enables a computer to perform the computer adaptive testing method provided by any of the above methods, which comprises: determining current ability information of a target subject for a target subject; inputting historical answering information of the target subject for the target subject and the current ability information into a question selection model to obtain a test resource output by the question selection model; wherein the question selection model is obtained by training a base model based on a first training sample, the first training sample comprising an ability information sample of a first reference subject for the target subject, historical answering information of the first reference subject for the target subject, and a first test resource sample sequence; a target loss function used when training the base model is determined based on at least one of a first loss function and a second loss function, the first loss function being a loss function corresponding to coverage of an output result of the base model on a knowledge graph of the target subject, and the second loss function being a loss function corresponding to a knowledge point chain distance of the output result in the knowledge graph.

[0158] The system embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0159] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0160] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A computer-adaptive testing method, characterized in that, include: Determine the target subjects' current competency information regarding the target subject; The target subject's historical answer information and current ability information for the target subject are input into the question selection model to obtain the test resources output by the question selection model; The topic selection model is obtained by training a basic model based on a first training sample. The first training sample includes a sample of the ability information of a first reference subject for the target subject, the historical answer information of the first reference subject for the target subject, and a sequence of first test resource samples. The target loss function used when training the base model is determined based on at least one of a first loss function and a second loss function. The first loss function is the loss function corresponding to the coverage of the base model's output results with the knowledge graph of the target subject, and the second loss function is the loss function corresponding to the distance of the knowledge point chain of the output results in the knowledge graph. The first loss function represents that one of the tasks of training the base model is to ensure that the test resources output by the topic selection model as a whole cover different knowledge points of the target subject, thus ensuring the overall knowledge coverage of each test resource. The second loss function represents that another task of training the base model is to ensure that each test resource output by the topic selection model corresponds to different knowledge points of the target subject with low relevance, thus ensuring the range of difficulty variation of each test resource.

2. The computer-adaptive testing method according to claim 1, characterized in that, The determination of the target subject's current competence information in the target subject includes: Obtain the current answer information of the target subject regarding the target subject; The current answer information is input into the learning assessment model to obtain the current ability information output by the learning assessment model; The learning assessment model is constructed by training a recurrent neural network model based on the answer information sequence of the target subject for the target subject; the answer information sequence includes multiple answer questions and the answer result corresponding to each answer question, or includes multiple answer questions, the answer result corresponding to each answer question, and knowledge points.

3. The computer-adaptive testing method according to claim 1, characterized in that, The base model is trained based on the following steps: Based on the second training samples, the initial model is trained to obtain the candidate model; the second training samples include the ability information samples of the second reference subjects for the target subject, the historical answer information of the second reference subjects for the target subject, and the second test resource sample sequence, the second test resource sample sequence carrying scores. A test sample is determined, which includes a sample of the ability information of a third reference subject in the target subject, the historical answer information of the third reference subject in the target subject, and a third test resource sample sequence, wherein the third test resource sample sequence does not carry a score. The ability information sample of the third reference subject for the target subject and the historical answer information of the third reference subject for the target subject are input into the alternative model to obtain the confidence level of the test resource prediction sequence output by the alternative model, and based on the confidence level, the manual score of the target test resource prediction result in the test resource prediction sequence is collected. Based on the target test resource prediction results and their human ratings, the second training sample is expanded, and based on the expanded second training sample, the candidate model is iteratively trained until the candidate model converges to obtain the base model.

4. The computer-adaptive testing method according to claim 3, characterized in that, Each test resource prediction result in the test resource prediction sequence corresponds to a confidence level; the target test resource prediction result is determined based on the following steps: Based on the confidence level corresponding to each test resource prediction result, the test resource prediction result with a confidence level within a preset range or closest to a preset threshold is selected as the target test resource prediction result.

5. The computer-adaptive testing method according to any one of claims 1-4, characterized in that, The target loss function is determined based on at least one of the first loss function and the second loss function, and at least one of the third loss function and the fourth loss function. The third loss function is the loss function corresponding to the information content of the output result, and the fourth loss function is the loss function corresponding to the information divergence of the output result.

6. The computer-adaptive testing method according to claim 5, characterized in that, The target loss function is determined based on the following steps: The target loss function is obtained by weighted summation of the first loss function, the second loss function, the third loss function, and the fourth loss function.

7. The computer-adaptive testing method according to any one of claims 1-4, characterized in that, The process involves inputting the target subject's historical answer information and current ability information for the target subject into the question selection model to obtain the test resources output by the question selection model, followed by: The test resources are shown to the target subjects; Receive the target subjects' responses to the target test resources; Based on the target response results, learning resources are recommended to the target subject.

8. A computer-adaptive testing system, characterized in that, include: The competency determination module is used to determine the current competency information of the target subject in the target subject. The topic selection module is used to input the target subject's historical answer information and current ability information into the topic selection model to obtain the test resources output by the topic selection model; The topic selection model is obtained by training a basic model based on a first training sample. The first training sample includes a sample of the ability information of a first reference subject for the target subject, the historical answer information of the first reference subject for the target subject, and a sequence of first test resource samples. The target loss function used when training the base model is determined based on at least one of a first loss function and a second loss function. The first loss function is the loss function corresponding to the coverage of the base model's output results with the knowledge graph of the target subject, and the second loss function is the loss function corresponding to the distance of the knowledge point chain of the output results in the knowledge graph. The first loss function represents that one of the tasks of training the base model is to ensure that the test resources output by the topic selection model as a whole cover different knowledge points of the target subject, thus ensuring the overall knowledge coverage of each test resource. The second loss function represents that another task of training the base model is to ensure that each test resource output by the topic selection model corresponds to different knowledge points of the target subject with low relevance, thus ensuring the range of difficulty variation of each test resource.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the computer adaptive testing method as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the computer adaptive testing method as described in any one of claims 1-7.

Images