Cognitive model training, exercise recommendation methods, devices, equipment, and media

Abstract

This disclosure provides a cognitive model training and exercise recommendation method, apparatus, device, and medium, relating to the field of data processing technology, and particularly to the fields of artificial intelligence and deep learning technology. The specific implementation scheme is as follows: Cognitive Model Training Method: From the historical answer interaction records of each sample exercise, multiple interaction representation vectors between each sample exercise and the tester are extracted. The training set and validation set of each sample exercise respectively include partial interaction representation vectors corresponding to each sample exercise. Based on the training set of each sample exercise, an inner layer training is performed on a preset cognitive model. Based on the validation set of multiple sample exercises, an outer layer training is performed on the preset cognitive model after the inner layer training. Exercise Recommendation Method: The representation parameters of each sample exercise obtained after training the preset cognitive model according to the above method are used to initialize the representation parameters of each sample exercise in the exercise library. Based on the representation parameters of each exercise in the exercise library, exercises are recommended to the target user.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to the fields of artificial intelligence and deep learning technology. Background Technology

[0002] Online intelligent testing is an emerging intelligent technology that aims to help assess users' professional level and comprehensive abilities by recommending suitable exercises. Summary of the Invention

[0003] This disclosure provides a method, apparatus, device, and medium for training cognitive models and recommending exercises.

[0004] According to a first aspect of this disclosure, a cognitive model training method is provided, comprising:

[0005] Extract multiple interaction representation vectors between each sample exercise and the tester from the historical response interaction records of each sample exercise;

[0006] Based on the training set of each sample exercise, the inner layer of the pre-defined cognitive model is trained. The training set of each sample exercise includes the partial interaction representation vector corresponding to each sample exercise.

[0007] Based on the validation set of multiple sample exercises, the pre-set cognitive model trained in the inner layer is trained in the outer layer. The validation set of each sample exercise includes another part of the interaction representation vector corresponding to each sample exercise.

[0008] According to a second aspect of this disclosure, a method for recommending exercises is provided, comprising:

[0009] The representation parameters of each sample exercise in the exercise library will be initialized using the representation parameters of each sample exercise obtained after training the preset cognitive model according to the method provided in the first aspect.

[0010] Based on the representation parameters of each exercise in the exercise library, exercises are recommended to the target user.

[0011] According to a third aspect of this disclosure, a cognitive model training apparatus is provided, comprising:

[0012] The first extraction module is used to extract multiple interaction representation vectors between each sample exercise and the tester from the historical response interaction records of each sample exercise;

[0013] The inner training module is used to perform inner training on the pre-defined cognitive model based on the training set of each sample exercise. The training set of each sample exercise includes the partial interaction representation vector corresponding to each sample exercise.

[0014] The outer training module is used to perform outer training on the preset cognitive model trained in the inner layer based on the validation set of multiple sample exercises. The validation set of each sample exercise includes another part of the interaction representation vector corresponding to each sample exercise.

[0015] According to a fourth aspect of this disclosure, an exercise recommendation device is provided, comprising:

[0016] An initialization module is used to initialize the representation parameters of each sample exercise in the exercise library with the representation parameters of each sample exercise obtained after training a preset cognitive model according to the device provided by the third party.

[0017] The recommendation module is used to recommend exercises to the target user based on the representation parameters of each exercise in the exercise bank.

[0018] According to a fifth aspect of this disclosure, an electronic device is provided, comprising:

[0019] At least one processor; and

[0020] A memory communicatively connected to the at least one processor; wherein,

[0021] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method provided in the first aspect above, or to perform the method provided in the second aspect above.

[0022] According to a sixth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method provided in the first aspect above, or to perform the method provided in the second aspect above.

[0023] According to a seventh aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method provided in the first aspect above, or performs the method provided in the second aspect above.

[0024] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0025] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0026] Figure 1 This is a schematic diagram of the first flowchart of the cognitive model training method provided in this embodiment;

[0027] Figure 2 This is a schematic diagram of the second process of the cognitive model training method provided in this embodiment of the disclosure;

[0028] Figure 3 This is a schematic diagram of the third process of the cognitive model training method provided in this embodiment of the disclosure;

[0029] Figure 4 This is a schematic diagram of the fourth process of the cognitive model training method provided in this embodiment of the disclosure;

[0030] Figure 5 This is a schematic diagram of the fifth process of the cognitive model training method provided in this embodiment of the disclosure;

[0031] Figure 6 This is a schematic flowchart of the first exercise recommendation method provided in this embodiment of the disclosure;

[0032] Figure 7 This is a second flowchart illustrating the exercise recommendation method provided in this embodiment of the disclosure;

[0033] Figure 8 This is a schematic flowchart of a problem characterization parameter update method provided in an embodiment of this disclosure;

[0034] Figure 9 This is a schematic diagram of a cold start framework for exercises provided in an embodiment of this disclosure;

[0035] Figure 10 This is a schematic diagram of a cognitive model training device provided in an embodiment of this disclosure;

[0036] Figure 11 This is a schematic diagram of a problem recommendation device provided in an embodiment of this disclosure;

[0037] Figure 12 This is a block diagram of an electronic device used to implement the cognitive model training method or exercise recommendation method provided in the embodiments of this disclosure;

[0038] Figure 13 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0039] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0040] Online intelligent testing is an emerging intelligent technology that aims to assist in assessing users' professional level and comprehensive abilities by recommending suitable exercises. Online testing systems often use a fixed question bank, which can lead to question leakage. Therefore, new questions need to be continuously added to the question bank to enhance the system's robustness. However, these new questions lack response interaction records, preventing the online testing system from effectively learning from them and determining whether they are suitable for recommendation. This results in a decline in the performance of the online testing system's question recommendation model. Therefore, the cold start problem of exercises has become a critical issue in online intelligent testing scenarios. Existing technologies mainly employ the following three methods to address the cold start problem of exercises:

[0041] Method 1: Offline Data Collection. Specifically, test takers are assigned to answer new exercises offline, and a large number of interaction records are collected for these new exercises, turning them into "old" exercises with a large number of interaction records. These collected interaction records are then used to train a cognitive model of the exercises, allowing the model to fully learn the representation parameters of the new exercises. The exercise recommendation model then uses these learned representation parameters to accurately recommend new exercises to users.

[0042] Method 1 requires manually generating and collecting interaction records for new exercises, resulting in high human and time costs. Furthermore, the online testing system needs to repeatedly collect interaction records every time a new exercise is added, leading to low persistence and scalability for Method 1.

[0043] Method 2: Transfer of exercise representation parameters. Specifically, identify existing exercises similar to the new exercises and initialize their representation parameters as those of the new exercises. In the online testing system, the exercise recommendation model uses the initialized representation parameters of the new exercises to accurately recommend new exercises to the user.

[0044] In Method 2, parameter transfer is performed using similar exercises. However, this method suffers from the problem that the representation parameters of the exercises differ greatly, which also reduces the diversity of the exercises and results in poor representation of the parameters of the new exercises.

[0045] Method 3: Periodic Updates. Specifically, based on online response records, the cognitive model is retrained periodically, allowing it to fully learn the representation parameters of new exercises. In the online testing system, the exercise recommendation model uses the representation parameters of the new exercises learned by the cognitive model to accurately recommend new exercises to users.

[0046] In Method 3, the phased updating of the cognitive model and the representation parameters of the exercises requires a large amount of computing resources to retrain the cognitive model offline, which results in low efficiency and poor timeliness.

[0047] To address the cold start problem of exercises, this disclosure provides a cognitive model training and exercise recommendation method. For ease of description, the following description uses an electronic device as the execution subject, without limitation. In this disclosure, the electronic device executing the cognitive model training method and the device executing the exercise recommendation method can be the same or different.

[0048] like Figure 1 As shown in the embodiments of this disclosure, a cognitive model training method is provided, including the following steps:

[0049] Step S11: Extract multiple interaction representation vectors between each sample exercise and the tester from the historical response interaction records of each sample exercise;

[0050] Step S12: Based on the training set of each sample exercise, perform inner-layer training on the preset cognitive model. The training set of each sample exercise includes the partial interaction representation vector corresponding to each sample exercise.

[0051] Step S13: Based on the validation set of multiple sample exercises, perform outer layer training on the preset cognitive model after inner layer training. The validation set of each sample exercise includes another part of the interaction representation vector corresponding to each sample exercise.

[0052] In the cognitive model training method provided in this disclosure, the electronic device extracts the interaction representation vector between the sample exercise and the tester from the historical answer interaction records. Using the interaction representation vector, the cognitive model is trained to learn the representation parameters of the sample exercise. This solves the cold start problem of the exercise and eliminates the need for manual generation and collection of answer interaction records for new exercises, reducing human resources and time costs. Even if new exercises are added, the technical solution provided in this disclosure is still applicable, improving the durability and scalability of the online testing system.

[0053] Furthermore, in the cognitive model training method provided in this embodiment, the electronic device utilizes interactive representation vectors to perform inner-layer and outer-layer training on the cognitive model. This enables the cognitive model to fully learn both the features of the sample exercises themselves and the features between sample exercises, thereby improving the representation effect of the exercise representation parameters. Using representation parameters with better representation effects, the online testing system can accurately recommend suitable exercises to users, improving the generalization ability and recommendation performance of the online testing system while ensuring the cold-start effect of the exercises.

[0054] In step S11 above, the sample exercises can be any exercise in the exercise bank, and the number of sample exercises can be one or more. The historical response interaction record is a record of the tester's historical responses to sample exercises. The interaction representation vector consists of the representation parameters of the exercises and the representation parameters of the tester. For example, the representation parameters of the exercises include the representation parameters of knowledge points 1-3, and the representation parameters of the tester include the parameters of feature dimensions 1-3. For Exercise 1, which involves knowledge points 1 and 2 but not knowledge point 3, the representation parameter of knowledge point 1 in Exercise 1 is x1, and the representation parameter of knowledge point 2 in Exercise 1 is x2. Testers a and b answered Exercise 1. Tester a's parameters in feature dimensions 1-3 are y1-3, and tester b's parameters in feature dimensions 1-3 are y4-6. Therefore, the interaction representation vector 1 between Exercise 1 and tester a can be represented as {x1, x2, 0, y1, y2, y3}, and the interaction representation vector 2 between Exercise 1 and tester b can be represented as {x1, x2, 0, y4, y5, y6}; or the interaction representation vector 1 and interaction representation vector 2 can be represented as:

[0055]

[0056] After acquiring the historical interaction records of sample exercises, the electronic device extracts the interaction representation vector between the sample exercise and each tester from the historical interaction records of each tester for that sample exercise. For a sample exercise, there may be one or more testers who have answered it. When multiple testers have answered the sample exercise, the electronic device can acquire the interaction representation vector corresponding to that sample exercise.

[0057] In this embodiment, the electronic device can divide all the interaction representation vectors corresponding to a sample exercise into a task data set. The task data includes interaction representation vectors divided into a training set and a validation set. The training set of the sample exercise includes a portion of the interaction representation vectors corresponding to the sample exercise, and the validation set includes another portion of the interaction representation vectors corresponding to the sample exercise. Subsequently, the electronic device performs steps S12 and S13 based on the training set and validation set to perform internal and external double-layer training on the preset cognitive model.

[0058] In this embodiment, the preset cognitive model can be a neural network model built based on item response theory in psychometrics. This makes the cognitive model more reasonable and interpretable, giving it a clear, symbolic internal knowledge representation to match the user's own knowledge framework. This allows the user to diagnose and modify the cognitive model at the semantic level. In this embodiment, the preset cognitive model can also be other cognitive models, such as pattern recognition, attention selection models, attention filter models, semantic network models, etc., and is not limited thereto.

[0059] In some embodiments, step S11 above may be: obtaining exercises from the exercise library whose number of historical response interaction records is less than a preset number of times threshold, as sample exercises. In this embodiment of the disclosure, the size of the preset number of times threshold can be set according to actual needs, for example, the preset number of times threshold can be 2, 3 or 4, etc. If the number of historical response interaction records of an exercise is less than the preset number of times threshold, the exercise is a new exercise; otherwise, it is an old exercise.

[0060] In this embodiment of the disclosure, the electronic device trains the preset cognitive model using only new exercises, which reduces the number of samples in the cognitive model, improves the training speed of the cognitive model, reduces the hardware performance requirements of the electronic device, and reduces the training cost.

[0061] In this embodiment of the disclosure, in order to improve the accuracy of the cognitive model, that is, to improve the accuracy of the representation parameters of the exercises, the electronic device can also acquire exercises in which the number of historical answer interaction records is less than a preset number of times threshold, and acquire a specified number of exercises in which the number of historical answer interaction records is greater than or equal to the preset number of times threshold, and use the acquired exercises as sample exercises.

[0062] In step S12 above, the electronic device performs inner-layer training on the preset cognitive model based on the training set of each sample exercise, and adjusts the cognitive parameters of the preset cognitive model so that the cognitive model can fully learn the features of the sample exercise itself.

[0063] In step S13 above, the electronic device performs outer layer training on the preset cognitive model after inner layer training based on the validation set of multiple sample exercises, and adjusts the cognitive parameters of the preset cognitive model and the representation parameters of the sample exercises so that the cognitive model can fully learn the features between the sample exercises.

[0064] In this embodiment of the disclosure, the electronic device can repeatedly execute the above steps S12 and S13 until the model training end condition is met.

[0065] In some embodiments, such as Figure 2 As shown in the embodiments of this disclosure, a cognitive model training method is also provided, which may include the following steps:

[0066] Step S21: Extract the first representation parameter of each sample exercise and the second representation parameter of each tester from the historical response interaction record of each sample exercise;

[0067] Step S22: Combine the first representation parameter of each sample exercise with the second representation parameter of each tester who answered each sample exercise to obtain multiple interaction representation vectors between each sample exercise and the tester.

[0068] Step S23: Based on the training set of each sample exercise, perform inner-layer training on the preset cognitive model. The training set of each sample exercise includes the partial interaction representation vector corresponding to each sample exercise.

[0069] Step S24: Based on the validation set of multiple sample exercises, perform outer layer training on the preset cognitive model after inner layer training. The validation set of each sample exercise includes another part of the interaction representation vector corresponding to each sample exercise.

[0070] By applying the technical solution provided in this disclosure, the electronic device extracts the first representation parameter of the sample exercise and the second representation parameter of the tester based on the historical interaction records. The first and second representation parameters are combined to obtain an interaction representation vector, which can more accurately express the relationship between the exercise and the tester. The cognitive model can then be trained using this interaction representation vector, enabling the cognitive model to more accurately learn the influence of the tester on the exercise, and thus accurately learn the representation parameters of the exercise, thereby improving the generalization ability and recommendation performance of the online testing system in subsequent applications.

[0071] Steps S23 and S24 are the same as steps S12 and S13, and will not be repeated here.

[0072] In step S21 above, the electronic device extracts the representation parameters of each sample exercise, namely the first representation parameter, and extracts the representation parameters of each tester, namely the second representation parameter. The first representation parameter may include the knowledge point information of the sample exercise, such as the difficulty coefficient and discrimination coefficient of the knowledge point, etc. The second representation parameter may include the tester's current job information, interview job information, self-skill information and job requirement skill information, etc.

[0073] In step S22 above, for each sample exercise, the electronic device obtains multiple interaction representation vectors between the sample exercise and the tester based on the first representation parameter of the sample exercise extracted in step S21 and the second representation parameter of each tester who answers the sample exercise.

[0074] In this embodiment of the disclosure, the electronic device may also obtain the interaction representation vector corresponding to the sample exercise in other ways. For example, the electronic device may extract the first representation parameter of each sample exercise from the historical answer interaction record of each sample exercise as the interaction representation vector, without limitation.

[0075] In some embodiments, such as Figure 3 As shown in the embodiments of this disclosure, a cognitive model training method is also provided, which may include the following steps:

[0076] Step S31: Extract multiple interaction representation vectors between each sample exercise and the tester from the historical response interaction records of each sample exercise;

[0077] Step S32: Select one sample exercise from the multiple sample exercises that have not been selected;

[0078] Step S33: Input each interaction representation vector in the training set of the selected sample exercises into the preset cognitive model to obtain the first probability that the tester correctly answers the selected sample exercises corresponding to each interaction representation vector.

[0079] Step S34: Determine the first inner-layer cognitive loss corresponding to the selected sample exercise based on the first probability and the true label corresponding to each interaction representation vector;

[0080] Step S35: Update the cognitive parameters of the preset cognitive model based on the first inner layer cognitive loss;

[0081] Step S36: Determine if there are still sample exercises that have not been selected. If yes, proceed to step S32; otherwise, proceed to step S37.

[0082] Step S37: Based on the validation set of multiple sample exercises, perform outer layer training on the preset cognitive model after inner layer training. The validation set of each sample exercise includes another part of the interaction representation vector corresponding to each sample exercise.

[0083] By applying the technical solution provided in this disclosure, the electronic device treats the training set of each sample exercise as task data. In one training iteration, a preset cognitive model is trained using the training set of each sample exercise. This allows the preset cognitive model to sequentially learn the representation parameters of each sample exercise, thereby enabling it to fully learn the representation parameters of the sample exercise itself. In this way, even if the historical interaction records of a sample exercise are limited, such as with a new exercise, the preset cognitive model can repeatedly mine the interaction information of these sample exercises, improving the cold start performance of the exercises.

[0084] The steps S31 and S37 described above are the same as steps S11 and S13, respectively.

[0085] In step S32 above, after acquiring multiple sample exercises, the electronic device can randomly select one sample exercise from the multiple unselected sample exercises. This selected sample exercise will not be selected again in subsequent selections. Through step S32, the repeated selection of the same sample exercise can be avoided, allowing the preset cognitive model to fully learn the representation parameters of each sample exercise.

[0086] In step S33 above, the training set of the selected sample exercises includes one or more interaction representation vectors. For each interaction representation vector in the training set of the selected sample exercises, the electronic device inputs the interaction representation vector into a preset cognitive model to obtain the probability that the tester correctly answers the selected sample exercise corresponding to the interaction representation vector, i.e., the first probability. The first probability represents the probability output by the cognitive model that the tester correctly answers the selected sample exercise, and can be any value between 0 and 1.0, for example, 0.5, 0.7, or 0.9. The tester corresponding to the interaction representation vector is the tester represented by the tester representation parameter in the interaction representation vector.

[0087] In step S34 above, the true label corresponding to the interaction representation vector is the actual answer of the tester to the sample exercise in the historical answer interaction record. For example, if the sample exercise is answered correctly, the true label is 1; if the sample exercise is answered incorrectly, the true label is 0.

[0088] For each interaction representation vector in the training set of the selected sample exercises, the electronic device determines the inner-layer cognitive loss corresponding to the selected sample exercise, i.e., the first inner-layer cognitive loss, based on the first probability and the true label obtained in step S33. In one example, the first inner-layer cognitive loss can be calculated using the following formula:

[0089]

[0090] In formula (1), loss inner This indicates the loss of the first inner layer of cognition; y i This represents the true label of the sample exercise selected by test taker i. The first probability represents the probability that tester i correctly answers the selected sample question, as output by the cognitive model; K is the number of interaction representation vectors in the training set of the selected sample question.

[0091] For example, if there are 3 interaction representation vectors in the training set of the selected sample exercises, and the true labels of the answers given by the 3 testers are 0, 1 and 1 respectively, and the first probabilities output by the cognitive model are 0.2, 0.7 and 0.8 respectively, then the first inner cognitive loss is -(1 / 3)(log0.8+log0.7+log0.8)=-(1 / 3)log0.448.

[0092] In this embodiment of the disclosure, the electronic device may also use other methods to determine the first inner layer cognitive loss, such as using mean square error to determine the first inner layer cognitive loss, and there is no limitation on this.

[0093] In step S35 above, the electronic device can update the network parameters of the preset cognitive model by using gradient descent algorithm and backpropagation algorithm, etc., based on the first inner-layer cognitive loss obtained in step S34. In this embodiment of the present disclosure, the network parameters are referred to as cognitive parameters, so that the cognitive model can fully learn the representation parameter information of the sample exercises.

[0094] After the electronic device updates the cognitive parameters of the preset cognitive model once, it executes step S36 above to determine whether there are still sample exercises that have not been selected. If there are still sample exercises that have not been selected, it returns to step S32 and selects a sample exercise from the multiple unselected sample exercises to continue training the preset cognitive model. If all sample exercises are selected, it executes step S37 to perform outer layer training on the preset cognitive model.

[0095] In some embodiments, such as Figure 4 As shown in the embodiments of this disclosure, a cognitive model training method is also provided, which may include the following steps:

[0096] Step S41: Extract multiple interaction representation vectors between each sample exercise and the tester from the historical response interaction records of each sample exercise;

[0097] Step S42: Input each interaction representation vector in the training set of each sample exercise into the preset cognitive model to obtain the second probability that the tester correctly answers the corresponding sample exercise for each interaction representation vector;

[0098] Step S43: For each sample exercise, determine the second inner-layer cognitive loss corresponding to the sample exercise based on the second probability and the true label corresponding to each interaction representation vector in the training set of the sample exercise.

[0099] Step S44: Update the cognitive parameters of the preset cognitive model based on multiple second inner layer cognitive losses;

[0100] Step S45: Based on the validation set of multiple sample exercises, perform outer layer training on the preset cognitive model after inner layer training. The validation set of each sample exercise includes another part of the interaction representation vector corresponding to each sample exercise.

[0101] By applying the technical solution provided in this disclosure, the electronic device treats the training set of each sample exercise as task data. In one training iteration, a preset cognitive model is trained using the training set of each sample exercise, determining the second inner-layer cognitive loss corresponding to each sample exercise, and then updating the cognitive parameters of the preset cognitive model. This allows the preset cognitive model to sequentially learn the representation parameters of each sample exercise, thereby enabling the preset cognitive model to fully learn the representation parameters of the sample exercise itself. In this way, even if the historical response interaction records of a sample exercise are few, such as with a new exercise, the preset cognitive model can repeatedly mine the interaction information of these sample exercises, improving the cold-start performance of the exercises.

[0102] The steps S41 and S45 described above are the same as steps S11 and S13, respectively.

[0103] In step S42 above, for each sample exercise, the electronic device inputs each interaction representation vector in the training set of the sample exercise into a preset cognitive model to obtain the probability that the tester correctly answers the corresponding sample exercise for each interaction representation vector, i.e., the second probability. The second probability represents the probability output by the cognitive model that the tester correctly answers the selected sample exercise, and can be any value between 0 and 1.0, for example, it can be 0.6, 0.8 or 0.95, etc.

[0104] In step S43 above, for each interaction representation vector in the training set of each sample exercise, the electronic device determines the inner cognitive loss corresponding to that sample exercise, i.e., the second inner cognitive loss, based on the second probability corresponding to the interaction representation vector obtained in step S43 and the true label. The specific method for determining the second inner cognitive loss can be found in the description of the method for determining the first inner cognitive loss in step S34 above, and will not be repeated here.

[0105] In step S44 above, after obtaining the second inner-layer cognitive loss corresponding to multiple sample exercises, the electronic device updates the cognitive parameters of the preset cognitive model based on the multiple second inner-layer cognitive losses.

[0106] In some embodiments, step S44 above may be: performing a preset processing on multiple second inner-layer cognitive losses to obtain processed losses; and updating the cognitive parameters of the preset cognitive model based on the processed losses. The preset processing may be any of the following: weighted processing; summation processing; product processing; maximum value processing, etc. Specifically, weighted processing, summation processing, and product processing represent performing weighted processing, summation processing, and product processing on multiple second inner-layer cognitive losses, respectively, while maximum value processing represents selecting the largest second inner-layer cognitive loss from multiple second inner-layer cognitive losses.

[0107] In this embodiment of the disclosure, the electronic device performs one inner-layer training on the preset cognitive model, and then performs one outer-layer training on the preset cognitive model, compared to the above... Figure 3 The illustrated embodiment saves computing resources.

[0108] In some embodiments, such as Figure 5 As shown in the embodiments of this disclosure, a cognitive model training method is also provided, which may include the following steps:

[0109] Step S51: Extract multiple interaction representation vectors between each sample exercise and the tester from the historical response interaction records of each sample exercise;

[0110] Step S52: Based on the training set of each sample exercise, perform inner-layer training on the preset cognitive model. The training set of each sample exercise includes the partial interaction representation vector corresponding to each sample exercise.

[0111] Step S53: Input each interaction representation vector in the validation set of multiple sample exercises into the preset cognitive model after inner training to obtain the third probability of the tester correctly answering the corresponding sample exercise for each interaction representation vector.

[0112] Step S54: Determine the outer cognitive loss corresponding to the multiple sample exercises based on the third probability and the true label corresponding to each interaction representation vector in the validation set of multiple sample exercises.

[0113] Step S55: Based on the outer cognitive loss, update the cognitive parameters of the preset cognitive model and the representation parameters of multiple sample exercises.

[0114] By applying the technical solution provided in this disclosure, an electronic device trains a preset cognitive model using a validation set of multiple sample exercises. This allows the preset cognitive model to learn the representation parameters of each sample exercise and then learn the features between sample exercises. In this way, even if the historical interaction records of sample exercises are few, such as with new exercises, the preset cognitive model can repeatedly mine the interaction information of these sample exercises, improving the cold start performance of the exercises.

[0115] The steps S51 and S52 described above are the same as steps S11 and S12, respectively.

[0116] In step S53 above, after the electronic device performs an inner-layer training on the preset cognitive model, it inputs each interaction representation vector from the validation set of multiple sample exercises into the inner-layer trained preset cognitive model. The preset cognitive model outputs the probability that the tester correctly answers the corresponding sample exercise for each interaction representation vector, i.e., the third probability. The third probability represents the probability that the tester correctly answers the selected sample exercise, which can be any value between 0 and 1.0, such as 0.3, 0.5, or 0.85.

[0117] In step S54 above, after obtaining the third probability corresponding to each interaction representation vector in the validation set of multiple sample exercises, the electronic device determines the outer cognitive loss corresponding to the multiple sample exercises based on the third probability corresponding to each interaction representation vector and the true label. In one example, the outer cognitive loss can be calculated using the following formula:

[0118]

[0119] In formula (2), loss outer Indicates loss of outer cognitive layer; y ij This represents the true label of the sample exercise i answered by test taker j; The third probability represents the probability that tester j correctly answers sample exercise i, as output by the cognitive model; N is the number of sample exercises, and L is the number of interaction representation vectors in the validation set of sample exercise i.

[0120] For example, an electronic device inputs each interaction representation vector from the validation set of three sample exercises into a pre-trained inner cognitive model. These three sample exercises correspond to three interaction representation vectors, with their respective true labels being: 0, 0, 1; 0, 1, 1; 0, 0, 1. The third probability of a tester correctly answering the corresponding sample exercise for each interaction representation vector output by the pre-trained cognitive model is: 0.1, 0.2, 0.8; 0.15, 0.7, 0.9; 0.2, 0.13, 0.85. Then, the outer cognitive loss is:

[0121] -(1 / (3*3))(log0.9+log0.8+log0.8+log0.85+log0.7+log0.9+log0.8+log0.87+log0.85)=-(1 / 9)log0.182.

[0122] In step S55 above, the electronic device can use gradient descent and backpropagation algorithms, etc., to update the cognitive parameters of the preset cognitive model and the representation parameters of multiple sample exercises based on the outer cognitive loss, so that the model can fully learn the representation parameter information of the sample exercises.

[0123] In this embodiment of the disclosure, the electronic device may also use other methods to perform outer training on the preset cognitive model, such as dividing the validation set of multiple sample exercises into a preset number of parts, and using one of the validation sets each time to perform outer training on the preset cognitive model. This can reduce the number of samples during each outer training and save the computing resources of the electronic device.

[0124] Based on the cognitive model obtained above, this disclosure also provides a problem recommendation method, see [link to relevant documentation]. Figure 6 , Figure 6 This is a schematic flowchart of the first exercise recommendation method provided in this embodiment, which includes the following steps:

[0125] Step S61: Initialize the representation parameters of each sample exercise in the exercise library with the representation parameters of each sample exercise obtained after training the preset cognitive model;

[0126] Step S62: Recommend exercises to the target user based on the representation parameters of each exercise in the exercise bank.

[0127] In the exercise recommendation method provided in this embodiment, during the training of the preset cognitive model, the electronic device can accurately learn the representation parameters of the sample exercises. The electronic device initializes the representation parameters of each sample exercise in the exercise library using the representation parameters of each sample exercise obtained after training the preset cognitive model, thus solving the cold start problem of exercises and improving the accuracy of exercise recommendation.

[0128] In step S61 above, the electronic device initializes the representation parameters of the online exercises with the representation parameters of each sample exercise obtained after offline training of the preset cognitive model, that is, initializes the representation parameters of each sample exercise in the exercise library of the online testing system.

[0129] In some embodiments, step S61 above may be: averaging multiple representation parameters of each sample exercise obtained after training the preset cognitive model to obtain the averaged representation parameter of each sample exercise; updating the representation parameters of each sample exercise in the exercise library to the averaged representation parameter of each sample exercise.

[0130] The representative parameters of exercises can include the difficulty coefficient and discrimination coefficient of each knowledge point. For example, the difficulty coefficient of a knowledge point can be between 0 and 1, where 0 represents easy and 1 represents difficult. If the difficulty coefficient of knowledge point 1 is 0.6 and the difficulty coefficient of knowledge point 2 is 0.8, then knowledge point 1 has a lower difficulty coefficient than knowledge point 2, and knowledge point 1 is easier. The discrimination coefficient of a knowledge point can also be between 0 and 1. For example, a discrimination coefficient below 0.19 indicates poor discrimination, a discrimination coefficient between 0.20 and 0.29 indicates fair discrimination, a discrimination coefficient between 0.30 and 0.39 indicates good discrimination, and a discrimination coefficient above 0.40 indicates excellent discrimination.

[0131] In this embodiment of the disclosure, for each sample exercise, the electronic device performs mean processing on multiple representation parameters of the sample exercise obtained after training a preset cognitive model, and the resulting representation parameters are the averaged representation parameters of the sample exercise. This enables the application of the exercise's representation parameters online, achieving accurate exercise recommendations to users.

[0132] For example, the representation parameters of exercises can include the difficulty coefficient and discrimination coefficient of each knowledge point. For each knowledge point included in each sample exercise, the electronic device calculates the mean difficulty coefficient and the mean discrimination coefficient of that knowledge point, and updates the representation parameters of that knowledge point included in the sample exercises in the exercise bank with the calculated mean difficulty coefficient and the mean discrimination coefficient of that knowledge point.

[0133] In this embodiment of the disclosure, the electronic device may also perform median processing, maximum processing, or minimum processing on multiple representation parameters of the sample exercise obtained after training the preset cognitive model, without limitation.

[0134] In step S62 above, the electronic device can recommend exercises to the target user based on the representation parameters of each exercise in the exercise bank and in combination with actual needs. For example, the electronic device can recommend exercises with a difficulty coefficient in the range of 0.6-0.7 to the target user based on the difficulty coefficient of the exercises.

[0135] In some embodiments, the representation parameters may include the difficulty coefficient and discrimination coefficient of each knowledge point. In this case, such as... Figure 7 As shown in the embodiments of this disclosure, a method for recommending exercises is also provided, which may include the following steps:

[0136] Step S71: Initialize the representation parameters of each sample exercise in the exercise library with the representation parameters of each sample exercise obtained after training the preset cognitive model;

[0137] Step S72: Determine the skill points to be examined for the target user;

[0138] Step S73: Based on the pre-stored mapping relationship between skill points and knowledge points, determine the knowledge points to be examined corresponding to the skill points to be examined;

[0139] Step S74: Based on the knowledge points to be examined, and the difficulty coefficient and discrimination coefficient of each knowledge point included in each exercise, determine a preset number of exercises from the exercise bank;

[0140] Step S75: Recommend selected exercises to the target user.

[0141] By applying the technical solution provided in this disclosure, the electronic device maps the skill points to be tested of the target user to the knowledge points to be tested. By combining the knowledge points to be tested with the difficulty coefficient and discrimination coefficient of each knowledge point included in each exercise, the electronic device can accurately select exercises suitable for the target user and then accurately recommend suitable exercises to the target user, thereby improving the recommendation performance of the online testing system.

[0142] The above steps S71 are the same as steps 61.

[0143] In step S72 above, the electronic device can determine the skill points to be assessed for the target user based on the actual application scenario. For example, in an online quiz test scenario, the target user is the job applicant. The electronic device can extract the target user's strengths from their resume and obtain the skill points involved in the job the target user is applying for. Both the obtained strengths and related skill points are used as the skill points to be assessed.

[0144] For example, in a simulated test paper generation scenario, the target user is an employee in a certain position. The electronic device can obtain the required skill points for the target user's position and the skill points that need to be assessed in the position, and use the obtained skill points as the skill points to be assessed.

[0145] In step S73 above, the electronic device pre-stores the mapping relationship between skill points and knowledge points. Based on the pre-stored mapping relationship between skill points and knowledge points, the mapping relationship including the skill points to be examined is determined, and the knowledge points included in the determined mapping relationship are the knowledge points to be examined.

[0146] In step S74 above, after obtaining the knowledge points to be tested, the electronic device determines the exercises to be recommended to the target user based on the knowledge points to be tested and the difficulty coefficient and discrimination coefficient of each knowledge point included in each exercise.

[0147] In some embodiments, step S74 may be: obtaining exercises from the exercise bank that include the knowledge points to be tested as candidate exercises; performing weighted summation on the difficulty coefficient and discrimination coefficient of the knowledge points to be tested included in each candidate exercise to obtain the test score of each candidate exercise; and determining a preset number of exercises with the highest test scores.

[0148] In other embodiments, step S74 may also be: obtaining exercises from a question bank that include the knowledge points to be tested, as candidate exercises; and from the obtained candidate exercises, selecting exercises whose difficulty coefficient of the knowledge points to be tested is in a first preset range and whose discrimination coefficient is in a second preset range. The first preset range and the second preset range can be set according to actual needs.

[0149] In this embodiment of the disclosure, the electronic device can filter out exercises suitable for the target user, and can test the target user's professional level, comprehensive ability and the degree of matching between the target user and the job to the greatest extent.

[0150] In this embodiment of the disclosure, the online testing system containing the aforementioned question bank may further include an online cognitive model. The online cognitive model can adjust the representation parameters of the questions based on real-time response and interaction records. Specifically, see [link to relevant documentation]. Figure 8 , Figure 8This is a flowchart illustrating a method for updating exercise representation parameters provided in an embodiment of this disclosure. The method may include the following steps:

[0151] Step S81: Transfer the cognitive parameters of the trained preset cognitive model to the online cognitive model;

[0152] Step S82: Obtain the real-time response and interaction records of the sample exercises;

[0153] Step S83: Extract the target interaction representation vector between the sample exercises and the target user from the real-time response interaction record;

[0154] Step S84: Input the target interaction representation vector into the online cognitive model to obtain the target probability of the target user correctly answering the sample questions;

[0155] Step S85: Adjust the representation parameters of the sample exercises in the exercise bank according to the target probability and true label of the sample exercises.

[0156] By applying the technical solution provided in this disclosure, when exercises are used in an online testing system, as the exercises are recommended to users for answering, the electronic device can acquire corresponding real-time answer interaction records. The electronic device uses these real-time answer interaction records to fine-tune the representation parameters of the exercises, thereby further improving the recommendation performance of the online testing system. Furthermore, in this disclosure, there is no need to retrain offline using accumulated historical answer records, thus improving the efficiency of acquiring representation parameters.

[0157] In step S81 above, the electronic device will... Figures 1-5 The cognitive parameters of the pre-trained cognitive model are transferred to the online cognitive model. The online cognitive model can be a neural network model based on item response theory in psychometrics, or other cognitive models, such as pattern recognition, attention selection models, attention filter models, semantic network models, etc. The online cognitive model has the same structure as the pre-trained cognitive model.

[0158] After the electronic device recommends exercises to the target user according to the above steps S61-S62, it can execute step S82 to obtain the target user's real-time response and interaction records for the sample exercises.

[0159] Steps S83-S85 can refer to the relevant descriptions in the above sections S53-S55.

[0160] The following is combined Figure 9 The exercise cold start framework shown illustrates the cognitive model training method and exercise recommendation method provided in the embodiments of this disclosure.

[0161] (1) Exercise data modeling: Electronic devices acquire the interaction records of answering exercises, extract the interaction representation vector between the exercise and the tester from the interaction records, and divide the extracted interaction representation vector of each exercise into a training set and a validation set;

[0162] (2) Cognitive diagnostic modeling: Using item response theory in psychometrics, a cognitive model of the interaction information between learning exercises and test takers is established.

[0163] (3) Loss function calculation: The electronic device inputs the training set and validation set of each exercise into the cognitive model, and the cognitive model outputs the prediction result, that is, the probability that the tester correctly answers the corresponding sample exercise corresponding to each interaction representation vector; the electronic device calculates the loss of the cognitive model based on the prediction result of the exercise and the true label.

[0164] (4) Two-layer parameter update: The electronic device optimizes the cognitive model by updating the cognitive parameters of the cognitive model and the representation parameters of each sample exercise based on the loss of the cognitive model during training.

[0165] (5) New exercise hot start: The electronic device initializes the representation parameters of the new exercise based on the representation parameters of each exercise obtained after training the cognitive model, and transfers the cognitive parameters in the cognitive model to the online cognitive model. Based on the real-time answer interaction records of the new exercise, the electronic device fine-tunes and updates the representation parameters of the exercise in the online cognitive model.

[0166] Subsequently, the online testing system can be used for specific applications, as follows:

[0167] a) Description of exercise information.

[0168] Electronic devices can quickly update and obtain the latent parameter representations of new questions based on the hot-start strategy of the exercises and online few-sample interaction, and learn the knowledge point attributes of the exercises.

[0169] After obtaining the latent parameter representation of the new exercises, the electronic device can obtain the difficulty coefficient and discrimination coefficient corresponding to the knowledge points involved in the exercises based on the mapping of knowledge point dimensions. At the same time, based on the attribute values ​​of each knowledge point, it can obtain the overall difficulty value and overall discrimination value of the exercises. The attribute values ​​of the knowledge points are the difficulty coefficient and discrimination coefficient corresponding to the aforementioned knowledge points.

[0170] like Figure 9 The exercises q1 and q2 shown involve knowledge points such as Python, SVM, Java, and HTML, respectively. The difficulty coefficients of these knowledge points are 0.6, 0.4, 0.5, and 0.5, respectively, and the discrimination coefficients of these knowledge points are 0.7, 0.5, 0.3, and 0.6, respectively.

[0171] b) Online quiz tests.

[0172] In online testing systems, electronic devices can quickly learn the knowledge point attribute values ​​of new exercises, thus enabling their use in online testing. For example, based on the applicant's strengths in their resume and the skills required for the corresponding position, the system utilizes the mapping relationship between knowledge points and skills, as well as difficulty and discrimination coefficients, to select the k most suitable candidate test questions—the optimal candidate questions—to maximize the assessment of the applicant's professional level and comprehensive abilities.

[0173] like Figure 9 The applicants u1 and u2 in table b) shown are required to be assessed on skills including machine learning, Python and data structures, and Java. The electronic device determines the best candidate exercises q4, q32, q11 and q7, q13, q3, etc. for applicants u1 and u2 respectively, according to the exercise recommendation method provided in this embodiment of the disclosure.

[0174] c) Simulated test paper.

[0175] The cognitive model training method and exercise recommendation method provided in this embodiment can be applied to the simulated test paper generation stage of the testing system. That is, the electronic device constructs a set of best candidate exercises for the user based on the required skill points and the skill points to be tested for the job, according to the knowledge point attribute values ​​of the exercises in the exercise bank, for testing and assessment in simulated interview scenarios, and to measure the degree of matching between the user and the job to the greatest extent.

[0176] like Figure 9 As shown in table c), users u1 and u2 need to be tested on knowledge points including databases, deep learning and Go language, and Redis. The electronic device selects the best candidate exercises for them according to the exercise recommendation method provided in this embodiment of the disclosure, including q2, q11, q15 and q6, q9, q12, etc.

[0177] Corresponding to the above-described embodiments of the cognitive model training method, this disclosure also provides a cognitive model training apparatus, see below. Figure 10 , Figure 10 This is a schematic diagram of a cognitive model training device provided in an embodiment of the present disclosure. The device includes:

[0178] The first extraction module 101 is used to extract multiple interaction representation vectors between each sample exercise and the tester from the historical answer interaction records of each sample exercise;

[0179] The inner training module 102 is used to perform inner training on the preset cognitive model based on the training set of each sample exercise. The training set of each sample exercise includes the partial interaction representation vector corresponding to each sample exercise.

[0180] The outer training module 103 is used to perform outer training on the preset cognitive model after the inner training based on the validation set of multiple sample exercises. The validation set of each sample exercise includes another part of the interaction representation vector corresponding to each sample exercise.

[0181] In some embodiments, the first extraction module 101 may include:

[0182] The extraction submodule is used to extract the first representation parameter of each sample exercise and the second representation parameter of each tester from the historical response interaction records of each sample exercise;

[0183] The combination submodule is used to combine the first representation parameters of each sample exercise and the second representation parameters of each tester who answers each sample exercise to obtain multiple interaction representation vectors between each sample exercise and the tester.

[0184] In some embodiments, the inner training block 102 may include:

[0185] The selection submodule is used to select one sample exercise from multiple sample exercises that have not been selected before.

[0186] The first input submodule is used to input each interaction representation vector in the training set of the selected sample exercises into the preset cognitive model to obtain the first probability that the tester correctly answers the selected sample exercises corresponding to each interaction representation vector.

[0187] The first determining submodule is used to determine the first inner-layer cognitive loss corresponding to the selected sample exercise based on the first probability and the true label corresponding to each interaction representation vector.

[0188] The first update submodule is used to update the cognitive parameters of the preset cognitive model based on the first inner layer cognitive loss.

[0189] In some embodiments, the inner training module 102 may include:

[0190] The second input submodule is used to input each interaction representation vector in the training set of each sample exercise into the preset cognitive model to obtain the second probability that the tester correctly answers the corresponding sample exercise for each interaction representation vector.

[0191] The second determination submodule is used to determine the second inner-layer cognitive loss for each sample exercise based on the second probability and the true label corresponding to each interaction representation vector in the training set of the sample exercise.

[0192] The second update submodule is used to update the cognitive parameters of the preset cognitive model based on multiple second inner-layer cognitive losses.

[0193] In some embodiments, the second update submodule can be specifically used for:

[0194] Multiple second inner-layer cognitive losses are pre-processed to obtain the processed losses;

[0195] Based on the processed loss, update the cognitive parameters of the preset cognitive model;

[0196] The preset processing can be any of the following methods:

[0197] Weighted processing;

[0198] Summation processing;

[0199] Product processing;

[0200] Maximum value handling.

[0201] In some embodiments, the outer training module 103 may include:

[0202] The third input submodule is used to input each interaction representation vector in the validation set of multiple sample exercises into the inner trained preset cognitive model to obtain the third probability that the tester correctly answers the corresponding sample exercise for each interaction representation vector.

[0203] The third determination submodule is used to determine the outer cognitive loss corresponding to multiple sample exercises based on the third probability and the true label corresponding to each interaction representation vector in the validation set of multiple sample exercises.

[0204] The third update submodule is used to update the cognitive parameters of the preset cognitive model and the representation parameters of multiple sample exercises based on the outer cognitive loss.

[0205] In some embodiments, the above-described cognitive model training apparatus may further include:

[0206] The first acquisition module is used to acquire exercises from the exercise bank whose number of historical answer interaction records is less than a preset threshold, as sample exercises.

[0207] In some embodiments, the preset cognitive model can be a neural network model built based on item response theory in psychometrics.

[0208] In the cognitive model training device provided in this embodiment, the electronic device extracts the interaction representation vector between the sample exercises and the tester from the historical answer interaction records. Using the interaction representation vector, the cognitive model is trained to learn the representation parameters of the sample exercises. This solves the cold start problem of exercises and eliminates the need for manual generation and collection of answer interaction records for new exercises, reducing human resources and time costs. Even if new exercises are added, the technical solution provided in this embodiment is still applicable, improving the durability and scalability of the online testing system.

[0209] Furthermore, in the cognitive model training method provided in this embodiment, the electronic device utilizes interactive representation vectors to perform inner-layer and outer-layer training on the cognitive model. This enables the cognitive model to fully learn both the features of the sample exercises themselves and the features between sample exercises, thereby improving the representation effect of the exercise representation parameters. Using representation parameters with better representation effects, the online testing system can accurately recommend suitable exercises to users, improving the generalization ability and recommendation performance of the online testing system while ensuring the cold-start effect of the exercises.

[0210] Corresponding to the above-described exercise recommendation method, this disclosure also provides an exercise recommendation device, see [link to relevant documentation]. Figure 11 , Figure 11 This is a schematic diagram of a problem recommendation device provided in an embodiment of the present disclosure. The device includes:

[0211] Initialization module 111 is used to initialize the representation parameters of each sample exercise in the exercise library with the representation parameters of each sample exercise obtained after training the preset cognitive model;

[0212] The recommendation module 112 is used to recommend exercises to the target user based on the representation parameters of each exercise in the exercise bank.

[0213] In some embodiments, the characterization parameters may include the difficulty coefficient and the discrimination coefficient of each knowledge point;

[0214] The recommendation module may include:

[0215] The fourth submodule is used to determine the skill points to be assessed for the target user;

[0216] The fifth submodule is used to determine the knowledge points to be tested corresponding to the skill points to be tested, based on the pre-stored mapping relationship between skill points and knowledge points.

[0217] The sixth sub-module is used to determine a preset number of exercises from the question bank based on the knowledge points to be tested, as well as the difficulty coefficient and discrimination coefficient of each knowledge point included in each exercise.

[0218] The recommendation submodule is used to recommend specific exercises to the target users.

[0219] In some embodiments, the sixth determining submodule can specifically be used for:

[0220] Extract practice questions from the question bank that cover the knowledge points to be tested, and use them as candidate practice questions;

[0221] The difficulty coefficient and discrimination coefficient of the knowledge points to be tested included in each candidate exercise are weighted and summed to obtain the test score for each candidate exercise;

[0222] Determine the maximum number of exercises to maximize the assessment score.

[0223] In some embodiments, the initialization module 111 may include:

[0224] The mean processing submodule is used to perform mean processing on multiple representation parameters of each sample exercise obtained after training the preset cognitive model, so as to obtain the mean representation parameters of each sample exercise.

[0225] The fourth update submodule is used to update the representation parameters of each sample exercise in the exercise bank to the averaged representation parameters of each sample exercise.

[0226] In some embodiments, the above-described exercise recommendation device may further include:

[0227] The transfer module is used to transfer the cognitive parameters of the trained preset cognitive model to the online cognitive model;

[0228] The second acquisition module is used to acquire real-time interaction records of sample exercises;

[0229] The second extraction module is used to extract the target interaction representation vector between the sample exercises and the target user from the real-time response interaction records;

[0230] The input module is used to input the target interaction representation vector into the online cognitive model to obtain the target probability of the target user correctly answering the sample exercises.

[0231] The adjustment module is used to adjust the representation parameters of sample exercises in the exercise bank based on the target probability and true label of the sample exercises.

[0232] In the exercise recommendation method provided in this embodiment, during the training of the preset cognitive model, the electronic device can accurately learn the representation parameters of the sample exercises. The electronic device initializes the representation parameters of each sample exercise in the exercise library using the representation parameters of each sample exercise obtained after training the preset cognitive model, thus solving the cold start problem of exercises and improving the accuracy of exercise recommendation.

[0233] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0234] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0235] Figure 12A schematic block diagram of an example electronic device 120 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0236] like Figure 12 As shown, device 120 includes a computing unit 121, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 122 or a computer program loaded from storage unit 128 into random access memory (RAM) 123. The RAM 123 may also store various programs and data required for the operation of device 120. The computing unit 121, ROM 122, and RAM 123 are interconnected via bus 124. Input / output (I / O) interface 125 is also connected to bus 124.

[0237] Multiple components in device 120 are connected to I / O interface 125, including: input unit 126, such as keyboard, mouse, etc.; output unit 127, such as various types of monitors, speakers, etc.; storage unit 128, such as disk, optical disk, etc.; and communication unit 129, such as network card, modem, wireless transceiver, etc. Communication unit 129 allows device 120 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0238] The computing unit 121 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 121 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 121 performs the various methods and processes described above, such as cognitive model training methods or exercise recommendation methods. For example, in some embodiments, the cognitive model training method or exercise recommendation method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 128. In some embodiments, part or all of the computer program may be loaded and / or installed on device 120 via ROM 122 and / or communication unit 129. When the computer program is loaded into RAM 123 and executed by the computing unit 121, one or more steps of the cognitive model training method or exercise recommendation method described above may be performed. Alternatively, in other embodiments, computing unit 121 may be configured in any other suitable manner (e.g., by means of firmware) to perform a cognitive model training method or a problem recommendation method.

[0239] This disclosure also provides an electronic device, such as... Figure 13 As shown, it includes:

[0240] At least one processor 131; and a memory 132 communicatively connected to the at least one processor; wherein,

[0241] The memory 132 stores instructions that can be executed by at least one processor, such that the at least one processor can perform any of the above-described cognitive model training methods or any of the exercise recommendation methods.

[0242] This disclosure also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute any of the above-described cognitive model training methods or any exercise recommendation methods.

[0243] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements any of the above-described cognitive model training methods or any of the exercise recommendation methods.

[0244] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0245] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0246] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0247] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0248] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0249] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0250] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0251] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for recommending practice problems, comprising: The representation parameters of each sample exercise obtained after training the preset cognitive model are used to initialize the representation parameters of each sample exercise in the exercise library. The representation parameters include the difficulty coefficient and discrimination coefficient of each knowledge point. Based on the pre-stored mapping relationship between skill points and knowledge points, determine the knowledge points to be examined corresponding to the skill points to be examined for the target user; Based on the knowledge points to be examined and the representation parameters of each exercise, a preset number of exercises are determined from the exercise bank; Recommend the identified exercises to the target users; The pre-defined cognitive model is obtained by extracting multiple interaction representation vectors between each sample question and the tester from the historical response interaction records of each sample question, and then performing inner-layer training and outer-layer training: The inner layer training is as follows: input each interaction representation vector in the training set of each sample exercise into the preset cognitive model to obtain the second probability that the tester correctly answers the corresponding sample exercise for each interaction representation vector; determine the second inner layer cognitive loss for each sample exercise based on the second probability corresponding to each interaction representation vector and the true label; update the cognitive parameters of the preset cognitive model based on multiple second inner layer cognitive losses, so that the preset cognitive model learns the features of the sample exercise itself. The outer layer training is as follows: each interaction representation vector in the validation set of multiple sample exercises is input into the preset cognitive model after inner layer training to obtain the third probability of the tester correctly answering the corresponding sample exercise for each interaction representation vector; the outer layer cognitive loss corresponding to the multiple sample exercises is determined based on the third probability corresponding to each interaction representation vector and the true label; the cognitive parameters of the preset cognitive model and the representation parameters of the multiple sample exercises are updated based on the outer layer cognitive loss, so that the preset cognitive model learns the features between sample exercises.

2. The method according to claim 1, wherein, The training set includes a portion of the interaction representation vectors corresponding to the sample exercises; the validation set includes another portion of the interaction representation vectors corresponding to the sample exercises.

3. The method according to claim 1, wherein, The step of extracting multiple interaction representation vectors between each sample exercise and the tester from the historical response interaction records of each sample exercise includes: Extract the first representation parameter of each sample exercise and the second representation parameter of each tester from the historical response records of each sample exercise. By combining the first representation parameter of each sample exercise and the second representation parameter of each tester who answered each sample exercise, multiple interaction representation vectors between each sample exercise and the tester are obtained.

4. The method according to claim 1, wherein, The inner layer training is as follows: Select one sample exercise from among multiple sample exercises that have never been selected; Input each interaction representation vector in the training set of the selected sample exercises into the preset cognitive model to obtain the first probability that the tester correctly answers the selected sample exercises corresponding to each interaction representation vector; Based on the first probability and true label corresponding to each interaction representation vector, determine the first inner cognitive loss corresponding to the selected sample exercise; Based on the first inner-layer cognitive loss, update the cognitive parameters of the preset cognitive model; Repeat the step of selecting one sample exercise from the multiple sample exercises that were never selected, until all of the multiple sample exercises are selected.

5. The method according to claim 1, wherein, The step of updating the parameters of the preset cognitive model based on multiple second inner-layer cognitive losses includes: Multiple second inner-layer cognitive losses are pre-processed to obtain the processed loss; Based on the processed loss, update the cognitive parameters of the preset cognitive model; The preset processing can be any of the following methods: Weighted processing; Summation processing; Product processing; Maximum value handling.

6. The method of claim 1, further comprising, before the step of extracting multiple interaction representation vectors between each sample exercise and the tester: Select sample questions from the question bank whose number of historical response records is less than a preset threshold.

7. The method according to any one of claims 1-6, wherein the preset cognitive model is a neural network model constructed based on item response theory in psychometrics.

8. The method according to claim 1, wherein, The step of determining a preset number of exercises from the exercise bank based on the knowledge points to be examined and the representation parameters of each exercise includes: Obtain exercises from the exercise bank that include the knowledge points to be tested, as candidate exercises; The difficulty coefficient and discrimination coefficient of the knowledge points to be tested included in each candidate exercise are weighted and summed to obtain the test score for each candidate exercise; Determine the maximum number of exercises to maximize the assessment score.

9. The method according to claim 1, wherein, The step of initializing the representation parameters of each sample exercise in the exercise library with the representation parameters of each sample exercise obtained after training the preset cognitive model includes: The averaged representation parameters of each sample exercise are obtained by averaging multiple representation parameters obtained after training the preset cognitive model. Update the representation parameters of each sample exercise in the exercise bank to the averaged representation parameters of each sample exercise.

10. The method according to any one of claims 1-6 and 8-9, further comprising: The cognitive parameters of the preset cognitive model are transferred to the online cognitive model; Obtain the real-time interaction records of the sample exercises; Extract the target interaction representation vector between the sample exercise and the target user from the real-time response interaction record; The target interaction representation vector is input into the online cognitive model to obtain the target probability that the target user correctly answers the sample exercise. Based on the target probability and true label of the sample exercises, adjust the representation parameters of the sample exercises in the exercise library.

11. A problem recommendation device, comprising: An initialization module is used to initialize the representation parameters of each sample exercise in the exercise library with the representation parameters of each sample exercise obtained after training the preset cognitive model. The representation parameters include the difficulty coefficient and discrimination coefficient of each knowledge point. The recommendation module is used to recommend exercises to the target user based on the representation parameters of each exercise in the exercise bank; The recommendation module includes: a fourth determining submodule, used to determine the skill points to be tested for the target user; a fifth determining submodule, used to determine the knowledge points to be tested corresponding to the skill points to be tested based on a pre-stored mapping relationship between skill points and knowledge points; a sixth determining submodule, used to determine a preset number of exercises from the exercise bank based on the knowledge points to be tested and the representation parameters of each exercise; and a recommendation submodule, used to recommend the determined exercises to the target user. The preset cognitive model is obtained using the first extraction module, the inner training module, and the outer training module: The first extraction module is used to extract multiple interaction representation vectors between each sample exercise and the tester from the historical answer interaction records of each sample exercise; The inner training module includes: a second input submodule, used to input each interaction representation vector in the training set of each sample exercise into a preset cognitive model to obtain a second probability that the tester correctly answers the corresponding sample exercise for each interaction representation vector; a second determination submodule, used to determine the second inner cognitive loss corresponding to each sample exercise based on the second probability corresponding to each interaction representation vector and the true label; and a second update submodule, used to update the cognitive parameters of the preset cognitive model based on multiple second inner cognitive losses, so that the preset cognitive model learns the features of the sample exercise itself. The outer training module includes: a third input submodule, used to input each interaction representation vector from the validation set of multiple sample exercises into the inner-layer trained preset cognitive model to obtain the third probability that the tester correctly answers the corresponding sample exercise for each interaction representation vector; a third determination submodule, used to determine the outer cognitive loss corresponding to the multiple sample exercises based on the third probability and the true label of each interaction representation vector; and a third update submodule, used to update the cognitive parameters of the preset cognitive model and the representation parameters of the multiple sample exercises based on the outer cognitive loss, so that the preset cognitive model learns the features between the sample exercises.

12. The apparatus according to claim 11, wherein, The training set includes a portion of the interaction representation vectors corresponding to the sample exercises; the validation set includes another portion of the interaction representation vectors corresponding to the sample exercises.

13. The apparatus according to claim 11, wherein, The first extraction module includes: The extraction submodule is used to extract the first representation parameter of each sample exercise and the second representation parameter of each tester from the historical response interaction records of each sample exercise; The combination submodule is used to combine the first representation parameters of each sample exercise and the second representation parameters of each tester who answers each sample exercise to obtain multiple interaction representation vectors between each sample exercise and the tester.

14. The apparatus according to claim 11, wherein, The inner training module includes: The selection submodule is used to select one sample exercise from multiple sample exercises that have not been selected before. The first input submodule is used to input each interaction representation vector in the training set of the selected sample exercises into the preset cognitive model to obtain the first probability that the tester correctly answers the selected sample exercises corresponding to each interaction representation vector. The first determining submodule is used to determine the first inner-layer cognitive loss corresponding to the selected sample exercise based on the first probability and the true label corresponding to each interaction representation vector. The first update submodule is used to update the cognitive parameters of the preset cognitive model based on the first inner-layer cognitive loss.

15. The apparatus according to claim 11, wherein, The second update submodule is specifically used for: Multiple second inner-layer cognitive losses are pre-processed to obtain the processed loss; Based on the processed loss, update the cognitive parameters of the preset cognitive model; The preset processing can be any of the following methods: Weighted processing; Summation processing; Product processing; Maximum value handling.

16. The apparatus of claim 11, further comprising: The first acquisition module is used to acquire exercises from the exercise bank whose number of historical answer interaction records is less than a preset threshold, as sample exercises.

17. The apparatus according to any one of claims 11-16, wherein the preset cognitive model is a neural network model constructed based on item response theory in psychometrics.

18. The apparatus according to claim 11, wherein, The sixth determining submodule is specifically used for: Obtain exercises from the exercise bank that include the knowledge points to be tested, as candidate exercises; The difficulty coefficient and discrimination coefficient of the knowledge points to be tested included in each candidate exercise are weighted and summed to obtain the test score for each candidate exercise; Determine the maximum number of exercises to maximize the assessment score.

19. The apparatus according to claim 11, wherein, The initialization module includes: The mean processing submodule is used to perform mean processing on multiple representation parameters of each sample exercise obtained after training the preset cognitive model, so as to obtain the mean representation parameters of each sample exercise. The fourth update submodule is used to update the representation parameters of each sample exercise in the exercise bank to the averaged representation parameters of each sample exercise.

20. The apparatus according to any one of claims 11-16, 18-19, further comprising: The migration module is used to migrate the cognitive parameters of the preset cognitive model to the online cognitive model; The second acquisition module is used to acquire the real-time answer interaction records of the sample exercises; The second extraction module is used to extract the target interaction representation vector between the sample exercise and the target user from the real-time response interaction record; The input module is used to input the target interaction representation vector into the online cognitive model to obtain the target probability that the target user correctly answers the sample exercise. An adjustment module is used to adjust the representation parameters of the sample exercises in the exercise library based on the target probability and true label of the sample exercises.

21. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method of any one of claims 1-10.

22. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-10.

23. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-10.

Images