Paper Generation Method, Apparatus, Electronic Device, and Storage Medium

Abstract

The present invention provides a test paper generation method, apparatus, electronic device, and storage medium. The method includes: determining candidate test questions and a target object of the same object type as the object to be examined; based on the application information of the candidate test questions corresponding to the target object and the test question information of the candidate test questions, performing a recommendation ranking on the candidate test questions to obtain the recommendation weights of the candidate test questions; aiming at the highest sum of the recommendation weights of the test questions included in the test paper, generating a test paper based on the recommendation weights of the candidate test questions. The method, apparatus, electronic device, and storage medium provided by the present invention can apply the application information of similar users, thereby being able to maximize the excavation of the test paper generation preferences of the object to be examined; generating a test paper based on the recommendation weights of the candidate test questions can improve the efficiency of searching for the optimal test question combination, shorten the test paper generation time, improve the quality of the generated test paper, and ensure the pertinence of the test paper to the object to be examined.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular, to a test paper generation method, apparatus, electronic device, and storage medium. Background Art

[0002] It is very important to use examinations to detect students' learning effects and provide feedback on their learning processes. Currently, when generating test papers, proposition teachers usually search for relevant test questions from resources such as review materials, exercise sets, and past examination papers according to the current examination requirements, and then select the test questions that meet the examination requirements based on personal experience to assemble a test paper.

[0003] Considering that manual test paper generation requires manual search for test questions, which is time-consuming and laborious, automatic test paper generation technology has emerged. Currently, for automatic test paper generation, under the constraints of the test paper generation parameters input by the user, test questions are extracted from the question bank for test paper generation. For example, test paper generation can be achieved through a random extraction strategy or a backtracking heuristic method. The above-mentioned automatic test paper generation technology defines test paper generation as an optimization problem under constraints. However, since this combinatorial optimization is a non-deterministic polynomial (NP) problem, the optimization time efficiency is low, and it is difficult to search for the optimal solution from a vast question bank; moreover, the test papers generated based on the above technology are mostly general assessment questions and cannot meet the needs of targeted assessment of students, so they cannot well reflect students' knowledge mastery and teaching quality. Summary of the Invention

[0004] The present invention provides a test paper generation method, apparatus, electronic device, and storage medium to solve the problems in the prior art that it is difficult to find the optimal solution in automatic test paper generation and it cannot meet the needs of targeted assessment of students.

[0005] The present invention provides a test paper generation method, including:

[0006] Determine candidate test questions and a target object that belongs to the same object type as the object to be examined;

[0007] Based on the application information of the candidate test questions corresponding to the target object and the question information of the candidate test questions, perform a recommended ranking on the candidate test questions to obtain the recommended weights of the candidate test questions;

[0008] Taking the highest sum of the recommended weights of the test questions included in the test paper as the goal, perform test paper generation based on the recommended weights of the candidate test questions to obtain the test paper.

[0009] According to the test paper generation method provided by the present invention, the step of taking the highest sum of the recommended weights of the test questions included in the test paper as the goal and performing test paper generation based on the recommended weights of the candidate test questions to obtain the test paper includes:

[0010] Taking the maximum number of questions in each question category in the test paper as a constraint condition, and taking the highest sum of the recommended weights of the questions included in the test paper as the goal, a test paper is generated based on the question categories and recommended weights of the candidate questions.

[0011] According to a test paper generation method provided by the present invention, the question categories of the candidate questions are determined based on the following steps:

[0012] Based on the similar question relationships between the questions, the question representations of the questions are determined;

[0013] Based on the question representations of the questions, the questions are clustered to obtain multiple question categories and the questions included in each question category.

[0014] According to a test paper generation method provided by the present invention, the determining the question representations of the questions based on the similar question relationships between the questions includes:

[0015] Taking the questions as nodes and the similar question relationships between the questions as edges, a question interaction graph is constructed;

[0016] Feature extraction is performed on each node in the question interaction graph to obtain the question representations of the questions corresponding to each node.

[0017] According to a test paper generation method provided by the present invention, the performing feature extraction on each node in the question interaction graph to obtain the question representations of the questions corresponding to each node includes:

[0018] The node interaction graphs of the nodes are input into a node feature extraction model to obtain the question representations of the questions corresponding to each node output by the node feature extraction model;

[0019] The node interaction graph is a representation form of the question interaction graph centered around the corresponding node, and the node feature extraction model is trained based on positive sample pairs and negative sample pairs. The positive sample pairs include two questions with a similar question relationship, and the negative sample pairs include two questions without a similar question relationship.

[0020] According to a test paper generation method provided by the present invention, the taking the maximum number of questions in each question category in the test paper as a constraint condition, and taking the highest sum of the recommended weights of the questions included in the test paper as the goal, and generating a test paper based on the question categories and recommended weights of the candidate questions includes:

[0021] Taking the maximum number of questions in each question category in the test paper as a constraint condition, and taking the highest sum of the recommended weights of the questions included in the test paper as the goal, based on the question categories and recommended weights of the candidate questions, test paper generation is performed on a preset test paper structure to obtain the test paper;

[0022] The preset test paper structure is determined based on the historical test paper structure of the object to be examined or the target object.

[0023] According to a test paper compilation method provided by the present invention, the application information of the target object corresponding to the candidate questions is determined based on the following steps:

[0024] Query the application information of the target object corresponding to the candidate questions in the usage information library. If it does not exist, determine the application information of the target object corresponding to the candidate questions based on the application information of each question under the question category to which the candidate questions belong for the target object.

[0025] According to a test paper compilation method provided by the present invention, the application information includes the number of times of adoption and / or the number of times of browsing but not adoption.

[0026] According to a test paper compilation method provided by the present invention, the determination of candidate questions includes:

[0027] Based on the knowledge points of each question and the knowledge points included in the assessment scope, select the questions within the assessment scope as candidate questions.

[0028] The present invention also provides a test paper compilation device, including:

[0029] A processing unit for determining candidate questions and a target object of the same object type as the object to be examined;

[0030] A sorting unit for performing recommended sorting on the candidate questions based on the application information of the target object corresponding to the candidate questions and the question information of the candidate questions to obtain the recommended weights of the candidate questions;

[0031] A test paper compilation unit for compiling a test paper based on the recommended weights of the candidate questions with the goal of maximizing the sum of the recommended weights of the questions included in the test paper to obtain the test paper.

[0032] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of any of the above-mentioned test paper compilation methods are implemented.

[0033] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned test paper compilation methods are implemented.

[0034] The test paper compilation method, device, electronic device, and storage medium provided by the present invention obtain the recommended weights of candidate questions based on the application information of the candidate questions corresponding to the target object belonging to the same object type as the object to be examined, enabling the recommendation and compilation of candidate questions to apply the application information of similar users, thereby maximizing the exploration of the test paper compilation preferences of the object to be examined; aiming at the highest sum of the recommended weights of the questions included in the test paper, compiling the test paper based on the recommended weights of the candidate questions can improve the efficiency of searching for the optimal combination of questions, shorten the test paper compilation time, improve the quality of the compiled test paper, and ensure the pertinence of the test paper to the object to be examined. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly describe the accompanying drawings required for use in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0036] Figure 1 It is a flowchart of the test paper compilation method provided by the present invention;

[0037] Figure 2 It is a flowchart of the question type determination method provided by the present invention;

[0038] Figure 3 It is a flowchart of step 210 in the question type determination method provided by the present invention;

[0039] Figure 4 It is an interaction diagram of questions provided by the present invention;

[0040] Figure 5 It is a node interaction diagram provided by the present invention;

[0041] Figure 6 It is a flowchart of the test paper compilation method provided by the present invention;

[0042] Figure 7 It is a structural diagram of the test paper compilation device provided by the present invention;

[0043] Figure 8 It is a structural diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] To make the objectives, technical solutions and advantages of the present invention more clear, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the protection scope of the present invention.

[0045] When compiling an exam paper, the proposition teacher usually searches for relevant test questions from resources such as review materials, exercise sets, and past exam papers according to the current exam requirements, and then selects the test questions that meet the exam requirements according to personal experience to assemble into a test paper. The above-mentioned manual paper compilation process mainly has three disadvantages: First, it is time-consuming and laborious to search for test questions manually, and it is difficult to make full use of the massive test question bank resources due to human limitations; second, the quality of the assembled test paper depends heavily on the personal intuition and experience of the teacher. A high-quality test paper generally needs to be completed by some education experts with relatively high levels after a long time of research. Third, it is difficult to separate teaching from examination in this process, so it cannot well reflect the students' knowledge mastery and teaching quality.

[0046] Automatic paper compilation is to automatically generate a test paper according to the paper compilation parameters input by the user. In the process of automatic paper compilation, both local constraint conditions (i.e., the objectives of the test question granularity) and global constraint conditions (i.e., the objectives of the test paper granularity) need to be taken into account to form a test paper that satisfies the user. Among them, the test question difficulty, score, knowledge points, etc. can be used to construct local constraint conditions, and the total number of questions, total score, total difficulty, proportion of different question quantities, different question knowledge point levels, different question knowledge point repetition problems, etc. in the test paper can be used to construct global constraint conditions. The common paper compilation algorithms in related technologies are generally the following two types:

[0047] 1) Paper compilation algorithm based on random extraction strategy: A method of randomly extracting test questions from the test question bank using a random function or variable;

[0048] 2) Paper compilation algorithm based on backtracking heuristic method: Before applying the backtracking heuristic method to extract a certain test question from the test question bank, first judge whether it meets the constraint conditions. If it meets, then extract the test question from the test question bank.

[0049] The above automatic paper compilation technology defines paper compilation as an optimization problem under constraint combinations. However, since this combinatorial optimization is a non-deterministic polynomial problem, the optimization time efficiency is low, and it is difficult to search for the optimal solution from the massive test question bank; and the test papers generated based on the above technology are mostly general assessment questions, which cannot meet the needs of targeted assessment of students, so it cannot well reflect the students' knowledge mastery and teaching quality.

[0050] In view of the above problems, the embodiments of the present invention provide a paper compilation method, Figure 1It is a schematic flowchart of the test paper compilation method provided by the present invention. As Figure 1 shown, the method includes:

[0051] Step 110, determining candidate test questions and a target object belonging to the same object type as the object to be examined.

[0052] Specifically, the candidate test questions are the test questions that can be used for test paper compilation. The candidate test questions can cover all the test questions in the test question bank, or can be the test questions within the assessment scope screened from the test question bank according to the assessment scope input by the user. The embodiments of the present invention do not make specific limitations on this.

[0053] The object to be examined is the object that needs to be examined through the test paper obtained by test paper compilation. The object to be examined can be a school as a unit or an education area as a unit. The embodiments of the present invention do not make specific limitations on this. Each object can be classified in advance to obtain the object type of each object. For example, in the case of determining the object with a school as a unit, each school can be classified according to factors such as the region where each school is located, the type of teaching materials used, and the pre-assessed teaching level. Thus, the learning situations of each school under the same school type are similar, and the preferences in test question selection and application are also similar.

[0054] Here, the target object belonging to the same object type as the object to be examined should be similar to the object to be examined in terms of learning situation. Specifically, in the test question selection and application during test paper compilation, it is also similar to the object to be examined. Therefore, the application information of the target object for the candidate test questions has reference value for the test paper compilation for the object to be examined.

[0055] Step 120, based on the application information of the target object corresponding to the candidate test questions and the test question information of the candidate test questions, performing a recommended ranking on the candidate test questions to obtain the recommended weights of the candidate test questions.

[0056] Specifically, the application information of the target object corresponding to the candidate test questions is used to characterize the application situation of the target object corresponding to the candidate test questions in test paper compilation, or is used to characterize the application situation of the combination of the test question attributes of the target object corresponding to the candidate test questions in test paper compilation. For example, it can include the number of times the target object uses the candidate test questions for test paper compilation, the number of times the target object browses the candidate test questions, and the number of times the target object browses the candidate test questions but does not use the candidate test questions for test paper compilation. The application information of the target object corresponding to the candidate test questions can reflect the preferences of the target object in test paper compilation, that is, whether the target object prefers to use the candidate test questions for test paper compilation, or the preference degree or recommendation degree of the target object for using the candidate test questions for test paper compilation.

[0057] The question information of the candidate questions reflects the information of the candidate questions themselves, which may specifically include the assessment content of the candidate questions, such as the semantics of the stem, answer, and analysis of the candidate questions, or may also include various attributes of the candidate questions, such as one or more of the question type, difficulty level, main knowledge points, secondary knowledge points, examination methods, scenarios, novelty, question categories, etc. Among them, the examination method refers to the assessment method of the analysis idea corresponding to the question, the scenario refers to the type of scenario described in the stem of the corresponding question, and the novelty is used to indicate whether the corresponding question is novel or to indicate the degree of novelty of the corresponding question.

[0058] In order to facilitate obtaining the optimal combination of questions for paper generation, it is possible to perform a recommended sorting for all current candidate questions, so as to obtain the recommended weight of each candidate question. Here, the recommended weight is used to indicate the recommended degree of the corresponding candidate question among all candidate questions. The higher the recommended weight, the higher the probability that the candidate question is a member of the optimal combination of questions for paper generation, and the higher the probability that the candidate question appears in the finally formed test paper.

[0059] Performing a recommended sorting for all current candidate questions needs to be carried out based on the application information of the target object corresponding to each candidate question and the question information of each candidate question. Further, if the application information of the target object corresponding to a candidate question reflects that the target object has a stronger preference for the candidate question in terms of paper generation, and the question information of the candidate question is more consistent with the question information to be examined, then the recommended sorting of the candidate question is more forward, and the recommended weight of the candidate question is higher; if the application information of the target object corresponding to a candidate question reflects that the target object has a weaker preference for the candidate question in terms of paper generation, and the degree of consistency between the question information of the candidate question and the question information to be examined is lower, then the recommended sorting of the candidate question is more backward, and the recommended weight of the candidate question is lower.

[0060] Here, based on the application information of the target object corresponding to the candidate questions and the question information of the candidate questions, performing a recommended sorting for the candidate questions can be implemented by any one or more of recommended algorithms such as Thompson sampling, UCB (Upper Confidence Bound) algorithm, and Epsilon-greedy algorithm.

[0061] In addition, it should be noted that the application information of the target object corresponding to the candidate questions can be updated in real time as each question in the question bank is browsed or used for paper generation by the user. Therefore, the recommended weights of each candidate question can also be updated in real time, which can further improve the user experience.

[0062] Step 130: Aim at the highest sum of the recommended weights of the questions included in the test paper, and perform paper generation based on the recommended weights of the candidate questions to obtain the test paper.

[0063] Specifically, after obtaining the recommended weights of each candidate question, the questions for forming a test paper can be screened from each candidate question based on the recommended weights of each candidate question, so as to realize test paper formation and obtain a test paper containing multiple questions.

[0064] In this process, in order to obtain the optimal combination of questions for forming a test paper, when screening the questions for forming a test paper from each candidate question, in addition to following various constraint conditions preset by the user, it is also necessary to screen with the goal of maximizing the sum of the recommended weights of the questions included in the test paper. In other words, there is an optimal combination goal when screening questions, that is, the sum of the recommended weights of the questions included in the test paper is the highest. The higher the recommended weight of the questions included in the test paper, the stronger the pertinence of the overall test paper to the object to be examined, and the more consistent it is with the content expected to be examined, and the more in line with the user's test paper formation requirements the overall test paper is.

[0065] The method provided by the embodiment of the present invention obtains the recommended weights of candidate questions based on the application information of candidate questions corresponding to a target object belonging to the same object type as the object to be examined, so that the recommendation and test paper formation for candidate questions can apply the application information of similar users, thereby being able to maximize the excavation of the test paper formation preferences of the object to be examined; with the goal of maximizing the sum of the recommended weights of the questions included in the test paper, forming a test paper based on the recommended weights of candidate questions can improve the efficiency of searching for the optimal combination of questions, shorten the test paper formation time, improve the quality of the test paper obtained by test paper formation, and ensure the pertinence of the test paper to the object to be examined.

[0066] In the process of automatic test paper formation, homogeneous questions may be introduced, affecting the quality of the test paper obtained by test paper formation. To solve this problem, based on the above embodiment, step 130 includes:

[0067] Taking the maximum value of the number of questions of each question category in the test paper as a constraint condition, and taking the sum of the recommended weights of the questions included in the test paper as the highest goal, form a test paper based on the question categories and recommended weights of the candidate questions to obtain the test paper.

[0068] Specifically, to solve the problem of homogeneous questions in the test paper, each question can be pre-classified by question type, so as to obtain the question types of each question. The question types here are used to realize the classification of homogeneous questions, that is, questions belonging to the same question type are homogeneous questions.

[0069] After determining the question types of each candidate question, during the test paper compilation process, the maximum number of questions of each question type in the test paper can be used as a constraint condition to limit the number of questions of the same question type in the optimal combination of questions selected for test paper compilation, thereby avoiding the problem of homogenization of questions in the test paper. For example, the maximum number of questions of each question type in the constraint condition can be set to 1. Under this constraint condition, the number of questions of each question type in the test paper obtained by test paper compilation will not exceed 1, that is, there are no homogenized questions among the questions in the test paper.

[0070] It should be noted that the maximum number of questions of each question type in the constraint condition here can be set to 1, or can be set to 2 or 3, etc. Specifically, it can be adjusted according to factors such as the overall number of questions required for the test paper and the user's acceptance degree of homogenized questions. The embodiments of the present invention do not make specific limitations.

[0071] The method provided by the embodiments of the present invention solves the problem of homogenization in test paper compilation by using the maximum number of questions of each question type in the test paper as a constraint condition, and further ensures the quality of test paper compilation.

[0072] Based on any of the above embodiments, Figure 2 is a flowchart of the method for determining question types provided by the present invention. As Figure 2 shown, the question types of the candidate questions are determined based on the following steps:

[0073] Step 210, based on the similar question relationship between each question, determine the question representation of each question;

[0074] Step 220, based on the question representations of each question, cluster each question to obtain multiple question types and the questions included in each question type.

[0075] Specifically, the similar question relationship is used to represent that the corresponding two questions are similar questions to each other, and the similar question relationship can be obtained by pre-artificial annotation. For any two questions, if there is a similar question relationship between these two questions, it can be considered that these two questions are similar questions to each other. If there is no similar question relationship between these two similar questions, it can be considered that these two questions are independent and have no association.

[0076] Considering that two test questions that are similar to each other must have the same or similar aspects in one or more of the attributes such as question type, difficulty level, main knowledge points, secondary knowledge points, test methods, scenarios, novelty, and question categories, one test question can be regarded as another manifestation of the other test question. Therefore, when extracting the test question representation for each test question, not only the information of the test question itself but also the information of the test questions similar to it can be considered to improve the reliability and accuracy of the test question representation. And because similar test questions draw on each other during the extraction of the test question representation, the test question representations of similar test questions will also be closer to each other.

[0077] After obtaining the test question representations of each test question, clustering can be performed on each test question based on its test question representation. During the clustering process, test questions with similar test question representations are grouped into the same test question category, and test questions with significantly different test question representations are grouped into different test question categories, thereby achieving unsupervised clustering based on the test question representation, obtaining multiple test question categories, and the test questions included in each test question category. It should be noted that the method of test question clustering here can be the hierarchical clustering algorithm, or the K - MEANS clustering algorithm, the DBSCAN (Density - Based Spatial Clustering of Applications with Noise) clustering algorithm, etc. The embodiments of the present invention do not make specific limitations on this.

[0078] For example, when applying the hierarchical clustering algorithm for test question clustering, for a new test question, it is necessary to calculate the distance between the test question representation of the new test question and the centers of each existing test question category. If all distances exceed the threshold, a new test question category needs to be added. If there is a distance less than the threshold, the new test question is assigned to the test question category corresponding to that distance.

[0079] Based on any of the above embodiments, Figure 3 is the flowchart of step 210 in the method for determining test question categories provided by the present invention, as Figure 3 shown, step 210 includes:

[0080] Step 211, constructing a test question interaction graph with each test question as a node and the similar question relationship between each test question as an edge;

[0081] Step 212, extracting features from each node in the test question interaction graph to obtain the test question representation of the test question corresponding to each node.

[0082] Specifically, when manually annotating the similarity relationship between questions, the data usually focused on is partial, that is, more attention is paid to whether two questions are similar. The question interaction graph constructed based on the similarity relationship between questions can view the similarity relationship between questions from a global perspective. For example, Figure 4 is the question interaction graph provided by the present invention, Figure 4 each circle in it represents a node, that is, a question. For example, i1 - i7 represent questions 1 to 7 respectively. Among them, there are similarity relationships between questions 1 and 2, 1 and 3, 2 and 3, 2 and 7, 3 and 4, 3 and 5, 4 and 5. Correspondingly, Figure 4 in it, there are connected edges between nodes i1, i2, i1, i3, i2, i3, i2, i7, i3, i4, i3, i5, i4, i5.

[0083] After obtaining the question interaction graph, the question interaction graph can be feature - extracted by means of graph convolution (Graph Convolutional Network, GCN), spectral clustering, etc., so as to obtain the features of each node in the question interaction graph, that is, the question representation of the question corresponding to each node. The question representations of each question obtained based on the question interaction graph can not only cover the information of the question itself, but also cover the information of the questions that have direct or indirect similarity relationships with the question. Thus, the question representations of similar questions will also be closer, which helps to improve the reliability of subsequent question classification.

[0084] Based on any of the above - mentioned embodiments, step 212 includes:

[0085] Input the node interaction graph of each node into the node feature extraction model, and obtain the question representation of the question corresponding to each node output by the node feature extraction model;

[0086] The node interaction graph is a manifestation form of the question interaction graph centered around the corresponding node. The node feature extraction model is trained based on positive sample pairs and negative sample pairs. The positive sample pairs include two questions with a similarity relationship, and the negative sample pairs include two questions without a similarity relationship.

[0087] Specifically, the extraction of the question representation of each question can be achieved through a node feature extraction model. Here, the node feature extraction model can aggregate the neighbor information of a node in the node interaction graph based on the node interaction graph of any input node, so as to obtain the question representation of the question corresponding to the node.

[0088] Here, the node interaction graph is for a single node. The relationship between the node interaction graph and the question interaction graph regarding the similar question relationship among questions can be understood as two different forms of representation. Specifically, the node interaction graph is a question interaction graph in an expanded form centered around a single node. For example, Figure 5 is the node interaction graph provided by the present invention, Figure 5 specifically, it is the node interaction graph of node i2 corresponding to question 2, Figure 5 in which the question interaction graph in it expands around i2, and then the high-order interaction information of node i2 is obtained through the node feature extraction model and used as the question representation of question 2 corresponding to node i2.

[0089] Before performing step 212, the node feature extraction model can also be pre-trained. Specifically, the model training can be carried out through the following method:

[0090] First, a large number of positive sample pairs and negative sample pairs are collected; among them, the positive sample pairs cover two questions with a similar question relationship, and the negative sample pairs cover two questions without a similar question relationship. Subsequently, the node interaction graphs of the two questions in the positive sample pairs can be input into the initial model for training. During the training process of the initial model, the common features between the node interaction graphs of the two questions in the positive sample pairs can be magnified and learned; in addition, the node interaction graphs of the two questions in the negative sample pairs are input into the initial model for training. During the training process of the initial model, the differential features between the node interaction graphs of the two questions in the negative sample pairs can be magnified and learned. The node feature extraction model trained in this way can better represent the common features of questions with a similar question relationship in terms of question features, and the differential features of questions without a similar question relationship in terms of question features.

[0091] Specifically in the training operation, for the question representations of the two questions in the positive sample pairs output by the initial model, the initial model parameters can be iteratively adjusted to make the difference between the question representations of the two questions in the positive sample pairs as small as possible. For the question representations of the two questions in the negative sample pairs output by the initial model, the initial model parameters can be iteratively adjusted to make the difference between the question representations of the two questions in the negative sample pairs as large as possible, thereby achieving the effect of magnifying the common features of questions with a similar question relationship in terms of question features, and the differential features of questions without a similar question relationship in terms of question features.

[0092] Furthermore, the positive sample pairs for training can be determined based on the similar question relationship among questions, and the negative sample pairs for training can be obtained by randomly sampling questions.

[0093] The method provided by the embodiments of the present invention can ensure the reliability of question representation extraction based on the node feature extraction model trained with positive sample pairs and negative sample pairs.

[0094] Based on any of the above embodiments, referring to Figure 5 the node interaction diagram shown in Figure 5 , the process of extracting test question representations based on the node feature extraction model may include the following steps:

[0095] For layer = 0, that is, Figure 5 for the node i2 in the inner circle of the two virtual circles in Figure 5 , its initial node vector can be encoded based on the test question information of the test question 2 corresponding to i2; for layer = k, where k is an integer greater than or equal to 1, the node vectors of the neighbor nodes of the node in the upper layer and the node vector of the node in the upper layer can be aggregated to obtain the node vector of the node in the current layer. Taking layer = 1 as an example, the corresponding nodes in layer = 1 are Figure 5 the nodes i1, i3, and i7 in the inner circle of the two virtual circles in Figure 5 . The neighbor node of node i1 in the upper layer is i2, and the node vector of node i1 at layer = 1 can be determined based on the node vector of node i2 at layer = 0, that is, the initial node vector and the node vector of node i1 at layer = 0 to determine.

[0096] Further, for layer = k, the node vectors of all neighbor nodes in the upper layer and the node vector of the node itself in the previous layer can be added, and through regularization and activation functions, the node vector of the node at layer = k is determined, which can be specifically expressed as the following formula:

[0097]

[0098] In the formula, is the node vector of the i-th node at layer = k, σ(·) represents the activation function, is the node vector of the i-th node at layer = k - 1, is the node vector of the j-th node at layer = k - 1, N(i) is the set of neighbor nodes of the i-th node, j ∈ N(i) means that the j-th node belongs to the set of neighbor nodes of the i-th node, and |N(i)| represents the number of nodes in the set of neighbor nodes, and are the pre-trained weight parameters of layer = k.

[0099] Based on any of the above embodiments, step 130 includes:

[0100] Taking the maximum value of the number of test questions in each test question category in the test paper as a constraint condition, and taking the highest sum of the recommended weights of the test questions included in the test paper as the goal, based on the test question categories and recommended weights of the candidate test questions, a test paper is generated on the preset test paper structure to obtain the test paper;

[0101] The preset test paper structure is determined based on the historical test paper structure of the object to be assessed or the target object.

[0102] Specifically, generally, constructing a complete test paper needs to meet certain structures, such as the total number of questions, the difficulty ratio, the ratio of different question types, etc. If these parameters are all input by the user, it will inevitably increase the user's usage cost. To address this issue, the embodiments of the present invention propose to determine the preset test paper structure applied to this test paper compilation based on the historical test paper structure of the object to be assessed or the target object.

[0103] For example, it can first be determined whether there is a historical test paper structure of the object to be assessed. If there is, the historical test paper structure of the object to be assessed can be provided for the user to select, or the historical test paper structure with the highest usage times or the most recently used can be directly selected and recommended to the user for test paper compilation. If there is no historical test paper structure of the object to be assessed, that is, the object to be assessed may itself be a new user, the historical test paper structure of the target object belonging to the same object category as the object to be assessed can be provided to facilitate the user to determine the preset test paper structure.

[0104] Based on any of the above embodiments, combined with the question types, recommended weights of each candidate question, and the test paper compilation algorithm of the preset test paper structure, it can be optimized into the following maximization problem:

[0105]

[0106] In the formula, R m,n is a binary matrix of size m*n, where m is the number of candidate questions and n is the number of target questions for this test paper compilation. w ij is the weight value of candidate question i placed at position j, and w i1 is the recommended weight value of candidate question i, and L kl and U kl respectively represent the lower limit quantity and the upper limit quantity of attribute l at position k. Taking the following table as an example, there are 7 candidate questions with the difficulty levels of easy, easy, difficult, easy, easy, difficult, and difficult in sequence. The total number of questions indicated in the preset test paper structure is 6 questions, the lower limit quantity of easy questions is 3 questions, the lower limit quantity of difficult questions is 3 questions, and the lower limit quantity of easy questions is 2 questions and the lower limit quantity of difficult questions is 2 questions for the first 4 questions. If there are no difficulty condition restrictions, the sum of the weight values on the diagonal is the optimal result. If there are difficulty L and U restrictions, the questions at positions 4-6 correspond to the 6th, 4th, and 7th questions in the sorting layer:

[0107] 1 2 3 4 5 6 Easy 0.97 0.94 0.90 0.78 0.74 0.71 Easy 0.93 0.90 0.82 0.74 0.71 0.68 Difficult 0.89 0.86 0.82 0.71 0.58 0.65 Easy 0.81 0.79 0.75 0.65 0.62 0.59 Easy 0.73 0.71 0.68 0.58 0.56 0.53 Difficult 0.72 0.69 0.66 0.57 0.55 0.52 Difficult 0.64 0.61 0.59 0.51 0.48 0.46

[0108] It should be noted that the above table is only an example under the difficulty limit. In actual operation, there can be multiple restrictive conditions for attributes in China, which will not be elaborated here. During the above-mentioned test paper compilation process, the test questions selected at the later positions and the previously selected test questions cannot belong to the same test question category to avoid the problem of homogenization.

[0109] Based on any of the above embodiments, the application information of the target object corresponding to the candidate test question is determined based on the following steps:

[0110] Query the application information of the target object corresponding to the candidate test question in the usage information database. If it does not exist, determine the application information of the target object corresponding to the candidate test question based on the application information of each test question under the test question category to which the candidate test question to which the target object belongs.

[0111] Specifically, the usage information database can be used to store the application information of each object corresponding to each test question. After determining the target object and the candidate test question, the application information of the target object corresponding to the candidate test question can be queried from the usage information database, and the application information of the target object corresponding to the candidate test question can be determined based on the query result:

[0112] If the usage information database stores the application information of the target object corresponding to the candidate test question, that is, the information can be directly queried, then the information can be directly obtained;

[0113] If the usage information database does not store the application information of the target object corresponding to the candidate test question, that is, the information cannot be queried, at this time, the test question category of the candidate test question can be determined based on the test question categories pre-constructed for each test question, so as to locate other test questions belonging to the same test question category as the candidate test question, and query the application information of the target object corresponding to other test questions belonging to the same test question category as the candidate test question in the usage information database. After obtaining the application information of other test questions, the application information of the target object corresponding to the candidate test question can be estimated therefrom. Here, when estimating the application information of the target object corresponding to the candidate test question, the application information of each test question under the same test question category can be averaged, or the median can be obtained and other methods to obtain the application information of the target object corresponding to the candidate test question.

[0114] Based on any of the above embodiments, the application information includes the number of times of adoption and / or the number of times of browsing but not adopting.

[0115] Among them, for any candidate test question, the number of times of adoption specifically refers to the number of times the target object uses the candidate test question to compile a test paper. The number of times of adoption can directly reflect the positive preference of the target object for the candidate test question in terms of test paper compilation. The higher the number of times of adoption, the stronger the preference of the target object for the candidate test question in terms of test paper compilation. The lower the number of times of adoption, the weaker the preference of the target object for the candidate test question in terms of test paper compilation.

[0116] The number of times of browsing but not adopting specifically refers to the number of times that the target object browses the candidate question during the process of generating a test paper but does not adopt the candidate question for generating the test paper. The number of times of browsing but not adopting can also reflect the preference of the target object for the candidate question in terms of generating a test paper. The higher the number of times of browsing but not adopting, the weaker the preference of the target object for the candidate question in terms of generating a test paper; the lower the number of times of browsing but not adopting, the stronger the preference of the target object for the candidate question in terms of generating a test paper.

[0117] It should be noted that the application information of the target object corresponding to the candidate question can be updated in real time as the questions in the question bank are browsed or used for generating test papers by users. For example, if any user under the target object browses any candidate question and uses the candidate question for generating a test paper, the adoption times of the candidate question will be incremented by 1; otherwise, the number of times of browsing but not adopting will be incremented by 1.

[0118] Based on any of the above embodiments, in step 110, determining the candidate questions includes:

[0119] Selecting the questions within the assessment scope as candidate questions based on the knowledge points of each question and the knowledge points included in the assessment scope.

[0120] Specifically, the assessment scope usually increases continuously with the progress of teaching. Questions beyond the assessment scope are meaningless to the object to be assessed and will directly affect the user experience. Therefore, when generating a test paper, users usually directly limit the assessment scope. Here, the assessment scope can be the specific knowledge points to be assessed, or the specific teaching chapters to be assessed. And since there is often a corresponding relationship between teaching chapters and knowledge points, under the above two forms of expression, they can be unified into the form of limiting the scope by knowledge points.

[0121] After determining the assessment scope, the knowledge points covered by the assessment scope can be determined, and then questions with knowledge points within the assessment scope can be screened out from the question bank as candidate questions for further screening in generating a test paper. The candidate questions obtained in this way are all within the assessment scope. By generating a test paper based on this, the situation that the test paper generated contains questions beyond the syllabus can be avoided, thus achieving the purpose of optimizing the user experience.

[0122] Based on any of the above embodiments, Figure 6 is a schematic flowchart of the test paper generation method provided by the present invention. As Figure 6 shown, the test paper generation method can be implemented through the following four steps:

[0123] 610, preprocessing:

[0124] The preprocessing stage is used to sort out and summarize the relevant attributes of each question in the question bank and organize the user information of each user.

[0125] Here, the relevant attributes of the test questions can be understood in the form of a test question portrait. The relevant attributes can cover manually labeled attributes such as the question type, difficulty level, main knowledge points, and secondary knowledge points of the test questions, and can also cover attributes such as the test method, context, novelty, and test question category that need to be obtained through natural language processing (NLP).

[0126] For example, for the test question information of each test question pre-stored in the test question bank, that is, the corpus of each test question, the corpus can be uniformly processed into a text format through a text processing module, and then the processed test question text can be used as a sample and input into a pre-trained Bert language model. Combining the test method, context, and novelty annotation data corresponding to the test question text used as the sample, a test method, context, and novelty prediction model can be trained on the basis of the Bert model, so that the attributes of the test method, context, and novelty of a large number of test questions in the test question bank can be annotated through the test method, context, and novelty prediction model. For example, when the test question corpus is stored in html format, html parsing, formula parsing, and text word segmentation can be performed through a text processing module to obtain the test question in text format.

[0127] For another example, for the test question category of each test question, each test question can be used as a node, and the relationship between similar questions among the test questions can be used as an edge to construct a test question interaction graph, and feature extraction can be performed on each node in the test question interaction graph to obtain the test question representation corresponding to each node. On this basis, based on the test question representation of each test question, clustering is performed on each test question to obtain multiple test question categories and the test questions included in each test question category.

[0128] The user information of each user can include attributes such as the user's school, region, and school category collected.

[0129] 620, Recall:

[0130] The knowledge points covered by the assessment scope can be determined, and then, from the test question bank, test questions whose knowledge points are within the assessment scope can be screened out as candidate test questions for further test paper screening, so as to avoid the problem of knowledge points exceeding the syllabus.

[0131] 630, Sorting:

[0132] Considering that the degree of mastery of each school in each knowledge and the focus of the exam will vary, it is necessary to sort the candidate test questions retrieved by recall differently according to the object:

[0133] In an embodiment of the present invention, the object to be evaluated is any school. Specifically, according to the school types divided by each school, a school belonging to the same school type as the school to be evaluated can be selected as the target object. By counting the application information of each user under the target object, the application information of the target object corresponding to each candidate question can be obtained. Specifically, it is manifested as the adoption times a and the browsing but non - adoption times b of the question - attribute combinations corresponding to the target object for each candidate question, which can be specifically expressed in the form shown in the following table:

[0134] Question type m Difficulty n Main knowledge point k …… Examination method l a b m1 n1 k1 …… l1 a1 b1 …… …… …… …… …… …… …… mi ni ki …… li ai bi

[0135] After obtaining the a - value and b - value of each candidate question corresponding to the target object, the following formula can be applied to form the recommendation weight of each candidate question:

[0136]

[0137] In the formula, beta(x; a, b) is the recommendation weight of the candidate question, x is the question information of the candidate question, Γ represents the gamma function; the mean of the beta distribution is and the variance is If the number of occurrences of the question - attribute combination is very large, that is, a + b is very large, then its sampling value will be very stable and close to the mean; if the question - attribute combination not only has a + b very large but also a very large, then this combination often appears in the exam, which exactly coincides with the key and difficult points in the education field; on the contrary, if a is very small, it belongs to very basic points and will not often appear in the exam; if a + b is very small, it is possible to get a relatively large random number or a relatively small random number when generating a random number; thus, the questions that coincide with the key and difficult points in the education field may be preferentially output when sorting.

[0138] 640, post - processing:

[0139] Taking the maximum value of the number of questions in each question category in the test paper as a constraint condition, and taking the highest sum of the recommendation weights of the questions included in the test paper as the goal, based on the question categories and recommendation weights of the selected questions, a test paper is assembled on the preset test - paper structure to obtain a complete test paper.

[0140] Based on any of the above embodiments, Figure 7 is a schematic structural diagram of the test - paper assembling device provided by the present invention, as Figure 7 shown. The device includes:

[0141] A processing unit 710, configured to determine candidate questions and a target object belonging to the same object type as the object to be evaluated;

[0142] A sorting unit 720, configured to perform a recommendation sort on the candidate questions based on the application information of the target object corresponding to the candidate questions and the question information of the candidate questions, so as to obtain the recommendation weights of the candidate questions;

[0143] A test paper compilation unit 730 is configured to compile a test paper based on the recommended weights of the candidate questions with the goal of maximizing the sum of the recommended weights of the questions included in the test paper.

[0144] The device provided by the embodiment of the present invention obtains the recommended weights of candidate questions based on the application information of the candidate questions corresponding to the target objects belonging to the same object type as the object to be examined, enabling the recommendation and compilation of candidate questions to apply the application information of similar users, thereby being able to maximize the exploration of the test paper compilation preferences of the object to be examined; with the goal of maximizing the sum of the recommended weights of the questions included in the test paper, compiling a test paper based on the recommended weights of the candidate questions can improve the efficiency of searching for the optimal combination of questions, shorten the test paper compilation time, improve the quality of the compiled test paper, and ensure the pertinence of the test paper to the object to be examined.

[0145] Based on any of the above embodiments, the test paper compilation unit 730 is configured to:

[0146] With the maximum number of questions in each question category in the test paper as a constraint condition and the goal of maximizing the sum of the recommended weights of the questions included in the test paper, compile a test paper based on the question categories and recommended weights of the candidate questions to obtain the test paper.

[0147] Based on any of the above embodiments, the device further includes a question category determination unit configured to:

[0148] Determine the question representations of each question based on the similar question relationships between the questions;

[0149] Cluster each question based on the question representations of each question to obtain multiple question categories and the questions included in each question category.

[0150] Based on any of the above embodiments, the question category determination unit is configured to:

[0151] Construct a question interaction graph with each question as a node and the similar question relationships between the questions as edges;

[0152] Extract features from each node in the question interaction graph to obtain the question representations of the questions corresponding to each node.

[0153] Based on any of the above embodiments, the question category determination unit is configured to:

[0154] Input the node interaction graph of each node into a node feature extraction model to obtain the question representations of the questions corresponding to each node output by the node feature extraction model;

[0155] The node interaction graph is a manifestation form of the test question interaction graph centered around the corresponding node. The node feature extraction model is trained based on positive sample pairs and negative sample pairs. The positive sample pairs include two test questions with a similar question relationship, and the negative sample pairs include two test questions without a similar question relationship.

[0156] Based on any of the above embodiments, the test paper compilation unit 730 is configured to:

[0157] Taking the maximum value of the number of test questions of each test question category in the test paper as a constraint condition, and taking the highest sum of the recommended weights of the test questions included in the test paper as a goal, based on the test question categories and recommended weights of the candidate test questions, compile a test paper on a preset test paper structure to obtain the test paper;

[0158] The preset test paper structure is determined based on the historical test paper structure of the object to be assessed or the target object.

[0159] Based on any of the above embodiments, the processing unit is configured to:

[0160] Query the application information of the target object corresponding to the candidate test question in the usage information library. If it does not exist, determine the application information of the target object corresponding to the candidate test question based on the application information of each test question under the test question category to which the candidate test question belongs for the target object.

[0161] Based on any of the above embodiments, the application information includes the number of adoption times and / or the number of views without adoption.

[0162] Based on any of the above embodiments, the processing unit is configured to:

[0163] Based on the knowledge points of each test question and the knowledge points included in the assessment scope, select the test questions within the assessment scope as candidate test questions.

[0164] Figure 8 Illustrates a schematic physical structure diagram of an electronic device, as Figure 8 shown. The electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840. Among them, the processor 810, the communication interface 820, and the memory 830 complete mutual communication through the communication bus 840. The processor 810 can call the logical instructions in the memory 830 to execute the test paper compilation method, and the method includes:

[0165] Determine candidate test questions and a target object of the same object type as the object to be assessed;

[0166] Based on the application information of the candidate questions corresponding to the target object and the question information of the candidate questions, perform a recommended ranking on the candidate questions to obtain the recommended weights of the candidate questions;

[0167] Aiming at the highest sum of the recommended weights of the questions included in the test paper, perform test paper compilation based on the recommended weights of the candidate questions to obtain the test paper.

[0168] In addition, when the logical instructions in the above-mentioned memory 830 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0169] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the test paper compilation method provided by the above-mentioned various methods. The method includes:

[0170] Determine candidate questions and a target object of the same object type as the object to be examined;

[0171] Based on the application information of the candidate questions corresponding to the target object and the question information of the candidate questions, perform a recommended ranking on the candidate questions to obtain the recommended weights of the candidate questions;

[0172] Aiming at the highest sum of the recommended weights of the questions included in the test paper, perform test paper compilation based on the recommended weights of the candidate questions to obtain the test paper.

[0173] On yet another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the test paper compilation method provided by the above-mentioned various methods. The method includes:

[0174] Determine candidate questions and a target object of the same object type as the object to be examined;

[0175] Based on the application information of the candidate test questions corresponding to the target object and the test question information of the candidate test questions, perform a recommended sorting on the candidate test questions to obtain the recommended weights of the candidate test questions;

[0176] Aiming at the highest sum of the recommended weights of the test questions included in the test paper, form a test paper based on the recommended weights of the candidate test questions to obtain the test paper.

[0177] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0178] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course also by hardware. Based on such an understanding, the above technical solutions, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0179] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A test paper compilation method, characterized in that, it includes: determining candidate test questions and a target object of the same object type as the object to be examined; recommending and ranking the candidate test questions based on the application information of the target object corresponding to the candidate test questions and the test question information of the candidate test questions, to obtain the recommended weights of the candidate test questions. The stronger the preference of the target object for the candidate test questions in test paper compilation reflected by the application information, and the more consistent the test question information with the test question information to be examined, the higher the recommended ranking of the candidate test questions; taking the highest sum of the recommended weights of the test questions included in the test paper as the goal, and compiling a test paper based on the recommended weights of the candidate test questions to obtain the test paper; wherein, the step of taking the highest sum of the recommended weights of the test questions included in the test paper as the goal, and compiling a test paper based on the recommended weights of the candidate test questions to obtain the test paper includes: taking the maximum value of the number of test questions of each test question category in the test paper as a constraint condition, taking the highest sum of the recommended weights of the test questions included in the test paper as the goal, and compiling a test paper based on the test question categories and recommended weights of the candidate test questions to obtain the test paper; wherein, the test question category is represented based on the similarity relationship between test questions.

2. The test paper compilation method according to claim 1, characterized in that, the test question category of the candidate test questions is determined based on the following steps: determining the test question representations of each test question based on the similar question relationship between each test question; clustering each test question based on the test question representations of each test question to obtain multiple test question categories and the test questions included in each test question category.

3. The test paper compilation method according to claim 2, characterized in that, the step of determining the test question representations of each test question based on the similar question relationship between each test question includes: constructing a test question interaction graph with each test question as a node and the similar question relationship between each test question as an edge; extracting features from the test question interaction graph to obtain the test question representations of the test questions corresponding to each node.

4. The test paper compilation method according to claim 3, characterized in that, the step of extracting features from the test question interaction graph to obtain the test question representations of the test questions corresponding to each node includes: inputting the node interaction graph of each node into a node feature extraction model to obtain the test question representations of the test questions corresponding to each node output by the node feature extraction model; the node interaction graph is a representation form of the test question interaction graph centered around the corresponding node, and the node feature extraction model is trained based on positive sample pairs and negative sample pairs. The positive sample pairs include two test questions with a similar question relationship, and the negative sample pairs include two test questions without a similar question relationship.

5. The test paper compilation method according to claim 1, characterized in that, the step of taking the maximum value of the number of test questions of each test question category in the test paper as a constraint condition, taking the highest sum of the recommended weights of the test questions included in the test paper as the goal, and compiling a test paper based on the test question categories and recommended weights of the candidate test questions to obtain the test paper includes: Taking the maximum number of questions in each question category in the test paper as a constraint condition, and taking the highest sum of the recommended weights of the questions included in the test paper as the goal, based on the question categories and recommended weights of the candidate questions, a test paper is generated on a preset test paper structure to obtain the test paper; The preset test paper structure is determined based on the historical test paper structure of the object to be assessed or the target object.

6. The test paper generation method according to any one of claims 1 to 5, characterized in that, The application information of the target object corresponding to the candidate questions is determined based on the following steps: Query the application information of the target object corresponding to the candidate questions in the usage information library. If it does not exist, then based on the application information of each question under the question category to which the candidate questions to which the target object belongs, determine the application information of the target object corresponding to the candidate questions.

7. The test paper generation method according to any one of claims 1 to 5, characterized in that, The application information includes the number of times of adoption and / or the number of times of browsing but not adoption.

8. The test paper generation method according to any one of claims 1 to 5, characterized in that, The determination of candidate questions includes: Based on the knowledge points of each question and the knowledge points included in the assessment scope, select the questions within the assessment scope as candidate questions.

9. A test paper generation device, characterized in that, including: A processing unit for determining candidate questions and a target object of the same object type as the object to be assessed; A sorting unit for recommending and sorting the candidate questions based on the application information of the target object corresponding to the candidate questions and the question information of the candidate questions to obtain the recommended weights of the candidate questions. The stronger the preference of the target object for the candidate questions in terms of test paper generation reflected by the application information, and the more consistent the question information with the question information to be assessed, the higher the recommended ranking of the candidate questions; A test paper generation unit for taking the highest sum of the recommended weights of the questions included in the test paper as the goal and generating a test paper based on the recommended weights of the candidate questions to obtain the test paper; wherein, taking the highest sum of the recommended weights of the questions included in the test paper as the goal and generating a test paper based on the recommended weights of the candidate questions to obtain the test paper includes: Taking the maximum number of questions in each question category in the test paper as a constraint condition, and taking the highest sum of the recommended weights of the questions included in the test paper as the goal, and generating a test paper based on the question categories and recommended weights of the candidate questions to obtain the test paper; wherein, the question category is represented based on the similarity relationship between questions.

10. An electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the test paper generation method according to any one of claims 1 to 8.

11. A non-transitory computer-readable storage medium, on which a computer program is stored, characterized in that, When the computer program is executed by a processor, it implements the steps of the test paper generation method according to any one of claims 1 to 8.

Images