A method and system for checking duplicate questions in a computer-based examination question bank

Abstract

The present application provides a method and system for checking duplicate content of a computer-based test question bank, including: using questions in the computer-based test question bank to perform semantic understanding and context association analysis on the question content to obtain a multi-dimensional similarity map between the questions; based on the multi-dimensional similarity map, using advanced graph analysis technology to process the association relationship between the questions to obtain a similarity matrix; using the similarity matrix to cluster the questions in the computer-based test question bank through a clustering algorithm to obtain question groups; using the question groups to check duplicate content of the questions in the question bank, generate a question bank duplicate checking report, and improve the accuracy and efficiency of checking duplicate content of the computer-based test question bank.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of computer-based examinations, and more particularly to a method and system for checking duplicate questions in a computer-based examination question bank. Background Art

[0002] With the popularity of online education and computer-assisted instruction, the management and maintenance of computer-based exam question banks have become increasingly important. Computer-based exam question banks usually contain a large number of programming questions, multiple-choice questions, fill-in-the-blank questions, etc. These questions need to be updated and maintained regularly to ensure the fairness and effectiveness of the exam. In question bank management, duplicate checking is a key task, the purpose of which is to discover and eliminate duplicate or highly similar questions, avoid the repetition of exam content, and improve the diversity and challenge of the question bank.

[0003] At present, the duplicate checking of computer-based test question banks mainly relies on manual review and simple text matching methods. Manual review is time-consuming and labor-intensive, and may lead to misjudgment due to subjective factors; while simple text matching methods can only detect completely consistent or highly similar text content. Therefore, the existing duplicate checking methods cannot meet the needs of rapid updating of computer-based test question banks, and there are high false positive and false negative rates. Summary of the invention

[0004] The embodiments of the present application provide a method and system for checking duplicate examination question banks in a computer-based examination, so as to solve the problems existing in the background technology.

[0005] In a first aspect, an embodiment of the present application provides a method for checking for duplicate content in a computer-based test question bank, comprising: using questions in the computer-based test question bank to perform semantic understanding and context association analysis on the content of the questions, and obtain a multidimensional similarity map between the questions, wherein the multidimensional similarity map is used to characterize the similarity of the textual expressions of the questions, as well as the similarity of the problem-solving ideas, knowledge fields, and logical structures of the questions; Based on the multidimensional similarity map, using advanced graph analysis technology to process the correlation between the questions, and obtain a similarity matrix, wherein the similarity matrix is ​​used to characterize the similarity between the questions; Using the similarity matrix, clustering the questions in the computer-based test question bank through a clustering algorithm to obtain question groups, wherein the question groups are a set of questions whose similarities in multiple dimensions are greater than a preset threshold; Using the question groups, checking for duplicate content in the questions in the question bank, and generating a question bank duplicate checking report, wherein the duplicate checking report includes: a list of questions in the question group and similarity scores of the questions in the question group.

[0006] Furthermore, using the questions in the computer-based test question bank, semantic understanding and context association analysis are performed on the question content to obtain a multi-dimensional similarity map between the questions, including: performing natural language processing on the questions in the computer-based test question bank, and combining the non-text elements in the questions to generate basic semantic information of the questions, wherein the basic semantic information includes: word segmentation results, part-of-speech tags, entity names, grammatical relationships between word segmentations and the non-text elements; based on the basic semantic information of the questions, using a deep learning model to perform semantic understanding on the questions to generate a semantic vector representation of the questions; based on the semantic vector representation of the questions and the metadata information of the questions, using graph embedding technology, multi-dimensional similarity analysis is performed on the relationship between the questions to generate an initial multi-dimensional similarity map between the questions, wherein the metadata includes: the knowledge field corresponding to the questions and the difficulty level of the questions; using a graph neural network algorithm, the initial multi-dimensional similarity map is analyzed to obtain the multi-dimensional similarity map.

[0007] Furthermore, the initial multidimensional similarity map is analyzed using a graph neural network algorithm to obtain the multidimensional similarity map, including: using the initial multidimensional similarity map to train a preset graph neural network model to obtain a trained graph neural network model; using the trained graph neural network model to update the feature vectors of the nodes in the initial multidimensional similarity map, and calculating the similarity between the nodes through the updated node feature vectors to obtain a similarity score; using the similarity score to generate the multidimensional similarity map.

[0008] Furthermore, based on the multi-dimensional similarity map, an advanced graph analysis technology is used to process the association relationship between the questions to obtain a similarity matrix, including: based on the multi-dimensional similarity map, a correlation map between the questions is constructed, wherein the nodes in the correlation map are the questions, the similarities between the questions are used as the edges of the correlation map, and the weights of the edges of the correlation map are determined by the similarities represented by the multimodal semantic vectors of the questions; based on the correlation map, a graph convolutional network algorithm is applied to iteratively process the nodes and edges in the correlation map, and information of neighboring nodes is aggregated through multiple convolution operations to generate the enhanced correlation map; based on the enhanced correlation map, and using the attention mechanism, the weights of the similarities between the nodes in the enhanced correlation map are dynamically adjusted, and based on the adjusted weights, comprehensive similarity information is generated; using the comprehensive similarity information, the final similarity scores between the questions are calculated to generate the similarity matrix.

[0009] Furthermore, the similarity matrix is ​​used to cluster the questions in the computer-based test question bank through a clustering algorithm to obtain question groups, including: clustering the questions in the computer-based test question bank using a preset clustering algorithm and the similarity matrix to obtain initial question groups; and optimizing the initial question groups using an adaptive threshold adjustment mechanism to obtain the question groups.

[0010] Further, it is characterized in that, according to the enhanced association graph, and using the attention mechanism, the weights of the similarities between nodes in the enhanced association graph are dynamically adjusted, and based on the adjusted weights, comprehensive similarity information is generated, including: using the enhanced association graph, initializing the attention mechanism to generate an attention weight matrix and a bias item; based on the attention weight matrix and the bias item, evaluating the similarity between each node and the adjacent nodes in the enhanced association graph to generate the attention weights between the nodes; based on the attention weights between the nodes, dynamically adjusting the similarity weights between the nodes in the enhanced association graph to generate the adjusted similarity weights; using the adjusted similarity weights, updating the similarity weights in the enhanced association graph to generate a similarity weight matrix; based on the similarity weight matrix, analyzing the similarities between the questions to generate comprehensive similarity information;

[0011] Based on the comprehensive similarity information, the comprehensive similarity information matrix is ​​constructed.

[0012] Furthermore, the calculation formula of the comprehensive similarity information is:

[0013] ;

[0014] in, Indicate the topic and topic The comprehensive similarity between For the The weights of the similarity factors, For the topic and topic In the Similarity scores under similarity factors, For the attention mechanism Factors on the topic and topic The influence coefficient of similarity, , is the activation function, topic and topic In the Similarity scores under similarity factors, is the adaptive threshold adjustment parameter, The number of similarity factors.

[0015] In a second aspect, an embodiment of the present application provides a system for checking duplicate content in a computer-based test question bank, comprising: a first analysis unit, for using questions in the computer-based test question bank to perform semantic understanding and context association analysis on the content of the questions, and obtain a multi-dimensional similarity map between the questions, wherein the multi-dimensional similarity map is used to characterize the similarity of the textual expressions of the questions, as well as the similarity of the problem-solving ideas, knowledge fields, and logical structures of the questions; a second analysis unit, for processing the association relationship between the questions based on the multi-dimensional similarity map using advanced graph analysis technology to obtain a similarity matrix, wherein the similarity matrix is ​​used to characterize the similarity between the questions; a clustering unit, for using the similarity matrix to cluster the questions in the computer-based test question bank through a clustering algorithm to obtain question groups, wherein the question groups are a set of questions whose similarities in multiple dimensions are greater than a preset threshold; a checking duplicate unit, for using the question groups to check duplicate content in the questions in the question bank, and generate a question bank checking duplicate content report, wherein the checking duplicate content report includes: a list of questions in the question group and similarity scores of the questions in the question group.

[0016] In a third aspect, an embodiment of the present application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the method described in the first aspect above.

[0017] In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program, which, when executed by a computer, implements the method described in the first aspect above.

[0018] In an embodiment of the present invention, questions in a computer-based test question bank are used to perform semantic understanding and contextual association analysis on the question content to obtain a multi-dimensional similarity map between the questions; based on the multi-dimensional similarity map, advanced graph analysis technology is used to process the association relationship between the questions to obtain a similarity matrix; using the similarity matrix, the questions in the computer-based test question bank are clustered through a clustering algorithm to obtain question groups; using the question groups, the questions in the question bank are checked for duplicates, and a question bank duplicate checking report is generated to improve the accuracy and efficiency of the computer-based test question bank duplicate checking.

[0019] These and other aspects of the present application will become more clearly understood in the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 A flowchart of a method for checking duplicate examination question bank on a computer provided in an embodiment of the present application;

[0022] Figure 2 A schematic diagram of a computer-based examination question bank duplication checking system provided in an embodiment of the present application;

[0023] Figure 3 A schematic diagram of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] Embodiment 1:

[0026] According to an embodiment of the present invention, an embodiment of a method for checking duplicate examination question banks on a computer is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0027] Figure 1 is a flow chart of a method for checking duplicate questions in a computer-based examination question bank according to an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0028] Step S102, using the questions in the computer-based test question bank, perform semantic understanding and context association analysis on the question content to obtain a multi-dimensional similarity map between the questions, wherein the multi-dimensional similarity map is used to represent the similarity of the textual expressions of the questions, as well as the similarity of the problem-solving ideas, knowledge fields, and logical structures of the questions;

[0029] Step S104, based on the multi-dimensional similarity graph, using advanced graph analysis technology, processing the correlation between the questions to obtain a similarity matrix, wherein the similarity matrix is ​​used to represent the similarity between the questions;

[0030] Step S106, using the similarity matrix, clustering the questions in the computer-based test question bank through a clustering algorithm to obtain question groups, wherein the question group is a set of questions whose similarities in multiple dimensions are greater than a preset threshold;

[0031] Step S108, using the question group, checking for duplicates in the question bank, and generating a question bank duplicate checking report, wherein the duplicate checking report includes: a list of questions in the question group and similarity scores of the questions in the question group.

[0032] With the popularity of online education and computer-assisted instruction, the management and maintenance of computer-based examination question banks have become increasingly important. Computer-based examination question banks usually contain a large number of programming questions, multiple-choice questions, fill-in-the-blank questions, etc. These questions need to be updated and maintained regularly to ensure the fairness and effectiveness of the examination. In question bank management, duplicate checking is a key task, the purpose of which is to discover and eliminate duplicate or highly similar questions, avoid the repetition of examination content, and improve the diversity and challenge of the question bank. Therefore, an efficient and accurate duplicate checking method is needed that can comprehensively evaluate the similarity between questions to ensure the quality of the question bank and the fairness of the examination.

[0033] At present, the duplication check of computer-based examination question banks mainly relies on the following methods:

[0034] Text-based duplicate checking method: Use text matching technology, such as TF-IDF, cosine similarity, etc. to compare the text content of the questions and determine the similarity between the questions. This method is simple and easy to use, but it has limited effect when dealing with complex questions and it is difficult to capture the deep semantics and logical structure of the questions.

[0035] Metadata-based duplicate checking method: Duplicate checking is performed by analyzing the metadata information of the question, such as question type, difficulty level, knowledge points, etc. This method can partially solve the shortcomings of the text matching method, but there is still the problem of insufficient understanding of the question content.

[0036] Machine learning-based duplicate checking method: Use deep learning models, such as BERT and RoBERTa, to perform semantic understanding and contextual analysis on the questions, generate semantic vector representations of the questions, and then check for duplicates using similarity calculation methods. This method improves the accuracy and reliability of duplicate checking to a certain extent, but it still has some limitations.

[0037] Although the existing duplicate checking methods can meet the needs of question bank management to a certain extent, they still have the following defects:

[0038] Single-dimensional similarity assessment: Existing duplicate checking methods often only focus on the similarity of the textual expressions of the questions, ignoring the multi-dimensional similarities of the problem-solving ideas, knowledge fields, and logical structures. This results in incomplete duplicate checking results and easily misses some deeply similar questions.

[0039] Lack of advanced graph analysis technology: Most existing duplicate checking methods are based on simple text matching or metadata analysis, and fail to make full use of graph analysis technology to deeply explore the complex relationships between questions. This limits the accuracy and robustness of duplicate checking and makes it difficult to cope with complex question bank management needs.

[0040] Insufficient efficiency and accuracy of duplicate checking: Existing duplicate checking methods have low efficiency when dealing with large-scale question banks and are easily affected by noise and abnormal data, resulting in low accuracy of duplicate checking results. This affects the efficiency and quality of question bank management and increases the workload of manual review.

[0041] Because the prior art has the problems of single-dimensional similarity assessment, lack of advanced graph analysis technology, and insufficient efficiency and accuracy in checking for duplicate content, the embodiment of the present invention proposes a method for checking for duplicate content in a computer-based test question bank based on a multi-dimensional similarity graph and advanced graph analysis technology to solve the above technical problems.

[0042] Multi-dimensional similarity map: a map used to represent the similarity of the textual expressions of questions, as well as the similarity of the problem-solving ideas, knowledge fields, and logical structures of the questions.

[0043] Advanced graph analysis technology: including graph convolutional networks (GCN), graph attention networks (GAT) and other technologies, which are used to perform in-depth analysis of nodes and edges in graphs.

[0044] Similarity matrix: A matrix used to represent the similarity between questions.

[0045] Clustering algorithm: an algorithm used to group questions according to similarity, such as K-means, DBSCAN, etc.

[0046] Question group: a collection of questions whose similarity in multiple dimensions is greater than a preset threshold.

[0047] Specifically, for example, there is a computer test question bank containing 100 programming questions. It needs to be checked for duplicates to ensure the diversity and challenge of the questions.

[0048] Perform word segmentation, part-of-speech tagging and named entity recognition on each question to extract basic semantic information.

[0049] Combine the code and diagrams in the question to generate a multimodal semantic vector representation of the question.

[0050] Based on the semantic vector representation and metadata information (such as question type, difficulty level, knowledge points, etc.), an initial multi-dimensional similarity map is generated.

[0051] Graph Convolutional Network: GCN is used to process the initial multi-dimensional similarity graph, aggregate the information of neighboring nodes through multiple convolution operations, and generate an enhanced association graph.

[0052] Based on the enhanced association graph, the attention mechanism is used to dynamically adjust the weights of similarities between nodes to generate comprehensive similarity information.

[0053] Based on the comprehensive similarity information, the final similarity score between each pair of questions is calculated to generate a similarity matrix.

[0054] Select the K-means clustering algorithm, cluster the questions according to the similarity matrix, and generate question groups.

[0055] Assume that the clustering results show that there are 5 question groups, and the similarity of the questions in each group in multiple dimensions is greater than a preset threshold (such as 0.8).

[0056] Identify and mark duplicate or highly similar questions based on question groups. For example, there are 3 highly similar questions in Group 1 and 2 highly similar questions in Group 2.

[0057] Generate a duplicate check report, listing the list of questions in each question group and the similarity score of the questions. For example:

[0058] Group 1: Question 1, Question 2, and Question 3, with similarity scores of 0.95, 0.92, and 0.88, respectively.

[0059] Group 2: Question 4 and Question 5, with similarity scores of 0.90 and 0.87 respectively.

[0060] Through a multi-dimensional similarity map, this method not only considers the similarity of the textual expression of the question, but also considers the similarity of problem-solving ideas, knowledge fields and logical structures, thereby improving the accuracy and reliability of duplicate checking.

[0061] Advanced graph analysis techniques, such as graph convolutional networks and attention mechanisms, are used to process the correlation between questions and generate a similarity matrix. These techniques can effectively capture the complex relationships between questions and improve the accuracy and robustness of similarity calculations.

[0062] The questions are clustered by clustering algorithms to generate question groups. This method can quickly classify questions with high similarity into the same group, improving the efficiency of duplicate checking. At the same time, the adaptive threshold adjustment mechanism ensures the rationality of the clustering results and avoids misjudgment caused by improper threshold setting.

[0063] The generated duplicate check report includes a list of questions in a question group and the similarity score of the questions, providing the question bank manager with a clear and intuitive duplicate check result. This helps to timely discover and deal with duplicate questions, improve the quality and diversity of the question bank, and ensure the fairness and impartiality of the exam.

[0064] In summary, this method realizes comprehensive duplication checking of computer-based examination question banks through multi-dimensional similarity evaluation, advanced graph analysis technology and efficient clustering algorithms, generates detailed duplication checking reports, significantly improves the efficiency and quality of question bank management, and ensures the fairness and impartiality of the examination.

[0065] In the embodiment of the present invention, the questions in the computer-based test question bank are used to perform semantic understanding and context association analysis on the question content to obtain a multi-dimensional similarity map between the questions, including:

[0066] Performing natural language processing on the questions in the computer-based test question bank, and combining the non-text elements in the questions to generate basic semantic information of the questions, wherein the basic semantic information includes: word segmentation results, part-of-speech tags, entity names, grammatical relations between word segmentations and the non-text elements;

[0067] Based on the basic semantic information of the question, a deep learning model is used to perform semantic understanding on the question and generate a semantic vector representation of the question;

[0068] Based on the semantic vector representation of the question and the metadata information of the question, a multi-dimensional similarity analysis is performed on the relationship between the questions using graph embedding technology to generate an initial multi-dimensional similarity graph between the questions, wherein the metadata includes: the knowledge field corresponding to the question and the difficulty level of the question;

[0069] The initial multi-dimensional similarity map is analyzed using a graph neural network algorithm to obtain the multi-dimensional similarity map.

[0070] In the embodiment of the present invention, each question is segmented, POS tagged and named entity recognized to extract basic semantic information. For example, the segmentation result of question 1 is ["write", "a", "program", "calculate", "Fibonacci sequence"], the POS tagged is ["verb", "quantifier", "noun", "verb", "noun"], and the entity name is ["Fibonacci sequence"].

[0071] Combined with the codes in the title (i.e., non-text elements), multimodal basic semantic information is generated.

[0072] Use the pre-trained BERT model to understand the basic semantic information of question 1 and generate a semantic vector representation.

[0073] For example, the knowledge area of ​​question 1 is “recursion” and the difficulty level is “medium”.

[0074] Using the Node2Vec algorithm, the similarity relationship between question 1 and other questions is converted into a low-dimensional vector representation to generate an initial multi-dimensional similarity graph.

[0075] The GCN model is used to analyze the initial multi-dimensional similarity graph, and the information of neighboring nodes is aggregated through multiple convolution operations to generate an enhanced association graph.

[0076] Based on the enhanced association graph, the attention mechanism is used to dynamically adjust the weights of similarities between nodes to generate comprehensive similarity information.

[0077] Finally, based on the comprehensive similarity information, a multi-dimensional similarity map is generated for subsequent duplication checking.

[0078] Through a multi-dimensional similarity map, this method not only considers the similarity of the textual expression of the question, but also considers the similarity of problem-solving ideas, knowledge fields and logical structures, thereby improving the accuracy and reliability of duplicate checking.

[0079] Advanced graph analysis techniques, such as graph convolutional networks and attention mechanisms, are used to process the correlation between questions and generate a similarity matrix. These techniques can effectively capture the complex relationships between questions and improve the accuracy and robustness of similarity calculations.

[0080] The questions are clustered by clustering algorithms to generate question groups. This method can quickly classify questions with high similarity into the same group, improving the efficiency of duplicate checking. At the same time, the adaptive threshold adjustment mechanism ensures the rationality of the clustering results and avoids misjudgment caused by improper threshold setting.

[0081] The generated duplicate check report includes a list of questions in a question group and the similarity score of the questions, providing the question bank manager with a clear and intuitive duplicate check result. This helps to timely discover and deal with duplicate questions, improve the quality and diversity of the question bank, and ensure the fairness and impartiality of the exam.

[0082] In summary, this method realizes comprehensive duplication checking of computer-based examination question banks through multi-dimensional similarity evaluation, advanced graph analysis technology and efficient clustering algorithms, generates detailed duplication checking reports, significantly improves the efficiency and quality of question bank management, and ensures the fairness and impartiality of the examination.

[0083] In the embodiment of the present invention, the initial multi-dimensional similarity map is analyzed by using a graph neural network algorithm to obtain the multi-dimensional similarity map, including:

[0084] Using the initial multi-dimensional similarity graph, training a preset graph neural network model to obtain a trained graph neural network model;

[0085] Using the trained graph neural network model, the feature vectors of the nodes in the initial multi-dimensional similarity graph are updated, and the similarity between the nodes is calculated by using the updated node feature vectors to obtain a similarity score;

[0086] Using the similarity score, generate the multi-dimensional similarity map

[0087] Graph Neural Network (GNN) is a deep learning model that processes graph data. It updates the feature vector of a node by iteratively aggregating the information of neighboring nodes, thereby capturing the complex relationship between nodes. The initial multi-dimensional similarity graph refers to a graph data structure in which nodes represent questions, edges represent the similarity relationship between questions, and the weight of the edge reflects the strength of this similarity.

[0088] First, an initial multi-dimensional similarity graph is constructed, where each node represents a question in the question bank, and the edge represents the similarity relationship between questions. The weight of the edge is calculated by the multi-dimensional features of the question (such as text content, problem-solving ideas, knowledge points, etc.). Then, this graph data structure is used to train a preset graph neural network model. The goal of the training is to let the model learn the true similarity relationship between questions, so that it can more accurately identify similar questions in subsequent steps.

[0089] The trained graph neural network model can be used to update the feature vectors of each node in the graph. These updated feature vectors better reflect the characteristics of the questions in multiple dimensions. Based on the updated feature vectors, the similarity scores between the questions can be recalculated, and these scores will be used to generate the final multi-dimensional similarity graph.

[0090] After the model training is completed, the trained model will be used to update the feature vectors of all nodes in the initial graph. This process is achieved through the forward propagation of the graph neural network. The new feature vector of each node is the comprehensive result of its own features and the features of its neighboring nodes. Subsequently, the similarity scores between questions are calculated based on the updated feature vectors. This step can be achieved through a variety of similarity measurement methods, such as cosine similarity, Euclidean distance, etc.

[0091] Finally, using the calculated similarity scores, a new multi-dimensional similarity graph can be generated. The new graph not only contains the direct similarity relationship between questions, but also reflects the position and role of the questions in the network structure, which is very helpful for discovering implicit patterns and potential problems in the question bank.

[0092] The generated multi-dimensional similarity graph will be provided to the question bank manager as part of the duplicate checking results. The manager can use this graph to quickly identify possible duplicate or similar questions in the question bank and take measures to optimize the question bank content. Compared with traditional methods, this method can not only identify similar questions more accurately, but also provide more insights into the relationship between questions, which helps to improve the overall quality of the question bank and the fairness of the examination.

[0093] Through the above three steps, the problems existing in the traditional duplicate checking method are effectively solved. This method not only improves the accuracy and efficiency of duplicate checking, but also provides valuable insights for question bank managers, which helps to further optimize the question bank management and maintenance process.

[0094] In the embodiment of the present invention, based on the multi-dimensional similarity graph, an advanced graph analysis technology is used to process the correlation between the questions to obtain a similarity matrix, including:

[0095] Based on the multi-dimensional similarity graph, a correlation graph between the topics is constructed, wherein the nodes in the correlation graph are the topics, the similarities between the topics are used as the edges of the correlation graph, and the weights of the edges of the correlation graph are determined by the similarities represented by the multimodal semantic vectors of the topics;

[0096] Based on the association graph, applying a graph convolutional network algorithm, iteratively processing the nodes and edges in the association graph, aggregating information of neighboring nodes through multiple convolution operations, and generating the enhanced association graph;

[0097] According to the enhanced association graph, and using an attention mechanism, dynamically adjusting the weights of similarities between nodes in the enhanced association graph, and generating comprehensive similarity information based on the adjusted weights;

[0098] The comprehensive similarity information is used to calculate the final similarity scores between the questions and generate the similarity matrix.

[0099] In an embodiment of the present invention, a multi-dimensional similarity graph is a graph data structure in which nodes represent topics, edges represent similarity relationships between topics, and edge weights reflect the strength of the similarity.

[0100] Association graph: A graph built based on a multi-dimensional similarity graph, where nodes are questions, edges are similarity relationships between questions, and edge weights are determined by the similarity represented by the multimodal semantic vectors of the questions.

[0101] First, we construct a correlation graph based on the multi-dimensional similarity graph of the questions. In this graph, each node represents a question, and the edge represents the similarity relationship between questions. The weight of the edge is determined by the similarity represented by the multi-modal semantic vector of the question.

[0102] Design reason: The purpose of building a correlation graph is to represent the similarity relationship between questions in the form of a graph, which is convenient for subsequent graph analysis and similarity calculation. This step ensures that all relevant information is integrated together, providing a basis for subsequent processing.

[0103] Graph Convolutional Network (GCN): A deep learning model specifically designed for processing graph data. It transmits node feature information through a multi-layer neural network to achieve information exchange and aggregation between nodes.

[0104] Enhanced association graph: After the association graph is processed by the graph convolutional network algorithm, the node feature vectors are updated and the edge weights are adjusted, which can more accurately reflect the similarity relationship between questions.

[0105] The graph convolutional network algorithm is used to iteratively process the nodes and edges in the association graph, aggregate the information of neighboring nodes through multiple convolution operations, and generate an enhanced association graph.

[0106] The graph convolutional network algorithm can transmit and aggregate node feature information through multi-layer neural networks, so that the feature vector of each node contains not only its own information, but also the information of its neighboring nodes. This step can enhance the feature representation of nodes in the graph and improve the accuracy of similarity calculation.

[0107] Attention mechanism: A technique commonly used in deep learning that dynamically adjusts the weight of similarity between nodes by calculating the attention weight between each node and its neighboring nodes.

[0108] Comprehensive similarity information: Based on the adjusted weights, the generated comprehensive similarity information can more accurately reflect the similarities between questions.

[0109] Based on the enhanced association graph, the attention mechanism is used to dynamically adjust the weight of the similarity between nodes. Specifically, by calculating the attention weight between each node and its neighboring nodes, the weight of the edge is adjusted to generate comprehensive similarity information.

[0110] The attention mechanism can dynamically adjust the weight of the similarity between nodes, so that when calculating the similarity, it can pay more attention to the important similarity relationships and ignore the unimportant relationships. This step can further improve the accuracy and robustness of the similarity calculation.

[0111] Final similarity score: The final similarity score between questions calculated based on the comprehensive similarity information.

[0112] Similarity matrix: A matrix where each element represents the similarity score between two questions.

[0113] Using the comprehensive similarity information, the final similarity score between the questions is calculated. This step can be achieved through a variety of similarity measurement methods, such as cosine similarity, Euclidean distance, etc.

[0114] The calculated final similarity scores are organized into a matrix to generate a similarity matrix. Each element in the similarity matrix represents the similarity score between two questions.

[0115] The purpose of generating a similarity matrix is ​​to provide a structured similarity information to facilitate subsequent clustering analysis and duplicate checking. The similarity matrix can clearly show the similarity relationship between questions and provide strong support for question bank management.

[0116] In an embodiment of the present invention, the similarity matrix is ​​used to cluster the questions in the computer-based test question bank through a clustering algorithm to obtain question groups, including:

[0117] Clustering the questions in the computer-based test question bank using a preset clustering algorithm and the similarity matrix to obtain an initial question group;

[0118] The initial question group is optimized by using an adaptive threshold adjustment mechanism to obtain the question group.

[0119] In the embodiment of the present invention, a K-means clustering algorithm is selected, and the questions in the question bank are clustered using a similarity matrix.

[0120] The initial question groups are generated by the K-means clustering algorithm. For example, suppose the clustering results show that there are 5 initial question groups, and the questions in each group have high similarity in multiple dimensions.

[0121] The initial question groups are optimized and the similarity threshold is dynamically adjusted through the adaptive threshold adjustment mechanism. Specifically, according to the internal similarity distribution of each initial question group, the similarity threshold is dynamically adjusted to ensure that the similarity of the questions in each question group in multiple dimensions is higher than the preset threshold.

[0122] Optimized question groups: After optimization by the adaptive threshold adjustment mechanism, the final question groups are generated. For example, the optimized question groups may be as follows:

[0123] Group 1: Question 1, Question 2, and Question 3, with similarity scores of 0.95, 0.92, and 0.88, respectively.

[0124] Group 2: Question 4 and Question 5, with similarity scores of 0.90 and 0.87 respectively.

[0125] Group 3: Question 6, Question 7, and Question 8, with similarity scores of 0.93, 0.91, and 0.89, respectively.

[0126] Group 4: Question 9 and Question 10, with similarity scores of 0.92 and 0.88 respectively.

[0127] Group 5: Question 11 and Question 12, with similarity scores of 0.91 and 0.89 respectively.

[0128] By using a preset clustering algorithm (such as K-means, DBSCAN, etc.), the questions in the question bank are quickly clustered using a similarity matrix to generate initial question groups. This step can quickly identify similar questions that may exist in the question bank and improve the efficiency of duplicate checking.

[0129] Through the adaptive threshold adjustment mechanism, the similarity threshold is dynamically adjusted to ensure that the similarity of questions in each question group in multiple dimensions is higher than the preset threshold. This step can avoid misjudgment or omission caused by fixed thresholds and improve the accuracy and rationality of duplicate checking.

[0130] The generated duplicate check report includes a list of questions in a question group and the similarity score of the questions, providing the question bank manager with a clear and intuitive duplicate check result. This helps to timely discover and deal with duplicate questions, improve the quality and diversity of the question bank, and ensure the fairness and impartiality of the exam.

[0131] Through efficient clustering processing and adaptive threshold adjustment mechanism, this method can automatically check the questions in the question bank for duplicates, reducing the workload of manual review and improving the efficiency of question bank management.

[0132] In an embodiment of the present invention, according to the enhanced association graph, and using the attention mechanism, the weights of the similarities between nodes in the enhanced association graph are dynamically adjusted, and based on the adjusted weights, comprehensive similarity information is generated, including:

[0133] Initialize the attention mechanism using the enhanced association graph to generate an attention weight matrix and a bias term;

[0134] Based on the attention weight matrix and the bias term, the similarity between each node and the adjacent nodes in the enhanced association graph is evaluated to generate attention weights between the nodes;

[0135] Based on the attention weights between the nodes, dynamically adjusting the similarity weights between the nodes in the enhanced association graph to generate adjusted similarity weights;

[0136] Using the adjusted similarity weights, updating the similarity weights in the enhanced association graph to generate a similarity weight matrix;

[0137] According to the similarity weight matrix, the similarities between the questions are analyzed and processed to generate comprehensive similarity information;

[0138] Based on the comprehensive similarity information, the comprehensive similarity information matrix is ​​constructed.

[0139] In an embodiment of the present invention, the enhanced association graph: the association graph processed by the graph convolutional network algorithm, the node feature vectors are updated, and the edge weights are adjusted, which can more accurately reflect the similarity relationship between questions.

[0140] Attention mechanism: A technique commonly used in deep learning that dynamically adjusts the weight of similarity between nodes by calculating the attention weight between each node and its neighboring nodes.

[0141] Attention weight matrix: A matrix where each element represents the attention weight between a node and its neighbor nodes.

[0142] Bias term: A constant term used to adjust the calculation of attention weights.

[0143] First, the attention mechanism is initialized to generate the attention weight matrix and bias term. This step includes initializing the values ​​of the attention weight matrix and bias term.

[0144] The initialization of the attention mechanism is to ensure that in the subsequent similarity adjustment process, the calculation of the attention weight has a reasonable starting point, so that the similarity weights between nodes can be adjusted more accurately.

[0145] Similarity evaluation: Generate attention weights between nodes by calculating the similarity between each node and its neighboring nodes.

[0146] Attention weight between nodes: A numerical value that indicates the importance of the similarity between a node and its neighboring nodes.

[0147] Based on the attention weight matrix and the bias term, the similarity between each node and its neighboring nodes in the enhanced association graph is evaluated to generate the attention weights between nodes. Specifically, the attention weights are generated by calculating the similarity between each node and its neighboring nodes.

[0148] By evaluating the similarity between nodes and generating attention weights, the weights of the similarities between nodes can be dynamically adjusted, so that when calculating the similarity, more attention can be paid to important similarity relationships and unimportant relationships can be ignored.

[0149] Dynamic adjustment: Dynamically adjust the similarity weights between nodes based on the attention weights between nodes.

[0150] Adjusted similarity weight: The dynamically adjusted similarity weight can more accurately reflect the similarity relationship between nodes.

[0151] Based on the attention weights between nodes, the similarity weights between nodes in the enhanced association graph are dynamically adjusted to generate adjusted similarity weights. Specifically, the similarity weights between nodes are adjusted by multiplying the attention weights.

[0152] Dynamically adjusting the similarity weights can make similarity calculations more accurate, avoid misjudgments or omissions caused by fixed weights, and improve the accuracy of duplicate checking.

[0153] Similarity weight matrix: A matrix where each element represents the similarity weight between two nodes.

[0154] The similarity weights in the enhanced association graph are updated using the adjusted similarity weights to generate a similarity weight matrix. Specifically, the adjusted similarity weights are organized into a matrix.

[0155] The purpose of generating a similarity weight matrix is ​​to provide a structured similarity information to facilitate subsequent similarity analysis and duplicate checking.

[0156] According to the similarity weight matrix, the similarities between the questions are analyzed and processed to generate comprehensive similarity information. Specifically, the comprehensive similarity information is generated by calculating the similarity score between each question and other questions.

[0157] Generating comprehensive similarity information can provide a comprehensive similarity assessment result and provide support for subsequent duplicate checking.

[0158] Among them, the calculation formula for comprehensive similarity information is:

[0159] ;

[0160] in, Indicate the topic and topic The comprehensive similarity between For the The weights of the similarity factors, For the topic and topic In the Similarity scores under similarity factors, For the attention mechanism Factors on the topic and topic The influence coefficient of similarity, , is the activation function, topic and topic In the Similarity scores under similarity factors, is the adaptive threshold adjustment parameter, The number of similarity factors.

[0161] Based on the comprehensive similarity information, a comprehensive similarity information matrix is ​​constructed. Specifically, the comprehensive similarity information is organized into a matrix, in which each element represents the comprehensive similarity information between two topics.

[0162] The purpose of constructing a comprehensive similarity information matrix is ​​to provide a structured similarity assessment result to facilitate subsequent duplicate checking and question bank management.

[0163] Embodiment 2:

[0164] An embodiment of the present invention also provides a computer-based examination question bank duplication checking system, which is used to execute the computer-based examination question bank duplication checking method provided by the above content of the embodiment of the present invention. The following is a specific introduction to the computer-based examination question bank duplication checking system provided by the embodiment of the present invention.

[0165] like Figure 2 As shown, Figure 2 This is a schematic diagram of the above-mentioned computer-based examination question bank duplicate checking system, which includes:

[0166] The first analysis unit 10 is used to use the questions in the computer-based test question bank to perform semantic understanding and context association analysis on the question content to obtain a multi-dimensional similarity map between the questions, wherein the multi-dimensional similarity map is used to represent the similarity of the textual expressions of the questions, as well as the similarity of the problem-solving ideas, knowledge fields and logical structures of the questions;

[0167] A second analysis unit 20 is used to process the correlation between the questions based on the multi-dimensional similarity graph and adopt advanced graph analysis technology to obtain a similarity matrix, wherein the similarity matrix is ​​used to represent the similarity between the questions;

[0168] The clustering unit 30 is used to cluster the questions in the computer-based test question bank using the similarity matrix through a clustering algorithm to obtain question groups, wherein the question groups are a set of questions whose similarities in multiple dimensions are greater than a preset threshold;

[0169] The duplicate checking unit 40 is used to use the question group to check the questions in the question bank for duplicates and generate a question bank duplicate checking report, wherein the duplicate checking report includes: a list of questions in the question group and similarity scores of the questions in the question group.

[0170] Embodiment three:

[0171] The embodiment of the present invention further provides a computing device for executing a program of the method described in the first embodiment, such as Figure 3 As shown, the computing device may include a storage component 41 and a processing component 42;

[0172] The storage component 41 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 42 .

[0173] The processing component 42 may include one or more processors to execute computer instructions to complete all or part of the steps in the method of embodiment 1. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0174] The storage component 41 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0175] The display component 43 may be an electroluminescent (EL) element, a liquid crystal display or a micro display having a similar structure, or a retinal direct display or a similar laser scanning display.

[0176] Of course, a computing device may also include other components, such as input / output interfaces, communication components, etc.

[0177] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0178] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0179] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0180] Embodiment 4:

[0181] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The method of the illustrated embodiment.

[0182] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0183] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0184] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for 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.

[0185] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for checking duplicate questions in a computer-based examination question bank, characterized in that: include: Using the questions in the computer-based test question bank, semantic understanding and context association analysis are performed on the question content to obtain a multi-dimensional similarity map between the questions, wherein the multi-dimensional similarity map is used to represent the similarity of the textual expressions of the questions, as well as the similarity of the problem-solving ideas, knowledge fields, and logical structures of the questions; Based on the multi-dimensional similarity graph, the association relationship between the questions is processed by using advanced graph analysis technology to obtain a similarity matrix, wherein the similarity matrix is ​​used to represent the similarity between the questions; Using the similarity matrix, clustering the questions in the computer-based test question bank through a clustering algorithm to obtain question groups, wherein the question group is a set of questions whose similarities in multiple dimensions are greater than a preset threshold; Using the question group, checking for duplicates in the question bank and generating a question bank duplicate checking report, the duplicate checking report including: a list of questions in the question group and similarity scores of the questions in the question group; Among them, using the questions in the computer-based test question bank, the semantic understanding and context association analysis of the question content are carried out to obtain a multi-dimensional similarity map between the questions, including: Performing natural language processing on the questions in the computer-based test question bank, and combining the non-text elements in the questions to generate basic semantic information of the questions, wherein the basic semantic information includes: word segmentation results, part-of-speech tags, entity names, grammatical relations between word segmentations and the non-text elements; Based on the basic semantic information of the question, a deep learning model is used to perform semantic understanding on the question and generate a semantic vector representation of the question; Based on the semantic vector representation of the question and the metadata information of the question, a multi-dimensional similarity analysis is performed on the relationship between the questions using graph embedding technology to generate an initial multi-dimensional similarity graph between the questions, wherein the metadata includes: the knowledge field corresponding to the question and the difficulty level of the question; Analyzing the initial multi-dimensional similarity map using a graph neural network algorithm to obtain the multi-dimensional similarity map; Among them, based on the multi-dimensional similarity graph, advanced graph analysis technology is used to process the correlation between the questions to obtain a similarity matrix, including: Based on the multi-dimensional similarity graph, a correlation graph between the topics is constructed, wherein the nodes in the correlation graph are the topics, the similarities between the topics are used as the edges of the correlation graph, and the weights of the edges of the correlation graph are determined by the similarities represented by the multimodal semantic vectors of the topics; Based on the association graph, applying a graph convolutional network algorithm, iteratively processing the nodes and edges in the association graph, aggregating information of neighboring nodes through multiple convolution operations, and generating an enhanced association graph; According to the enhanced association graph, and using an attention mechanism, dynamically adjusting the weights of similarities between nodes in the enhanced association graph, and generating comprehensive similarity information based on the adjusted weights; The comprehensive similarity information is used to calculate the final similarity scores between the questions and generate the similarity matrix.

2. The method according to claim 1, characterized in that The initial multi-dimensional similarity map is analyzed by using a graph neural network algorithm to obtain the multi-dimensional similarity map, including: Using the initial multi-dimensional similarity graph, training a preset graph neural network model to obtain a trained graph neural network model; Using the trained graph neural network model, the feature vectors of the nodes in the initial multi-dimensional similarity graph are updated, and the similarity between the nodes is calculated by using the updated node feature vectors to obtain a similarity score; The multi-dimensional similarity map is generated using the similarity scores.

3. The method according to claim 1, characterized in that The similarity matrix is ​​used to cluster the questions in the computer-based test question bank through a clustering algorithm to obtain question groups, including: Clustering the questions in the computer-based test question bank using a preset clustering algorithm and the similarity matrix to obtain an initial question group; The initial question group is optimized by using an adaptive threshold adjustment mechanism to obtain the question group.

4. The method according to claim 2, characterized in that: According to the enhanced association graph, and using the attention mechanism, dynamically adjusting the weights of similarities between nodes in the enhanced association graph, and generating comprehensive similarity information based on the adjusted weights, including: Initialize the attention mechanism using the enhanced association graph to generate an attention weight matrix and a bias term; Based on the attention weight matrix and the bias term, the similarity between each node and the adjacent nodes in the enhanced association graph is evaluated to generate attention weights between the nodes; Based on the attention weights between the nodes, dynamically adjusting the similarity weights between the nodes in the enhanced association graph to generate adjusted similarity weights; Using the adjusted similarity weights, updating the similarity weights in the enhanced association graph to generate a similarity weight matrix; According to the similarity weight matrix, the similarities between the questions are analyzed and processed to generate comprehensive similarity information; Based on the comprehensive similarity information, a comprehensive similarity information matrix is ​​constructed.

5. The method according to claim 4, characterized in that The calculation formula of the comprehensive similarity information is: ； in, Indicate the topic and topic The comprehensive similarity between For the The weights of the similarity factors, For the topic and topic In the Similarity scores under similarity factors, For the attention mechanism Factors on the topic and topic The influence coefficient of similarity, , is the activation function, topic and topic In the Similarity scores under similarity factors, is the adaptive threshold adjustment parameter, The number of similarity factors.

6. A system for checking duplicate questions in a computer-based examination question bank, characterized in that: include: The first analysis unit is used to use the questions in the computer-based test question bank to perform semantic understanding and context association analysis on the question content to obtain a multi-dimensional similarity map between the questions, wherein the multi-dimensional similarity map is used to represent the similarity of the textual expressions of the questions, as well as the similarity of the problem-solving ideas, knowledge fields and logical structures of the questions; A second analysis unit is used to process the correlation between the questions based on the multi-dimensional similarity graph and adopt advanced graph analysis technology to obtain a similarity matrix, wherein the similarity matrix is ​​used to represent the similarity between the questions; A clustering unit, configured to cluster the questions in the computer-based test question bank using the similarity matrix through a clustering algorithm to obtain question groups, wherein the question groups are sets of questions whose similarities in multiple dimensions are greater than a preset threshold; A duplicate checking unit, used to use the question group to check the questions in the question bank for duplicate processing, and generate a question bank duplicate checking report, wherein the duplicate checking report includes: a list of questions in the question group and similarity scores of the questions in the question group; Among them, using the questions in the computer-based test question bank, the semantic understanding and context association analysis of the question content are carried out to obtain a multi-dimensional similarity map between the questions, including: Performing natural language processing on the questions in the computer-based test question bank, and combining the non-text elements in the questions to generate basic semantic information of the questions, wherein the basic semantic information includes: word segmentation results, part-of-speech tags, entity names, grammatical relations between word segmentations and the non-text elements; Based on the basic semantic information of the question, a deep learning model is used to perform semantic understanding on the question and generate a semantic vector representation of the question; Based on the semantic vector representation of the question and the metadata information of the question, a multi-dimensional similarity analysis is performed on the relationship between the questions using graph embedding technology to generate an initial multi-dimensional similarity graph between the questions, wherein the metadata includes: the knowledge field corresponding to the question and the difficulty level of the question; Analyzing the initial multi-dimensional similarity map using a graph neural network algorithm to obtain the multi-dimensional similarity map; Among them, based on the multi-dimensional similarity graph, advanced graph analysis technology is used to process the correlation between the questions to obtain a similarity matrix, including: Based on the multi-dimensional similarity graph, a correlation graph between the topics is constructed, wherein the nodes in the correlation graph are the topics, the similarities between the topics are used as the edges of the correlation graph, and the weights of the edges of the correlation graph are determined by the similarities represented by the multimodal semantic vectors of the topics; Based on the association graph, applying a graph convolutional network algorithm, iteratively processing the nodes and edges in the association graph, aggregating information of neighboring nodes through multiple convolution operations, and generating an enhanced association graph; According to the enhanced association graph, and using an attention mechanism, dynamically adjusting the weights of similarities between nodes in the enhanced association graph, and generating comprehensive similarity information based on the adjusted weights; The comprehensive similarity information is used to calculate the final similarity scores between the questions and generate the similarity matrix.

7. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the method according to any one of claims 1 to 5.

8. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, the method according to any one of claims 1 to 5 is implemented.

Images