Test question repeatability detection method and system based on multi-modal large language model

Abstract

The invention discloses a test question repeatability detection method and system based on a multi-modal large language model, and the method comprises the steps: collecting multi-modal test question data, taking the multi-modal test question data as a question bank data set, and carrying out the data cleaning of the question bank data set, and obtaining a cleaned data set; for the cleaned data set, generating a multi-modal high-dimensional semantic vector representation of the test question in combination with a multi-modal large language model, and pooling the multi-modal high-dimensional semantic vector representation to obtain a final vector representation of the question; representing the final vectors of the questions, and constructing a Milvus set and an HNSW index in combination with test question data exchange format JSON (JavaScript Object Notation) data; and the multi-modal representation of the to-be-searched set is combined with the HNSW index of the Milvus, and a response result is visually displayed. Based on a Milvus vector database and a multi-modal representation technology, the method can support similarity retrieval scoring of millisecond-level more accurate response of a million-level question bank.

Description

Technical Field

[0001] The present invention belongs to the field of multimodal retrieval in educational examination scenarios, and specifically relates to a method and system for detecting test question repetitiveness based on a multimodal large language model. Background Art

[0002] Test question duplication detection technology aims to identify semantically or logically repeated questions through a comprehensive understanding of multimodal information such as test text, formulas, and charts, thereby improving duplication detection efficiency and optimizing the allocation of educational resources. However, existing technologies face severe challenges in modal coverage, depth of multimodal semantic understanding, and efficiency, as shown in the following: 1. Insufficient modal coverage: Modal information such as formulas and diagrams must be manually annotated or processed separately, lacking a unified representation. Furthermore, the test question's text description, schematic diagrams, and knowledge point structure cannot be effectively integrated, resulting in insufficient cross-modal similarity detection capabilities.

[0003] 2. Weak multimodal semantic understanding depth: Traditional models have limited semantic understanding capabilities for multimodal information such as complex mathematical symbols and geometric figures.

[0004] 3. Inefficiency: Traditional relational databases such as MySQL are unable to support the real-time retrieval needs of massive test questions and have slow response speeds. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for detecting test question repetitiveness based on a multimodal large language model to solve the above problems.

[0006] To achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for detecting test question repetitiveness based on a multimodal large language model, comprising: Collect multimodal test data as a question bank data set, and perform data cleaning on the question bank data set to obtain a cleaned data set; For the cleaned dataset, a multimodal large language model is used to generate a multimodal high-dimensional semantic vector representation of the test question. The multimodal high-dimensional semantic vector representation is then pooled to obtain the final vector representation of the question. The final vector representation of the question is combined with the JSON data in the test data exchange format to build the Milvus collection and HNSW index; The multimodal representation of the query set is combined with Milvus's HNSW index to visualize the response results.

[0007] Furthermore, the multimodal test data is collected as a question bank data set, including: Collect the subject textbooks and Internet question banks of the open test questions as the question bank dataset, build a tree index structure according to the subject knowledge point level, and use JSON format to store the original data .

[0008] Furthermore, the data cleaning of the question bank data set to obtain a cleaned data set includes: For the test question dataset, use regular expressions to remove HTML tags, use LaTeX parsing tools to unify the formula format, unify the option number format, and obtain the cleaned question bank dataset .

[0009] Furthermore, the multimodal high-dimensional semantic vector representation of the test questions is generated by combining the cleaned dataset with the multimodal large language model, including: The cleaned test data is split into three independent modal channels: text, image, and knowledge points. The text modality contains the pure text content of the question stem and options, the image modality is associated with the schematic diagram or formula image in the test question, and the knowledge point modality is mapped to the leaf node level in the tree index structure; a 7B parameter multimodal large language model is used to input the three modal data into the model through a preset instruction template, and the multimodal features are cross-modally aligned and fused based on the multi-head attention mechanism to calculate and generate a high-dimensional semantic vector representation .

[0010] Furthermore, the final vector representation of the question obtained by pooling the multimodal high-dimensional semantic vector representation includes: Representation of the generated high-dimensional semantic vector conduct Operation, compressing the high-dimensional vector into a low-dimensional vector of fixed length to retain the core features of the semantic information and obtain the final vector representation of the question .

[0011] Furthermore, the final vector representation of the question is combined with the test question data exchange format JSON data to construct a Milvus collection and HNSW index, including: Multimodal representation vector set based on pooling , combined with the test question JSON metadata to build a Milvus database collection, where the fields are defined as test question ID, question text, image path, knowledge point, JSON metadata and multimodal vector; the vector field is retrieved and optimized through HNSW index, and the index parameters are set as the distance measurement method COSINE, the index construction parameters of the number of adjacent nodes M=64, and the search width.

[0012] Furthermore, the multimodal representation of the query set is combined with Milvus's HNSW index to visualize the response results, including: The multimodal test questions are quickly deduplicated by calculating the vector cosine similarity. HNSW index of Milvus database is used to set parameters for query; the search results are generated into an interactive interface through the visualization engine, showing the contents of the TOP100 similar test questions.

[0013] In a second aspect, the present invention provides a test question repetitiveness detection system based on a multimodal large language model, comprising: The data acquisition module is used to collect multimodal test data as a question bank data set, and perform data cleaning on the question bank data set to obtain a cleaned data set; The cleaning and pooling module is used to generate a multimodal high-dimensional semantic vector representation of the test question based on the cleaned dataset and combine it with a multimodal large language model. The final vector representation of the question is obtained by pooling the multimodal high-dimensional semantic vector representation. The optimization module is used to represent the final vector of the question and build the Milvus collection and HNSW index based on the JSON data in the question data exchange format. The display output module is used to combine the multimodal representation of the query set with Milvus's HNSW index to visualize the response results.

[0014] In a third aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for detecting test question repetitiveness based on a multimodal large language model are implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the test question repetitiveness detection method based on a multimodal large language model are implemented.

[0016] Compared with the prior art, the present invention has the following technical effects: The purpose of the present invention is to provide a method for detecting test question repeatability based on a multimodal large language model. By integrating multimodal data such as text, images, and knowledge points, and combining cross-modal semantic alignment with efficient indexing technology, the method can achieve rapid repeatability detection of large-scale question banks. First, the test question data is multimodally cleaned, and a high-dimensional multimodal semantic vector representation of the test question is generated using a 7B-level multimodal large language model. This is then compressed into a low-dimensional vector through mean pooling. At the same time, an HNSW index based on the Milvus database is constructed. Combined with similarity calculation, the top 100 similar test questions are retrieved within millisecond response time, addressing the defects of traditional methods that ignore image semantics and modality associations, and significantly improving the efficiency and accuracy of repeated test question detection.

[0017] This method utilizes a large, multimodal model with 7B parameters to automatically generate high-dimensional semantic vectors, eliminating the need for manual rule design or vocabulary maintenance, and adapting to formula variations and evolving question types. Based on the Milvus vector database and multimodal representation technology, this method supports similarity retrieval and scoring for question banks with millions of questions, with millisecond-level precision. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a block diagram of the test question repetitiveness detection method based on the multimodal large language model of the present invention.

[0019] Figure 2 It is a flowchart of multimodal data crawling and cleaning.

[0020] Figure 3 It is a flowchart of multimodal data representation.

[0021] Figure 4 This is the flow chart of pooling representation.

[0022] Figure 5 It is a flowchart of vector database index optimization.

[0023] Figure 6 This is the multimodal similarity retrieval flowchart. DETAILED DESCRIPTION

[0024] The following describes the embodiments of the present invention in detail with reference to the accompanying drawings and examples. It should be noted that the embodiments described herein are only intended to explain the present invention and are not intended to limit the present invention. In addition, the technical features of the embodiments of the present invention may be combined with each other unless there is a conflict.

[0025] The specific implementation process of the present invention includes multimodal test data collection, multimodal data cleaning, multimodal data representation, pooling representation, vector database index optimization, and multimodal similarity retrieval process. Figure 1 This is a block diagram of the test question repetitiveness detection method based on the multimodal large language model of the present invention.

[0026] Multimodal test data collection process In order to build a high-quality multimodal test question bank, it is necessary to collect data from multiple channels, covering various forms such as text and images, and at the same time obtain structured data related to the test questions, such as difficulty, year, subject labels, etc. First, collect textbooks for the subjects to which the open test questions belong, covering various subjects from elementary school to high school, including Chinese, mathematics, English, physics, chemistry, biology, history, geography, politics, etc., and integrate the test content in the textbooks. Next, use web crawler technology to collect information from multiple sources of data. These data sources include subject open source question banks, educational resource libraries, e-book libraries, and academic resource libraries. The data includes: Text: Question, options, analysis Image: function image, geometric graph, etc. Structured data: difficulty level, year, region, knowledge point labels, etc.

[0027] The obtained test question data is divided into subject knowledge point levels to construct a subject knowledge point tree index as follows: Mathematics (Roots) ├─ Algebra and Elementary Functions │└─ Function Basics │├─ Linear function │└─ Quadratic function → Vertex form / graph property / extreme value problem (leaf node) └─ Geometry and analytic geometry ├─ Plane geometry └─ Solid Geometry → Space Vector / Sphere Surface Area (Leaf Node) The JSON format storage content of each question is as follows: { "question_id": "0002091928132ebfc181a3299413df54", "paper_id": "1020#", "question_type": "Answer question", "source": "2022 Autumn·Zhangjiakou, Hebei·Senior High School Entrance Examination Final Semester", "stem": "Given a quadrangular pyramid \( P-ABCD \). Find the cosine of the angle between plane \( BEC \) and plane \( PAD \)", "image": [], "choices": [], "url": "https: / / xxx.com / 11q17801498.html", "province": "Hebei", "md5": "f5ee32366c4be748ed21850d1fdca84a", "question_year": "2024", "knowledge": "Prove that a line is perpendicular to a plane, prove that lines are perpendicular to each other, and find the vector method of the angle between two planes." "difficulty": "moderate", "difficulty_score": "0.65", "sub_questions": [{"stem": "(1) Prove the line-plane perpendicular relationship;", "choices": []}], "sub_questions_count": 1, "knowledge_expanding": "Applications of spatial vectors, Vector methods for finding spatial angles, Vector methods for finding dihedral angles, Spatial vectors and solid geometry", "item_id": 84879 } The original data fields are stored in JSON format, including question ID, question stem, options, knowledge point tags, image path, difficulty, year, etc., to obtain the original question bank dataset .

[0028] Multimodal data cleaning process S In the data preprocessing stage, a series of cleaning operations were performed on the question dataset obtained in step 1 to ensure the quality and consistency of the data. First, regular expressions were used to match and remove HTML tags to extract the plain text content. Next, a LaTeX parsing tool was used to uniformly format the formulas, and a conversion function was defined , where S is the original text, is the standard text:

[0029] Rule set is as follows: <00001,38>

[0030] In terms of processing the option number format, the number formats of all options were unified to ensure their consistency. Different number formats such as \"1.\", \"(1)\", \"One,\" etc. were unified into the \"1.\" standard format. The cleaned question bank dataset was obtained .

[0031] The above flowchart of the multimodal data crawling and cleaning process is as Figure 2 shown Multimodal data representation process The cleaned question data is split into three independent modal channels: text, image, and knowledge point Text modality: Extract the plain text content of the question stem and options Image modality: Extract the images related to the questions, such as geometric diagrams, function images, etc Knowledge point modality: Extract the knowledge point tags related to the questions

[0032] Then, a multimodal standard prompt template is constructed, integrating text, images, and knowledge point information into a single prompt. The image is then fed into the image alignment ViT model of the 7B-level multimodal large language model for processing and image features extraction:

[0033] in, is the input image, It is the ViT visual encoder of MLLM, which outputs visual features Next, the visual features are projected into the language embedding space through a linear layer:

[0034] here, Is a trainable projection matrix that transforms visual features Convert to visual markup aligned with the language model embedding space .

[0035] Embed text and visual markers Aggregated together to form a unified multimodal embedding:

[0036] The aggregated multimodal embedding Input into the base large language model of MLLM, which aligns multimodal information through self-attention and cross-attention mechanisms:

[0037] in, Through the parameter matrix Multimodal information interaction. The specific prompt template is as follows: Observe the geometric figure in the image provided. Carefully analyze and describe all important geometric elements, including: 1. Shapes: Identify all geometric shapes such as circles, semicircles, triangles, quadrilaterals, etc. 2. Special points: Describe the positions of vertices, intersections,midpoints, and any other notable points. 3. Lines and angles: Discuss the relative positions of line segments,parallel or perpendicular lines, and the measures or relationships of angles. 4. Additional properties: Identify any symmetries, congruencies, orother notable geometric relationships. Question Context: <stem> Image: Knowledge: <knowledge> Based on the provided image, context, and knowledge points, provide adetailed analysis of the geometric figure. Your response should be structured and logically explain all observations in relation to the question. The above multimodal data representation process flow chart is as follows Figure 3 shown.

[0038] Pooling representation process High-dimensional semantic vectors generated by the multimodal large language model (MLLM) It contains rich semantic information, but has high dimension and high computational complexity. First, we remove the padding tokens and retain the valid semantic information. The pooling operation generates a low-dimensional vector of fixed length by calculating the element-by-element average of the vector sequence. It is specifically defined as:

[0039] in, is a high-dimensional semantic vector No. elements, is the dimension of the vector, after After the operation, the low-dimensional vector This is the final vector representation of the question.

[0040] The above pooling characterization process flow chart is as follows Figure 4 shown.

[0041] Vector database index optimization process The vector database index optimization process takes low-dimensional pooling representation as the core input and combines the structured metadata of the test questions to build an efficient retrieval system. First, the data storage structure is defined, and the multimodal pooling representation vector of the question bank dataset is , combined with the test question JSON data to build a Milvus collection, the field definition includes the test question ID, question text, image path, knowledge point, JSON metadata and multimodal representation vector. Let the fields of the Milvus collection be , a one-to-one corresponding storage unit is formed for the semantic vector and metadata of each test question, which satisfies the following mapping relationship:

[0042] Construct an efficient graph-based approximate nearest neighbor search algorithm HNSW index, firstly realize fast search through the layer graph structure, and insert the number of layers of each node Obeying the geometric distribution, the probability formula is:

[0043] in is the interlayer attenuation factor (usually taken as ), is the maximum number of neighbors per layer, In the graph, nodes are only connected to nodes at the same layer or lower layers. The density of high-level nodes is low but the connection span is large. The bottom layer contains all the data and has dense connections. The connection rules between layers are constructed. Then the graph connectivity is optimized by dynamically adjusting the neighbor relationship. , select in layer l neighbors when:

[0044] The search process uses a greedy algorithm to achieve multi-layer navigation. , from the highest level First, go down layer by layer to find the nearest neighbor and insert it:

[0045] During construction, the maximum number of neighbors per layer is set to M = 64, the distance metric uses cosine similarity COSINE, and the candidate pool size during construction is controlled to efConstruction = 80.

[0046] The flowchart of the above vector database index optimization process is as follows Figure 5 shown.

[0047] Multimodal similarity retrieval process Obtain a new set of test questions to be searched, including multimodal data such as text, images, and knowledge points. Vectorize the test questions using the multimodal pooling representation described above, converting them to the same format as the vectors stored in the database and integrating them into a unified query vector. Using the HNSW index in Milvus, search parameter efSearch = 200, and the cosine similarity between the query vector and the vectors in the database:

[0048] Through efficient index query, the top 100 most similar results are retrieved.

[0049] The above multimodal similarity retrieval process flow chart is as follows Figure 6 shown.

[0050] In yet another embodiment of the present invention, a system for detecting test question repetitiveness based on a multimodal large language model is provided, which can be used to implement the above-mentioned method for detecting test question repetitiveness based on a multimodal large language model. Specifically, the system includes: The data acquisition module is used to collect multimodal test data as a question bank data set, and perform data cleaning on the question bank data set to obtain a cleaned data set; The cleaning and pooling module is used to generate a multimodal high-dimensional semantic vector representation of the test question based on the cleaned dataset and combine it with a multimodal large language model. The final vector representation of the question is obtained by pooling the multimodal high-dimensional semantic vector representation. The optimization module is used to represent the final vector of the question and build the Milvus collection and HNSW index based on the JSON data in the question data exchange format. The display output module is used to combine the multimodal representation of the query set with Milvus's HNSW index to visualize the response results.

[0051] The module division in the embodiments of the present invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in various embodiments of the present invention may be integrated into a single processor, exist physically as separate modules, or two or more modules may be integrated into a single module. The integrated modules may be implemented in either hardware or software functional modules.

[0052] In another embodiment of the present invention, a computer device is provided, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the test question repetitiveness detection method based on the multimodal large language model.

[0053] In another embodiment of the present invention, the present invention further provides a storage medium, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in a computer device, used to store programs and data. It is understandable that the computer-readable storage medium herein may include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides storage space, which stores the terminal's operating system. In addition, the storage space also stores one or more instructions suitable for being loaded and executed by a processor. These instructions may be one or more computer programs (including program code). It should be noted that the computer-readable storage medium herein may be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage device. The processor may load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the test question repetitiveness detection method based on a multimodal large language model in the above-mentioned embodiment.

[0054] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0055] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0056] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0057] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.< / knowledge> < / stem>

Claims

1. A method for detecting test question repetitiveness based on a multimodal large language model, characterized in that: include: Collect multimodal test data as a question bank data set, and perform data cleaning on the question bank data set to obtain a cleaned data set; For the cleaned dataset, a multimodal large language model is used to generate a multimodal high-dimensional semantic vector representation of the test question. The multimodal high-dimensional semantic vector representation is then pooled to obtain the final vector representation of the question. The final vector representation of the question is combined with the JSON data in the test data exchange format to build the Milvus collection and HNSW index; The multimodal representation of the query set is combined with Milvus's HNSW index to visualize the response results.

2. The method for detecting test question repetitiveness based on a multimodal large language model according to claim 1, characterized in that: The multimodal test data is collected as a question bank data set, including: Collect the subject textbooks and Internet question banks of the open test questions as the question bank dataset, build a tree index structure according to the subject knowledge point level, and use JSON format to store the original data .

3. The method for detecting test question repetitiveness based on a multimodal large language model according to claim 1, characterized in that: The data cleaning of the question bank data set to obtain a cleaned data set includes: For the test question dataset, use regular expressions to remove HTML tags, use LaTeX parsing tools to unify the formula format, unify the option number format, and obtain the cleaned question bank dataset .

4. The method for detecting test question repetitiveness based on a multimodal large language model according to claim 1, characterized in that: The method of generating a multimodal high-dimensional semantic vector representation of the test questions by combining the cleaned dataset with a multimodal large language model includes: The cleaned test data is split into three independent modal channels: text, image, and knowledge points. The text modality contains the pure text content of the question stem and options, the image modality is associated with the schematic diagram or formula image in the test question, and the knowledge point modality is mapped to the leaf node level in the tree index structure; a 7B parameter multimodal large language model is used to input the three modal data into the model through a preset instruction template, and the multimodal features are cross-modally aligned and fused based on the multi-head attention mechanism to calculate and generate a high-dimensional semantic vector representation .

5. The method for detecting test question repetitiveness based on a multimodal large language model according to claim 4, characterized in that: The final vector representation of the question obtained by pooling the multimodal high-dimensional semantic vector representation includes: Representation of the generated high-dimensional semantic vector conduct Operation, compressing the high-dimensional vector into a low-dimensional vector of fixed length to retain the core features of the semantic information and obtain the final vector representation of the question .

6. The method for detecting test question repetitiveness based on a multimodal large language model according to claim 1, wherein: The final vector representation of the question is combined with the test question data exchange format JSON data to construct the Milvus collection and HNSW index, including: Multimodal representation vector set based on pooling , combined with the test question JSON metadata to build a Milvus database collection, where the fields are defined as test question ID, question text, image path, knowledge point, JSON metadata and multimodal vector; the vector field is retrieved and optimized through HNSW index, and the index parameters are set as the distance measurement method COSINE, the index construction parameters of the number of adjacent nodes M=64, and the search width.

7. The method for detecting test question repetitiveness based on a multimodal large language model according to claim 1, characterized in that: The multimodal representation of the query set is combined with Milvus's HNSW index to visualize the response results, including: The multimodal test questions are quickly deduplicated by calculating the vector cosine similarity. HNSW index of Milvus database is used to set parameters for query; the search results are generated into an interactive interface through the visualization engine, showing the contents of the TOP100 similar test questions.

8. The test question repetitiveness detection system based on a multimodal large language model is characterized by: include: The data acquisition module is used to collect multimodal test data as a question bank data set, and perform data cleaning on the question bank data set to obtain a cleaned data set; The cleaning and pooling module is used to generate a multimodal high-dimensional semantic vector representation of the test question based on the cleaned dataset and combine it with a multimodal large language model. The final vector representation of the question is obtained by pooling the multimodal high-dimensional semantic vector representation. The optimization module is used to represent the final vector of the question and build the Milvus collection and HNSW index based on the JSON data in the question data exchange format. The display output module is used to combine the multimodal representation of the query set with Milvus's HNSW index to visualize the response results.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the test question repetitiveness detection method based on a multimodal large language model as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the test question repetitiveness detection method based on a multimodal large language model as described in any one of claims 1 to 7 are implemented.

Images