Multi-modal selection question interference term generation method and system based on knowledge guidance thinking chain

Abstract

The invention discloses a multi-modal selection question interference item generation method and system based on a knowledge-guided thinking chain, and the method comprises the steps: collecting corpora, obtaining a rule feature set, and constructing an interference item setting rule library based on the rule feature set in a structured manner; feature extraction is carried out on the multi-modal topic information, a topic vector database is constructed, and examples similar to a target topic are retrieved and are used as few sample prompts; based on the constructed interference item, setting a rule base and a few-sample prompt of retrieval, and generating a knowledge guidance thinking chain with rule driving, retrieval enhancement and interference item setting guidance by utilizing a large reasoning model; and generating a corresponding interference item candidate set according to the knowledge guidance thinking chain and the multi-modal selection question by using a multi-modal large model, and outputting a final interference item. According to the method, a multi-mode processing mode is adopted, text and visual information of the questions can be comprehensively understood, and the method is suitable for more application scenes while understanding the questions more deeply.

Description

Technical Field

[0001] The present invention relates to the field of educational examination question setting, and in particular to a method and system for generating interference items for multimodal multiple-choice questions based on knowledge-guided thinking chains. Background Art

[0002] In the field of education, the interference items of high-quality multiple-choice questions need to have characteristics such as rationality, confusion, and inspiration, so as to effectively examine the students' knowledge mastery. In recent years, with the continuous advancement of educational informatization, educational propositions have gradually shifted towards automation and intelligence. In terms of the proposition of multiple-choice questions, the traditional proposition method is inefficient and requires strong ability of the propositioner. It is impossible to produce high-quality and high-quantity multiple-choice questions, and there are problems such as incomplete examination scope due to the subjective influence of the propositioner. The existing multiple-choice question interference item generation technology based on large models for instruction fine-tuning has problems such as lack of image information of question features, over-generalization of interference items, and unstable quality. Therefore, a new multimodal multiple-choice question interference item generation technology is urgently needed. In response to the above problems, the present invention proposes a multimodal multiple-choice question interference item generation method based on knowledge-guided thinking chain.

[0003] A new method for generating distractor items for Chinese multiple-choice questions based on free text has been proposed. The patent application is CN112686025A. This method proposes a method for generating distractor items for Chinese multiple-choice questions based on free text. The method involves: first, preprocessing the text to parse core keywords; second, searching the parsed core keywords in a knowledge graph, selecting the sentence that best matches the core keywords in the knowledge graph as the information sentence, and generating the correct answer and question stem; then, expanding the core keywords, calculating the semantic similarity between the expanded words and the knowledge graph triples, and filtering out distractor items; and finally, combining the question stem and distractor items to construct a complete multiple-choice question.

[0004] The above-mentioned method for generating distractor items for Chinese multiple-choice questions based on free text relies on traditional knowledge graphs and only enhances knowledge through dependency relationships. It lacks the guidance of educational rules for the generation of distractor items, and uses traditional machine learning methods to generate distractor items, which cannot make breakthroughs in the quality and diversity of distractor items.

[0005] A method for generating multiple distractors based on a large language model and reinforcement learning, with patent application publication number CN118760765A, is proposed for formulating multiple-choice questions. The method mainly includes: first, obtaining English multiple-choice question data and dividing the data into a training set and a validation set; second, establishing a distractor generation model based on text instructions, using the training set as input for supervised fine-tuning of the distractor generation model, and using the validation set as input for the supervised fine-tuned distractor generation model, and ranking the generated results; then, constructing a reward model, obtaining a training dataset for the reward model based on the ranking results, and calculating the pairwise ranking loss; finally, further training the supervised fine-tuned distractor generation model based on the reward model and reinforcement learning to obtain the highest-quality multiple distractors.

[0006] The above-mentioned multi-interference item generation method based on large language models and reinforcement learning relies on the basic text generation capabilities of the large language model, ignores the understanding of the image modal information of the question, lacks the guidance of the thinking chain on the generation of interference items, and the model based on reinforcement learning of a single dataset has poor generalization ability and cannot generate high-quality interference items in different scenarios. Summary of the Invention

[0007] The purpose of the present invention is to provide a method and system for generating interference items for multimodal multiple-choice questions based on knowledge-guided thinking chains to solve the above problems.

[0008] To achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for generating distractor items for a multimodal multiple-choice question based on knowledge-guided thinking chains, comprising: Collect corpus, obtain rule feature sets, and construct a structured interference item setting rule library based on the rule feature sets; Extract features from multimodal question information, build a question vector database, and retrieve examples similar to the target question as few-shot prompts; Based on the constructed distractor setting rule base and the few-sample retrieval prompts, a large inference model is used to generate a knowledge-guided thinking chain with rule-driven and retrieval-enhanced functions to guide distractor setting. Use a multimodal large model to generate corresponding interference item candidate sets based on knowledge-guided thinking chains and multimodal multiple-choice questions, and output the final interference items.

[0009] Furthermore, the collecting of corpus, obtaining a rule feature set, and constructing a structured interference item setting rule base based on the rule feature set include: Collect test questions and their parsed texts as corpus, and obtain rule feature sets through rule mining ;Through screening and structuring based on educational and psychological theories, a rule library for interference item settings is constructed .

[0010] Furthermore, specifically: in the construction of the interference item setting rule base, by collecting test questions and their parsed texts as corpus, construct implicit rule data samples ; In the data sample Based on the entity recognition and syntactic analysis rule mining method, a rule feature set is constructed Through educational and psychological theories Screen and structure to obtain the final interference item setting rule base .

[0011] Furthermore, the feature extraction of multimodal question information, the construction of a question vector database, and the retrieval of examples similar to the target question as a few-sample prompt include: Extract information from multimodal questions, capture the text semantic features and image visual features of the questions, and build a total question vector database ; Then use cosine similarity as the retrieval basis, in the question vector database Retrieve the most relevant examples from the query as a few-shot prompt .

[0012] Furthermore, based on the constructed distractor setting rule base and the retrieval of a few sample prompts, the inference model is used to generate a knowledge-guided thinking chain with rule-driven and retrieval-enhanced functions to guide distractor setting, including: In the generation of knowledge-guided thinking chains, by using the inference model, combining the interference item setting rule library and the retrieved few sample prompts, a prompt template is constructed for the knowledge-guided thinking chain; Few-shot hints for retrieval Construct knowledge-guided prompts to generate chains of thought using reasoning models with specific prompts , which makes the thinking chain have the knowledge guidance effect of rule-driven and retrieval-enhanced.

[0013] Furthermore, the multimodal large model is used to generate corresponding interference item candidate sets based on the knowledge-guided thinking chain and the multimodal multiple-choice questions, and output the final interference items, including: By using a large multimodal model, candidate distractors are generated based on the knowledge-guided thinking chain and the retrieved few-sample prompts, and they are screened for rule compliance and domain consistency to obtain the distractors that ultimately meet the requirements; by utilizing the retrieved few-sample prompts I and the generated knowledge-guided thinking chain C to construct prompts for distractor generation, the multimodal large model is used to specifically prompt the generation of a candidate set D^' of distractor items; after checking the rule compliance of the candidate set and screening it for domain consistency, the distractor D that meets the requirements of the target proposition is obtained.

[0014] In a second aspect, the present invention provides a multimodal multiple-choice question distractor generation system based on knowledge-guided thinking chains, comprising: The data acquisition module is used to collect corpus, obtain rule feature sets, and construct a structured interference item setting rule library based on the rule feature sets; The prompt generation module is used to extract features from multimodal question information, build a question vector database, and retrieve examples similar to the target question as few-shot prompts; The guided thinking chain generation module is used to generate a knowledge-guided thinking chain with rule-driven and retrieval-enhanced functions, and to guide the setting of distractor items based on the constructed distractor setting rule base and retrieval hints. The output module is used to use the multimodal large model to generate corresponding interference item candidate sets based on the knowledge-guided thinking chain and multimodal multiple-choice questions, and output the final interference items.

[0015] Furthermore, based on the constructed distractor setting rule base and the retrieval of a few sample prompts, the inference model is used to generate a knowledge-guided thinking chain with rule-driven and retrieval-enhanced functions to guide distractor setting, including: In the generation of knowledge-guided thinking chains, by using the inference model, combining the interference item setting rule library and the retrieved few sample prompts, a prompt template is constructed for the knowledge-guided thinking chain; Few-shot hints for retrieval Construct knowledge-guided prompts to generate chains of thought using reasoning models with specific prompts , which makes the thinking chain have rule-driven and retrieval-enhanced knowledge guidance effects; The multimodal large model is used to generate corresponding interference item candidate sets based on the knowledge-guided thinking chain and the multimodal multiple-choice questions, and output the final interference items, including: By using a large multimodal model, candidate distractors are generated based on the knowledge-guided thinking chain and the retrieved few-sample prompts, and they are screened for rule compliance and domain consistency to obtain the distractors that ultimately meet the requirements; by utilizing the retrieved few-sample prompts I and the generated knowledge-guided thinking chain C to construct prompts for distractor generation, the multimodal large model is used to specifically prompt the generation of a candidate set D^' of distractor items; after checking the rule compliance of the candidate set and screening it for domain consistency, the distractor D that meets the requirements of the target proposition is obtained.

[0016] 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 processor implements the steps of the method for generating interference items for multimodal multiple-choice questions based on knowledge-guided thinking chains.

[0017] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for generating interference items for multimodal multiple-choice questions based on knowledge-guided thinking chains.

[0018] 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 generating distractor items for multimodal multiple-choice questions based on knowledge-guided thought chains. The method realizes the generation of distractor items for multimodal multiple-choice questions by constructing a distractor item setting rule library, retrieving knowledge-related information, generating knowledge-guided thought chains, and generating distractor items. First, based on educational and psychological theories, the collected and processed rule feature library is screened and structured to construct a distractor item setting rule library; second, feature extraction is performed on multimodal question information, a question vector database is constructed, and the retrieved examples are used as few-sample prompts; then, a prompt template for generating knowledge-guided thought chains is designed, and a knowledge-guided thought chain for distractor item setting is produced using a large inference model; finally, a multimodal large model is used to generate and screen distractor items for multimodal multiple-choice questions based on knowledge-guided thought chains, few-sample prompts, and the designed distractor item generation prompt template. The present invention fully explores the proposition rules of high-quality test questions, and condenses them into the form of a interference item setting rule library under the guidance of educational and psychological theories, providing sufficient proposition basis, with the advantages of accurate proposition and rules close to reality; the present invention fully utilizes knowledge-related retrieval and multimodal processing capabilities, and through the guidance of knowledge-guided thinking chains, has the advantages of consistency in proposition fields, accurate understanding of question information, and in-depth thinking on how to proposition; these advantages make the present invention have obvious advantages over other methods based on large models to generate interference items.

[0019] This method uses multimodal processing to comprehensively understand the text and visual information of the question, providing a deeper understanding of the question's meaning and adapting it to more application scenarios. The retrieval of relevant knowledge enables the method to understand the specific domain of the question and achieve more accurate distractor generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a block diagram of the method for generating interference items for multimodal multiple-choice questions based on knowledge-guided thinking chains of the present invention.

[0021] Figure 2 It is a flowchart for constructing a rule base for setting interference items.

[0022] Figure 3 It is a flow chart of knowledge related information retrieval.

[0023] Figure 4 It is a flowchart generated by knowledge-guided thinking chain Figure 5 It is a flow chart of interference item generation and screening. 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 the process of constructing a interference item setting rule base, the process of retrieving knowledge-related information, the process of generating a knowledge-guided thinking chain, and the process of generating and screening interference items. Figure 1 It is a block diagram of the method for generating interference items for multimodal multiple-choice questions based on knowledge-guided thinking chains of the present invention.

[0026] 1. Process of building a rule base for interference setting This process aims to build a rule base that includes explicit rules and implicit knowledge for distractor setting, providing educational and psychological knowledge guidance for subsequent distractor generation. The specific process of building a distractor setting rule base is as follows: (1) For the data collection of interference item setting rule base, we collect high-quality test questions and their parsing texts as original corpus. The sources can include public education resource libraries, standardized test questions, high-quality teaching materials, etc. Each piece of original data should contain multimodal question information, question options and official test question parsing text. First, we clean and standardize the collected original corpus, remove non-text information such as advertisements and format tags, and construct a corpus containing The initial database of data samples , where each sample Then, we extract key information related to the logic of setting distractors from the parsed text, such as error cause classification, common misunderstandings, confusion of knowledge points, etc., and associate them with the corresponding questions, answers, and distractors to construct a sample set of implicit rule data for rule mining. .

[0027] (2) For the extracted implicit rule sample library , using natural language processing technology to mine potential interference item setting rule features. First, use the pre-trained entity recognition model (Named Entity Recognition, NER) to identify key entities in the question stem, options, and parsed text in the sample library, such as subject concepts, proper nouns, numerical values, and formula symbols, and obtain the entity set extracted from the sample library Then, we use dependency syntactic analysis to analyze the grammatical structure of the sample and identify the relationship between words, such as negation, comparison, causality, and conditional relationships. This helps us understand the structural or logical differences between the interference items and the correct answers, and obtain a syntactic analysis set extracted from the sample library. Finally, based on and The results of the above analysis are combined with statistical analysis and pattern matching methods to automatically extract regular patterns that describe the characteristics of interference items. For example, “the formula Variables in Replace with variables that are similar in form, sound, or meaning ", "For key values conduct Operation", "reversal of logical relationship ", "Confusing concepts and ", etc. Through this process, the construction contains Rule feature set of preliminary rules

[0028] (3) For the rule feature set By inviting experts in education, psychology and related disciplines to conduct manual review, screening and optimization. First, experts evaluate the characteristics of each rule based on educational theories such as Bloom's Cognitive Objectives Taxonomy and Cognitive Load Theory and psychological theories such as Cognitive Bias Theory and Learning Error Theory. The effectiveness, universality and educational value of the rules should be evaluated; then, redundant, erroneous or unrepresentative rules should be eliminated, similar rules should be merged, and the rules should be classified and graded according to the error category, difficulty level, discipline of use and other dimensions, and described using standardized language.

[0029] The above steps are as follows Figure 2 As shown, the final result contains A distractor setting rule base with high-quality structured rules ,The rule base is stored in a structured data format, which is suitable for subsequent calls.

[0030] 2. Knowledge-related information retrieval process This process aims to build a vector database containing multimodal question information and retrieve the most similar examples based on the input target question as few-shot prompting information. This is used by large models to improve the domain consistency and accuracy of the interference terms generated by the large models. The specific process of knowledge-related information retrieval is as follows: (1) For each multimodal question in the question bank (Contains text and images ), using the CLIP (Contrastive Language-Image Pre-training) model, specifically using Clip-Vit-Base-patch32 to extract the feature vectors of text and image in the shared semantic space Specifically, the CLIP model consists of a text encoder and an image encoder ; For text feature extraction, , for image feature extraction, ; Then, each question The extracted feature vector pairs ( ) and the unique identifier of the title Associative storage, building a total question vector database ,in Represents the identification index, text feature vector, and image feature vector of a single topic.

[0031] (2) For the target question with distractor items to be generated (Contains text and possible images ), first use the same CLIP model and processing flow to extract its text feature vector and image feature vector ; Secondly, traverse the vector database Each entry in , calculate the target question The comprehensive similarity with the questions in the database.

[0032] For text similarity, cosine similarity is used for calculation. The actual calculation formula is as follows:

[0033] in represents the vector dot product, Represents the Euclidean length of a vector.

[0034] For image features, only when the target title and database issues When both contain valid image vectors, the cosine similarity of their image vectors is calculated. If either side has no image (the corresponding vector is empty), the image similarity is considered to be 0 or not included in the calculation. Image similarity is calculated in the same way as text similarity.

[0035] Then, the weighted average method is used to fuse the text similarity and image similarity to obtain the final comprehensive similarity score. is a hyperparameter used to balance the importance of text and image information, which can be set empirically (e.g. indicates equal importance) or through experimental tuning. The final similarity score can be expressed as follows:

[0036]

[0037] Finally, calculate the target question and all After calculating the comprehensive similarity scores of the questions, sort the scores in descending order; select the identifier corresponding to the data entry with the highest ranking , retrieve the complete example information including question text, image, options and analysis according to the identifier as a few sample prompts , used to guide the subsequent generation of thought chains and interference items.

[0038] The process flow chart of retrieving the above knowledge related information is as follows Figure 3 shown.

[0039] 3. Knowledge guides the generation process of thinking chain This process aims to use the large inference model, combined with the distractor setting rule library and retrieved similar examples, to generate a detailed thinking process to guide the distractor setting. The specific process of generating a knowledge-guided thinking chain is as follows: (1) Construct a structured prompt template and input it into the reasoning model. Select DeepSeek-R1-7B as the base model and define the prompt template generated by the knowledge-guided thinking chain as follows: (2) Combine the interference rule base , Few sample prompts Target topic information Specific prompts for input reasoning large models , its output can be expressed as follows:

[0040] The output of the model is That is, the knowledge-guided thinking chain, which details the reasoning process of generating distractors based on distractor setting rules, as well as the reference basis for generating distractors for target questions, reflecting the characteristics of rule-driven and retrieval enhancement.

[0041] The flowchart of the generation process of the above knowledge-guided thinking chain is as follows Figure 4 shown.

[0042] 4. Interference item generation and screening process This process aims to use the multimodal large model to and few-shot tips , generate candidate interference items and screen them to obtain the interference items that meet the requirements. The specific process of interference item generation is as follows: (1) Construct a structured prompt template and input it to the multimodal large model. Select Qwen2.5-VL-7B as the base large model and define the prompt template for candidate distractor generation as follows: Will include thought chain , Few sample prompts and target topic information Specific prompts for input of multimodal large models , its output can be expressed as follows:

[0043] Among them, the interference candidate set is , the candidate set contains m candidate interference items.

[0044] (2) For the obtained interference candidate set ,Through post-processing screening, the quality requirements of the final output interference items are ensured. First, each candidate interference item is checked by rule compliance Compliance with the rule base The rules or chains of thought defined in Specifically, using rule-based matching, keyword checking, and defining a compliance scoring function , whose output can be a Boolean value (conform or not) or a continuous score. Secondly, the domain consistency check is used to ensure that the distractors are consistent with the subject area and knowledge point scope of the question. Specifically, the candidate distractors are calculated. and target topic Cosine similarity of Then, by setting the rule compliance threshold and domain consistency threshold To select the candidate interference set that meets all the check criteria. Select the one with the highest score Satisfaction and The candidate distractors (p is the number of distractors required by the target question, usually 3) constitute the final set of high-quality distractors that meet the requirements. .

[0045] The flow chart of the above interference items is as follows Figure 5 shown.

[0046] In yet another embodiment of the present invention, a system for generating distractor items for multimodal multiple-choice questions based on knowledge-guided thinking chains is provided, which can be used to implement the above-mentioned method for generating distractor items for multimodal multiple-choice questions based on knowledge-guided thinking chains. Specifically, the system includes: The data acquisition module is used to collect corpus, obtain rule feature sets, and construct a structured interference item setting rule library based on the rule feature sets; The prompt generation module is used to extract features from multimodal question information, build a question vector database, and retrieve examples similar to the target question as few-shot prompts; The guided thinking chain generation module is used to generate a knowledge-guided thinking chain with rule-driven and retrieval-enhanced functions, and to guide the setting of distractor items based on the constructed distractor setting rule base and retrieval hints. The output module is used to use the multimodal large model to generate corresponding interference item candidate sets based on the knowledge-guided thinking chain and multimodal multiple-choice questions, and output the final interference items.

[0047] 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.

[0048] In another embodiment of the present invention, a computer device is provided, comprising 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 may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate 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 multimodal multiple-choice question interference item generation method based on knowledge-guided thinking chain.

[0049] In another embodiment of the present invention, a storage medium is provided, 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 understood 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. Furthermore, 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 high-speed RAM memory or non-volatile memory, such as at least one disk storage device. The processor may load and execute the one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the method for generating distractor items for multimodal multiple-choice questions based on knowledge-guided thinking chains in the above-mentioned embodiment.

[0050] 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.

[0051] 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.

[0052] 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 1a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0053] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational 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.

[0054] 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.

Claims

1. A method for generating distractor items for multimodal multiple-choice questions based on knowledge-guided thinking chains, characterized by: include: Collect corpus, obtain rule feature sets, and construct a structured interference item setting rule library based on the rule feature sets; Extract features from multimodal question information, build a question vector database, and retrieve examples similar to the target question as few-shot prompts; Based on the constructed distractor setting rule base and the few-sample retrieval prompts, a large inference model is used to generate a knowledge-guided thinking chain with rule-driven and retrieval-enhanced functions to guide distractor setting. Use a multimodal large model to generate corresponding interference item candidate sets based on knowledge-guided thinking chains and multimodal multiple-choice questions, and output the final interference items.

2. The method for generating distractor items for multimodal multiple-choice questions based on knowledge-guided thinking chains according to claim 1 is characterized in that: The collecting of corpus, obtaining a rule feature set, and constructing a structured interference item setting rule base based on the rule feature set include: Collect test questions and their parsed texts as corpus, and obtain rule feature sets through rule mining ;Through screening and structuring based on educational and psychological theories, a rule library for interference item settings is constructed .

3. The method for generating distractor items for multimodal multiple-choice questions based on knowledge-guided thinking chains according to claim 2, characterized in that: Specifically: In the construction of the interference item setting rule library, by collecting test questions and their parsed texts as corpus, construct implicit rule data samples ; In the data sample Based on the entity recognition and syntactic analysis rule mining method, a rule feature set is constructed Through educational and psychological theories Screen and structure to obtain the final interference item setting rule base .

4. The method for generating distractor items for multimodal multiple-choice questions based on knowledge-guided thinking chains according to claim 1, characterized in that: The feature extraction of multimodal question information, the construction of a question vector database, and the retrieval of examples similar to the target question as a few-sample prompt include: Extract information from multimodal questions, capture the text semantic features and image visual features of the questions, and build a total question vector database ; Then use cosine similarity as the retrieval basis, in the question vector database Retrieve the most relevant examples from the query as a few-shot prompt .

5. The method for generating distractor items for multimodal multiple-choice questions based on knowledge-guided thinking chains according to claim 1, characterized in that: The method, based on the constructed distractor setting rule base and the few-sample retrieval prompts, utilizes the inference model to generate a knowledge-guided thinking chain with rule-driven and retrieval-enhanced functions to guide distractor setting, including: In the generation of knowledge-guided thinking chains, by using the inference model, combining the interference item setting rule library and the retrieved few sample prompts, a prompt template is constructed for the knowledge-guided thinking chain; Few-shot hints for retrieval Construct knowledge-guided prompts to generate chains of thought using reasoning models with specific prompts , which makes the thinking chain have the knowledge guidance effect of rule-driven and retrieval-enhanced.

6. The method for generating distractor items for multimodal multiple-choice questions based on knowledge-guided thinking chains according to claim 5, characterized in that: The multimodal large model is used to generate corresponding interference item candidate sets based on the knowledge-guided thinking chain and the multimodal multiple-choice questions, and output the final interference items, including: By using a large multimodal model, candidate distractors are generated based on the knowledge-guided thinking chain and the retrieved few-sample prompts, and they are screened for rule compliance and domain consistency to obtain the distractors that ultimately meet the requirements; by utilizing the retrieved few-sample prompts I and the generated knowledge-guided thinking chain C to construct prompts for distractor generation, the multimodal large model is used to specifically prompt the generation of a candidate set D^' of distractor items; after checking the rule compliance of the candidate set and screening it for domain consistency, the distractor D that meets the requirements of the target proposition is obtained.

7. A multimodal multiple-choice question distractor generation system based on knowledge-guided thinking chains, characterized by: include: The data acquisition module is used to collect corpus, obtain rule feature sets, and construct a structured interference item setting rule library based on the rule feature sets; The prompt generation module is used to extract features from multimodal question information, build a question vector database, and retrieve examples similar to the target question as few-shot prompts; The guided thinking chain generation module is used to generate a knowledge-guided thinking chain with rule-driven and retrieval-enhanced functions, and to guide the setting of distractor items based on the constructed distractor setting rule base and retrieval hints. The output module is used to use the multimodal large model to generate corresponding interference item candidate sets based on the knowledge-guided thinking chain and multimodal multiple-choice questions, and output the final interference items.

8. The multimodal multiple-choice question distractor generation system based on knowledge-guided thinking chain according to claim 7 is characterized in that: The method, based on the constructed distractor setting rule base and the few-sample retrieval prompts, utilizes the inference model to generate a knowledge-guided thinking chain with rule-driven and retrieval-enhanced functions to guide distractor setting, including: In the generation of knowledge-guided thinking chains, by using the inference model, combining the interference item setting rule library and the retrieved few sample prompts, a prompt template is constructed for the knowledge-guided thinking chain; Few-shot hints for retrieval Construct knowledge-guided prompts to generate chains of thought using reasoning models with specific prompts , which makes the thinking chain have rule-driven and retrieval-enhanced knowledge guidance effects; The multimodal large model is used to generate corresponding interference item candidate sets based on the knowledge-guided thinking chain and the multimodal multiple-choice questions, and output the final interference items, including: By using a large multimodal model, candidate distractors are generated based on the knowledge-guided thinking chain and the retrieved few-sample prompts, and they are screened for rule compliance and domain consistency to obtain the distractors that ultimately meet the requirements; by utilizing the retrieved few-sample prompts I and the generated knowledge-guided thinking chain C to construct prompts for distractor generation, the multimodal large model is used to specifically prompt the generation of a candidate set D^' of distractor items; after checking the rule compliance of the candidate set and screening it for domain consistency, the distractor D that meets the requirements of the target proposition is obtained.

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 method for generating distractor items for a multimodal multiple-choice question based on knowledge-guided thinking chain 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 method for generating distractor items for a multimodal multiple-choice question based on knowledge-guided thinking chain as described in any one of claims 1 to 7 are implemented.

Images