Student comprehensive quality evaluation method and system based on mixed large model

Abstract

The invention provides a student comprehensive quality evaluation method and system based on a mixed large model, and belongs to the technical field of education management. The method comprises the following steps: S1, constructing an evaluation index system for student comprehensive quality evaluation based on an evaluation target; s2, constructing a large model knowledge base based on the evaluation index system through an RAG technology; s3, constructing a mixed evaluation large model comprising an input layer, an analysis layer, a discussion layer and an aggregation layer; s4, acquiring to-be-evaluated student data, and preprocessing the to-be-evaluated student data; and S5, inputting the preprocessed to-be-evaluated student data into a mixed evaluation large model, and outputting a comprehensive quality evaluation result. The method has the advantages that objectivity, comprehensiveness and flexibility of student comprehensive quality evaluation are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of educational management technology, and in particular to a method and system for evaluating the comprehensive quality of students based on a hybrid large model. Background Art

[0002] In order to teach students in accordance with their aptitude, fill in their gaps, and assess their learning outcomes, it is necessary to evaluate the students' comprehensive qualities. The evaluation of students' comprehensive qualities has traditionally relied on teachers' subjective judgment, standardized tests, and scales (questionnaires). Traditional methods have limitations in the comprehensiveness and objectivity of evaluation and can no longer meet current evaluation needs.

[0003] As large language models have shown strong capabilities in natural language understanding and generation tasks, using large language models to evaluate students' comprehensive qualities has become a future development direction. However, the existing large language models have the following shortcomings when applied to the evaluation of students' comprehensive qualities:

[0004] 1. A single large-scale language model is often limited by its training data and model structure, and it is difficult to fully cover the multiple dimensions and multimodal data of students' comprehensive qualities. In addition, student data from different sources and formats are difficult to effectively integrate, which makes the objectivity and comprehensiveness of the evaluation need to be improved; 2. When faced with complex and diverse evaluation scenarios, it is often affected by factors such as model capacity limitations and training data bias, making it difficult to fully capture the comprehensive qualities of individuals; 3. Different students have unique growth backgrounds and development needs. A single large-scale language model is difficult to provide highly personalized evaluations and suggestions, and has poor flexibility.

[0005] Therefore, how to provide a method and system for evaluating students' comprehensive qualities based on a hybrid large model to improve the objectivity, comprehensiveness and flexibility of students' comprehensive quality evaluation has become a technical problem that needs to be solved urgently. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for evaluating the comprehensive quality of students based on a hybrid large model, so as to improve the objectivity, comprehensiveness and flexibility of the evaluation of the comprehensive quality of students.

[0007] In a first aspect, the present invention provides a method for evaluating the comprehensive quality of students based on a hybrid large model, comprising the following steps:

[0008] Step S1: construct an evaluation index system for evaluating students' comprehensive quality based on the evaluation objectives;

[0009] Step S2: constructing a large model knowledge base based on the evaluation index system through RAG technology;

[0010] Step S3, construct a hybrid evaluation large model including an input layer, an analysis layer, a discussion layer and an aggregation layer; the input layer includes three first large language models for inputting student data and prompt words; the analysis layer includes three second large language models for invoking the large model knowledge base according to the prompt words to perform multi-dimensional evaluation on the student data to obtain preliminary evaluation results of each dimension; the discussion layer includes three third large language models for comparing and integrating the preliminary evaluation results to obtain a number of optimized evaluation results; the aggregation layer includes a fourth large language model for outputting comprehensive quality evaluation results according to the optimized evaluation results;

[0011] Step S4: obtaining the student data to be evaluated, and preprocessing the student data to be evaluated;

[0012] Step S5: input the preprocessed student data to be evaluated into the hybrid evaluation model and output the comprehensive quality evaluation results.

[0013] Furthermore, the step S1 is specifically as follows:

[0014] Evaluation goals are set based on the six dimensions of ideology and morality, academic level, physical and mental health, artistic literacy, social practice, and labor literacy. Evaluation indicators are set for each evaluation goal, including at least indicator definition, scoring criteria, and weights. The evaluation indicators are organized according to a preset logical structure to construct an evaluation indicator system for evaluating students' comprehensive qualities.

[0015] Furthermore, the step S2 is specifically as follows:

[0016] Obtain a large amount of student historical data including at least ideological and moral data, academic level data, physical and mental health data, artistic literacy data, social practice data and labor literacy data, perform preprocessing on each of the student historical data including at least abnormal data detection, data cleaning and format conversion, and then map each of the student historical data to the evaluation index corresponding to the evaluation index system through RAG technology to construct a large model knowledge base for the analysis layer to call, and continuously update the large model knowledge base based on the latest student data;

[0017] The ideological and moral data at least include activity project data, reward project data and violation of rules and regulations data; the academic level data at least include subject competition data, elective course data and test scores; the physical and mental health data at least include mental health data, sports scores and sports hobbies; the artistic literacy data at least include art teams, art activities, art hobbies and art scores; the social practice data at least include practical activities, research results, scientific and technological innovation and creative inventions; the labor literacy data at least include labor competition data, labor results, ability and technical certificates and practical projects.

[0018] Furthermore, in step S3, the three first large-scale language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner; the three second large-scale language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner; the three third large-scale language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner.

[0019] Furthermore, the step S4 is specifically as follows:

[0020] Acquire the student data to be evaluated through a database, and perform preprocessing on the student data to be evaluated, including at least abnormal data detection, data cleaning, and format conversion;

[0021] The step S5 is specifically as follows:

[0022] The pre-processed student data to be evaluated is input into the hybrid evaluation model in real time, and the comprehensive quality evaluation results are output. The comprehensive quality evaluation results are displayed in the form of charts or graphs through a visual interface, and the comprehensive quality evaluation results are stored and backed up.

[0023] In a second aspect, the present invention provides a student comprehensive quality evaluation system based on a hybrid large model, comprising the following modules:

[0024] An evaluation index system construction module is used to construct an evaluation index system for evaluating students' comprehensive qualities based on evaluation objectives;

[0025] A large model knowledge base construction module is used to construct a large model knowledge base based on the evaluation index system through RAG technology;

[0026] A hybrid evaluation large model construction module is used to construct a hybrid evaluation large model including an input layer, an analysis layer, a discussion layer and an aggregation layer; the input layer includes three first large language models for inputting student data and prompt words; the analysis layer includes three second large language models for invoking the large model knowledge base according to the prompt words to perform multi-dimensional evaluation on the student data to obtain preliminary evaluation results of each dimension; the discussion layer includes three third large language models for comparing and integrating the preliminary evaluation results to obtain a number of optimized evaluation results; the aggregation layer includes a fourth large language model for outputting comprehensive quality evaluation results according to the optimized evaluation results;

[0027] A preprocessing module for student data to be evaluated, used to obtain the student data to be evaluated and preprocess the student data to be evaluated;

[0028] The comprehensive quality evaluation module is used to input the pre-processed student data to be evaluated into the hybrid evaluation model and output the comprehensive quality evaluation results.

[0029] Furthermore, the evaluation index system construction module is specifically used for:

[0030] Evaluation goals are set based on the six dimensions of ideology and morality, academic level, physical and mental health, artistic literacy, social practice, and labor literacy. Evaluation indicators are set for each evaluation goal, including at least indicator definition, scoring criteria, and weights. The evaluation indicators are organized according to a preset logical structure to construct an evaluation indicator system for evaluating students' comprehensive qualities.

[0031] Furthermore, the large model knowledge base construction module is specifically used for:

[0032] Obtain a large amount of student historical data including at least ideological and moral data, academic level data, physical and mental health data, artistic literacy data, social practice data and labor literacy data, perform preprocessing on each of the student historical data including at least abnormal data detection, data cleaning and format conversion, and then map each of the student historical data to the evaluation index corresponding to the evaluation index system through RAG technology to construct a large model knowledge base for the analysis layer to call, and continuously update the large model knowledge base based on the latest student data;

[0033] The ideological and moral data at least include activity project data, reward project data and violation of rules and regulations data; the academic level data at least include subject competition data, elective course data and test scores; the physical and mental health data at least include mental health data, sports scores and sports hobbies; the artistic literacy data at least include art teams, art activities, art hobbies and art scores; the social practice data at least include practical activities, research results, scientific and technological innovation and creative inventions; the labor literacy data at least include labor competition data, labor results, ability and technical certificates and practical projects.

[0034] Furthermore, in the hybrid evaluation large model construction module, the three first large language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner; the three second large language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner; the three third large language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner.

[0035] Furthermore, the student data preprocessing module to be evaluated is specifically used for:

[0036] Acquire the student data to be evaluated through a database, and perform preprocessing on the student data to be evaluated, including at least abnormal data detection, data cleaning, and format conversion;

[0037] The comprehensive quality evaluation module is specifically used for:

[0038] The pre-processed student data to be evaluated is input into the hybrid evaluation model in real time, and the comprehensive quality evaluation results are output. The comprehensive quality evaluation results are displayed in the form of charts or graphs through a visual interface, and the comprehensive quality evaluation results are stored and backed up.

[0039] The advantages of the present invention are:

[0040] By constructing an evaluation index system for evaluating students' comprehensive quality, and then using RAG technology to build a large model knowledge base based on the evaluation index system, and then constructing a hybrid evaluation large model including an input layer, an analysis layer, a discussion layer and an aggregation layer; the input layer includes three first large language models for inputting student data and prompt words; the analysis layer includes three second large language models for calling the large model knowledge base according to the prompt words to conduct multi-dimensional evaluation of student data and obtain preliminary evaluation results of each dimension; the discussion layer includes three third large language models for comparing and integrating the preliminary evaluation results to obtain several optimized evaluation results; the aggregation layer includes a fourth large language model for outputting comprehensive evaluation results according to the optimized evaluation results. Quality evaluation results; finally, the student data to be evaluated is obtained and pre-processed before being input into the hybrid evaluation large model to output the comprehensive quality evaluation results; that is, the present invention performs a multi-dimensional comprehensive quality evaluation on the input student data to be evaluated by a hybrid evaluation large model that combines a number of first large language models, second large language models, third large language models, and fourth large language models, and combines the advantages of different large language models to provide highly personalized evaluations and suggestions, and the large model knowledge base constructed through the evaluation index system assists the comprehensive quality evaluation of the hybrid evaluation large model to improve the accuracy and reliability of the comprehensive quality evaluation results, and ultimately greatly improves the objectivity, comprehensiveness and flexibility of the comprehensive quality evaluation of students. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention will be further described below in conjunction with embodiments with reference to the accompanying drawings.

[0042] Figure 1 It is a flow chart of a method for evaluating comprehensive quality of students based on a hybrid large model of the present invention.

[0043] Figure 2 It is a structural schematic diagram of a student comprehensive quality evaluation system based on a hybrid large model of the present invention.

[0044] Figure 3 It is a structural schematic diagram of the hybrid evaluation large model of the present invention. DETAILED DESCRIPTION

[0045] The technical solution in the embodiment of the present application has the following overall idea: by integrating a hybrid evaluation large model that includes a plurality of first large-scale language models, a second large-scale language model, a third large-scale language model, and a fourth large-scale language model, a multi-dimensional comprehensive quality evaluation is performed on the input student data to be evaluated. By combining the advantages of different large-scale language models, highly personalized evaluations and suggestions can be provided. In addition, a large-model knowledge base constructed through an evaluation index system assists the comprehensive quality evaluation of the hybrid evaluation large model to improve the accuracy and reliability of the comprehensive quality evaluation results, thereby improving the objectivity, comprehensiveness, and flexibility of the comprehensive quality evaluation of students.

[0046] Please refer to Figures 1 to 3 As shown, a preferred embodiment of a method for evaluating comprehensive quality of students based on a hybrid large model of the present invention comprises the following steps:

[0047] Step S1: construct an evaluation index system for evaluating students' comprehensive quality based on the evaluation objectives;

[0048] Step S2, constructing a large model knowledge base based on the evaluation index system through RAG technology (Retrieval-Augmented Generation technology);

[0049] Step S3, construct a hybrid evaluation large model including an input layer, an analysis layer, a discussion layer and an aggregation layer; the input layer includes three first large language models for inputting student data and prompt words (Prompt); the analysis layer includes three second large language models for calling the large model knowledge base according to the prompt words to perform multi-dimensional evaluation on the student data and obtain preliminary evaluation results of each dimension; the discussion layer includes three third large language models for comparing and integrating the preliminary evaluation results to obtain a number of optimized evaluation results, that is, simulating the expert discussion process; the aggregation layer includes a fourth large language model for outputting comprehensive quality evaluation results according to the optimized evaluation results, that is, acting as an aggregator for high-quality output;

[0050] The hybrid evaluation model consists of a four-layer network structure, each layer contains a different large language model, and the j-th output Y of the i-th layer is expressed as:

[0051]

[0052] in, Represents an aggregation operation; A i,j represents the j-th large language model at the i-th layer; x i represents the prompt word of the i-th layer input; + represents the text connection operation; x1 represents the prompt word of the 1st layer input; x i+1Represents the prompt word input at the i+1th layer;

[0053] The prompt word refers to a piece of text or instruction input by the user to the hybrid evaluation model, which is used to guide the hybrid evaluation model to generate specific outputs or perform specific tasks. The prompt word is crucial to the performance and output quality of the hybrid evaluation model, and directly affects the way the hybrid evaluation model understands and processes user requests. The format of the prompt word is as follows: "The student submitted a record in the {indicator} module of the student comprehensive evaluation system, the content is: {title} {description}, and uploaded this picture, the content of the picture is: {picture content}. As a class teacher, please start the evaluation with the first person "you". Please make sure that the evaluation content is concise and clear, highlight the student's achievements and progress, and give specific evaluations and suggestions. The word count is limited to 100 words."

[0054] Step S4: obtaining the student data to be evaluated, and preprocessing the student data to be evaluated;

[0055] Step S5: input the preprocessed student data to be evaluated into the hybrid evaluation model and output the comprehensive quality evaluation results.

[0056] The present invention optimizes the generated answers layer by layer through the collaboration of multi-level large language models (LLMs). Each large language model not only generates its own answers, but also refers to the output of the large language model of the previous layer to gradually improve the quality of the generated results. That is, the accuracy of the overall comprehensive quality evaluation is improved through the collaboration of multiple large language models.

[0057] The step S1 is specifically as follows:

[0058] Evaluation goals are set based on the six dimensions of ideology and morality, academic level, physical and mental health, artistic literacy, social practice, and labor literacy. Evaluation indicators are set for each evaluation goal, including at least indicator definition, scoring standards, and weights to ensure the accuracy and fairness of the comprehensive quality evaluation. The evaluation indicators are organized according to a preset logical structure to construct an evaluation indicator system for the comprehensive quality evaluation of students.

[0059] The evaluation index system described above is the basis for comprehensive quality evaluation and helps to ensure the accuracy and fairness of comprehensive quality evaluation.

[0060] The step S2 is specifically as follows:

[0061] Obtain a large amount of student historical data including at least ideological and moral data, academic level data, physical and mental health data, artistic literacy data, social practice data and labor literacy data, and after preprocessing each of the student historical data including at least abnormal data detection, data cleaning and format conversion, map each of the student historical data to the evaluation index corresponding to the evaluation index system through RAG technology to construct a large model knowledge base for the analysis layer to call, and continuously update the large model knowledge base based on the latest student data to ensure that the large model knowledge base keeps pace with the times; the large model knowledge base is used to provide knowledge support for the comprehensive quality evaluation of the hybrid evaluation large model;

[0062] The ideological and moral data at least include activity project data, reward project data and violation of rules and regulations data; the academic level data at least include subject competition data, elective course data and test scores; the physical and mental health data at least include mental health data, sports scores and sports hobbies; the artistic literacy data at least include art teams, art activities, art hobbies and art scores; the social practice data at least include practical activities, research results, scientific and technological innovation and creative inventions; the labor literacy data at least include labor competition data, labor results, ability and technical certificates and practical projects.

[0063] RAG technology is a natural language processing technology that combines retrieval and generation capabilities. It retrieves relevant information from the knowledge base and provides it to a large language model to generate high-quality text output. RAG technology introduces an external knowledge base on the basis of a large language model, which improves the ability of a large language model to handle complex tasks. The purpose of building a large model knowledge base based on RAG technology is to support the hybrid evaluation large model to conduct a comprehensive quality evaluation of student data, provide query and understanding of indicators of different dimensions, and use it to establish indexes, so that the hybrid evaluation large model can dynamically understand the evaluation indicator system and the corresponding knowledge observation points, and retrieve faster. The specific approach is: analyze and extract key information from the evaluation indicator system and student historical data, classify and organize according to the primary dimension, secondary dimension and observation point, and encode the student historical data and save it in the database in the form of vectors to facilitate the hybrid evaluation large model to call and process.

[0064] In step S3, the three first large language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner; the three second large language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner; the three third large language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner. By defining a unified input, it is ensured that different large language models can smoothly exchange information and collaborate.

[0065] The first large-scale language model, the second large-scale language model, the third large-scale language model and the fourth large-scale language model are pre-trained; the accuracy of comprehensive quality evaluation is improved through the collaboration of multiple pre-trained large-scale language models, and higher quality responses are generated based on the outputs of different large-scale language models. Through the interaction and information sharing between large-scale language models, the accuracy and reliability of comprehensive quality evaluation are improved, while reducing human subjective bias, thereby achieving a multi-dimensional and refined evaluation of comprehensive quality.

[0066] The first large language model adopts the Tongyi Qianwen Qwen-VL large language model, which belongs to the visual language large model. It can not only understand and generate text, but also understand and generate information related to vision, such as objects, scenes and activities in images, which is suitable for analyzing student data; the second large language model adopts the Zhipu Qingyan ChatGLM large language model. The ChatGLM large language model is combined with the Langcha in application framework to realize the offline deployment of RAG application projects. Langcha in is a framework for LLM application development, which includes functions such as model calling and data source connection. The combination of the two can realize the evaluation system based on the knowledge base; the third large language model adopts the iFlytek Spark Cognitive Large Language Model, which is good at Chinese language processing and friendly to educational scenarios; the fourth large language model adopts the Wenxin Yiyan large language model, which has strong Chinese natural language processing capabilities, integrates the results of the discussion layer, and forms a high-quality and unified evaluation output. Each of the major language models (large language models) has its own characteristics in text understanding, reasoning, generation, etc., and can provide a variety of perspectives for comprehensive quality evaluation.

[0067] The step S4 is specifically as follows:

[0068] Acquire the student data to be evaluated through a database, and perform preprocessing on the student data to be evaluated, including at least abnormal data detection, data cleaning, and format conversion;

[0069] The step S5 is specifically as follows:

[0070] The pre-processed student data to be evaluated is input into the hybrid evaluation model in real time, and the comprehensive quality evaluation results are output. The comprehensive quality evaluation results are displayed in the form of charts or graphs through a visual interface, and the comprehensive quality evaluation results are stored and backed up. In specific implementation, a student portrait can be output based on the comprehensive quality evaluation results to intuitively understand the comprehensive quality.

[0071] The storage and backup of the comprehensive quality evaluation results are specifically as follows:

[0072] The comprehensive quality evaluation result is encrypted into first-level encrypted data through the RC6 algorithm, each character of the first-level encrypted data is shifted right by 9 bits to obtain second-level encrypted data, the second-level encrypted data is encrypted through the 3DES algorithm to obtain third-level encrypted data, the third-level encrypted data is mapped based on a preset mapping rule to obtain fourth-level encrypted data, the number 8 and the letter B in the fourth-level encrypted data are swapped to obtain fifth-level encrypted data, the fifth-level encrypted data is encrypted through the ChaCha20 algorithm to obtain an encrypted result, and the encrypted result is stored and distributedly backed up.

[0073] The comprehensive quality evaluation results are encrypted into first-level encrypted data through the RC6 algorithm, and each character of the first-level encrypted data is shifted right by 9 bits to obtain second-level encrypted data. The second-level encrypted data is encrypted through the 3DES algorithm to obtain third-level encrypted data. The third-level encrypted data is encrypted based on the mapping rule to obtain fourth-level encrypted data. The number 8 and the letter B in the fourth-level encrypted data are swapped to obtain fifth-level encrypted data. The fifth-level encrypted data is encrypted through the ChaCha20 algorithm to obtain the encrypted result. If the corresponding encryption algorithm or data transformation rule is not known, the encryption result cannot be cracked. At least 7 security measures are taken before and after (RC6 algorithm, character shift right by 9 bits, 3DES algorithm, mapping rule, swapping number 8 and letter B, ChaCha20 algorithm, distributed backup), which greatly improves the security of the storage of comprehensive quality evaluation results, thereby effectively protecting privacy.

[0074] The decryption process of the encryption result is: decrypt the encryption result through the ChaCha20 algorithm to obtain level-five encrypted data, swap the number 8 and the letter B in the level-five encrypted data to obtain level-four encrypted data, map the level-four encrypted data based on a preset mapping rule to obtain level-three encrypted data, decrypt the level-three encrypted data through the 3DES algorithm to obtain level-two encrypted data, shift each character of the level-two encrypted data to the left by 9 bits to obtain level-one encrypted data, and decrypt the level-one encrypted data into a comprehensive quality evaluation result through the RC6 algorithm.

[0075] A preferred embodiment of a student comprehensive quality evaluation system based on a hybrid large model of the present invention includes the following modules:

[0076] An evaluation index system construction module is used to construct an evaluation index system for evaluating students' comprehensive qualities based on evaluation objectives;

[0077] A large model knowledge base construction module, used to construct a large model knowledge base based on the evaluation index system through RAG technology (Retrieval-Augmented Generation technology);

[0078] A hybrid evaluation large model construction module is used to construct a hybrid evaluation large model including an input layer, an analysis layer, a discussion layer and an aggregation layer; the input layer includes three first large language models for inputting student data and prompt words (Prompt); the analysis layer includes three second large language models for invoking the large model knowledge base according to the prompt words to perform multi-dimensional evaluation on student data and obtain preliminary evaluation results of each dimension; the discussion layer includes three third large language models for comparing and integrating the preliminary evaluation results to obtain a number of optimized evaluation results, that is, simulating the expert discussion process; the aggregation layer includes a fourth large language model for outputting comprehensive quality evaluation results according to the optimized evaluation results, that is, acting as an aggregator for high-quality output;

[0079] The hybrid evaluation model consists of a four-layer network structure, each layer contains a different large language model, and the j-th output Y of the i-th layer is expressed as:

[0080]

[0081] in, Represents an aggregation operation; A i,j represents the j-th large language model at the i-th layer; x i represents the prompt word of the i-th layer input; + represents the text connection operation; x1 represents the prompt word of the 1st layer input; x i+1 Represents the prompt word input at the i+1th layer;

[0082] The prompt word refers to a piece of text or instruction input by the user to the hybrid evaluation model, which is used to guide the hybrid evaluation model to generate specific outputs or perform specific tasks. The prompt word is crucial to the performance and output quality of the hybrid evaluation model, and directly affects the way the hybrid evaluation model understands and processes user requests. The format of the prompt word is as follows: "The student submitted a record in the {indicator} module of the student comprehensive evaluation system, the content is: {title} {description}, and uploaded this picture, the content of the picture is: {picture content}. As a class teacher, please start the evaluation with the first person "you". Please make sure that the evaluation content is concise and clear, highlight the student's achievements and progress, and give specific evaluations and suggestions. The word count is limited to 100 words."

[0083] A preprocessing module for student data to be evaluated, used to obtain the student data to be evaluated and preprocess the student data to be evaluated;

[0084] The comprehensive quality evaluation module is used to input the pre-processed student data to be evaluated into the hybrid evaluation model and output the comprehensive quality evaluation results.

[0085] The present invention optimizes the generated answers layer by layer through the collaboration of multi-level large language models (LLMs). Each large language model not only generates its own answers, but also refers to the output of the large language model of the previous layer to gradually improve the quality of the generated results. That is, the accuracy of the overall comprehensive quality evaluation is improved through the collaboration of multiple large language models.

[0086] The evaluation index system construction module is specifically used for:

[0087] Evaluation goals are set based on the six dimensions of ideology and morality, academic level, physical and mental health, artistic literacy, social practice, and labor literacy. Evaluation indicators are set for each evaluation goal, including at least indicator definition, scoring standards, and weights to ensure the accuracy and fairness of the comprehensive quality evaluation. The evaluation indicators are organized according to a preset logical structure to construct an evaluation indicator system for the comprehensive quality evaluation of students.

[0088] The evaluation index system described above is the basis for comprehensive quality evaluation and helps to ensure the accuracy and fairness of comprehensive quality evaluation.

[0089] The large model knowledge base construction module is specifically used for:

[0090] Obtain a large amount of student historical data including at least ideological and moral data, academic level data, physical and mental health data, artistic literacy data, social practice data and labor literacy data, and after preprocessing each of the student historical data including at least abnormal data detection, data cleaning and format conversion, map each of the student historical data to the evaluation index corresponding to the evaluation index system through RAG technology to construct a large model knowledge base for the analysis layer to call, and continuously update the large model knowledge base based on the latest student data to ensure that the large model knowledge base keeps pace with the times; the large model knowledge base is used to provide knowledge support for the comprehensive quality evaluation of the hybrid evaluation large model;

[0091] The ideological and moral data at least include activity project data, reward project data and violation of rules and regulations data; the academic level data at least include subject competition data, elective course data and test scores; the physical and mental health data at least include mental health data, sports scores and sports hobbies; the artistic literacy data at least include art teams, art activities, art hobbies and art scores; the social practice data at least include practical activities, research results, scientific and technological innovation and creative inventions; the labor literacy data at least include labor competition data, labor results, ability and technical certificates and practical projects.

[0092] RAG technology is a natural language processing technology that combines retrieval and generation capabilities. It retrieves relevant information from the knowledge base and provides it to a large language model to generate high-quality text output. RAG technology introduces an external knowledge base on the basis of a large language model, which improves the ability of a large language model to handle complex tasks. The purpose of building a large model knowledge base based on RAG technology is to support the hybrid evaluation large model to conduct a comprehensive quality evaluation of student data, provide query and understanding of indicators of different dimensions, and use it to establish indexes, so that the hybrid evaluation large model can dynamically understand the evaluation indicator system and the corresponding knowledge observation points, and retrieve faster. The specific approach is: analyze and extract key information from the evaluation indicator system and student historical data, classify and organize according to the primary dimension, secondary dimension and observation point, and encode the student historical data and save it in the database in the form of vectors to facilitate the hybrid evaluation large model to call and process.

[0093] In the hybrid evaluation large model construction module, the three first large language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner; the three second large language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner; the three third large language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner. By defining a unified input, it is ensured that different large language models can exchange information and collaborate smoothly.

[0094] The first large-scale language model, the second large-scale language model, the third large-scale language model and the fourth large-scale language model are pre-trained; the accuracy of comprehensive quality evaluation is improved through the collaboration of multiple pre-trained large-scale language models, and higher quality responses are generated based on the outputs of different large-scale language models. Through the interaction and information sharing between large-scale language models, the accuracy and reliability of comprehensive quality evaluation are improved, while reducing human subjective bias, thereby achieving a multi-dimensional and refined evaluation of comprehensive quality.

[0095] The first large language model adopts the Tongyi Qianwen Qwen-VL large language model, which belongs to the visual language large model. It can not only understand and generate text, but also understand and generate information related to vision, such as objects, scenes and activities in images, which is suitable for analyzing student data; the second large language model adopts the Zhipu Qingyan ChatGLM large language model. The ChatGLM large language model is combined with the Langcha in application framework to realize the offline deployment of RAG application projects. Langcha in is a framework for LLM application development, which includes functions such as model calling and data source connection. The combination of the two can realize the evaluation system based on the knowledge base; the third large language model adopts the iFlytek Spark Cognitive Large Language Model, which is good at Chinese language processing and friendly to educational scenarios; the fourth large language model adopts the Wenxin Yiyan large language model, which has strong Chinese natural language processing capabilities, integrates the results of the discussion layer, and forms a high-quality and unified evaluation output. Each of the major language models (large language models) has its own characteristics in text understanding, reasoning, generation, etc., and can provide a variety of perspectives for comprehensive quality evaluation.

[0096] The student data preprocessing module to be evaluated is specifically used for:

[0097] Acquire the student data to be evaluated through a database, and perform preprocessing on the student data to be evaluated, including at least abnormal data detection, data cleaning, and format conversion;

[0098] The comprehensive quality evaluation module is specifically used for:

[0099] The pre-processed student data to be evaluated is input into the hybrid evaluation model in real time, and the comprehensive quality evaluation results are output. The comprehensive quality evaluation results are displayed in the form of charts or graphs through a visual interface, and the comprehensive quality evaluation results are stored and backed up. In specific implementation, a student portrait can be output based on the comprehensive quality evaluation results to intuitively understand the comprehensive quality.

[0100] The storage and backup of the comprehensive quality evaluation results are specifically as follows:

[0101] The comprehensive quality evaluation result is encrypted into first-level encrypted data through the RC6 algorithm, each character of the first-level encrypted data is shifted right by 9 bits to obtain second-level encrypted data, the second-level encrypted data is encrypted through the 3DES algorithm to obtain third-level encrypted data, the third-level encrypted data is mapped based on a preset mapping rule to obtain fourth-level encrypted data, the number 8 and the letter B in the fourth-level encrypted data are swapped to obtain fifth-level encrypted data, the fifth-level encrypted data is encrypted through the ChaCha20 algorithm to obtain an encrypted result, and the encrypted result is stored and distributedly backed up.

[0102] The comprehensive quality evaluation results are encrypted into first-level encrypted data through the RC6 algorithm, and each character of the first-level encrypted data is shifted right by 9 bits to obtain second-level encrypted data. The second-level encrypted data is encrypted through the 3DES algorithm to obtain third-level encrypted data. The third-level encrypted data is encrypted based on the mapping rule to obtain fourth-level encrypted data. The number 8 and the letter B in the fourth-level encrypted data are swapped to obtain fifth-level encrypted data. The fifth-level encrypted data is encrypted through the ChaCha20 algorithm to obtain the encrypted result. If the corresponding encryption algorithm or data transformation rule is not known, the encryption result cannot be cracked. At least 7 security measures are taken before and after (RC6 algorithm, character shift right by 9 bits, 3DES algorithm, mapping rule, swapping number 8 and letter B, ChaCha20 algorithm, distributed backup), which greatly improves the security of the storage of comprehensive quality evaluation results, thereby effectively protecting privacy.

[0103] The decryption process of the encryption result is: decrypt the encryption result through the ChaCha20 algorithm to obtain level-five encrypted data, swap the number 8 and the letter B in the level-five encrypted data to obtain level-four encrypted data, map the level-four encrypted data based on a preset mapping rule to obtain level-three encrypted data, decrypt the level-three encrypted data through the 3DES algorithm to obtain level-two encrypted data, shift each character of the level-two encrypted data to the left by 9 bits to obtain level-one encrypted data, and decrypt the level-one encrypted data into a comprehensive quality evaluation result through the RC6 algorithm.

[0104] In summary, the advantages of the present invention are:

[0105] By constructing an evaluation index system for evaluating students' comprehensive quality, and then using RAG technology to build a large model knowledge base based on the evaluation index system, and then constructing a hybrid evaluation large model including an input layer, an analysis layer, a discussion layer and an aggregation layer; the input layer includes three first large language models for inputting student data and prompt words; the analysis layer includes three second large language models for calling the large model knowledge base according to the prompt words to conduct multi-dimensional evaluation of student data and obtain preliminary evaluation results of each dimension; the discussion layer includes three third large language models for comparing and integrating the preliminary evaluation results to obtain several optimized evaluation results; the aggregation layer includes a fourth large language model for outputting comprehensive evaluation results according to the optimized evaluation results. Quality evaluation results; finally, the student data to be evaluated is obtained and pre-processed before being input into the hybrid evaluation large model to output the comprehensive quality evaluation results; that is, the present invention performs a multi-dimensional comprehensive quality evaluation on the input student data to be evaluated by a hybrid evaluation large model that combines a number of first large language models, second large language models, third large language models, and fourth large language models, and combines the advantages of different large language models to provide highly personalized evaluations and suggestions, and the large model knowledge base constructed through the evaluation index system assists the comprehensive quality evaluation of the hybrid evaluation large model to improve the accuracy and reliability of the comprehensive quality evaluation results, and ultimately greatly improves the objectivity, comprehensiveness and flexibility of the comprehensive quality evaluation of students.

[0106] Although the specific implementation modes of the present invention are described above, those skilled in the art should understand that the specific implementation modes described are only illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A method for evaluating the comprehensive quality of students based on a hybrid large model, characterized by: The steps include: Step S1: construct an evaluation index system for evaluating students' comprehensive quality based on the evaluation objectives; Step S2: constructing a large model knowledge base based on the evaluation index system through RAG technology; Step S3, construct a hybrid evaluation large model including an input layer, an analysis layer, a discussion layer and an aggregation layer; the input layer includes three first large language models for inputting student data and prompt words; the analysis layer includes three second large language models for invoking the large model knowledge base according to the prompt words to perform multi-dimensional evaluation on the student data to obtain preliminary evaluation results of each dimension; the discussion layer includes three third large language models for comparing and integrating the preliminary evaluation results to obtain a number of optimized evaluation results; the aggregation layer includes a fourth large language model for outputting comprehensive quality evaluation results according to the optimized evaluation results; Step S4: obtaining the student data to be evaluated, and preprocessing the student data to be evaluated; Step S5: input the preprocessed student data to be evaluated into the hybrid evaluation model and output the comprehensive quality evaluation results.

2. A method for evaluating comprehensive quality of students based on a hybrid large model as claimed in claim 1, characterized in that: The step S1 is specifically as follows: Evaluation goals are set based on the six dimensions of ideology and morality, academic level, physical and mental health, artistic literacy, social practice, and labor literacy. Evaluation indicators are set for each evaluation goal, including at least indicator definition, scoring criteria, and weights. The evaluation indicators are organized according to a preset logical structure to construct an evaluation indicator system for evaluating students' comprehensive qualities.

3. The method for evaluating comprehensive quality of students based on a hybrid large model as claimed in claim 1, characterized in that: The step S2 is specifically as follows: Obtain a large amount of student historical data including at least ideological and moral data, academic level data, physical and mental health data, artistic literacy data, social practice data and labor literacy data, perform preprocessing on each of the student historical data including at least abnormal data detection, data cleaning and format conversion, and then map each of the student historical data to the evaluation index corresponding to the evaluation index system through RAG technology to construct a large model knowledge base for the analysis layer to call, and continuously update the large model knowledge base based on the latest student data; The ideological and moral data at least include activity project data, reward project data and violation of rules and regulations data; the academic level data at least include subject competition data, elective course data and test scores; the physical and mental health data at least include mental health data, sports scores and sports hobbies; the artistic literacy data at least include art teams, art activities, art hobbies and art scores; the social practice data at least include practical activities, research results, scientific and technological innovation and creative inventions; the labor literacy data at least include labor competition data, labor results, ability and technical certificates and practical projects.

4. The method for evaluating comprehensive quality of students based on a hybrid large model as claimed in claim 1, characterized in that: In step S3, the three first large-scale language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner; the three second large-scale language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner; the three third large-scale language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner.

5. The method for evaluating comprehensive quality of students based on a hybrid large model as claimed in claim 1, characterized in that: The step S4 is specifically as follows: Acquire the student data to be evaluated through a database, and perform preprocessing on the student data to be evaluated, including at least abnormal data detection, data cleaning, and format conversion; The step S5 is specifically as follows: The pre-processed student data to be evaluated is input into the hybrid evaluation model in real time, and the comprehensive quality evaluation results are output. The comprehensive quality evaluation results are displayed in the form of charts or graphs through a visual interface, and the comprehensive quality evaluation results are stored and backed up.

6. A student comprehensive quality evaluation system based on a hybrid large model, characterized by: Includes the following modules: An evaluation index system construction module is used to construct an evaluation index system for evaluating students' comprehensive qualities based on evaluation objectives; A large model knowledge base construction module is used to construct a large model knowledge base based on the evaluation index system through RAG technology; A hybrid evaluation large model construction module is used to construct a hybrid evaluation large model including an input layer, an analysis layer, a discussion layer and an aggregation layer; the input layer includes three first large language models for inputting student data and prompt words; the analysis layer includes three second large language models for invoking the large model knowledge base according to the prompt words to perform multi-dimensional evaluation on the student data to obtain preliminary evaluation results of each dimension; the discussion layer includes three third large language models for comparing and integrating the preliminary evaluation results to obtain a number of optimized evaluation results; the aggregation layer includes a fourth large language model for outputting comprehensive quality evaluation results according to the optimized evaluation results; A preprocessing module for student data to be evaluated, used to obtain the student data to be evaluated and preprocess the student data to be evaluated; The comprehensive quality evaluation module is used to input the pre-processed student data to be evaluated into the hybrid evaluation model and output the comprehensive quality evaluation results.

7. A student comprehensive quality evaluation system based on a hybrid large model as claimed in claim 6, characterized in that: The evaluation index system construction module is specifically used for: Evaluation goals are set based on the six dimensions of ideology and morality, academic level, physical and mental health, artistic literacy, social practice, and labor literacy. Evaluation indicators are set for each evaluation goal, including at least indicator definition, scoring criteria, and weights. The evaluation indicators are organized according to a preset logical structure to construct an evaluation indicator system for evaluating students' comprehensive qualities.

8. A student comprehensive quality evaluation system based on a hybrid large model as claimed in claim 6, characterized in that: The large model knowledge base construction module is specifically used for: Obtain a large amount of student historical data including at least ideological and moral data, academic level data, physical and mental health data, artistic literacy data, social practice data and labor literacy data, perform preprocessing on each of the student historical data including at least abnormal data detection, data cleaning and format conversion, and then map each of the student historical data to the evaluation index corresponding to the evaluation index system through RAG technology to construct a large model knowledge base for the analysis layer to call, and continuously update the large model knowledge base based on the latest student data; The ideological and moral data at least include activity project data, reward project data and violation of rules and regulations data; the academic level data at least include subject competition data, elective course data and test scores; the physical and mental health data at least include mental health data, sports scores and sports hobbies; the artistic literacy data at least include art teams, art activities, art hobbies and art scores; the social practice data at least include practical activities, research results, scientific and technological innovation and creative inventions; the labor literacy data at least include labor competition data, labor results, ability and technical certificates and practical projects.

9. A student comprehensive quality evaluation system based on a hybrid large model as claimed in claim 6, characterized in that: In the hybrid evaluation large model construction module, the three first large language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner; the three second large language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner; the three third large language models are connected in a serial collaboration, parallel fusion or hybrid collaboration manner.

10. A student comprehensive quality evaluation system based on a hybrid large model as claimed in claim 6, characterized in that: The student data preprocessing module to be evaluated is specifically used for: Acquire the student data to be evaluated through a database, and perform preprocessing on the student data to be evaluated, including at least abnormal data detection, data cleaning, and format conversion; The comprehensive quality evaluation module is specifically used for: The pre-processed student data to be evaluated is input into the hybrid evaluation model in real time, and the comprehensive quality evaluation results are output. The comprehensive quality evaluation results are displayed in the form of charts or graphs through a visual interface, and the comprehensive quality evaluation results are stored and backed up.

Images