An intelligent automatic examination system for university courses

Abstract

This invention relates to the field of computer technology combining artificial intelligence and educational measurement. It provides an intelligent automated examination system for university courses, comprising an automated examination center unit, a hardware and network construction unit, a hardware and software operation coordination unit, a software interface unit, an anti-cheating unit, a question bank unit, a marking unit, and an intelligent question bank management and maintenance unit. The system achieves student information binding through two-factor authentication; it ensures fair and randomized question paper generation by controlling the overall difficulty of the exam paper with a target average difficulty and tolerance; and it generates question bank management results using point-to-two correlation coefficients and two-column correlation coefficients as core statistical indicators, combined with multi-dimensional indicators, and constructs a closed-loop update mechanism. This invention enables full traceability and auditability of the examination process, effectively solving industry pain points and is suitable for the construction of automated examination centers in universities and daily course assessments.

Description

Technical Field

[0001] This invention belongs to the field of computer technology that combines artificial intelligence and educational measurement, and in particular relates to an intelligent automatic examination system for university courses. Background Technology

[0002] In the existing technology, the construction of automated examination centers in universities has formed a mature engineering foundation in individual aspects such as identity verification, computer room infrastructure, examination management, question bank construction, artificial intelligence evaluation, and intelligent operation and maintenance. Fingerprint / facial recognition devices can realize human-machine-seat binding, structured question banks can complete basic question sampling and automatic scoring, domestic large language models can assist in subjective question scoring, and the computer room's layered network and redundant power supply can provide basic hardware support. Each individual technical module can independently realize its corresponding function, providing the basic conditions for the intelligentization of the entire examination process.

[0003] However, existing technologies still face five core problems in practical applications of university course examinations. These problems are intertwined, leading to low examination organization efficiency, poor fairness, and high labor costs: First, the fairness of exam papers lacks a stable control mechanism. Most courses use fixed papers or weakly constrained random question selection, resulting in significant differences in the difficulty distribution of different students' papers, seriously affecting the fairness of grade evaluation. Second, the organization of make-up exams and re-examinations is costly and time-consuming. After a student fails the exam, the school needs to arrange a make-up exam and re-set the questions, consuming a large amount of time and labor costs. Third, teachers' workload is concentrated and their work content is difficult to standardize. Subjective question marking becomes a bottleneck at the end of the semester, and the effect of technology in reducing workload is limited. Fourth, the operation and management of the examination center highly relies on human experience, leading to frequent malfunctions during peak examination periods, affecting student experience and increasing disputes. Fifth, the question bank lacks a comprehensive intelligent governance system. There are no automated means for question quality assessment and optimization, making it difficult to support the long-term need for fair paper generation. Furthermore, a discrimination evaluation system based on point-to-two correlation coefficients and two-column correlation coefficients has not been established, and there is a lack of quantitative standards and engineering implementation paths for controlling difficulty drift, exposure, and similarity clusters.

[0004] Therefore, an intelligent university course automatic examination system is needed to solve the above problems, realize deep hardware and software collaboration across the entire examination process, precise control of the fairness of test paper generation, intelligent management of the question bank, and data-driven operation and maintenance of the examination center. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent university course automatic examination system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] An intelligent university course automated examination system includes an automated examination center unit, a hardware and network construction unit, a hardware and software operation coordination unit, a software interface unit, an anti-cheating unit, a question bank unit, a marking unit, and an intelligent question bank management and maintenance unit. The hardware and software operation coordination unit is communicatively connected to the automated examination center unit and the anti-cheating unit, performing two-factor authentication (fingerprint and facial recognition) and binding the student's identity with seat number, course number, and exam paper configuration. The anti-cheating unit provides auxiliary verification during the two-factor authentication process. The question bank unit is communicatively connected to the hardware and software operation coordination unit and the anti-cheating unit, storing a question bank constructed according to the course and maintaining the difficulty coefficient and question type attributes of the questions. The question bank unit includes a fairness-constrained random question generation module and a question bank storage module. The fairness-constrained random question generation module is based on a target average difficulty level. With tolerance Control the overall difficulty of each test paper satisfy The anti-cheating unit controls the cheating process during the test paper generation process. The intelligent question bank management and operation and maintenance unit communicates with the question bank unit, hardware and network construction unit, and realizes intelligent management of the question bank and intelligent operation and maintenance of the examination center through the indicator calculation module and the operation and maintenance management sub-module. It completes the test question quality assessment with point-to-two correlation coefficient and two-column correlation coefficient as core statistical indicators. The marking unit communicates with the software and hardware operation coordination unit and the question bank unit to complete the fully automatic scoring of objective questions and intelligent auxiliary scoring of subjective questions. It realizes deterministic verification and consistency arbitration for calculation questions, programming questions and graph questions. Each unit realizes data synchronization and visualization through the software interface unit. The system adopts a three-layer deployment form of examination center platform-examination room node-terminal to realize full-process data traceability and operation auditability.

[0008] A further technical solution is that the fair constraint random test paper generation module includes a question selection submodule, a difficulty calculation submodule, and a correction submodule. Through a closed-loop process of stratified question selection, difficulty calculation, and iterative correction, it ensures that the difficulty of each test paper meets the requirements. It also meets the constraints of question type quotas, chapter quotas, answering time, and similar clusters; the overall difficulty of the exam paper... Calculated using the weighted summation formula: ,in For the first Question type weighting For the first The difficulty level of the question This structure represents the total number of questions in the exam paper. It effectively addresses the issue of significant differences in the difficulty of exam papers among different candidates, ensuring fairness in the paper compilation process and providing a reliable basis for horizontal comparison of scores.

[0009] A further technical solution is provided, in which the anti-cheating unit includes a question order shuffling submodule and a similar cluster suppression submodule. The similar cluster verification avoids the concentration of similar questions in the same test paper, and the question order shuffling reduces the risk of plagiarism. The similar cluster suppression submodule establishes similar cluster identifiers for test questions, sets that test questions with a similarity of ≥80% are grouped into the same similar cluster, and the number of test questions in the same similar cluster in a single test paper does not exceed a preset limit. This structure does not require highly intrusive monitoring and achieves multi-dimensional anti-cheating without affecting the test experience, thereby enhancing the credibility of the test.

[0010] A further technical solution is that the question bank unit supports eleven major question types: multiple choice, true / false, fill-in-the-blank, definition, short answer, essay, calculation, graph / chart, material-based, and discussion questions. The question bank unit assigns weights to different question types, and these weights are used to calculate the overall difficulty of the exam paper. In addition, the proportion of each question type in the exam paper is controlled. All question type data and metadata are stored in the question bank storage module and synchronized to the anti-cheating unit and the intelligent question bank governance and operation and maintenance unit. The question metadata includes at least the unique identifier of the question, difficulty coefficient, point-to-two correlation coefficient / two-column correlation coefficient, similar cluster identifier, exposure, version number, chapter / ability tag, and estimated answering time. The design of multi-question type coverage and weight configuration is adapted to the assessment needs of multiple disciplines in arts, sciences and engineering, and ensures that the distribution of question types in the exam paper meets the course assessment requirements.

[0011] A further technical solution involves the indicator calculation module using a rolling sample window to estimate the point-to-bivariate correlation coefficient and the bivariate correlation coefficient, and calculating a 95% confidence interval for the correlation coefficient. For objective questions, the point-to-bivariate correlation coefficient is used to measure the differentiation between high-scoring and low-scoring candidates. For subjective questions, the bivariate correlation coefficient is used to measure the correlation between a candidate's achievement / failure and their total score. When the correlation coefficient is lower than a preset threshold or its lower confidence bound is lower than a preset threshold, the corresponding question is automatically marked as a review or replacement object, and a governance work order including an explanation of the cause is generated. The rolling sample window is configured by default to data from the most recent 1000 candidates / 3 exams. New questions are fixed to use the answer data of 200 candidates for the initial indicator calculation. Through dynamic monitoring of statistical indicators, the automatic identification and accurate positioning of substandard questions are achieved, providing data support for optimizing the question bank quality.

[0012] A further technical solution involves a health score calculation module that communicates with the indicator calculation module, question bank storage module, and software interface unit. It calculates the question health score based on a weighted average of point-to-two correlation coefficient, two-column correlation coefficient, difficulty coefficient fit, difficulty drift, reasonableness of answering time, and anomaly rate. The maximum question health score is 100 points. The default weighting configuration is: point-to-two correlation coefficient 40%, difficulty coefficient fit 20%, difficulty drift 15%, reasonableness of answering time 10%, anomaly rate 10%, and exposure 5%. The weighting can be customized by teachers according to course / subject. The intelligent question bank management and maintenance unit generates a question quality ranking list and a commendation list based on the question health score. The review and replacement lists are defined by default as follows: questions with a health score <60 need to be replaced, questions with a health score between 60 and 70 need to be reviewed, and questions with a health score ≥80 are considered high-quality questions. The software interface unit displays the lists to teachers in an operable manner and provides entry points for new question entry and version release. After new questions are entered, they are directly stored in the question bank storage module and synchronized to the fair constraint random test paper generation module and the anti-cheating unit. A closed-loop governance system of "indicator calculation - health score assessment - work order processing - gray-scale verification - version update" is constructed. After new questions are entered, they enter a small-scale gray-scale verification stage. The default question sampling ratio for test paper generation is 5%. After verification (health score ≥70), they are included in the official question bank, promoting the continuous evolution of the question bank quality as the examination progresses.

[0013] A further technical solution involves a question bank storage module that includes a similarity and exposure control module. This module is communicatively connected to an anti-cheating unit and a fair constraint random question generation module. It establishes similarity cluster identifiers for questions and sets an upper limit for the occurrence of similar clusters within the same question paper. High-exposure questions are weighted less, while low-exposure or new questions are weighted more. Exposure is quantified by the frequency of a question's use in the selection process. Questions reaching the preset exposure limit are temporarily removed from the question generation pool. Simultaneously, the question similarity cluster identifiers and exposure data are synchronized to the anti-cheating unit and the fair constraint random question generation module. This dynamic adjustment of similarity cluster control and exposure avoids assessment bias caused by question duplication and reduces the memory effect of high-frequency questions, ensuring the vitality of the question bank.

[0014] A further technical solution involves the operation and maintenance governance submodule communicating with the hardware and network construction unit and the software interface unit. The collected examination center operation and maintenance indicators include at least terminal online rate, peripheral availability, network latency / packet loss, identity verification success rate, exam paper distribution success rate, submission success rate, and submission queue length. The operation and maintenance governance submodule maps these indicators to exam room health scores and platform health scores according to preset weights. The health score is out of 100 points, and the weights are dynamically adjusted according to the examination stage. When the health score falls below a threshold, a three-level warning and handling strategy is triggered: Level 1 warning (80-90 points) implements a rate-limiting strategy; Level 2 warning (60-80 points) implements a batch submission and service degradation strategy; and Level 3 warning (<60 points) implements a fault node isolation and work order dispatch strategy. The handling process is written to the audit log and synchronized to the software interface unit. This achieves a shift from experience-driven to data-driven operation and maintenance, proactively addressing fault risks and reducing exam disputes and review workload.

[0015] A further technical solution involves a subjective question scoring module that is communicatively connected to a question bank storage module, a deterministic verification module, and a software interface unit. The standard answer is broken down into multiple weighted requirement points, which are stored in the question bank storage module. Based on a domestic large language model, the module determines the requirement point coverage of the examinee's answer and outputs the coverage rate, missing points, and explanatory text. A coverage rate threshold and a model confidence threshold are set; if either threshold is not met, a manual review or arbitration process is automatically triggered, and the review results are synchronized to the software interface unit and the question bank storage module. This combination of standardized scoring logic and manual review mechanism ensures both the efficiency of subjective question scoring and improves the fairness and interpretability of the scoring results.

[0016] A further technical solution involves a deterministic verification module that communicates with the subjective question scoring module, the question bank storage module, and the software interface unit. This module provides customized deterministic verification mechanisms for calculation, programming, and graph / chart questions. For programming questions, unit testing is performed in a restricted sandbox environment to verify the code execution results, logical correctness, and functional integrity. For calculation questions, symbolic computation or numerical verification of key intermediate quantities and final conclusions is used, with a numerical error tolerance of ≤5%. For graph / chart questions, structured extraction of coordinate axes, units, key points, and trends is performed and compared with the required points, which are retrieved from the question bank storage module. When the difference between the deterministic verification result and the model's suggested score exceeds a threshold, a review is automatically triggered, and the review data is synchronized to the software interface unit. Customized verification for high-difficulty question types compensates for the uncertainty of model scoring and minimizes scoring errors.

[0017] Compared with the prior art, the beneficial effects of the present invention are:

[0018] This invention, through fair constraints on the stratified question selection, difficulty calculation, and iterative correction process of the random test paper generation module, combined with... The core constraint, through The system accurately calculates the overall difficulty of the test paper, ensuring that each test paper has the same level of difficulty. At the same time, the anti-cheating unit suppresses similar clusters and breaks up the question order, ensuring the fairness of the performance evaluation from the source of test paper assembly. This solves the problem of the lack of stable control over the fairness of test papers in existing technologies.

[0019] This invention, relying on a standardized question bank and a fair test paper generation mechanism, supports candidates to take the test at any time. The marking unit completes intelligent scoring and generates scores in real time, eliminating the need for schools to set questions and arrange invigilation, reducing the pressure on teachers to set questions repeatedly, and significantly reducing the time and manpower costs of make-up exams. It solves the problems of high organization costs and long cycles for make-up exams.

[0020] This invention enables fully automated scoring without human intervention for objective questions, while subjective questions are scored intelligently using a large language model and deterministic verification. Only a small number of abnormal questions require manual review. At the same time, the question bank is automatically generated and the scores are automatically archived, standardizing and lightening the workload of teachers and effectively reducing their technical burden. This invention solves the problem of concentrated and difficult-to-standardize teacher workload.

[0021] This invention involves hardware and network construction units collecting comprehensive operation and maintenance indicators, which are then mapped to health scores by the operation and maintenance governance submodule. Early warning and response strategies are automatically triggered according to a three-level standard, generating traceable audit logs. This enables a shift in operation and maintenance from experience-driven to data-driven, allowing for proactive fault handling, improved student experience, reduced exam disputes and workload for score review, and solving the problem of exam center operation and maintenance relying on manual experience.

[0022] This invention uses point-to-two columns and two-column correlation coefficients as its core, and combines multiple indicators such as difficulty coefficient fit and difficulty drift to calculate the health score of questions. It automatically generates governance work orders and quality lists. The similarity and exposure control module realizes the quantitative management and control of similar clusters and exposure. The teacher end provides a standardized question bank maintenance entry, forming a complete intelligent decision chain for the question bank. This enables the question bank quality to continuously evolve with the operation of the exam, and solves the problem of the lack of full-dimensional intelligent governance of the question bank.

[0023] This invention adopts a three-layer deployment model of examination center platform - examination room node - terminal, clarifies the core data structure, interface agreement and algorithm key points, realizes the engineering implementation of deep software and hardware collaboration, and ensures that all data is traceable and operations are auditable, adapting to the needs of large-scale examination scenarios in multiple campuses of universities.

[0024] To more clearly illustrate the structural features and effects of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0025] Figure 1This is a schematic diagram of the overall architecture of the present invention, clearly showing the communication connection relationship of the eight core units: the automated examination center unit, the hardware and network construction unit, the software and hardware operation coordination unit, the software interface unit, the anti-cheating unit, the question bank unit, the marking unit, and the question bank management and intelligent operation and maintenance unit, and clarifying the core functional positioning and data interaction logic of each unit.

[0026] Figure 2 This is a schematic diagram of the anti-cheating unit of the present invention, which shows in detail the question order shuffling submodule, the similar cluster suppression submodule, and the relationship with the question extraction submodule and the question bank storage module included in the anti-cheating unit, intuitively demonstrating the core anti-cheating logic of similar cluster suppression and question order shuffling;

[0027] Figure 3 This is a structural diagram of the question bank unit of the present invention, showing the hierarchical architecture of the question selection submodule, difficulty calculation submodule, and correction submodule of the fair constraint random test paper generation module in the question bank unit, as well as the question bank storage module and the similarity and exposure control module, clarifying the linkage mechanism between the test paper generation process and similarity and exposure control.

[0028] Figure 4 This is a schematic diagram of the marking unit of the present invention, showing the communication relationship between the objective question judging module, subjective question scoring module, deterministic verification module, question bank storage module, and software interface unit of the marking unit, and clearly demonstrating the collaborative logic of tiered scoring and deterministic verification;

[0029] Figure 5 This is a structural diagram of the intelligent question bank governance and operation and maintenance unit of the present invention, showing the question bank governance module (including indicator calculation module and health score calculation module), operation and maintenance governance module (including operation and maintenance governance sub-module), and its relationship with the software interface unit, question bank storage module, hardware and network construction unit, reflecting the closed-loop process of intelligent question bank governance and intelligent operation and maintenance.

[0030] Figure 6 This is a schematic diagram of the three-layer deployment of the examination center of the present invention, which presents the three-layer architecture of the examination center platform (on-campus data center / private cloud), examination room nodes (examination room gateway / local service), and candidate terminal group (30-60 units), and clarifies the deployment logic of the core components, data flow direction and supporting equipment (camera / fingerprint collector) at each layer;

[0031] Figure 7 This is a functional zoning diagram of the automated examination center of this invention, divided into five major functional areas: entrance verification area, waiting area, examination answering area, invigilation and dispatching area, and equipment maintenance area. The core equipment, network links (network cable → access switch → aggregation switch → examination room node) and power circuits (UPS emergency power supply coverage) of each area are marked, reflecting the standardized design of the examination physical environment.

[0032] Figure 8 This is a schematic diagram of the hardware and network construction unit architecture of the present invention, showing the layered network architecture of access layer-aggregation layer-core layer, clarifying the hierarchical relationship of the examination center platform (question bank system, marking system, governance system, audit system), core switch, aggregation switch, examination room node, terminal group, as well as the fault isolation, rate limiting, and work order dispatch function logic of the intelligent operation and maintenance engine.

[0033] Figure 9 This is a schematic diagram of the examination business state machine and event flow pipeline of the present invention, showing the full state transition of the examination business, such as Draft, AutoSave, and Submitted, as well as the terminal submission, deduplication verification, scoring service, and result write-back process of the event flow pipeline, and marking the core technical features such as idempotent key (req_id) and replay protection;

[0034] Figure 10 This is a schematic diagram of the teacher-side question bank management visualization dashboard of the present invention. It displays four core dashboards: question bank quality ranking, TOP10 questions to be replaced, difficulty drift warning, exposure and similar cluster crowding. It presents key indicators such as point-to-two correlation coefficient, two-column correlation coefficient, and discrimination, as well as operation entry points such as generating work orders and entering new questions.

[0035] Figure 11 This invention presents an intelligent closed-loop flowchart for question bank management, clearly showing the entire closed-loop process of "question entry → small-scale gray-scale testing → indicator calculation → question health score → generating list and work order → teacher review / rewrite / replace → version release", and clearly defining the core indicators and operational logic of each link;

[0036] Figure 12 This invention presents an intelligent closed-loop operation and maintenance flowchart, illustrating the closed-loop operation and maintenance logic of "indicator collection (terminal / network / business indicators) → health score calculation + anomaly detection → execution of handling strategies (rate limiting / batch submission / degradation / isolation / work order) → record archiving → strategy optimization", reflecting the intelligent operation and maintenance of the entire process from data collection to strategy optimization.

[0037] Figure 13 The flowchart of the fair constraint random test paper generation mechanism of this invention details the entire test paper generation process: “Input (question bank + quota + μ / τ) → stratified random sampling → calculate the overall difficulty S of the test paper → determine S−μ≤τ → local exchange correction (iteration upper limit: MAX) → constraint check (similar cluster / duration / exposure) → output test paper”, and clarifies the implementation path of difficulty equivalence constraint.

[0038] In the diagram: 1. Automated Examination Center Unit; 2. Hardware and Network Construction Unit; 3. Hardware and Software Operation Coordination Unit; 4. Software Interface Unit; 5. Anti-cheating Unit; 51. Question Sequence Dispersal Submodule; 52. Similar Cluster Suppression Submodule; 6. Question Bank Unit; 61. Fairness Constraint Random Paper Generation Module; 611. Question Selection Submodule; 612. Difficulty Calculation Submodule; 613. Correction Submodule; 62. Question Bank Storage Module; 621. Similarity and Exposure Control Module; 7. Marking Unit; 71. Objective Question Judging Submodule; 72. Subjective Question Scoring Module; 73. Deterministic Verification Module; 8. Question Bank Management and Intelligent Operation and Maintenance Unit; 81. Question Bank Management Module; 811. Indicator Calculation Module; 812. Health Score Calculation Module; 82. Operation and Maintenance Management Module; 821. Operation and Maintenance Management Submodule. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0040] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0041] like Figure 1-5 As shown, this embodiment of the invention provides an intelligent university course automatic examination system, including an automated examination center unit 1, a hardware and network construction unit 2, a hardware and software operation coordination unit 3, a software interface unit 4, an anti-cheating unit 5, a question bank unit 6, a marking unit 7, and an intelligent question bank management and maintenance unit 8. The hardware and software operation coordination unit 3 performs two-factor identity verification using fingerprint recognition and facial recognition and completes multi-dimensional information binding. The question bank unit 6 has a built-in fair constraint random question generation mechanism to achieve equal difficulty levels in the exam papers. The intelligent question bank management and maintenance unit 8 completes full-dimensional management of the question bank and intelligent operation and maintenance of the examination center. The marking unit 7 implements... The system features fully automated scoring for general questions and intelligent assisted scoring for subjective questions. Data interaction and functional linkage between units are achieved through a preset encrypted communication protocol. The system adopts a three-tier deployment model: examination center platform - examination room nodes - terminals. The examination center platform is deployed in the school's data center or private cloud and is responsible for core businesses such as question bank management, test paper generation scheduling, and scoring governance. The examination room nodes are deployed in each standardized computer room and are responsible for identity verification, device access, and data aggregation. The terminals are the devices for candidates to answer questions, realizing interface rendering, local caching, and uploading of answer data. This deployment model balances the efficiency of centralized management with the fault tolerance of local failures and is suitable for multi-campus and large-scale examination scenarios in universities.

[0042] Specifically, the hardware and software operation coordination unit 3 is communicatively connected to the automated examination center unit 1 and the anti-cheating unit 5, respectively. It performs two-factor authentication (fingerprint and facial recognition) when candidates enter or log in, uniquely binding the candidate's identity to their seat number, course number, and exam paper configuration. The anti-cheating unit 5 assists in verifying the two-factor authentication process, identifying anomalies such as photo manipulation and fingerprint forgery through biometric liveness detection. The question bank unit 6 is communicatively connected to both the hardware and software operation coordination unit 3 and the anti-cheating unit 5, storing a question bank built according to courses and maintaining the difficulty level and question type attributes of the questions. The question bank unit 6 includes a fair constraint random question generation module, based on the target average difficulty... With tolerance Control the overall difficulty of each test paper satisfy The anti-cheating unit 5 implements anti-cheating controls throughout the test paper generation process. A two-factor identity verification and multi-information binding mechanism eliminates proxy testing and mis-testing at the source, ensuring the accuracy of the testing environment for candidates. The fair constraint random test paper generation mechanism achieves equal difficulty levels based on random question selection, solving the problem of significant difficulty differences in traditional test paper generation models. This provides a fair and reliable basis for horizontal comparison of scores and teaching evaluation. Furthermore, the full-process control by the anti-cheating unit further enhances the standardization of the test paper generation process.

[0043] Specifically, the fair constraint random question generation module selects a set of candidate questions under the constraints of question type quotas and chapter quotas. The candidate questions are then weighted and randomly selected from the set into three levels of difficulty (easy, medium, and difficult) to form a draft test paper. The difficulty calculation submodule then uses a formula... Calculate the overall difficulty of the first draft exam paper. ,when At that time, perform iterative correction by partially exchanging or replacing questions between adjacent difficulty levels, with an upper limit of 20 iterations, until the condition is met. Furthermore, it satisfies the constraints of answering time and similarity clusters. The closed-loop test paper generation process of question selection, calculation, and correction, combined with multi-dimensional constraints, ensures the coverage of question types and chapters in the test paper, accurately controls the overall difficulty of the test paper, and the iterative correction method ensures that the difficulty of each test paper is within the target range. At the same time, it avoids problems such as the concentration of similar questions and unreasonable answering time, thereby improving the scientific and reasonable nature of test paper generation.

[0044] Specifically, question bank unit 6 supports at least the following question types: multiple choice, true / false, fill-in-the-blank, definition, short answer, essay, calculation, graph / chart, material-based, and discussion questions. Weights are assigned to different question types, and these weights are used to calculate the overall difficulty of the exam. The system controls the proportion of each question type in the exam paper. All question type data and metadata are standardized, stored, and synchronized to the anti-cheating unit 5 and the intelligent question bank governance and operation and maintenance unit 8. Question metadata includes at least a unique question identifier, difficulty coefficient, point-to-two correlation coefficient / two-column correlation coefficient, similarity cluster identifier, exposure, version number, chapter / ability tag, and estimated answering time. The comprehensive coverage of eleven question types and personalized weighting configurations adapt to the assessment needs of multiple disciplines, including arts, sciences, and engineering, meeting the requirements for question type proportions and difficulty calculations for different courses. The standardized data storage and synchronization mechanism provides a unified and accurate data source for exam paper generation, anti-cheating, and question bank governance, reducing the cost of redundant question bank construction and improving the efficiency of system-wide collaboration.

[0045] Specifically, the intelligent question bank governance and maintenance unit 8 uses a rolling sample window to calculate the correlation coefficients of two columns for objective questions and two columns for subjective questions, and calculates a 95% confidence interval for the correlation coefficients. The rolling sample window is configured by default to use data from the most recent 1000 candidates / 3 exams, and new questions are fixed to use the answer data from 200 candidates. When the correlation coefficient is lower than a preset threshold or its lower confidence bound is lower than a preset threshold, the corresponding question is automatically marked as a review or replacement object, and a governance work order including an explanation of the reason is generated. The relevant information of the work order is synchronized to the question bank storage module and the software interface unit 4. Based on the correlation coefficient calculation method using the rolling sample window, the discrimination of the questions can be dynamically and accurately reflected. The evaluation of the confidence interval makes the judgment of question quality more scientific. The automatic marking of inferior questions and the generation of governance work orders replace the traditional manual screening mode, realize the dynamic monitoring and accurate positioning of question quality, significantly reduce the workload of teachers in maintaining the question bank, and provide data support for the optimization of question bank quality.

[0046] Specifically, the intelligent question bank governance and operation unit 8 calculates a question health score based on a weighted average of point-to-two correlation coefficient, two-column correlation coefficient, difficulty coefficient fit, difficulty drift, reasonableness of answering time, and anomaly rate. The question health score has a maximum of 100 points, and the default weighting configuration is: point-to-two correlation coefficient 40%, difficulty coefficient fit 20%, difficulty drift 15%, reasonableness of answering time 10%, anomaly rate 10%, and exposure 5%. Based on this, it generates a question quality ranking list, a list of commendations, a list of reviews, and a list of replacements. The question health score is categorized by health score: <60 points require replacement, 60-70 points require review, and ≥80 points are considered high-quality. The software interface unit 4 displays this list to teachers in an interactive manner and provides entry points for new question input and version release. New questions are directly stored in the question bank storage module and synchronized to the fair and randomized test paper generation module and anti-cheating unit 5. After entry, new questions enter a small-scale gray-scale verification phase, with a default question sampling ratio of 5%. Once verified (health score ≥70 points), they are included in the official question bank. The multi-dimensional weighted calculation of question health scores comprehensively assesses question quality. The generation of various lists makes teachers' question bank maintenance work more targeted. The standardized entry points for new question input and version release construct a closed-loop question bank governance system encompassing "indicator calculation - health score assessment - work order processing - gray-scale verification - version update," promoting continuous evolution of question bank quality with exam operation and providing stable support for long-term fair test paper generation.

[0047] Specifically, question bank unit 6 includes a similarity and exposure control module, used to establish similarity cluster identifiers for test questions and set an upper limit for the occurrence of similar clusters within the same test paper. Test questions with a similarity of ≥80% are grouped into the same similarity cluster. High-exposure questions are weighted less, while low-exposure or new questions are weighted more. Exposure is quantified by the frequency of a question being used in the selection process. Questions that reach the preset exposure limit are temporarily removed from the test paper pool. The question similarity cluster identifiers and exposure data are synchronized to anti-cheating unit 5 and the fairness-constrained random test paper generation module. The setting of similarity cluster identifiers and the upper limit for the occurrence of the same question within the same test paper effectively suppresses the concentrated appearance of variations of the same question in a single test paper, ensuring the breadth of the test paper's assessment. The dynamic weight adjustment of exposure reduces the memory effect of high-frequency questions and increases the assessment opportunities for low-exposure and new questions, thus avoiding assessment bias caused by question duplication and ensuring the vitality of the question bank and the effectiveness of the assessment.

[0048] Specifically, the intelligent question bank governance and operation and maintenance unit 8 performs intelligent governance of the examination center's operation and maintenance indicators. These indicators include at least terminal online rate, peripheral device availability, network latency / packet loss, identity verification success rate, paper distribution success rate, submission success rate, and submission queue length. The intelligent question bank governance and operation and maintenance unit 8 maps these operation and maintenance indicators to examination room health score and platform health score according to preset weights. The health score is out of 100 points, and the weights are dynamically adjusted according to the examination stage. When the health score is lower than the threshold, a three-level warning and handling strategy is triggered: Level 1 warning (80-90 points) implements a flow restriction strategy; Level 2 warning (60-80 points) implements a batch submission and service degradation strategy; and Level 3 warning (<60 points) implements a fault node isolation and work order dispatch strategy. The handling process is written to the audit log and synchronized to the software interface unit 4. The collection of comprehensive operation and maintenance indicators and the mapping of health scores make the operational status of the examination center quantifiable and visualized. The hierarchical early warning and automated handling strategies realize the transformation of operation and maintenance from experience-driven to data-driven. It can detect and handle fault risks in advance, reduce on-site failures during peak examination periods, improve the stability of system operation and the examination experience of candidates. At the same time, the recording of audit logs makes the operation and maintenance handling process traceable, reducing examination disputes and review workload.

[0049] Specifically, the scoring assistance for subjective questions in marking unit 7 includes: breaking down the standard answer into multiple weighted requirement points and storing them in question bank unit 6; using a domestic large language model to determine the requirement point coverage of the candidate's answer and outputting the coverage rate, missing points, and explanatory text; setting coverage rate thresholds and model confidence thresholds; automatically triggering manual review or arbitration when either threshold is not met, with the review results synchronized to software interface unit 4 and question bank unit 6. The scoring logic of requirement point breakdown and coverage determination makes subjective question scoring more standardized and objective; the application of the domestic large language model significantly improves the efficiency of subjective question scoring; the explanatory text provides a clear basis for the scoring results; and the dual-threshold triggered manual review mechanism compensates for the uncertainty of intelligent scoring, ensuring the fairness and interpretability of the scoring results while improving scoring efficiency.

[0050] Specifically, the marking unit 7 provides a deterministic verification mechanism for calculation, programming, and graph questions: For programming questions, unit testing is performed in a restricted sandbox environment to verify the code's execution results, logical correctness, and functional completeness; for calculation questions, symbolic computation or numerical verification of key intermediate quantities and final conclusions is used, with a numerical error tolerance of ≤5%; for graph questions, structured extraction of coordinate axes, units, key points, and trends is performed and compared with the required points, which are retrieved from the question bank storage module 62; when the difference between the deterministic verification result and the model's suggested score exceeds a threshold, a review is automatically triggered, and the review data is synchronized to the software interface unit 4. This customized deterministic verification mechanism for different question types accurately compensates for the uncertainty in scoring caused by large language models. Targeted verification of programming, calculation, and graph questions effectively identifies errors in answers, significantly reducing the scoring error rate. The linkage of the review mechanism further ensures the accuracy and auditability of the scoring results, significantly improving the reliability of intelligent marking.

[0051] Working principle and usage process of this invention:

[0052] This system is supported by core technologies of artificial intelligence, educational measurement, and statistical quality control. It adopts a three-tier deployment architecture of examination center platform – examination room node – terminal. Eight functional units work collaboratively and interconnectedly to form a closed-loop management system for the entire process, achieving intelligent management across the entire chain of university course examinations, from pre-exam preparation to post-exam question bank evolution and intelligent operation and maintenance. The core of the system revolves around the equivalence constraint of exam paper difficulty. To achieve fairness in test paper compilation, the system utilizes point-to-two / two-column correlation coefficients as the core for intelligent question bank governance, and data-driven intelligent operation and maintenance of the examination center. Ultimately, this transforms examination organization from being manually led to being intelligently led with manual review, addressing core pain points of traditional examinations such as insufficient fairness, high organizational costs, low efficiency, reliance on manual operation and maintenance, and lack of comprehensive question bank governance. The system adheres to six design principles: fairness, efficiency, reliability, evolution, traceability and auditability, and multi-disciplinary adaptability. With the hardware and software operation coordination unit 3 as the core hub, it connects the business processes and data interactions of the other seven units, ensuring clear rules, data interoperability, traceability, and auditability at every stage of the entire process.

[0053] The Automated Examination Center Unit 1, serving as the dedicated physical implementation platform for offline examinations, is divided into five functional areas: an entrance verification area, a waiting and security check area, an examination answering area, an invigilation and dispatching area, and an equipment maintenance area. It provides a standardized physical space for conducting examinations. This area facilitates on-site biometric information (fingerprint, facial recognition) collection for candidate identity verification, the establishment and management of the physical environment for examination answering, and synchronizes real-time examination room status data to the hardware and software operation coordination unit 3 and the software interface unit 4, enabling visualized monitoring of the examination room status. The terminal seat numbers in the examination answering area are traceably linked to network ports, switch partitions, and power circuits, facilitating accurate fault location during subsequent maintenance. The equipment maintenance area is equipped with backup terminals and a rapid replacement process, while the invigilation and dispatching area features a large monitoring screen and management terminals to support rapid handling of examination room malfunctions. Working in conjunction with the hardware and software operation coordination unit 3, it manages on-site candidate identity verification, guiding candidates with verification errors to the manual service desk for secondary verification, ensuring the standardization and accuracy of the identity verification process.

[0054] Hardware and network construction unit 2 adopts a layered network architecture of access layer-aggregation layer-core layer, dividing VLANs or security domains according to examination rooms to achieve fault and broadcast isolation between different examination rooms; equipped with UPS redundant power supply and dual power supply design for key servers, providing underlying network, power, and hardware equipment support for all system units to ensure stable operation at the physical level of the system; configured with dedicated links or policy routing for examination room nodes and examination center platform to ensure stable bandwidth for core services such as scoring tasks, question bank access, and log feedback during peak examination periods, avoiding network congestion; real-time data acquisition terminal The system collects comprehensive operational metrics, including online rate, network latency / packet loss, peripheral device availability, identity verification success rate, exam paper distribution success rate, submission success rate, and submission queue length. The collection frequency is dynamically adjusted according to the exam stage (5 seconds / time during peak periods and 30 seconds / time during regular periods). The metric data is transmitted in real time to the question bank governance and intelligent operation and maintenance unit 8 to provide data support for intelligent operation and maintenance. The system also performs comprehensive verification of hardware and network, such as network debugging and UPS redundant power supply testing. Verification anomalies are immediately generated into operation and maintenance work orders and pushed to the management terminal of software interface unit 4 for operation and maintenance personnel to handle within a limited time.

[0055] The hardware and software operation coordination unit 3, as the core business hub of the system, establishes communication connections with the other seven units, coordinating the functional linkage and data flow of each unit. It is the core carrier for the entire examination process's business connection, data interoperability, and traceability. It performs two-factor authentication using fingerprint recognition and facial recognition, with the anti-cheating unit 5 providing auxiliary verification of the authentication process. After successful authentication, a unique binding token is generated, binding the candidate's identity with their seat number, course number, exam paper configuration, and terminal device, eliminating proxy exams and mis-exams from the source. It triggers a paper generation request to the question bank unit 6, receives the corrected exam paper, and distributes it to the candidate using a hybrid RSA+AES encryption method. The system generates a terminal; it simultaneously handles the relay of answer data, the return of scoring results, and the standardized encrypted archiving of scores, returning all exam data to the question bank governance module 81; it adopts an event-driven mechanism to decouple answer submission from scoring, writing the candidate's submission event into the scoring queue and immediately returning submission success information with a unique receipt ID, avoiding platform overload caused by concentrated submissions; it generates idempotent keys and replay protection rules for each submission to prevent duplicate scoring caused by network fluctuations or repeated clicks; it assigns a dedicated exam session to each candidate and unlocks the corresponding terminal, and if a candidate experiences terminal failure, the answer progress and local cached data can be accurately restored through two-factor authentication, ensuring exam continuity.

[0056] Software interface unit 4 serves as the unified human-computer interaction entry point for the system, providing dedicated operation interfaces for candidates, teachers, teaching management departments, and maintenance personnel based on their roles. This achieves standardized operation and data visualization, simultaneously displaying the monitoring results of anti-cheating unit 5 and the operational data of each unit. For candidates: a low-interference, standardized answering interface is provided, displaying the remaining exam time, answering progress, network status, and automatically saving records in real time. Dedicated answering tools (calculator, code editor, formula input component) are provided for calculation and programming questions, with strict binding between tools and question types. Draft area data is only cached locally and does not participate in uploading or scoring. For teachers: a question bank management list, a question quality ranking, and praise are provided. The system includes a review / replacement list and an entry point for subjective question review; it provides a standardized entry point for new question entry and version release, with new question entry information synchronized in real time to question bank unit 6 and anti-cheating unit 5; it supports viewing statistical data such as class / course grade distribution and accuracy rate of each question; the teaching management / operation and maintenance end displays the status of the examination center, audit reports, and operation and maintenance alarm information, and supports filtering and querying operation and maintenance indicators and examination room / platform health scores by examination room / course / time dimensions; it receives governance work orders and operation and maintenance work orders, synchronizes the handling process and results, and supports one-click synchronization of grades to the school's academic affairs system; it receives alarm information, handling records, and audit logs from each unit, realizing the visualized display and operable management of the entire examination process data.

[0057] The anti-cheating unit 5 includes a question order shuffling submodule 51 and a similar cluster suppression submodule 52. This module ensures exam fairness and participates in all stages, from identity verification and test paper assembly to answering. It employs a non-intrusive monitoring mode, balancing anti-cheating effectiveness with the test-taker experience. It assists in verifying the two-factor identity verification process, identifying abnormal biometric features such as photo copying and fingerprint forgery, thus improving the accuracy of identity verification. The similar cluster suppression submodule 52 retrieves the question similarity cluster identifier data from the question bank storage module 62 to perform real-time verification of the test paper assembly and question selection process, preventing the concentrated appearance of similar questions in the same test paper. If the verification fails, it immediately triggers a re-selection of questions. It also enables the re-examination of the initial test paper and the re-examination of the original test paper. Similar clusters of exam papers are isolated; Question order shuffling submodule 51: After the exam paper assembly and correction are completed, the question order of all question types is randomly shuffled to reduce the risk of plagiarism and cheating from the source of exam paper assembly; Lightweight non-intrusive real-time monitoring is implemented during the answering stage to identify extreme behaviors such as excessive screen switching frequency (≥3 screen switching within 1 minute), abnormal answering speed, and abnormal external terminal devices; Minor anomalies are only recorded in the background, while severe anomalies generate alarms immediately and push them to the invigilator's terminal in software interface unit 4 for manual verification by invigilators; Monitoring data and anomaly alarms are synchronized to software and hardware operation coordination unit 3 and software interface unit 4, and all anti-cheating related records are included in the full-process audit log of the exam.

[0058] Question bank unit 6 includes a fair and constrained random test paper generation module 61 and a question bank storage module 62. The question bank storage module 62 contains a similarity and exposure control module 621, which is the core asset module of the system. It provides test question data sources and metadata support for test paper generation, marking, and anti-cheating. It builds and stores question banks according to courses. It supports eleven major question types: multiple choice, true / false, fill-in-the-blank, definition, short answer, essay, calculation, graph / chart, material-based, and discussion questions. It sets weights for different question types and is adapted to assessments in multiple disciplines, including arts, sciences, and engineering. Requirements; weights are used to control the proportion of each question type in the test paper and serve as the core parameter for calculating the overall difficulty of the test paper; maintain metadata such as the difficulty coefficient of the test questions (expert calibration + pre-test dual calibration), similar cluster identifiers, exposure, version number, chapter / ability tags, etc. All question type data and metadata are stored in a standardized manner and synchronized to the anti-cheating unit 5 and the intelligent question bank governance and operation and maintenance unit 8; the fair constraint random test paper generation module 61 includes a question selection submodule 611, a difficulty calculation submodule 612, and a correction submodule 613, to... As the core constraint, a closed-loop process of question selection, calculation, and correction is used to generate the test paper; Question selection submodule 611: Under the constraints of question type quota and chapter quota, candidate question sets are selected, divided into three levels of difficulty (easy, medium, and difficult), and weighted random selection is performed to form the first draft test paper; Difficulty calculation submodule 612: The weighted summation formula is used to calculate the difficulty of the test paper. (in For the first Question weighting For the first (Difficulty level) Calculate the overall difficulty of the first draft exam paper. ; Correction submodule 613: If Perform iterative correction by locally exchanging / replacing questions between adjacent difficulty levels (up to 20 iterations) until the condition is met. And it meets the answering time and similar cluster constraints; if the question re-selection fails twice in a row, an abnormal test paper work order is generated and pushed to the management terminal; similarity and exposure control module 621: establishes similar cluster identifiers for test questions (similarity ≥80% is the same cluster) and sets the upper limit of similar clusters appearing in the same test paper; implements deweighting for high-exposure test questions and increases weighting for low-exposure / new questions to reduce the memory effect of high-frequency test questions; at the same time, the similar cluster identifiers and exposure data are synchronized to the anti-cheating unit 5 and the fair constraint random test paper generation module 61; generate test papers for retake candidates that meet the requirements. The exam paper is designed with difficulty constraints and implements isolation of similar clusters between the make-up exam paper and the initial exam paper, as well as avoidance of high-profile questions, to ensure the fairness of the make-up exam.

[0059] The marking unit 7, comprising an objective question judging module 71, a subjective question scoring module 72, and a deterministic verification module 73, is the core module of the system's scoring. It communicates with the hardware and software operation coordination unit 3 and the question bank unit 6. It employs a tiered scoring model: fully automatic objective question scoring + intelligent assisted subjective question scoring + deterministic verification for specific question types + dual-threshold manual review. It outputs interpretable and traceable scoring criteria. The objective question judging module 71 directly compares the examinee's answers with the standard answers in the question bank storage module 62, completing fully automatic scoring without human intervention. The scoring results are written to the examinee's score database in real time and synchronized to the software interface unit 4. The subjective question scoring module 72 retrieves weighted subjective question requirement points from the question bank storage module 62, performs semantic analysis and requirement point coverage determination on the examinee's answers based on a domestic large language model, accurately calculates the coverage rate, identifies missing points, and generates standardized interpretable scoring criteria. When the coverage rate is lower than a preset threshold or the model confidence level is lowered, the scoring module will automatically correct the score. When the score is below a preset threshold, a manual review or arbitration process is automatically triggered; Deterministic verification module 73: provides a customized deterministic verification mechanism for calculation questions, programming questions, and graph questions to compensate for the uncertainty of model scoring; when the difference between the verification result and the model's suggested score exceeds a threshold, a review is automatically triggered; Programming questions: execute unit tests in a restricted sandbox environment to verify the code's execution results, logical correctness, and functional integrity; Calculation questions: use symbolic computation or numerical verification (numerical error allowable range ≤5%) of key intermediate quantities and final conclusions to accurately identify calculation errors; Graph questions: extract structured data of coordinate axes, units, key points, and trends, and compare them one by one with the required points to determine the accuracy of the answer; All review results are synchronized to the hardware and software operation coordination unit 3, the question bank unit 6, and the software interface unit 4, and the review records generate audit logs, which are incorporated into the entire examination process traceability system; Subjective question requirement data and verification standards are stored in the question bank storage module 62.

[0060] The intelligent question bank governance and operation and maintenance unit 8 includes a question bank governance module 81 and an operation and maintenance governance module 82. The question bank governance module 81 contains an indicator calculation module 811 and a health score calculation module 812. The operation and maintenance governance module 82 contains an operation and maintenance governance sub-module 821, which is the core module for system optimization, realizing intelligent governance of the question bank across all dimensions and intelligent operation and maintenance of the examination center from experience-driven to data-driven. The question bank governance module 81 – indicator calculation module 811: uses a rolling sample window (defaulting to the most recent 1000 candidates / 3 exams, with 200 candidates for new questions) to calculate the two-column correlation coefficient for objective questions and the two-column correlation coefficient for subjective questions, and calculates a 95% confidence interval for the correlation coefficient; when the correlation coefficient is lower than a preset threshold or... When the confidence lower bound is below a preset threshold, the corresponding question is automatically marked as a review / replacement object, a management work order including the reason explanation is generated, and synchronized to the question bank storage module 62 and the software interface unit 4; Question bank management module 81 - health score calculation module 812: calculates the question health score (full score 100 points) based on point-two / two-column correlation coefficient, difficulty coefficient fit, difficulty drift, answer time reasonableness, abnormality rate, and exposure weight. The indicator weights can be customized by teachers according to courses / subjects (default discrimination 40%, difficulty fit 20%, difficulty drift 15%, answer time 10%, abnormality rate 10%, exposure 5%); health score <60 points requires replacement, 60-70 points requires review, ≥80 points For high-quality questions; based on the question health score, a question quality ranking, commendation list, review list, and replacement list are generated and synchronized to the teacher's end in software interface unit 4; a closed-loop management system is constructed, consisting of indicator calculation, health score evaluation, work order processing, gray-scale verification, and version update: after teachers complete the test question processing and new question entry, the new questions enter a small-volume gray-scale verification (default test paper sampling ratio of 5%). If the verification is successful (health score ≥ 70 points), it is included in the official question bank; if it fails, it is returned to the teacher's end for modification; Operation and maintenance management submodule 821: collects the full-dimensional operation and maintenance indicators of hardware and network construction unit 2, maps the indicators to the test room health score and platform health score (maximum score 100 points) according to preset weights (dynamically adjusted according to the examination stage), and displays them in the software interface. Unit 4 provides real-time visualization; when the health score falls below the threshold, a three-tiered warning and handling strategy is automatically triggered, and all handling processes are written to the audit log and synchronized to Unit 4 in the software interface; Level 1 warning (80-90 points): minor anomaly, implementing a rate limiting strategy to restrict non-core requests; Level 2 warning (60-80 points): moderate anomaly, implementing a batch submission and service degradation strategy to suspend non-core services; Level 3 warning (<60 points): severe anomaly, implementing a fault node isolation and work order dispatch strategy to prevent the fault from spreading; after the exam, the exam data, question bank governance data, operation and maintenance data, and audit logs are compiled and permanently encrypted and archived to provide a scientific basis for subsequent exam organization optimization, system parameter adjustment, and question bank construction;Support the academic affairs management department in completing the entire post-exam review process, and review the overall situation of the examination center, the results of handling anomalies, and statistics on the fairness of the test paper compilation. Data such as value distribution.

[0061] System's overall operational logic:

[0062] Pre-exam preparation: Complete the fine-grained configuration of the question bank (question metadata, subjective question requirement point calibration), personalized settings of exam parameters (test paper generation, threshold, operation and maintenance parameters), and full verification of hardware and network; establish RSA+AES hybrid encrypted communication connection for each unit, complete the initialization of similar cluster identification and exposure weight for questions, generate operation and maintenance work orders for verification anomalies and handle them within a time limit, and the system enters the ready state;

[0063] Identity verification: Candidates complete two-factor identity verification and anti-cheating unit 5 auxiliary verification. After successful verification, multiple pieces of information are uniquely bound and an examination session is assigned. Verification failures guide candidates to a second manual verification. All records are generated into an audit log and permanently archived.

[0064] Intelligent test paper generation: The hardware and software operation coordination unit 3 triggers the test paper generation request, and the question bank unit 6 generates a test paper that meets the requirements through stratified question extraction, difficulty calculation, and iterative correction. The test paper; Anti-cheating unit 5 completes similar cluster suppression and question order shuffling, the test paper is encrypted and sent to the candidate's terminal, and the entire test paper assembly process generates an audit log;

[0065] Exam answering: Candidates answer on a standardized interface. The system provides dual encryption protection by caching answer data locally and synchronizing it with the cloud every 30 seconds. Anti-cheating unit 5 monitors answering behavior, and hardware and network units dynamically collect operation and maintenance indicators. It provides full-dimensional fault tolerance for network jitter and terminal / peripheral failures to ensure the continuous conduct of the exam.

[0066] Submission of Scoring: The answer data is encrypted with AES and submitted asynchronously to the scoring queue. The marking unit 7 completes the scoring according to the tiered scoring mode. When the double threshold is triggered or the difference between the verification results exceeds the threshold, manual review is pushed. The scoring results and the basis are encrypted and archived synchronously.

[0067] Results Archiving: Results and related data (answer details, scoring basis, review records) are archived in a standardized and encrypted manner, and personalized visual query services are provided according to roles; unqualified candidates can take a make-up exam at any time in a preset window. The make-up exam paper meets the difficulty constraints and is isolated from the initial exam paper by similar clusters, and high-exposure questions are avoided.

[0068] Question bank management: All exam data is fed back to the question bank management module 81 to complete indicator calculations, health score assessments, and generate management lists and work orders; teachers complete the processing of test questions and the entry of new questions, and the question bank version is updated after the new questions are verified in a gray-scale manner, thus promoting the continuous evolution of the question bank quality;

[0069] Intelligent Operation and Maintenance and Post-Exam Review: The Operation and Maintenance Governance Submodule 821 completes the collection of operation and maintenance indicators, health score mapping and hierarchical early warning and handling, and records and archives all handling processes; after the exam, it completes the full-process data review, and all data is permanently encrypted and archived to provide a basis for subsequent exam organization optimization.

[0070] Core unit collaboration logic:

[0071] The hardware and software operation coordination unit 3 serves as the core hub, connecting the business processes of the other seven units. It enables the synchronization, transfer, and feedback of data throughout the entire process, including identity verification, test paper generation, scoring, and grade archiving, ensuring process traceability. The question bank unit 6, as the core asset module, provides test question data sources and metadata support for intelligent test paper generation, grading, and anti-cheating, forming the foundation for achieving fairness in test paper generation and continuous evolution of the question bank. The software interface unit 4, as the human-computer interaction entry point, provides dedicated operation interfaces for each role, synchronously displaying the operational data, governance lists, and alarm information of each unit, achieving data visualization and standardized operation. The intelligent question bank governance and maintenance unit 8, as the core of system optimization, receives all examination data to achieve intelligent question bank governance, receives maintenance indicators from hardware and network units to achieve data-driven maintenance, and synchronizes various work orders generated to the software interface unit 4. The anti-cheating unit 5, as a fairness guarantee module, participates in the entire process of identity verification, test paper generation, and answering. It reduces the risk of cheating through similar cluster suppression, question order shuffling, and lightweight behavior monitoring. The monitoring results are synchronized to the core hub and software interface unit 4. The marking unit 7, as the core scoring module, receives answer data and retrieves standard answers and requirements from the question bank unit 6 to complete intelligent scoring and customized verification. Abnormal questions are pushed for manual review, and the scoring results are synchronized to the core hub. The automated examination center unit 1, as the physical implementation carrier, provides a dedicated examination space and on-site management, and synchronizes the examination room status data to the core hub and software interface unit 4. The hardware and network construction unit 2, as the underlying support module, provides network, power, and hardware support for all units, collects operation and maintenance indicators, and synchronizes them to the operation and maintenance management submodule 821 to ensure the stable operation of the system.

[0072] The circuits, electronic components, and modules involved are all existing technologies, which can be fully implemented by those skilled in the art, and need not be elaborated upon. The content protected by this invention does not involve any improvement to the software and methods.

[0073] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent university course automatic examination system, characterized in that: It includes an automated examination center unit (1), a hardware and network construction unit (2), a hardware and software operation coordination unit (3), a software interface unit (4), an anti-cheating unit (5), a question bank unit (6), a marking unit (7), and an intelligent question bank management and maintenance unit (8). The software and hardware operation coordination unit (3) is connected to the automated examination center unit (1) and the anti-cheating unit (5) respectively, performs two-factor identity verification of fingerprint recognition and face recognition, and binds the candidate's identity with seat number, course number and test paper configuration. The anti-cheating unit (5) performs auxiliary verification of the two-factor identity verification process. The question bank unit (6) is connected to the hardware and software operation coordination unit (3) and the anti-cheating unit (5) respectively. It stores the question bank constructed according to the course and maintains the difficulty coefficient and question type attributes of the test questions. The question bank unit (6) includes a fair constraint random test paper generation module (61) and a question bank storage module (62). The fair constraint random test paper generation module (61) is based on the target average difficulty With tolerance Control the overall difficulty of each test paper satisfy The anti-cheating unit (5) controls the cheating process during the test paper assembly process; The question bank governance and operation and maintenance intelligent unit (8) is connected to the question bank unit (6) and the hardware and network construction unit (2) respectively. The question bank governance and operation and maintenance intelligent unit (8) includes a question bank governance module (81) and an operation and maintenance governance module (82). The question bank governance module (81) includes an indicator calculation module (811) and a health score calculation module (812). The operation and maintenance governance module (82) includes an operation and maintenance governance sub-module (821). The indicator calculation module (811) calculates the two-column correlation coefficient for objective questions and the two-column correlation coefficient for subjective questions. The question bank governance results are generated by combining the difficulty coefficient, difficulty drift, exposure and similar clusters. The operation and maintenance governance sub-module (821) collects the operation and maintenance indicators of the hardware and network construction unit (2) and carries out intelligent governance. The marking unit (7) is connected to the hardware and software operation coordination unit (3) and the question bank unit (6) respectively. The marking unit (7) includes an objective question judging module (71), a subjective question scoring module (72) and a deterministic verification module (73). The objective question judging module (71) is used to complete the fully automatic scoring of objective questions, and to perform requirement point coverage judgment and scoring assistance for subjective questions based on the domestic big language model. The deterministic verification module (73) verifies specific question types and outputs explanatory scoring basis. The software interface unit (4) is connected to the automated examination center unit (1), the hardware and software operation coordination unit (3), the anti-cheating unit (5), the question bank unit (6), the marking unit (7), and the question bank management and operation and maintenance intelligent unit (8) respectively. It provides the answer interface, the question bank management list, the examination center status and audit report to the candidates, teachers and teaching management departments respectively, and simultaneously displays the monitoring results of the anti-cheating unit (5) and the operation data of each unit. The hardware and network construction unit (2) is connected to the automated examination center unit (1) and the question bank governance and operation and maintenance intelligent unit (8) respectively. It adopts a hierarchical network architecture of access layer-aggregation layer-core layer and redundant power supply design to provide underlying support for all units of the system, collect operation and maintenance indicators such as terminal online rate and network latency / packet loss and transmit them to the question bank governance and operation and maintenance intelligent unit (8). The automated examination center unit (1) is connected to the hardware and software operation coordination unit (3), the anti-cheating unit (5), the software interface unit (4), and the hardware and network construction unit (2) respectively. It is divided into five functional areas, including the entrance verification area and the waiting security check area. As the physical implementation carrier of the examination, it completes the on-site management of the candidate's identity verification and provides the answering environment, and synchronizes the examination room status data to the software interface unit (4). The anti-cheating unit (5) is also connected to the software interface unit (4) to monitor the candidate's answering process in real time and synchronize the monitoring data and abnormal alarms to the hardware and software operation coordination unit (3) and the software interface unit (4). The system adopts a three-tier deployment model of examination center platform - examination room node - terminal to achieve full-process data traceability and operation auditability.

2. The intelligent university course automatic examination system according to claim 1, characterized in that: The fair constraint random test paper generation module (61) includes a question selection submodule (611), a difficulty calculation submodule (612), and a correction submodule (613). The question selection submodule (611) is communicatively connected to the question bank storage module (62) and the anti-cheating unit (5). Under the constraints of question type quota and chapter quota, it selects a set of candidate questions from the question bank storage module (62), randomly selects questions according to difficulty level and weights to form a draft test paper, and receives the similarity cluster verification results from the anti-cheating unit (5). The difficulty calculation submodule (612) is communicatively connected to the question selection submodule (611) and the correction submodule (613), and calculates the difficulty using a formula. Calculate the overall difficulty of the first draft exam paper. ,in For the first Question weighting For the first Difficulty level of the question The total number of questions in the test paper is calculated, and the result is transmitted to the correction submodule (613); the correction submodule (613) is communicatively connected to the difficulty calculation submodule (612), the question selection submodule (611), and the question bank storage module (62). At that time, local swapping or replacement iterative correction is performed between adjacent difficulty levels, with an upper limit of 20 iterations, until the condition is met. And satisfying the answer time constraint and similar cluster constraint, the corrected test paper is sent to the candidate's terminal through the software and hardware operation coordination unit (3).

3. The intelligent university course automatic examination system according to claim 1, characterized in that: The anti-cheating unit (5) includes a question order shuffling submodule (51) and a similar cluster suppression submodule (52). The similar cluster suppression submodule (52) is connected to the question extraction submodule (611) and the question bank storage module (62) to retrieve the question similar cluster identification data to verify the question extraction process. Questions with a similarity of ≥80% are classified into the same similar cluster, and the number of questions in the same similar cluster in a single test paper does not exceed the preset upper limit to avoid the concentrated appearance of similar questions in the same test paper. The question order shuffling submodule (51) performs a question order random shuffling process on the test paper after the test paper is assembled.

4. The intelligent university course automatic examination system according to claim 1, characterized in that: The question bank unit (6) supports at least the following question types: multiple choice, true / false, fill-in-the-blank, definition, short answer, essay, calculation, graph / chart, material-based, and discussion questions. The question bank unit (6) assigns weights to different question types, and these weights are used to calculate the overall difficulty of the exam paper. In addition to controlling the proportion of each question type in the test paper, all question type data and metadata are stored in the question bank storage module (62) and synchronized to the anti-cheating unit (5) and the question bank governance and operation and maintenance intelligent unit (8); the question metadata includes at least the unique identifier of the question, the difficulty coefficient, the point-to-two correlation coefficient / two-column correlation coefficient, the similar cluster identifier, the exposure, the version number, the chapter / ability tag, and the expected answering time.

5. The intelligent university course automatic examination system according to claim 1, characterized in that: The indicator calculation module (811) is connected to the question bank storage module (62) and the health score calculation module (812). It uses a rolling sample window to estimate the point-to-two correlation coefficient and the two-column correlation coefficient, and calculates the 95% confidence interval for the correlation coefficient. The rolling sample window is configured by default to the data of the most recent 1000 candidates / 3 exams. New questions are fixed to the answer data of 200 candidates. When the correlation coefficient is lower than the preset threshold or its confidence lower bound is lower than the preset threshold, the question bank governance and operation and maintenance intelligent unit (8) automatically marks the corresponding test question as a review or replacement object, and generates a governance work order including the reason explanation. The work order related information is synchronized to the question bank storage module (62) and the software interface unit (4).

6. The intelligent university course automatic examination system according to claim 1, characterized in that: The health score calculation module (812) is connected to the indicator calculation module (811), the question bank storage module (62), and the software interface unit (4) for communication. It calculates the question health score based on a weighted average of point-to-two correlation coefficient, two-column correlation coefficient, difficulty coefficient fit, difficulty drift, reasonableness of answering time, and anomaly rate. The full score for the question health score is 100 points. The default weight configuration is: point-to-two / two-column correlation coefficient 40%, difficulty coefficient fit 20%, difficulty drift 15%, reasonableness of answering time 10%, anomaly rate 10%, and exposure 5%. The weights can be customized by teachers according to course / subject. The question bank governance and operation intelligent unit ( 8) Generate a question quality ranking list, a commendation list, a review list and a replacement list based on the question health score. By default, questions with a health score <60 are questions that need to be replaced, questions with a health score of 60-70 are questions that need to be reviewed, and questions with a health score ≥80 are high-quality questions. The software interface unit (4) displays the list to the teacher in an operable manner and provides an entry point for new question entry and version release. After the new question is entered, it is directly stored in the question bank storage module (62) and synchronized to the fair constraint random test paper generation module (61) and the anti-cheating unit (5). After the new question is entered, it enters the small-volume gray-scale verification stage. The default test paper generation sampling ratio is 5%. After verification (health score ≥70), it is included in the formal question bank.

7. The intelligent university course automatic examination system according to claim 1, characterized in that: The question bank storage module (62) includes a similarity and exposure control module (621). The similarity and exposure control module (621) is connected to the anti-cheating unit (5) and the fair constraint random test paper generation module (61). It establishes similar cluster identifiers for test questions and sets an upper limit for the occurrence of similar clusters in the same test paper. It also reduces the weight of high-exposure test questions and increases the weight of low-exposure or new questions. The exposure is quantified by the frequency of test questions being used in the selection. Test questions that reach the preset exposure limit are temporarily removed from the test paper generation pool. The test question similarity cluster identifiers and exposure data are synchronized to the anti-cheating unit (5) and the fair constraint random test paper generation module (61).

8. The intelligent university course automatic examination system according to claim 1, characterized in that: The operation and maintenance management submodule (821) is connected to the hardware and network construction unit (2) and the software interface unit (4) to collect the operation and maintenance indicators of the examination center, including at least the terminal online rate, peripheral availability, network latency / packet loss, identity verification success rate, paper distribution success rate, submission success rate and submission queue length. The operation and maintenance management submodule (821) maps the operation and maintenance indicators to the examination room health score and the platform health score according to the preset weight. The full score of the health score is 100 points. The weight is dynamically adjusted according to the examination stage. When the health score is lower than the threshold, the warning and handling strategy is triggered according to the three-level standard: the first-level warning (80-90 points) implements the flow restriction strategy, the second-level warning (60-80 points) implements the batch submission and service degradation strategy, and the third-level warning (<60 points) implements the fault node isolation and work order dispatch strategy. The handling process is written into the audit log and synchronized to the software interface unit (4).

9. The intelligent university course automatic examination system according to claim 1, characterized in that: The subjective question scoring module (72) is connected to the question bank storage module (62), the deterministic verification module (73), and the software interface unit (4) to split the standard answer into multiple weighted requirement points. The requirement point data is stored in the question bank storage module (62). Based on the domestic big language model, the requirement point coverage of the candidate's answer is judged and the coverage rate, missing points and explanation text are output. Set coverage threshold and model confidence threshold. If either threshold is not met, the manual review or arbitration process will be automatically triggered. The review results will be synchronized to the software interface unit (4) and the question bank storage module (62).

10. The intelligent university course automatic examination system according to claim 1, characterized in that: The deterministic verification module (73) is connected to the subjective question scoring module (72), the question bank storage module (62), and the software interface unit (4) to provide a customized deterministic verification mechanism for calculation questions, programming questions, and graph questions. For programming questions, unit testing is performed in a restricted sandbox environment to verify the code execution results, logical correctness, and functional integrity. For calculation questions, symbolic operation or numerical verification is used to verify key intermediate quantities and final conclusions, with the allowable range of numerical error ≤5%. For graph questions, the coordinate axes, units, key points, and data trends are extracted in a structured manner and compared with the required points one by one. The required point data is retrieved from the question bank storage module (62). When the difference between the deterministic verification result and the model suggested score exceeds the threshold, manual review is automatically triggered, and the review data is synchronized to the software interface unit (4).

Images