CN-122022553-A - Education evaluation method and system supported by domestic base

CN122022553ACN 122022553 ACN122022553 ACN 122022553ACN-122022553-A

Abstract

The invention discloses an education evaluation method and system supported by a domestic base, and relates to the technical field of intelligent education evaluation, wherein the method monitors software and hardware resources such as a CPU (Central processing Unit), a GPU (graphics processing Unit), a memory, a disk, a database and the like in real time through a base monitoring agent component, calculates a domestic deployment adaptation index, and judges resource environment suitability according to a threshold value; the method comprises the steps of collecting course knowledge structure, teaching targets and student learning behavior data, constructing an education cognitive map, calculating semantic consistency coefficients for evaluating the matching degree of a student learning path and the course targets, collecting course evaluation operator execution logs and calculation power utilization data, calculating operator collaborative acceleration indexes, judging operator collaborative optimization effects, calculating comprehensive education evaluation indexes, judging the overall learning effect of students and optimizing. The method realizes the efficient adaptation and performance optimization of the education evaluation on the domestic platform, and improves the level of intellectualization, dynamics and interpretability of the education evaluation.

Inventors

Liang Zuohuan
ZOU YUERONG
LI JUNJIE
HE LEI
XU DONGSHENG
REN MINGYUE

Assignees

广东讯飞启明科技发展有限公司

Dates

Publication Date: 20260512
Application Date: 20260106

Claims (10)

1. The education evaluation method supported by the domestic base is characterized by comprising the following steps of: The method comprises the steps of firstly, monitoring a software layer and a hardware layer in real time through a base monitoring agent component in the running process of an education evaluation platform, collecting the use states and performance indexes of CPU, GPU, memory, disk and database resources, performing data processing, and calculating a domestic deployment adaptation index NMI; Step two, during the operation of the education evaluation platform, acquiring a course knowledge structure, a teaching target text, a student learning behavior log and evaluation result data, constructing an education cognition map and generating a semantic index mapping set; Step three, collecting execution logs of course evaluation operators, CPU and GPU occupancy rates and cache hit conditions in real time through an education model monitoring component, and calculating operator collaborative acceleration indexes OAI, comparing and analyzing with a domestic operator optimization standard threshold Oth, judging whether the course evaluation operators collaborative optimization meets the standard, and executing operator path recompilation, thread and cache optimization and GPU and CPU core remapping strategies when the course evaluation operators collaborative optimization does not meet the standard; And fourthly, calculating a comprehensive score df of the student through multidimensional scoring of the answering behaviors, the learning duration and the interactive operation of the student, carrying out statistics of sliding windows, outlier rejection and standardization processing, calculating a comprehensive education evaluation index EEI, comparing with a threshold Eth, judging whether the overall learning effect of the student meets the standard, and executing operator optimization, task scheduling adjustment, learning path and content recommendation optimization strategies if the overall learning effect of the student does not meet the standard.
2. The method for educational evaluation of a domestic base support according to claim 1, wherein the first step comprises: S11, during the running of the education evaluation platform, the running states of a software layer and a hardware layer are acquired in real time through a base monitoring agent component, wherein a monitoring object comprises an evaluation server, an edge node and a CPU (central processing unit), a GPU (graphics processing unit), a memory and database service component in a domestic computing cluster, and monitoring content comprises a processor utilization rate, a thread switching frequency, a GPU core load, a video memory bandwidth occupancy rate, a disk delay and a database query response time; S111, acquiring core use condition, main frequency fluctuation, cache access delay and thread switching condition of a CPU in real time through a system API interface and a hardware performance counter PMU provided by a domestic operating system, and generating comprehensive CPU utilization Ccpu by adopting sliding window statistics and averaging; S112, collecting the core occupancy rate, the thread block execution efficiency and the video memory bandwidth occupancy condition of the GPU through a GPU driving interface, and performing time synchronization processing on collected data to generate comprehensive GPU utilization Cgpu; S113, acquiring the response time and throughput of disk read-write operation through a domestic file system interface, adopting a sliding window averaging and outlier filtering method to process, and generating file delay Lio; s114, acquiring average response time and hit rate of index query through a system-level timing function or query performance monitoring interface provided by the domestic database, and carrying out sample homogenization and outlier rejection processing on the acquired results to obtain database index response time Tdb.
3. The method for educational evaluation of a domestic base support according to claim 1, wherein said step one further comprises: s12, calculating and obtaining a domestic deployment adaptation index NMI after dimensionless processing through the obtained comprehensive CPU utilization Ccpu, the comprehensive GPU utilization Cgpu, the file delay Lio and the database index response time Tdb; s13, comparing and analyzing the homemade deployment adaptation index NMI with the homemade adaptation standard threshold Nth by presetting the homemade adaptation standard threshold Nth, and acquiring a first evaluation result comprises the following steps: When the localization deployment adaptation index NMI is more than or equal to the localization adaptation standard threshold Nth, judging that the localization resource environment does not reach the standard, automatically generating an adaptation confirmation record, and continuously monitoring; When the domestic deployment adaptation index NMI is smaller than the domestic adaptation standard threshold Nth, judging that the domestic resource environment is not up to standard, when the current domestic education evaluation platform has risks of excessively high evaluation algorithm running delay, insufficient CPU and GPU calculation power utilization or unbalanced load and low database index query efficiency in hardware and system resource adaptation, triggering a first early warning instruction to generate a first strategy, carrying out resource binding optimization, binding a key task with a high-performance CPU and a GPU core, avoiding resource competition, carrying out thread scheduling adjustment, optimizing thread priority, core affinity and task scheduling strategy, improving parallel processing efficiency, carrying out database index optimization, adjusting an index structure, parallel query strategy and buffer allocation, improving query throughput, and after adjustment, re-acquiring and updating the domestic deployment adaptation index NMI until the domestic deployment adaptation index NMI is larger than or equal to the domestic adaptation standard threshold Nth.
4. The method for educational evaluation of a domestic base support according to claim 1, wherein the second step comprises: S21, acquiring a course knowledge structure, a teaching target text, a student learning behavior log and evaluation result data in real time through an education data acquisition component during the operation of an education evaluation platform based on the standardized domestic resource environment, wherein the education evaluation platform is suitable for being deployed in a domestic database and an operating system environment; S22, constructing an education cognitive map based on a course target Node set Gt and a student knowledge Node set Gs, a student behavior vector set Vs and a target semantic vector Vt, modeling semantic relations between knowledge nodes and behavior nodes by adopting a domestic database graph storage engine and a graph structure analysis algorithm comprising a graph convolutional neural network GCN and a Node embedding algorithm Node2 Vec; S23, based on the generated semantic node set G and the node attribute vector set A, a localization database map index construction method is adopted to establish a mapping relation between semantic nodes and attribute vectors in a dream database, clustering indexing is carried out on the node attribute vectors through a hash division bucket and a vector space division algorithm, index weight dynamic updating is carried out by combining a sliding time window and a behavior change detection algorithm, and semantic similarity results are sequenced and de-duplicated by utilizing a localization embedded vector retrieval library to obtain a semantic index mapping set.
5. The method for educational evaluation of a domestic base support according to claim 1, wherein the second step further comprises: s24, based on a semantic index mapping set, extracting a corresponding course target node set Gt, a student knowledge node set Gs, a student behavior vector set Vs and a target semantic vector Vt through index retrieval, and calculating to obtain a semantic consistency coefficient SCI after dimensionless processing; s25, comparing and analyzing the semantic consistency coefficient SCI with the domestic semantic evaluation standard threshold Sth by presetting the domestic semantic evaluation standard threshold Sth, and acquiring a second evaluation result comprises the following steps: when the semantic consistency coefficient SCI is more than or equal to the domestic semantic evaluation standard threshold Sth, judging that the matching degree of the student learning path and the course target is qualified, generating a semantic evaluation confirmation record, continuously monitoring the student learning state, and entering a step III; When the semantic consistency coefficient SCI is smaller than the domestic semantic evaluation standard threshold Sth, judging that the matching degree of a student learning path and a course target is unqualified, triggering a second early warning instruction to generate a second strategy when the current education evaluation has risks of unbalanced knowledge mastering, learning path deviation or important knowledge point omission, carrying out personalized recommendation, recommending corresponding knowledge points, exercises and learning resources according to weak items and interests of students, carrying out knowledge node reordering, adjusting node priority in the learning path, advancing or increasing review frequency of key nodes, and updating the semantic consistency coefficient SCI until the domestic semantic evaluation standard threshold Sth is reached after execution.
6. The method for educational evaluation of a domestic base support according to claim 1, wherein the third step comprises: S31, operator execution logs, CPU and GPU occupancy rates and execution time sequence data in course evaluation tasks are acquired in real time through an education model monitoring component, acquisition objects comprise forward computation operators, key matrix computation operators and statistical analysis operators, and each operator execution time Th, CPU utilization rate Ucpu, GPU utilization rate Ugpu, and hit times Ehit and total times Etotal of each operator in a CPU and GPU cache.
7. The method for educational evaluation of a domestic base support according to claim 1, wherein the third step further comprises: s32, acquiring execution time Th, CPU utilization rate Ucpu, GPU utilization rate Ugpu of each operator, and hit number Ehit and total number Etotal of each operator in a CPU and GPU cache, and calculating to acquire an operator collaborative acceleration index OAI after dimensionless processing; S33, comparing and analyzing the operator collaborative acceleration index OAI with the localization operator optimization standard threshold Oth by presetting the localization operator optimization standard threshold Oth, and obtaining a third evaluation result comprises: When the operator collaborative acceleration index OAI is more than or equal to the localization operator optimization standard threshold Oth, judging that course evaluation operators reach the standard in collaborative optimization, generating operator optimization confirmation records, continuously monitoring operator execution performance, and entering a step four; When the operator collaborative acceleration index OAI is smaller than the domestic operator optimization standard threshold Oth, judging that course evaluation operator collaborative optimization does not reach the standard, triggering a third early warning instruction to generate a third strategy, namely executing operator path recompilation, automatically optimizing operator binding and execution sequence, adjusting key operator thread allocation and memory storage caching strategy, improving operator execution efficiency, remapping GPU and CPU core to realize operator collaborative acceleration, and recalculate the operator collaborative acceleration index OAI until the operator collaborative acceleration index is larger than or equal to the domestic operator optimization standard threshold Oth, wherein the operator execution delay is high, the GPU and CPU utilization is uneven or the risk of the key operator not hitting the optimal execution path exists in the current education evaluation.
8. The method for educational evaluation of a domestic base support according to claim 1, wherein the fourth step comprises: S41, based on operator optimization confirmation records, performing multi-dimensional scoring on student answer behaviors, learning duration, interaction operation and exercise completion conditions by using the optimized course evaluation operator, and comprehensively mapping the grasping degree of each knowledge node, learning path completion degree and operator execution optimization results into scores df of students on each index by using a weighted accumulation method; S42, adopting sliding window statistics, outlier rejection and standardization processing to clean and normalize the scores df on each index so that the scores of all dimensions are comparable under unified dimension, and summarizing the scores of all dimensions by an arithmetic average method to obtain a student comprehensive score average value 。
9. The method for educational evaluation of a domestic base support according to claim 1, wherein the fourth step further comprises: s43, obtaining the average value of the student comprehensive scores The method comprises the steps of combining a domestic deployment adaptation index NMI, a semantic consistency coefficient SCI and an operator collaborative acceleration index OAI, performing dimensionless treatment, and calculating to obtain a comprehensive education evaluation index EEI; S44, comparing and analyzing the comprehensive education evaluation index EEI with the comprehensive education evaluation threshold Eth by presetting the comprehensive education evaluation threshold Eth, and obtaining a fourth evaluation result comprises: When the comprehensive education evaluation index EEI is more than or equal to the comprehensive education evaluation threshold Eth, judging that the overall learning effect of the student reaches the standard, generating an education evaluation confirmation record, and outputting an individual education evaluation report of the student and a system performance report; When the comprehensive education evaluation index EEI is smaller than the comprehensive education evaluation threshold value Eth, judging that the overall learning effect of the student is not up to standard, triggering a fourth early warning instruction when the current education evaluation has risks of unbalance knowledge mastering and learning path deviation of the student, generating a fourth strategy, performing task scheduling and operator optimization, adjusting course evaluation operator execution sequence, batch processing size and parallel depth, improving operator cooperative efficiency and computing resource utilization rate, performing learning path and content recommendation optimization, re-adjusting key knowledge node priority in combination with semantic evaluation results, performing personalized learning resource pushing and path optimization, performing data and index dynamic update, and dynamically updating semantic index mapping and node weight based on latest behavior data and course adjustment, so that timeliness and accuracy of evaluation analysis results are achieved.
10. The education evaluation system supported by a localization base, which is applied to the education evaluation method supported by the localization base as claimed in any one of claims 1 to 9, is characterized by comprising: The resource monitoring adaptation module is used for monitoring a software layer and a hardware layer in real time during the running of the education and evaluation platform through the base monitoring agent component, collecting the use states and performance indexes of CPU, GPU, memory, disk and database resources, performing data processing, and calculating a localization deployment adaptation index NMI; The education and cognition semantic analysis module is used for collecting a course knowledge structure, a teaching target text, a student learning behavior log and evaluation result data during the operation of the education and evaluation platform, constructing an education and cognition map and generating a semantic index mapping set, calculating a semantic consistency coefficient SCI and comparing and analyzing the semantic consistency coefficient SCI with a domestic semantic evaluation standard threshold Sth, judging whether the matching degree of a student learning path and the course target is qualified or not, and triggering a personalized recommendation and knowledge node reordering strategy when the student learning path and the course target are unqualified; The operator monitoring collaborative optimization module is used for collecting the execution log of the course evaluation operator, the occupancy rate of the CPU and the GPU and the cache hit condition in real time through the education model monitoring component, calculating an operator collaborative acceleration index OAI, comparing and analyzing with a domestic operator optimization standard threshold Oth, judging whether the course evaluation operator collaborative optimization meets the standard, and executing operator path recompilation, thread and cache optimization and GPU and CPU core remapping strategies when the course evaluation operator collaborative optimization does not meet the standard; The score evaluation optimization module is used for calculating a student comprehensive score df through multi-dimensional scoring of student answer behaviors, learning duration and interactive operation, sliding window statistics, outlier rejection and standardization processing, calculating a comprehensive education evaluation index EEI, comparing the comprehensive education evaluation index EEI with a threshold Eth, judging whether the overall learning effect of the student meets the standard, and executing operator optimization, task scheduling adjustment, learning path and content recommendation optimization strategies if the overall learning effect of the student does not meet the standard.

Description

Education evaluation method and system supported by domestic base Technical Field The invention relates to the technical field of intelligent education evaluation, in particular to a domestic base supported education evaluation method and system. Background With the development of information technology, the education field is gradually developed to the direction of intelligence, digitization and individualization. Various education evaluation platforms and online learning systems begin to introduce artificial intelligence technology to analyze and evaluate learning behaviors, knowledge mastering degree and class performances of students in real time. However, the existing education evaluation system still has the following technical problems in practical application: The adaptability of domestic resources is insufficient, most of the conventional education and evaluation platforms depend on foreign general computing resources and operating systems, and the conventional education and evaluation platforms lack effective adaptation mechanisms for CPU and GPU computing power utilization, memory scheduling, database index efficiency and the like of domestic servers, edge nodes and computing clusters, so that computing power resources cannot be fully utilized, the execution delay of evaluation algorithms is high, and the performance of the overall system is unstable. The correlation analysis capability of the student behaviors and the course targets is limited, the current system collects the student learning behaviors more in simple log records or click stream analysis, lacks deep semantic understanding of the course knowledge structure and the teaching targets, lacks quantitative analysis methods for learning paths and knowledge nodes of the students and matching degree of the course targets, and causes that evaluation results cannot accurately reflect the learning effects and personalized requirements of the students. The operator execution optimization and cooperative efficiency is insufficient, and in educational evaluation, a large number of operators are involved in a deep learning model and course evaluation task, and comprise a forward calculation operator, a matrix calculation operator and a statistical analysis operator. The existing system lacks real-time monitoring and optimizing strategies for the execution time of operators, the occupancy rate of a CPU and a GPU and the buffer hit condition, and has low operator cooperation efficiency, so that key task delay and unbalanced resource allocation are caused, and the accuracy of student evaluation results and the stability of the system are affected. The comprehensive analysis capability of the education evaluation index is limited, the prior method usually only depends on a single index to evaluate the learning effect of students, such as knowledge point mastering rate or answering accuracy, lacks a multi-dimensional index fusion and quantitative analysis means, and simultaneously lacks a closed-loop mechanism for incorporating operator execution efficiency, semantic consistency and system resource suitability into the comprehensive evaluation index, so that comprehensive, accurate and sustainable optimization of the education evaluation result cannot be realized. The system self-adaptive evolution and the multi-scene deployment capability are insufficient, a large number of manual configurations are needed when the scene is expanded or upgraded by most of the existing education evaluation platforms, and the modularization and containerization deployment capability based on the domestic rapid development base is lacking, so that the system is low in efficiency of adapting to different education scenes such as language learning, classroom evaluation, examination training and the like, and rapid updating and continuous optimization cannot be realized. Disclosure of Invention Aiming at the defects of the prior art, the invention provides an education evaluating method and system supported by a domestic base, which are used for solving the problems in the background art. In order to achieve the purpose, the invention is realized by the following technical scheme that the education evaluating method supported by the domestic base comprises the following steps: The method comprises the steps of firstly, monitoring a software layer and a hardware layer in real time through a base monitoring agent component in the running process of an education evaluation platform, collecting the use states and performance indexes of CPU, GPU, memory, disk and database resources, performing data processing, and calculating a domestic deployment adaptation index NMI; Step two, during the operation of the education evaluation platform, acquiring a course knowledge structure, a teaching target text, a student learning behavior log and evaluation result data, constructing an education cognition map and generating a semantic index mapping set; Step three, collecting execution log