Personalized dynamic weight teaching evaluation model and parameter adjustment method thereof

Abstract

The invention discloses a personalized dynamic weight teaching evaluation model and a parameter adjustment method thereof, and relates to the technical field of teaching evaluation, and the model comprises a data acquisition and fusion module, a feature extraction module, a self-adaptive weight generation module and an evaluation model construction module. By constructing the personalized dynamic weight teaching evaluation model, teaching practice can be closely combined, evaluation key points can be adjusted in real time according to teaching stages and student learning states, evaluation results are more timely and accurate, more accurate teaching feedback is provided for teachers, the teachers are helped to adjust teaching strategies in time, and the teaching efficiency is improved. A multi-objective optimization parameter adjustment method is utilized, multiple important aspects of evaluation are comprehensively considered, the evaluation accuracy is improved, fairness and interpretability are guaranteed, the overall performance of a teaching evaluation model is improved through the comprehensive optimization method, the requirements of different teaching scenes and student groups are met, and the teaching evaluation model has good application prospects. And the improvement of education and teaching quality is facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of teaching evaluation, and in particular to a personalized dynamic weight teaching evaluation model and a parameter adjustment method thereof. Background Art

[0002] Teaching evaluation is an activity that makes value judgments on the teaching process and results based on the teaching objectives and serves the teaching decision-making. It is the process of judging the actual or potential value of teaching activities. Teaching evaluation is the process of studying the value of teachers' teaching and students' learning. Teaching evaluation generally includes the evaluation of teachers, students, teaching content, teaching methods and means, teaching environment, and teaching management factors in the teaching process, but it is mainly the evaluation of students' learning effects and the evaluation of teachers' teaching work process. Teaching evaluation is an important link in the educational process. It helps to improve the quality of teaching and promote students' learning and development.

[0003] The teaching evaluation model will be used in the process of teaching evaluation. The teaching evaluation model is an important tool in the field of education for measuring and evaluating teaching effectiveness and student learning outcomes. It collects and analyzes various teaching-related data, uses specific algorithms and indicator systems, and conducts quantitative evaluation of the teaching process and student performance. It provides a scientific basis for teachers to adjust their teaching strategies and students to improve their learning methods. It is of great significance to improving teaching quality, promoting educational equity and promoting the all-round development of students.

[0004] However, the existing teaching evaluation models still have certain defects when used. Most of the existing teaching evaluation models adopt a fixed weight mode, which cannot dynamically adjust the evaluation index weight according to the teaching stage, teaching objectives and students' real-time learning status, resulting in the evaluation results not accurately reflecting the changes in teaching focus and students' learning evolution. In addition, the existing teaching evaluation model parameter adjustment methods often only focus on a single goal, ignoring fairness and interpretability, and are difficult to meet the needs of diversified teaching practices. Therefore, it is necessary to propose a personalized dynamic weight teaching evaluation model and its parameter adjustment method to solve the problems in the existing technology. Summary of the invention

[0005] The purpose of the present invention is to make up for the shortcomings of the prior art and to provide a personalized dynamic weight teaching evaluation model and a parameter adjustment method thereof, which can realize the dynamics and personalization of teaching evaluation, improve the timeliness, accuracy, fairness and interpretability of evaluation results, and meet the diverse needs of teaching practice.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a personalized dynamic weight teaching evaluation model, the model includes: a data acquisition and fusion module, a feature extraction module, an adaptive weight generation module and an evaluation model construction module;

[0007] The data collection and fusion module collects multi-source heterogeneous data, including students' learning behavior data X b , learning ability data X a And sentiment attitude data X e , where the learning behavior data X b Including online learning time l , classroom interaction frequency f i , learning ability data X a Contains knowledge acquisition speed v k , Problem-solving ability assessment results p , emotional attitude data X e Including the learning interest change value Δi and the learning motivation intensity m, the collected data are fused to obtain the fused data X = [X b ,X a ,X e ];

[0008] The feature extraction module uses data mining and machine learning technology to extract key feature vectors F reflecting individual characteristics and learning status of students from the fused data X. The extraction formula is F=f(X), where f is the feature extraction function;

[0009] The adaptive weight generation module uses an adaptive weight generation algorithm to dynamically adjust the weight W of each evaluation index according to the teaching stage s, the teaching goal g, and the student's real-time learning state feature vector F. The teaching stage is divided into a course introduction period s1, a knowledge consolidation period s2, a capability improvement period s3, and a comprehensive application period s4. For different teaching stages, the calculation method of the weight W of each evaluation index is as follows:

[0010] In the course introduction period s1, the weight of the basic knowledge mastery indicator w1 is calculated as w1 = α1·f1(F,g), and the weights of other evaluation indicators w i Calculated by the corresponding function, where i>1;

[0011] In the knowledge consolidation period s2, the weight of the basic knowledge mastery index w1 is w1 = α2·f2(F,g), and the weight of the knowledge integration and application ability index w j Calculated by the corresponding function;

[0012] In the capability improvement period s3 and the comprehensive application period s4, the weight of each evaluation index is calculated by the specific function of the corresponding stage;

[0013] The evaluation model construction module constructs a hierarchical personalized teaching evaluation model based on the extracted feature vector F and the dynamically generated weight W. The bottom layer of the model is the evaluation index I, and the middle layer performs weighted summation of the indexes according to the weight W. The calculation formula is: Where S is the score of the middle layer, n is the number of evaluation indicators, and the top layer outputs the comprehensive evaluation result R, R = h(S), where h is the function that converts the middle layer score into the final evaluation result.

[0014] Furthermore, in the data acquisition and fusion module, a standardized processing method is used for different types of data to unify the data into the same numerical range. The processing formula is: Where X norm is the standardized data, and X is the original data.

[0015] Furthermore, in the feature extraction module, a convolutional neural network is used for feature extraction, and features of the fused data X are extracted through convolution layer and pooling layer operations. The convolution operation formula is: Where Y is the convolution output, W is the convolution kernel, b is the bias, k and l are the sizes of the convolution kernel.

[0016] Furthermore, in the adaptive weight generation module, the weight adjustment process takes into account the course difficulty coefficient d, and the course difficulty coefficient is introduced as a parameter in the weight calculation function of different teaching stages for adjustment. In the course introduction period s1, the calculation formula of the basic knowledge mastery index weight w1 is w1=α1·f1(F,g,d).

[0017] Furthermore, in the evaluation model building module, after the weighted summation of the intermediate layer, a regularization term is introduced to prevent the model from overfitting, and the intermediate layer score calculation formula becomes Where λ is the regularization coefficient.

[0018] A parameter adjustment method for a personalized dynamic weight teaching evaluation model, the method comprising the following steps:

[0019] For the personalized dynamic weight teaching evaluation model, set multi-objective optimization goals, including evaluation accuracy A, fairness E, and explainability I n , accuracy A is evaluated by model prediction result R p and students’ actual learning performance a The matching degree is measured by the calculation formula: Where m is the number of samples, and fairness E is evaluated by comparing the distribution of evaluation results of different student groups, using the variance σ 2 Measuring the discreteness of evaluation results of different groups, interpretability I n By defining the indicator contribution function c i To measure, where c i is the contribution of the i-th evaluation index;

[0020] The multi-objective evolutionary algorithm is used to globally optimize the model parameters. In the optimization process, the model parameters are regarded as individuals, and the optimal parameter combination that meets multiple objectives is continuously searched through selection, crossover and mutation genetic operations. In each iteration, the individuals are evaluated and screened according to the priority and weight of each objective, and the better parameter combination is retained.

[0021] Combining parameter adjustment with model structure optimization, the structure of the evaluation model was fine-tuned according to the actual teaching situation and data feedback. It was found that the discrimination degree of a certain evaluation indicator in different stages was not high. The model performance was changed by adjusting the position of the indicator in the model and the weight calculation method. The specific adjustment method was as follows: the evaluation indicator I j The discrimination D j When it is lower than the threshold θ, adjust its weight calculation function w j =β·f3(F,g).

[0022] Furthermore, in the step of setting the multi-objective optimization target, different weights ω are assigned to different targets. A ,ω E , The multi-objective comprehensive optimization function is

[0023] Furthermore, in the step of using the multi-objective evolutionary algorithm for global optimization, an elite retention strategy is adopted during the genetic operation process, and at the same time, the probabilities of crossover and mutation operations are adjusted. As the number of iterations increases, the crossover probability P c Gradually decrease from 0.8 to 0.6, the mutation probability P n Gradually increase from 0.01 to 0.03.

[0024] Furthermore, in the step of combining parameter adjustment with model structure optimization, by comparing and analyzing the accuracy, recall rate, and F1 value before and after the model structure adjustment, it is determined whether the model structure and parameters need to be further adjusted. When the model improves by no more than 3% on a certain performance indicator, a step-by-step backtracking adjustment strategy is adopted to return to the model state before the previous adjustment and readjust it.

[0025] Furthermore, after completing the parameter adjustment, the stability test of the optimized model was performed by inputting different test data sets 10 times and calculating the variance σ of the model output results. r 2 , variance σ r 2 If it is greater than 0.1, re-adjust the parameters and optimize the model.

[0026] Compared with the prior art, the personalized dynamic weight teaching evaluation model and its parameter adjustment method have the following beneficial effects:

[0027] The present invention constructs a personalized dynamic weight teaching evaluation model, which can be closely integrated with teaching practice, and the evaluation focus can be adjusted in real time according to the teaching stage and student learning status, so that the evaluation results are more timely and accurate, and more accurate teaching feedback can be provided to teachers, helping them to adjust teaching strategies in time. The parameter adjustment method of multi-objective optimization is used to comprehensively consider multiple important aspects of the evaluation, and fairness and interpretability are guaranteed while improving the accuracy of the evaluation. This comprehensive optimization method improves the overall performance of the teaching evaluation model, meets the needs of different teaching scenarios and student groups, and helps to promote the improvement of the quality of education and teaching.

[0028] Other advantages, objectives and features of the present invention will be set forth in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0030] Figure 1 It is a structural diagram of a personalized dynamic weight teaching evaluation model;

[0031] Figure 2 This is a flow chart of a parameter adjustment method for a personalized dynamic weight teaching evaluation model. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] Embodiment 1

[0034] In the mathematics course teaching of a middle school, the school is committed to improving the teaching quality, comprehensively and accurately understanding the students' learning status, providing teachers with targeted teaching feedback, and ensuring that the evaluation process is fair and just. The evaluation results can be clearly understood by teachers and students, thereby helping students adjust their learning strategies and achieve better learning results. The mathematics course covers multiple knowledge modules such as algebra and geometry. The teaching process is divided into four stages: course introduction, knowledge consolidation, ability improvement and comprehensive application. Different stages have different requirements for students' knowledge and abilities, and a flexible and accurate teaching evaluation model is needed to adapt to these changes.

[0035] First, we use the school's online learning platform and teaching management system to collect students' learning data and learning behavior data from all aspects. b The collection includes the duration of students' online learning mathematics course videos. l , the platform background records the start and end time of each student's login to study and makes statistics, and the frequency of interaction in the classroom in participating in mathematical problem discussions f i , quantified based on speech records in class discussion areas and participation in group activities.

[0036] Learning ability dataX a The acquisition includes: the speed at which students master new mathematical knowledge k , by analyzing the time it takes students to complete their new knowledge pre-study homework and their understanding and acceptance of new knowledge in class; the results of the math problem-solving ability test r p , using the results of regular math tests, exams, and special problem-solving ability tests to determine the emotional attitude data X e The data collected include: the change value Δi of students' interest in learning mathematics, the comprehensive judgment of the motivation intensity m for learning mathematics through periodic questionnaire surveys and students' active participation in the learning process (such as actively consulting mathematical materials, participating in mathematical interest groups, etc.), and quantitative analysis based on students' learning goal setting, degree of learning effort, and attitude towards mathematical learning.

[0037] The collected multi-source heterogeneous data are integrated and standardized, and the different types of data are unified into the same numerical range. The processing formula is: Where X norm is the standardized data, X is the original data, and after standardization, various types of data are integrated into fused data X=X b ,X a ,X e ], providing a basis for subsequent feature extraction.

[0038] Convolutional neural network is used to extract features from fused data X. First, the data is preprocessed to meet the input requirements of convolutional neural network. In convolutional neural network, multiple convolutional layers and pooling layers are used alternately to mine deep features of the data. The convolution operation is based on the formula: Among them, Y is the convolution output result, W is the convolution kernel, b is the bias, k and l are the sizes of the convolution kernel. By continuously adjusting the parameters of the convolution kernel and the network structure, the key feature vector F reflecting the individual characteristics and learning status of students is gradually extracted, that is, F = f(X). These feature vectors contain comprehensive information about students' learning behavior, ability, and emotional attitude, providing strong support for subsequent weight calculation and evaluation model construction.

[0039] According to the teaching stages s of the mathematics course (course introduction period s1, knowledge consolidation period s2, ability improvement period s3 and comprehensive application period s4), the teaching objectives g and the students' real-time learning status characteristic vector F, combined with the course difficulty coefficient d, the adaptive weight generation algorithm is used to dynamically adjust the weight W of each evaluation indicator. The course difficulty coefficient d is determined by the school's mathematics teaching experts through a comprehensive evaluation based on multiple factors such as the course syllabus, textbook content, and students' learning conditions in previous years.

[0040] In the course introduction period s1, the teaching goal is mainly to enable students to master basic knowledge such as basic mathematical concepts, theorems and formulas. At this time, the basic knowledge mastery index weight w1 = α1·f1(F,g,d(, where α1 is the weight adjustment coefficient of the basic knowledge mastery index in the course introduction period, f1 is the function for calculating the weight based on the student feature vector F, the teaching goal g and the course difficulty coefficient d, and the weights of other evaluation indicators w i It is also calculated through corresponding functional relationships, where i>1. These functions fully consider the learning characteristics and teaching focus of students at this stage to ensure that the weight distribution is reasonable.

[0041] In the knowledge consolidation period s2, the teaching goal focuses on students' integration and preliminary application of basic knowledge. The weight of the basic knowledge mastery index w1 = α2·f2(F,g,d(, compared with the course introduction period, the weight of the basic knowledge mastery index in this stage will be appropriately adjusted to highlight the cultivation of knowledge integration and application ability. The weight of the knowledge integration and application ability index w j The corresponding function is calculated according to the students' learning status and teaching objectives at that stage to ensure that the weights can accurately reflect the students' performance in knowledge consolidation and application.

[0042] In the capacity improvement period s3, the teaching focus shifts to cultivating students' comprehensive application ability and ability to solve complex mathematical problems. The weight of each evaluation indicator is calculated according to the specific functional relationship of this stage. These functions pay more attention to students' thinking ability, innovation ability and other aspects in solving practical problems. For example, for some comprehensive questions that require students to use a variety of mathematical knowledge and methods to solve, the weight of the relevant ability indicators will be increased accordingly.

[0043] In the comprehensive application period s4, the teaching goal is to enable students to comprehensively and flexibly apply the mathematical knowledge they have learned to practical situations, cultivate students' mathematical literacy and innovation ability, and the weight of each evaluation indicator is also calculated by the specific function of the corresponding stage to highlight the evaluation of students' comprehensive application ability and innovation ability. For example, in practical application projects such as mathematical modeling, the weights of indicators such as students' teamwork ability, problem analysis and solving ability will be improved.

[0044] Based on the extracted feature vector F and the dynamically generated weight W, a hierarchical personalized teaching evaluation model is constructed. The bottom layer of the model is the evaluation indicators I, which cover various aspects of mathematical knowledge and dimensions such as students' learning ability and attitude, such as algebraic knowledge mastery, geometric figure understanding, problem-solving speed, and learning interest.

[0045] The middle layer performs weighted summation of the indicators according to the weight W, and the calculation formula is: In order to prevent the model from overfitting, the intermediate layer score calculation formula becomes Where λ is the regularization coefficient, and the value of λ is adjusted to balance the model's fitting degree and generalization ability.

[0046] The top layer outputs the comprehensive evaluation result R, R=h(S), where h is the function that converts the middle layer score into the final evaluation result. This function will comprehensively consider the middle layer score as well as the teaching objectives and evaluation criteria, and convert the score into a specific evaluation level (such as excellent, good, medium, qualified, unqualified) or a specific evaluation score, providing teachers and students with intuitive evaluation results.

[0047] Set multi-objective optimization goals, including evaluation accuracy A, fairness E, and interpretability I n , accuracy A is evaluated by model prediction result R p and students’ actual learning performance a The matching degree is measured by the calculation formula: Where m is the number of samples, and fairness E is evaluated by comparing the distribution of evaluation results of different student groups (such as different learning backgrounds, different genders, etc.), using the variance σ 2 Measures the degree of dispersion of evaluation results of different groups. The smaller the variance, the higher the fairness. The more interpretable I nBy defining the indicator contribution function To measure, where c i is the contribution of the i-th evaluation index. This function can clearly show the contribution of each evaluation index to the final evaluation result.

[0048] Assign weights ω to different objectives A ,ω E , The multi-objective comprehensive optimization function is The weights of each objective are adjusted dynamically according to different teaching scenarios and needs. For example, in the comprehensive evaluation at the end of the semester, more emphasis is placed on the weight of accuracy; in the periodic evaluation during daily teaching, the distribution of fairness and explainability weights will be appropriately increased.

[0049] A multi-objective evolutionary algorithm is used to globally optimize the model parameters. In the optimization process, the model parameters are taken as individuals, and the optimal parameter combination that meets multiple objectives is continuously searched through genetic operations such as selection, crossover and mutation. In each iteration, the individuals are evaluated and screened according to the priority and weight of each objective, and the better parameter combination is retained. In the genetic operation process, the elite retention strategy is used to ensure that the best individuals in each iteration will not be eliminated to accelerate the convergence of the algorithm. At the same time, the probability of crossover and mutation operations is adjusted. As the number of iterations increases, the crossover probability P c Gradually decrease from the initial value, the mutation probability P m Gradually increase from the initial value to balance the global search and local search capabilities. Specifically, the crossover probability P c Gradually decrease from 0.8 to 0.6, the mutation probability P n Gradually increase from 0.01 to 0.03.

[0050] According to the actual teaching situation and data feedback, the parameter adjustment is combined with the model structure optimization. The feedback from teachers and students on the evaluation results is collected regularly. The evaluation data is analyzed and the performance of the model is observed in different stages and among different student groups. If it is found that a certain evaluation indicator has low discrimination in different stages, for example, a certain evaluation indicator I j The discrimination D j If the value is lower than the threshold value θ, the position of the indicator in the model or the weight calculation method is adjusted. The specific adjustment method is to adjust its weight calculation function w j =β·f3(F,g), where β is the adjustment coefficient and f3 is the new weight calculation function. In this way, the discrimination of the indicator is improved, thereby improving the overall performance of the model.

[0051] After completing the parameter adjustment, the stability test of the optimized model is performed. The variance of the model output results is calculated by inputting different test data sets 10 times. These test data sets cover math problems of different difficulty levels and knowledge modules, as well as data of students in different learning states. If it is greater than the set threshold (such as 0.1), it means that the performance of the model under different data inputs fluctuates greatly and is not stable enough. It is necessary to readjust the parameters and optimize the model until the variance of the model output meets the requirements, ensuring that the optimized model has reliable and stable evaluation capabilities in practical applications.

[0052] Effects brought by this embodiment: Through this embodiment, a comprehensive, dynamic and accurate evaluation of students' mathematics learning is achieved. The adaptive weight generation mechanism enables the evaluation index weight to be adjusted in real time with the teaching stage and the students' learning status, accurately reflecting the changes in teaching focus and students' learning situation, allowing teachers to timely understand students' learning progress and problems at different stages, and providing a strong basis for the adjustment of teaching strategies. The parameter adjustment method of multi-objective optimization takes into account the accuracy, fairness and interpretability of the evaluation, ensuring that students with different learning foundations and backgrounds can receive fair evaluation, and the evaluation results are clear and easy to understand, which helps students to clarify their learning advantages and disadvantages and adjust their learning strategies in a targeted manner. This series of measures ultimately effectively improved the quality of mathematics teaching and promoted the development of students' mathematics learning ability and comprehensive quality.

[0053] Embodiment 2

[0054] In the software development course teaching at a vocational and technical college, the course aims to cultivate students' practical software development capabilities so that students can quickly adapt to the development needs of enterprises after graduation. The course covers multiple teaching modules such as programming language basics, database applications, project development practice, etc. The teaching process is divided into four stages: basic theoretical learning, skill practice training, project actual application and comprehensive ability improvement. Each stage has different requirements for students' abilities. A scientific teaching evaluation model is needed to comprehensively evaluate students' learning outcomes and provide strong support for teaching improvement and students' career development.

[0055] Utilize the college's online teaching platform, training management system and enterprise internship feedback channels to comprehensively collect students' learning data and learning behavior data b Includes: The duration of students learning programming language course videos on online teaching platforms l , recorded and counted through the platform background; the frequency of interaction in classroom discussions and group collaborative development projects f i , quantified based on discussion forum speech records and group project collaboration.

[0056] Learning ability dataX a Acquisition includes: How quickly students master the syntax and features of new programming languages k, by analyzing the time and accuracy of students completing programming exercises; the problem-solving ability evaluation results in software development projects p , determined based on students’ performance in solving technical problems in actual project development, emotional attitude data X e The data collected include: the change value Δi of students' interest in software development, judged by periodic questionnaire surveys and students' active participation in extracurricular programming activities, their expectations for career development and their level of effort m, and quantitative analysis combined with students' career planning and performance in internships.

[0057] To fuse the collected multi-source heterogeneous data, we first use a standardized processing method to unify different types of data into the same numerical range. The processing formula is: Where X norm is the standardized data, X is the original data, and then all kinds of data are integrated into fusion data X = [X b ,X a ,X e ].

[0058] Convolutional neural network is used to extract features from fused data X. First, the fused data is preprocessed to meet the input requirements of convolutional neural network. In convolutional neural network, multiple convolutional layers and pooling layers are used to perform deep feature mining on the data. The convolution operation is based on the formula In this paper, Y is the convolution output result, W is the convolution kernel, b is the bias, k and l are the sizes of the convolution kernel. By continuously adjusting the parameters of the convolution kernel and the network structure, the key feature vector F reflecting the individual characteristics and learning status of students is gradually extracted, that is, F = f(X). These feature vectors contain comprehensive information about students' learning behavior, ability, and emotional attitude, which provides strong support for the subsequent weight calculation and evaluation model construction.

[0059] According to the teaching stages s of the software development course (basic theory learning s1, skill practice training s2, project application s3 and comprehensive ability improvement s4), the teaching objectives g and the students' real-time learning status feature vector F, combined with the course difficulty coefficient d (determined by industry experts and teachers based on the course content, the actual needs of the enterprise and the students' past learning situation), the adaptive weight generation algorithm is used to dynamically adjust the weight W of each evaluation indicator.

[0060] In the basic theory learning stage s1, the teaching goal is to enable students to master the basic concepts of software development, programming language basics and other knowledge. The weight of the basic knowledge mastery indicator w1 = α1·f1(F,g,d), and the weight of other evaluation indicators w iCalculated by corresponding functions, where i>1, these functions fully consider students’ learning of theoretical knowledge and teaching focus at this stage and reasonably allocate weights.

[0061] In the skill practice training stage s2, the teaching goal focuses on the cultivation of students' practice and application ability of programming skills. The weight of the basic knowledge mastery index w1 = α2·f2(F, g, d), and the weight of the programming skill application ability index w j The corresponding function is calculated based on the students' performance in the practical project and the teaching objectives, highlighting the evaluation of students' practical operation ability.

[0062] In the project practical application stage s3, the teaching focus is to enable students to comprehensively apply the knowledge and skills they have learned in actual projects, and cultivate teamwork and project management capabilities. The weight of each evaluation indicator is calculated according to the specific functional relationship of this stage. For example, the weights of indicators such as teamwork ability and project progress control ability will be increased accordingly to meet the requirements of project practice.

[0063] In the comprehensive ability improvement stage s4, the teaching goal is to improve students' comprehensive software development capabilities, innovation capabilities and mastery of cutting-edge industry technologies. The weights of each evaluation indicator are also calculated by the specific functions of the corresponding stage, focusing on the evaluation of students' innovative thinking, technology expansion capabilities, etc.

[0064] Based on the extracted feature vector F and the dynamically generated weight W, a hierarchical personalized teaching evaluation model is constructed. The bottom layer of the model is the evaluation index I, covering multiple aspects such as programming language mastery, database operation, project development process, team collaboration, etc. The middle layer performs weighted summation of the indexes according to the weight W. The calculation formula is: To prevent the model from overfitting, the intermediate layer score calculation formula becomes Where λ is the regularization coefficient.

[0065] The top layer outputs the comprehensive evaluation result R, R = h(S). Through this function, the middle layer scores are converted into specific evaluation levels (such as excellent, good, medium, qualified, unqualified) or specific evaluation scores, providing intuitive evaluation information for teachers and students.

[0066] Set multi-objective optimization goals, including evaluation accuracy A, fairness E, and interpretability I n , accuracy A is evaluated by model prediction result R p and students’ actual learning performance a The matching degree is measured by the calculation formula: Where m is the number of samples, and fairness E is evaluated by comparing the distribution of evaluation results of different student groups (such as different learning progress, different practical experience, etc.), using the variance σ 2Measuring the discreteness of evaluation results of different groups, interpretability I n By defining the indicator contribution function To measure and clearly show the contribution of each evaluation indicator to the final evaluation results.

[0067] Assign weights to different goals The multi-objective comprehensive optimization function is The weights of various objectives are adjusted dynamically according to different teaching scenarios and needs. For example, in the pre-assessment before corporate internships, more emphasis is placed on the accuracy weight to accurately assess students' actual abilities; in daily course evaluations, the fairness and explainability weights are appropriately increased.

[0068] A multi-objective evolutionary algorithm is used to globally optimize the model parameters. In the optimization process, the model parameters are taken as individuals, and the optimal parameter combination that meets multiple objectives is continuously searched through genetic operations such as selection, crossover and mutation. The elite retention strategy is adopted to ensure that the best individuals in each iteration will not be eliminated, and the crossover probability P is dynamically adjusted. c and mutation probability P m , as the number of iterations increases, the crossover probability P c Gradually decrease from 0.8 to 0.6, the mutation probability P n Gradually increase from 0.01 to 0.03.

[0069] According to the actual teaching situation and data feedback, the parameter adjustment is combined with the model structure optimization. Feedback from teachers, enterprise mentors and students is collected regularly. Evaluation data is analyzed. If it is found that a certain evaluation indicator has low differentiation in different stages, such as a certain evaluation indicator I j The discrimination D j If it is lower than the threshold value θ, the position of the indicator in the model or the weight calculation method is adjusted, and its weight calculation function w is adjusted. j =β·f3(F,g).

[0070] After completing the parameter adjustment, the stability test of the optimized model is carried out by inputting different test data sets 10 times (including different types of software development project cases, data of students at different learning stages, etc.) and calculating the variance of the model output results. If the variance If it is greater than the set threshold (such as 0.1), the parameters are readjusted and the model is optimized again until the model output results are stable and reliable.

[0071] Effects brought by this embodiment: Through the application of this embodiment in the software development course of vocational and technical colleges, a comprehensive, dynamic and accurate evaluation of students' software development capabilities is achieved. The adaptive weight generation mechanism enables the evaluation index weights to be adjusted in real time according to the teaching stage and students' learning status, accurately reflecting the teaching focus and changes in students' ability development, and providing a scientific basis for teachers to adjust teaching content and methods. The parameter adjustment method of multi-objective optimization takes into account the accuracy, fairness and interpretability of the evaluation, ensuring that students with different learning levels and backgrounds can receive fair evaluation. The evaluation results are clear and easy to understand, which helps students understand their strengths and weaknesses in the field of software development and clarify their career development direction. The application of convolutional neural networks in feature extraction effectively mines the deep features in the data and improves the evaluation accuracy of the model. This series of measures effectively improves the quality of course teaching, improves students' employment competitiveness, and lays a solid foundation for students' future career development.

[0072] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the same elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A personalized dynamic weight teaching evaluation model, characterized in that: The model includes: data collection and fusion module, feature extraction module, adaptive weight generation module and evaluation model construction module; The data collection and fusion module collects multi-source heterogeneous data, including students' learning behavior data X b , learning ability data X a And sentiment attitude data X e , where the learning behavior data X b Including online learning time l , classroom interaction frequency f i , learning ability data X a Contains knowledge acquisition speed v k , Problem-solving ability assessment results p , emotional attitude data X e Including the learning interest change value Δi and the learning motivation intensity m, the collected data are fused to obtain the fused data X = [X b ,X a ,X e ]; The feature extraction module uses data mining and machine learning technology to extract key feature vectors F reflecting individual characteristics and learning status of students from the fused data X. The extraction formula is F=f(X), where f is the feature extraction function; The adaptive weight generation module uses an adaptive weight generation algorithm to dynamically adjust the weight W of each evaluation index according to the teaching stage s, the teaching goal g, and the student's real-time learning state feature vector F. The teaching stage is divided into a course introduction period s1, a knowledge consolidation period s2, a capability improvement period s3, and a comprehensive application period s4. For different teaching stages, the calculation method of the weight W of each evaluation index is as follows: In the course introduction period s1, the weight of the basic knowledge mastery indicator w1 is calculated as w1 = α1·f1(F,g), and the weights of other evaluation indicators w i Calculated by the corresponding function, where i>1; In the knowledge consolidation period s2, the weight of the basic knowledge mastery index w1 is w1 = α2·f2(F,g), and the weight of the knowledge integration and application ability index w j Calculated by the corresponding function; In the capability improvement period s3 and the comprehensive application period s4, the weight of each evaluation index is calculated by the specific function of the corresponding stage; The evaluation model construction module constructs a hierarchical personalized teaching evaluation model based on the extracted feature vector F and the dynamically generated weight W. The bottom layer of the model is the evaluation index I, and the middle layer performs weighted summation of the indexes according to the weight W. The calculation formula is: Where S is the score of the middle layer, n is the number of evaluation indicators, and the top layer outputs the comprehensive evaluation result R, R = h(S), where h is the function that converts the middle layer score into the final evaluation result.

2. According to claim 1, a personalized dynamic weight teaching evaluation model is characterized in that: In the data acquisition and fusion module, a standardized processing method is used for different types of data to unify the data into the same numerical range. The processing formula is: Where X norm is the standardized data, and X is the original data.

3. According to claim 1, a personalized dynamic weight teaching evaluation model is characterized in that: In the feature extraction module, a convolutional neural network is used for feature extraction. The fusion data X is subjected to feature extraction through convolution layer and pooling layer operations. The convolution operation formula is: Where Y is the convolution output result, W is the convolution kernel, b is the bias, and k and l are the sizes of the convolution kernel.

4. According to claim 1, a personalized dynamic weight teaching evaluation model is characterized in that: In the adaptive weight generation module, the weight adjustment process takes into account the course difficulty coefficient d. The course difficulty coefficient is introduced as a parameter in the weight calculation function of different teaching stages for adjustment. In the course introduction period s1, the calculation formula of the basic knowledge mastery index weight w1 is w1=α1·f 1(F ,g,d).

5. According to claim 1, a personalized dynamic weight teaching evaluation model is characterized in that: In the evaluation model construction module, after the weighted summation of the intermediate layer, a regularization term is introduced to prevent the model from overfitting, and the intermediate layer score calculation formula becomes Where λ is the regularization coefficient.

6. A parameter adjustment method for a personalized dynamic weight teaching evaluation model, characterized in that: The method comprises the following steps: For the personalized dynamic weight teaching evaluation model, set multi-objective optimization goals, including evaluation accuracy A, fairness E, and explainability I n , accuracy A is evaluated by model prediction result R p and students’ actual learning performance a The matching degree is measured by the calculation formula: Where m is the number of samples, and fairness E is evaluated by comparing the distribution of evaluation results of different student groups, using the variance σ 2 Measuring the discreteness of evaluation results of different groups, interpretability I n By defining the indicator contribution function c i To measure, where c i is the contribution of the i-th evaluation index; The multi-objective evolutionary algorithm is used to globally optimize the model parameters. In the optimization process, the model parameters are regarded as individuals, and the optimal parameter combination that meets multiple objectives is continuously searched through selection, crossover and mutation genetic operations. In each iteration, the individuals are evaluated and screened according to the priority and weight of each objective, and the better parameter combination is retained. Combining parameter adjustment with model structure optimization, the structure of the evaluation model was fine-tuned according to the actual teaching situation and data feedback. It was found that the discrimination degree of a certain evaluation indicator in different stages was not high. The model performance was changed by adjusting the position of the indicator in the model and the weight calculation method. The specific adjustment method was as follows: the evaluation indicator I j The discrimination D j When it is lower than the threshold θ, adjust its weight calculation function w j =β·f3(F,g).

7. The parameter adjustment method of a personalized dynamic weight teaching evaluation model according to claim 6 is characterized in that: In the step of setting the multi-objective optimization objectives, different weights ω are assigned to different objectives. A ,ω E , The multi-objective comprehensive optimization function is 8. The parameter adjustment method of a personalized dynamic weight teaching evaluation model according to claim 6 is characterized in that: In the step of global optimization using the multi-objective evolutionary algorithm, an elite retention strategy is adopted during the genetic operation process. At the same time, the probabilities of crossover and mutation operations are adjusted. As the number of iterations increases, the crossover probability P c Gradually decrease from 0.8 to 0.6, the mutation probability P n Gradually increase from 0.01 to 0.

03.

9. The parameter adjustment method of a personalized dynamic weight teaching evaluation model according to claim 6 is characterized in that: In the step of combining parameter adjustment with model structure optimization, by comparing and analyzing the accuracy, recall rate, and F1 value before and after the model structure adjustment, it is determined whether the model structure and parameters need to be further adjusted. When the model does not improve by more than 3% on a certain performance indicator, a step-by-step backtracking adjustment strategy is adopted to return to the model state before the previous adjustment and readjust it.

10. The parameter adjustment method of a personalized dynamic weight teaching evaluation model according to claim 6, characterized in that: After completing the parameter adjustment, the stability test of the optimized model is carried out by inputting different test data sets 10 times and calculating the variance of the model output results. variance If it is greater than 0.1, re-adjust the parameters and optimize the model.

Images