Teaching data index-based multi-dimensional portrait construction method and system

Abstract

The invention provides a multi-dimensional portrait construction method based on a teaching data index. The method comprises the following steps: S1, extracting original data from a system related to teaching data, and preprocessing the original data to form a structured data set; s2, constructing a science and intelligence teacher model, and correcting the subjective score of the teacher to form a mixed data set; s3, carrying out feature dimension reduction on the data in the mixed data set; s4, training by adopting a multilayer perceptron model based on the key feature vector to generate a multi-dimensional index; s5, performing generalization ability evaluation on the trained multi-layer perceptron model, and establishing a real-time monitoring and automatic updating mechanism to synchronously update a multi-dimensional index; and S6, generating a multi-dimensional portrait based on the multi-dimensional index. Through big data and artificial intelligence technologies, the teaching quality is improved, and the workload is reduced. In addition, reasonable allocation of resources is realized by analyzing the resource use condition, the educational effect is improved, and the resource utilization efficiency is promoted.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, and specifically relates to a method and system for constructing a multidimensional profile based on teaching data index. Background Technology

[0002] Data mining is an advanced analytical method for discovering patterns, relationships, and insights from large datasets. By employing statistics, machine learning, and database techniques, it can extract valuable information from complex, incomplete, and noisy data. The advancement of data mining lies in its ability to automatically identify implicit patterns in data, predict future trends, and support decision-making. Its applications are wide-ranging, encompassing market analysis, bioinformatics, financial forecasting, fraud detection, and recommender systems, helping to optimize business processes, improve efficiency, and enhance competitiveness.

[0003] In the digital age, data is playing an increasingly important role in education. Data not only changes the way education is conducted, but also provides new possibilities for improving educational quality and personalized learning. It helps teachers optimize teaching methods and strategies, improve the assessment of educational quality, and promote the optimal allocation of educational resources.

[0004] Traditional teaching models often lack personalization and precision, making it difficult to meet the diverse learning needs of students. The emergence of knowledge graph-based intelligent teaching application systems aims to address this issue. This system, centered on knowledge graphs, uses intelligent recommendation algorithms to provide students with customized learning resources and pathways, while simultaneously supporting teachers in precise teaching and resource optimization. This system not only stimulates students' enthusiasm for self-directed learning but also promotes innovative applications of educational models such as academic year and credit system reforms. Against this backdrop, this project, based on multi-dimensional data, constructs a scientific and visual portrait of each major at Wuhan University. This portrait consists of four parts: a comprehensive learning portrait, a teaching process portrait, a management and support portrait, and a professional development portrait. Through these four portraits, students can directly understand the learning situation of students in their major, the teaching situation of teachers, the level of management and support provided by the administrative department, and the development status of the major itself within the Smart Luojia platform. The construction of these professional portraits helps newly enrolled students understand the situation of their major and also helps students intending to switch majors understand their target major.

[0005] Patent document CN113704610B discloses a method and system for generating learning style profiles based on learning and growth data. The method includes the following steps: collecting learning and growth data, which includes user attribute data and multidimensional user behavior data obtained from multiple data sources; performing characterization processing on the learning and growth data to obtain a user behavior representation matrix; inputting the user behavior representation matrix into a pre-trained user feature extraction model, which is based on a neural network and is used to output a user behavior feature vector matrix; multiplying the user behavior feature vector matrix with a learning feature vector, and using the resulting matrix as the user learning style profile. This invention cannot accurately assess the achievement of course and professional teaching objectives, and the system's evaluation report has limitations, failing to provide profiles of majors, teachers, courses, and students, and cannot offer multi-level improvement methods. Summary of the Invention

[0006] The purpose of this invention is to promote the digital transformation of school education and teaching by constructing a comprehensive and accurate multi-dimensional profile to support teaching reform and professional development. Simultaneously, it develops a big data-based digital teaching evaluation system and indicators. Through multi-dimensional profiles and a digital teaching index assessment system, it comprehensively evaluates the degree of digitalization in the educational process, thereby achieving personalized teaching and precise teaching evaluation. Furthermore, it will establish a comprehensive digital profile platform based on the indicator system to improve teaching quality and efficiency, and ensure the platform's openness and scalability to adapt to the needs of different disciplines and teaching models.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A method for constructing a multidimensional profile based on teaching data indices includes the following steps: Step S1. Extract raw data from the system related to teaching data and preprocess it to form a structured dataset, which includes teacher subjective rating data and objective behavior data; Step S2. Construct a rational teacher model, correct the teacher's subjective rating to generate an objective rating, and merge the objective rating with the objective behavioral data to form a hybrid dataset; Step S3. Perform feature dimensionality reduction on the data in the mixed dataset and extract key feature vectors; Step S4. Based on the key feature vectors, a multilayer perceptron model is used for training, with dynamic batch processing, learning rate scheduling and regularization strategies configured, and a multidimensional exponent is generated during training. Step S5. Evaluate the generalization ability of the trained multilayer perceptron model, introduce the learning progress rate and the continuous improvement rate of teaching effect indicators, and establish a real-time monitoring and automatic update mechanism to update the multidimensional index synchronously. Step S6. Generate a multidimensional profile based on the multidimensional index.

[0008] Furthermore, the systems related to teaching data include teaching evaluation systems, academic affairs systems, online course learning systems, OA systems, library systems, student affairs management systems, enrollment and employment systems, and teaching and research systems.

[0009] Furthermore, the preprocessing in step S1 includes: cleaning, standardizing, and uniformly identifying the raw data, wherein the uniform identification process includes: achieving multi-source heterogeneous data fusion through data warehouse technology to establish a unified identification system; performing semantic annotation and structured transformation on unstructured data to generate standardized fields.

[0010] Furthermore, the construction of the rational teacher model in S2 includes: Based on the SVM double-blind peak-shaving comparison method, the teacher's subjective ratings were standardized and key features were extracted. The weight of each key feature is determined using the analytic hierarchy process (AHP) or expert scoring. Based on the SVM double-blind peak-shaving control method, the classification boundary is generated by utilizing teaching outcomes, teaching process, educational background and training experience, academic research results, teaching attitude, and work pressure coping ability. Cluster analysis was performed on the generated objective scores to remove extreme value data.

[0011] Furthermore, the feature dimensionality reduction in step S3 further includes: Dimensionality reduction of high-dimensional data in mixed datasets: Principal component analysis is used for initial dimensionality compression; t-SNE algorithm is combined to further reduce the dimensionality of nonlinear features, and cosine distance is used to optimize the similarity measure; sliding window statistics are performed on time series data to extract periodic features; Extracting key feature vectors: Ranking the importance of features in structured numerical data and retaining the Top-N features; using convolutional neural networks to extract teaching behavior features from unstructured data.

[0012] Furthermore, generating the multidimensional index in step S4 includes: Constructing a multilayer perceptron model based on key feature vectors: The input layer dimension matches the number of key features, and the hidden layer uses the Swish activation function. The output layer generates multidimensional indices, including teaching competence index, teaching digitization index, teacher-student interaction index, and learning challenge index. Configure dynamic training strategies, including dynamic batch processing, learning rate scheduling, and regularization strategies.

[0013] Furthermore, the training of the multilayer perceptron model in step S4 further includes: A multi-task learning architecture is adopted to simultaneously optimize sub-tasks related to teaching ability, digital literacy, and scientific research level. Recurrent neural networks are used to extract temporal features from unstructured data.

[0014] Furthermore, in step S5, alternating K-fold cross-validation is used to evaluate the generalization ability of the trained multilayer perceptron model, specifically including: The key feature vectors are divided into K×K subsets. K×K-2 subsets are used for each training session, and the remaining 1 subset is used as the validation set. Introduce learning progress rate and continuous improvement rate of teaching effectiveness as supplementary evaluation indicators; The sensitivity of the model to the generation of different multidimensional indices was verified by dynamically adjusting the weights.

[0015] Furthermore, the dynamic profile generation in step S6 further includes: dynamically configuring the weight ratios based on the generated multidimensional indices to generate a comprehensive multidimensional profile of teachers, students, courses, and majors, and outputting the results through a visualization interface.

[0016] On the other hand, the present invention provides a multi-dimensional profile construction system based on teaching data index, comprising: The data acquisition module is used to extract raw data from systems related to teaching data and preprocess it to form a structured dataset, which includes teacher subjective rating data and objective behavioral data. The data correction module is used to construct a rational teacher model, correct the teacher's subjective rating to generate an objective rating, and merge the objective rating with the objective behavioral data to form a hybrid dataset. The feature processing module is used to perform feature dimensionality reduction on the data in the mixed dataset and extract key feature vectors; the index generation module is used to train a multilayer perceptron model based on the key feature vectors, configure dynamic batch processing, learning rate scheduling and regularization strategies, and train to generate a multidimensional index. Evaluation module. It is used to evaluate the generalization ability of the trained multilayer perceptron model, introduces the learning progress rate and the continuous improvement rate of teaching effect as indicators, and establishes a real-time monitoring and automatic update mechanism to update the multidimensional index synchronously. The portrait generation module is used to generate a multidimensional portrait based on the multidimensional index.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention boasts high efficiency and intelligence in feature processing. By combining principal component analysis (PCA), t-SNE algorithm, and sliding window techniques, it achieves efficient dimensionality reduction and periodic feature extraction from high-dimensional data. Simultaneously, it employs convolutional neural networks and recurrent neural networks for deep processing of unstructured data, enabling more effective capture of potential patterns within the data. Furthermore, it introduces a multi-dimensional indicator system and dynamically configured weight ratios to ensure the flexibility of the profiling results and accurately reflect the achievement of teaching objectives. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the multi-dimensional portrait construction method of the present invention.

[0020] Figure 2 This is a schematic diagram illustrating the construction of a multi-dimensional profile based on the original data source in this invention.

[0021] Figure 3 This is a schematic diagram of student scores before the teacher's subjective scoring is corrected according to the present invention.

[0022] Figure 4 This is a diagram illustrating student scores after correction of the teacher's subjective scoring in this invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0024] Example 1 like Figure 1 As shown, this embodiment provides a method for constructing a multi-dimensional profile based on teaching data indices, characterized by the following steps: Step S1. Extract raw data from the system related to teaching data and preprocess it to form a structured dataset, which includes teacher subjective rating data and objective behavior data; like Figure 2As shown, the teaching process profile data comes from the teaching evaluation system, academic affairs system, online course learning system, and OA system, from which key data fields have been extracted. These data originate from the teaching evaluation system, academic affairs system, online course learning system, and OA system. Key data fields cover information related to teaching competence index, teaching digitization index, teacher-student interaction index, and teaching research activity index, such as basic teacher information, various evaluation scores, the quantity and quality of teaching achievements, and the use of teaching resources.

[0025] (1) Teaching competence index. The data input fields include: teacher, college, major, course name, student evaluation score, peer evaluation score, teaching supervision evaluation score, teaching quality evaluation score, number of academic achievements, number of teaching-related achievements, proportion of teaching-related achievements, score of proportion of teaching-related achievements, status of teaching case or textbook writing, score of status of teaching case or textbook writing, score of integration of academic achievements with teaching, number of teaching awards, score of number of teaching awards, highest level of teaching awards, score of highest level of teaching awards, score of teaching awards and honors, number of course design projects, score of number of course design projects, quality of course design projects, score of quality of course design projects, score of teaching design and course development, teaching years, score of teaching years, number of times teaching the same course, score of teaching years and experience.

[0026] (2) Teaching Digitalization Index. The data input fields include: teacher, college, major, number of resources, resource quantity score, resource access, resource downloads, total resource access and downloads, school-wide average of total resource access and downloads, resource access and download score, digital teaching resource usage score, number of course announcements per semester, number of assignments per semester, course announcement and assignment posting score, number of weekly discussions, number of weekly Q&A sessions, discussion and Q&A score, digital teaching tool application score, number of blended learning courses, blended learning model implementation score, whether digital tools are used for student learning evaluation and feedback, digital teaching evaluation and feedback score, whether social media academic platforms and other channels are used to promote courses, whether courses are widely recognized, such as winning awards, being adopted or cited by other institutions, etc., and digital teaching evaluation and feedback score.

[0027] (3) Teacher-student interaction index. The data input fields include: teacher, college, major, response rate to student questions, student question response rate score, whether at least one group discussion is organized in each class and the average duration is not less than 10 minutes, number of group discussions and duration score, number of interactive methods in class, diversity of interactive forms score, number of replies after class, number of replies score, response time, response time score, post-class communication score, whether there is an interactive feedback mechanism, feedback mechanism score, reply ratio, implementation effect score, and interactive feedback mechanism score.

[0028] (4) Teaching and research activity index. The data input fields include: teacher, college, major, frequency of participation in teaching and research activities, score of participation in teaching and research meetings and activities, role of teaching and research team, score of teaching and research team building, highest level of teaching competition / teaching achievement award, score of teaching research and teaching achievement award, frequency of participation in teaching and research, score of participation in teaching and research, level of teaching and research projects participated in by teachers, and score of participation in teaching and research projects and topics.

[0029] 2. The learning profile data comes from the library system, student affairs management system, online course learning system, and OA system, from which key data fields are extracted. Data is collected from these systems and includes fields related to learning challenge index, learning engagement index, interdisciplinary index, and digital literacy index, including course performance data, learning resource utilization, and subject background information.

[0030] (1) Learning Challenge Index. The data input fields include: college, major, average course score, average score, high score rate, high score rate score, failure rate, failure rate score, course difficulty, percentage of required course credits, percentage of required course credits score, teaching class name, number of reading materials, number of reading materials score, length of reading materials, length of reading materials score, source diversity, source diversity score, course reading volume and richness of learning materials score, internship batch, comprehensive score, and internship practice difficulty score.

[0031] (2) Learning input index. The data input fields include: college, student major, number of weekly library check-ins, average number of library check-ins across the university, check-in ratio, learning time input score, attendance rate, attendance rate score, number of classroom interactions, classroom interaction score, course participation score, average monthly library visit frequency, visit frequency score, average monthly library visit duration, visit duration score, extracurricular learning resource utilization score, number of self-study sessions, number of self-study sessions score, whether there are self-study achievements, highest level of achievements, self-study achievement score, and self-study project and activity score.

[0032] (3) Interdisciplinary index. The data input fields include: teachers, colleges, majors, disciplines, number of teachers with disciplinary backgrounds, teachers' background diversity score, interdisciplinary nature of teachers' research directions, teachers' research direction interdisciplinary score, interdisciplinary nature of curriculum design, interdisciplinary score of curriculum design, proportion of practical courses, and diversity score of course nature in professional teaching plans.

[0033] (4) Digital literacy index. The data input fields include: college, major, average monthly library visit frequency, visit frequency score, average monthly library visit duration, visit duration score, digital tool and platform usage ability score, course name, credits, information technology course grades, weighted average grade score, information technology course grade score, digital learning participation frequency, digital learning participation score, number of times digital literacy training and education is participated, number of times of participation score, number of types, type diversity score, and digital literacy training and education experience score.

[0034] 3. Management decision-making profile data is derived from the student affairs management system, online course learning system, academic affairs system, teaching evaluation system, teaching and research system, and OA system. Key data fields are extracted. This includes data related to student learning trajectory tracking index, teacher teaching quality improvement index, educational resource management index, and professional development history index, such as graduate employment data, teacher teaching ability evaluation data, and teaching resource management data.

[0035] (1) Student learning trajectory tracking index. The data input fields include: college, major, graduate employment rate, graduate employment rate score, number of graduate employment industries, graduate employment field diversity score, average salary of graduates, graduate employment quality score, graduate continuing education rate, graduate continuing education rate score, graduate entrepreneurship rate, graduate entrepreneurship rate score, employer feedback satisfaction, employer feedback satisfaction score, number of internships, cumulative duration, number of internships and duration score, internship unit type, internship object score, graduate employment status score, number of courses studied, number of courses studied score, course pass rate, course pass rate score, number of course assessment methods, proportion of course assessment methods score, course examination method type, course examination method score, course teaching format type, course teaching format score, and course learning status score.

[0036] (2) Teacher teaching quality improvement index. The data input fields include: college, major, teacher course design ability, teacher course design ability score, teacher-student interaction frequency, teacher-student interaction frequency score, classroom teaching effect evaluation, classroom teaching effect evaluation score, teaching innovation application status, teaching innovation application evaluation score, teacher continuing education participation rate, teacher continuing education participation rate score, teaching innovation application status, teaching innovation application evaluation score, teacher continuing education participation rate, teacher continuing education participation rate score, course evaluation feedback application status, course evaluation feedback application score, number of teaching awards, teaching award acquisition status score, teaching material update frequency, teaching material update frequency score, teaching observation and peer evaluation status, teaching observation and peer evaluation score.

[0037] (3) Education and teaching resource management index. The data input fields include: college, major, textbook and resource update rate, textbook and resource update rate score, number of library borrowings, number of library visits, resource utilization rate score, number of practice bases, average number of practice bases across the school, highest level of base, rationality score of course resource allocation, frequency of teaching equipment maintenance and update, frequency of teaching equipment maintenance and update, number of activated courses, percentage of activated courses, digital resource coverage score, number of course visits, frequency of teaching platform use score, efficiency of resource survey and procurement, and efficiency score of resource survey and procurement.

[0038] (4) Professional Development Historical Development Index. The data input fields include: college, major, teacher, college, teacher title ratio, teacher title ratio score, professional leader influence, professional leader influence score, teacher research achievements, teacher research achievements score, interdisciplinary nature of curriculum, interdisciplinary nature of curriculum, teacher teaching workload, average workload of the whole school, teacher teaching workload score, teaching syllabus update frequency, teaching syllabus iteration score, textbook compilation and revision, textbook compilation and revision score, student academic situation, student academic performance score, professional development funding investment, professional development funding investment score, number of cooperative enterprises, number of practice bases, and school-enterprise cooperation and practice base score.

[0039] 4. The evaluation and development profile data comes from the teaching evaluation system, enrollment and employment system, student affairs management system, and OA system. Key data fields are extracted. This data is obtained from the teaching evaluation system, enrollment and employment system, student affairs management system, and OA system. It involves fields related to professional growth index, teacher excellence index, career potential index, and social impact index, such as teacher competence scores, teaching achievement awards, and professional social influence data.

[0040] (1) Professional growth index. The data input fields include: college, major, average score of teachers’ knowledge and literacy, average score of teachers’ ability, average score of personal traits, teaching achievement awards, teaching achievement award score, number of teachers’ published textbooks and monographs, score of number of teachers’ published textbooks and monographs, professional achievement, average achievement of the whole school, score of students’ required learning ability, score of employment adaptability, score of mastery of professional knowledge, score of teamwork and communication ability, score of professional ethics and sense of responsibility, retention rate score, and employer feedback score.

[0041] (2) Teacher Excellence Index. The data input fields include: teacher, college, major, teaching competence score, number of core journal papers, number of teacher research publications score, teacher-led projects, teacher-participated projects, teacher research project hosting score, number of teaching awards, number of important teaching awards, teacher teaching awards received score, number of student awards guided by teachers, number of student competition awards guided by teachers score, number of training sessions attended, whether there are any achievements, and teacher training and professional development score.

[0042] (3) Career potential index. Data input fields include: teacher, college, major, number of international academic conferences attended by teachers, number of publications, score of teachers' participation in international academic conferences, and satisfaction score of teacher-student interaction. (4) Social Impact Index. The data input fields include: college, major, number of times the promotional video is viewed, positive feedback and comment rate, professional promotional video influence score, number of times the professional media is exposed, professional media exposure score, number of school-enterprise cooperation, number of school-enterprise cooperation scores, number of social service projects participated in, social service project participation rate score, number of alumni in important positions, alumni social achievement score, public evaluation score, professional social reputation score, number of professional academic conferences held, number of professional academic conferences held scores, number of patents or innovative achievements obtained, number of patents and innovative achievements in the professional field scores, amount of social donations and sponsorships, social donations and sponsorships scores, number of participations in policy formulation, and professional participation in government or industry policy formulation scores.

[0043] In addition to the currently mentioned teaching evaluation systems, academic affairs systems, online course learning systems, OA systems, library systems, student affairs management systems, admissions and employment systems, and teaching and research systems, more data sources related to teaching and learning can be considered for inclusion. These could include campus card systems (used to analyze students' consumption patterns on campus, indirectly reflecting their lifestyle and economic status, and thus relating to their learning engagement) and campus network usage logs (to understand students' use of online resources for learning or entertainment during their free time, supplementing assessments of digital literacy, etc.), to obtain more comprehensive information and enrich the student profile. Specifically, the campus card system (analyzing the correlation between student consumption behavior and learning engagement) and campus network usage logs (assisting in assessing students' digital literacy) can be incorporated to enrich the student profile. Simultaneously, data integration and correlation should be strengthened, a unified identification system established, and data warehouse technology utilized to improve data management and analysis efficiency.

[0044] Current data integration primarily involves extracting key data fields, which can further strengthen the correlation and fusion between data. For example, establishing a unified system of student, teacher, course, and major identifiers ensures that the same entity in different data sources can be accurately matched and associated. Simultaneously, data warehouse technology is used to centrally manage and store the integrated data, improving the efficiency of data querying and analysis.

[0045] Two key issues need to be addressed at this stage: Define the data source: List the key data fields provided by each data source (such as teaching evaluation system, academic affairs system, online course learning system, etc.) to ensure that the data is comprehensive and without redundancy.

[0046] Data field standardization: Use a unified naming and format standard for data fields of the same type (such as evaluation scores, resource quantity, etc.) to facilitate subsequent data processing and analysis.

[0047] The standards for data processing are: Establish a unified identification system (Teacher ID / Student ID / Course Code) Implementing data warehouse technology to achieve multi-source heterogeneous data fusion Establish metadata management standards (naming rules / format standards / quality verification). Data Acquisition: Data is collected from all relevant subsystems of the school using data interfaces, providing incremental and full data synchronization interfaces. Data storage: A distributed data storage file system will be used to store the collected data. The plan is to use the Hadoop file system to support the storage of structured, semi-structured, and unstructured data. It will also support high-throughput data access requirements.

[0048] Data cleaning involves removing invalid, duplicate, or erroneous data to ensure accuracy and reliability. Besides filtering based on obvious rules (such as formatting errors or excessive null values), additional data verification mechanisms can be introduced. For example, comparing data with historical data or other reliable data sources can identify potential anomalies. For data that appears erroneous but is difficult to determine directly, further analysis using manual review or data mining algorithms can be employed.

[0049] Data standardization: Normalization eliminates differences in dimensions and magnitudes of data to facilitate unified analysis; it converts non-structured data (such as text and dates) into a unified structured format to support subsequent analysis and processing. When normalizing numerical data, in addition to common linear normalization methods, more appropriate normalization methods are selected based on the data's distribution characteristics. For example, for data with uneven distribution and many extreme values, non-linear normalization methods such as logarithmic transformation can be used to better reflect the data's characteristics. For the structured transformation of text data, besides simple format unification, semantic analysis and annotation can also be performed to improve data usability.

[0050] Numerical transformation: Establish unified numerical transformation rules to transform non-numerical data to support subsequent analysis and processing. Utilize cluster analysis or decision tree algorithms to automatically generate appropriate numerical transformation strategies based on data characteristics and distribution, improving the flexibility and accuracy of data processing.

[0051] Step S2. Construct a rational teacher model, correct the teacher's subjective ratings to generate objective ratings, and merge the objective ratings with the objective behavioral data to form a hybrid dataset; specifically: To construct a "rational teacher" model within the entire evaluation system, it is necessary to first build a "rational teacher" model based on various data from teachers. This is because each teacher in a school has differences in teaching outcomes, teaching processes, educational background and training experience, academic research achievements, teaching attitudes, and ability to cope with work pressure. These differences can lead to subjective preferences and biases in the process of scoring and evaluating students.

[0052] like Figure 4 As shown, through Figure 4 The final student scores do not necessarily reflect their true performance. A relatively rational teacher data structure is constructed by using an algorithm based on the average Cartesian distance to evaluate teachers' subjective scores. Then, the subjective scores of each student are reassessed using this rational teacher model. After constructing the "rational teacher" model, the composition of student scores is as follows: Figure 4 As shown, after constructing the rational model, the comparison between student scores before and after correction can be clearly displayed through the correction mechanism and the transparency mechanism, as well as how the correction mechanism reduces the influence of subjective evaluation.

[0053] The specific methods for constructing the rational teacher model are as follows: Identify key characteristics: Based on educational theories and practical needs, select key characteristics that embody a "rational teacher." These include, for example, the extent of student academic improvement in teaching outcomes, the effectiveness of classroom interaction during the teaching process, the relevance of training experience to teaching content in professional competence, and the ability to cope with work stress in professional ethics.

[0054] Weighting: The weight of each feature can be determined using methods such as the analytic hierarchy process (AHP) or expert scoring. For example, teaching outcomes account for 40%, professional competence for 30%, and vocational competence for 30%. Within teaching outcomes, student grade improvement accounts for 60%, pass rate for 30%, and excellent rate for 10%.

[0055] Algorithm Construction: The algorithm used to construct the teacher model was a double-blind, peak-shaving, control-based method using SVM. SVM can effectively handle high-dimensional data and find the optimal classification boundary in large-scale datasets. For example, in a dataset containing thousands of teachers, SVM can utilize data from multiple dimensions, including teaching outcomes, teaching processes, educational background and training experience, academic research achievements, teaching attitudes, work stress coping abilities, professional background, and career attitudes, to construct a classification model. Double-blind involves using multiple rounds of discrete observation and control groups to process the data using SVM, while peak-shaving removes extreme values ​​from the final calculation results and discards data that deviates too far from the center.

[0056] Data Processing: For structured data (such as teachers' educational background, years of teaching experience, and student exam scores), traditional machine learning algorithms such as linear regression, logistic regression, and decision trees can handle it well. For example, linear regression can be used to analyze the relationship between teachers' years of teaching experience and student performance improvement to determine whether teachers' teaching experience aligns with the characteristics of a rational teacher.

[0057] If the data contains unstructured data, such as teacher evaluations of instructional videos (text descriptions) or audio recordings of classroom teaching, then Natural Language Processing (NLP) techniques combined with machine learning algorithms are needed. For example, Convolutional Neural Networks (CNNs) in deep learning can be used to process the image information of the instructional videos, extracting features related to the teacher's teaching behavior, and then combining this with other structured data to build a model. Step S3. Perform feature dimensionality reduction on the data in the mixed dataset to extract key feature vectors; specifically: Feature Extraction: Dimensionality reduction is performed using t-SNE (t-Distributed Stochastic Neighbor Embedding). The core idea of ​​t-SNE is to define a probability distribution for data points in a high-dimensional space, representing the similarity between points, and then create a similar probability distribution in a low-dimensional space. By minimizing the difference between these two distributions (using KL divergence), the algorithm maps high-dimensional data to a low-dimensional space. We use a Gaussian distribution (normal distribution) to calculate the similarity between points. The Gaussian distribution is a common probability distribution with a bell-shaped shape, determined by its mean and variance (the square of the standard deviation). A key property of the Gaussian distribution is that its shape is determined by its mean (center point) and variance (width of the distribution). When we draw a Gaussian distribution around a data point x, this distribution assigns higher probability values ​​to nearby points and lower probability values ​​to points that are farther away. This aligns with our intuitive understanding of "similarity": points that are close are more similar, and points that are far apart are dissimilar.

[0058] For each data point X i We calculate X for all other points. j Its conditional probability P jji This probability reflects point X. j It is point X i The probability of its nearest neighbors. The calculation formula is:

[0059] Here, the molecule part is calculated. and The square of the Euclidean distance between them (i.e. Then, it is transformed into a probability using a Gaussian distribution. The denominator is a normalization factor, ensuring that all... The sum of is 1.

[0060] ó i The variance of the Gaussian distribution determines the range of nearest neighbors. Different points may have different densities, therefore ó i For each point They may be different and need to be determined using a quantity called "confusion level".

[0061] Finally, in order to obtain a symmetric similarity matrix, we take... and The average value is obtained :

[0062] Preprocessing with other dimensionality reduction algorithms (such as PCA) first reduces the data dimensionality, and then t-SNE is used for further feature extraction to improve computational efficiency. Simultaneously, when calculating the similarity between points, different distance metrics (such as cosine distance, Manhattan distance, etc.) can be tried, and the optimal metric can be selected based on the data characteristics to more accurately reflect the relationships between data points.

[0063] The basic characteristic data of teachers may include the average score of teachers' knowledge literacy, the average score of teachers' ability, the average score of teachers' personal traits, the number of teaching achievement awards, the score of teaching achievement awards, the number of textbooks and monographs published by teachers, the score of the number of textbooks and monographs published by teachers, the score of teaching competence, the number of core journal articles, the score of the number of research results published by teachers, the number of projects led by teachers, the number of projects participated in by teachers, the score of the research project led by teachers, the number of teaching awards, the number of important teaching awards, the score of the teaching awards won by teachers, the number of awards won by students guided by teachers, the score of competition awards won by students guided by teachers, the number of training sessions attended, and the number of international academic conferences attended by teachers.

[0064] The basic characteristics of students may include student major, student grade, student attendance rate, attendance rate score, number of classroom interactions, classroom interaction score, course participation score, number of weekly library check-ins, library visit frequency, library visit frequency score, library visit duration, library visit duration score, study time investment score, extracurricular learning resource utilization score, number of self-study sessions, self-study session score, self-study outcomes, self-study outcome score, digital learning participation frequency, and digital learning participation score.

[0065] The basic characteristic data of a course may include course name, student evaluation scores, peer evaluation scores, teaching supervision evaluation scores, teaching quality evaluation scores, proportion of teaching-related achievements, scores for the proportion of teaching-related achievements, status of teaching case studies or textbooks, scores for the status of teaching case studies or textbooks, scores for the integration of academic achievements with teaching, number of teaching awards, highest level of teaching awards, scores for teaching awards and honors, number of course design projects, quality of course design projects, scores for instructional design and course development, years of teaching experience, scores for years of teaching experience, number of times the same course has been taught, and scores for years of teaching experience.

[0066] The basic characteristic data of a major may include the major name, the proportion of faculty members with professional titles, the score of the proportion of faculty members with professional titles, the influence of the major leader, the score of the influence of the major leader, the status of teachers' scientific research achievements, the score of teachers' scientific research achievements, the interdisciplinary nature of the curriculum, the score of the interdisciplinary nature of the curriculum, the status of textbook compilation and revision, the score of textbook compilation and revision, the academic status of students, the score of students' academic performance, the investment in major development funds, the score of investment in major development funds, the score of school-enterprise cooperation and practice bases, the score of professional media exposure, the score of alumni social achievements, the score of professional social reputation, and the score of the number of patents and innovation achievements in the professional field. Step S4. Based on the key feature vectors, a multilayer perceptron model is used for training, configuring dynamic batch processing, learning rate scheduling and regularization strategies, and training to generate multidimensional indices; specifically: Dataset partitioning: Divide the preprocessed dataset into training, validation, and test sets. Typically, 70%-80% of the data is used as the training set, 10%-15% as the validation set, and 10%-20% as the test set.

[0067] Choosing a loss function and optimizer: Select an appropriate loss function based on the type of portrait task. For classification tasks, the cross-entropy loss function can be chosen; for regression tasks, the mean squared error (MSE) loss function can be chosen. Simultaneously, select a suitable optimizer, such as stochastic gradient descent (SGD), Adagrad, Adadelta, Adam, etc. The optimizer is used to update the model's weights and biases to minimize the loss function.

[0068] Training process: Input the training set data into the MLP model, calculate the output through forward propagation, then calculate the loss according to the loss function, and then update the model's weights and biases through backpropagation. Repeat this process until the model's performance on the validation set no longer improves (e.g., the validation set loss no longer decreases) or the preset number of training rounds is reached.

[0069] Configure training: Customize the relevant training parameters based on the teaching data of students and teachers, such as learning rate, batch size and number of iterations. These parameters have an important impact on the efficiency and effectiveness of model training.

[0070] Model training and parameter tuning: The multilayer perceptron (MLP) model is trained using the training dataset, and the internal parameters of the model are continuously adjusted through the backpropagation algorithm to minimize the value of the loss function.

[0071] During training, model parameters and training configurations may need to be adjusted multiple times to achieve better training results. During parameter tuning, L2 regularization, data augmentation, early termination, or dropout techniques can be used to prevent overfitting. Ensemble learning can be attempted by combining multiple different types of models (such as decision trees, neural networks, and Bayesian classifiers) to improve the accuracy and stability of predictions. For example, fusion strategies such as voting, weighted averaging, or stacking can be used to combine the prediction results of multiple models.

[0072] Introducing automated hyperparameter optimization algorithms, such as grid search, random search, or model-based optimization algorithms (such as Bayesian optimization), during model training can improve the efficiency and accuracy of hyperparameter tuning and find a better model configuration.

[0073] Model Loading: After model training is complete, the model is saved for later use. Code is written to load the model and apply it to practical tasks. Model interpretation techniques (such as feature importance analysis and locally interpretable model-independent interpretation (LIME)) are employed to improve model interpretability, helping teachers and administrators better understand the model's predictions and decision-making basis, thereby enabling more targeted improvements to teaching and management strategies. Step S5. The generalization ability of the trained multilayer perceptron model is evaluated, introducing learning progress rate and continuous improvement rate of teaching effectiveness indicators. A real-time monitoring and automatic update mechanism is established to synchronously update the multidimensional indices; specifically: Divide the dataset into training, validation, and test sets, and select appropriate loss functions (cross-entropy loss for classification tasks, mean squared error loss for regression tasks) and optimizers (such as stochastic gradient descent, Adagrad, etc.). Configure training parameters, train and tune the model, employ various methods to prevent overfitting, introduce automated hyperparameter optimization algorithms to improve efficiency, save and load the model, and use model interpretation techniques to enhance interpretability.

[0074] Training the model on the training set: First, input the training data into the multilayer perceptron model, calculate the predicted output through forward propagation, and then calculate the loss according to the loss function (such as cross-entropy loss in classification tasks, MSE in regression tasks, etc.). Use the backpropagation algorithm to update the model's weights and biases, and repeat this process until the model's loss on the training set converges or the preset number of training rounds is reached.

[0075] Evaluation and tuning on the validation set: After each training cycle, input the validation set data into the model and calculate the selected evaluation metric. Observe the changes in the evaluation metric on the validation set. If the validation set loss starts to rise or the evaluation metric deteriorates, overfitting may have occurred. In this case, measures can be taken, such as stopping training early, adding regularization terms, or adjusting the model complexity (e.g., reducing the number of hidden layer nodes or hidden layers).

[0076] Final evaluation on the test set: After model training and hyperparameter tuning are complete, the model is finally evaluated using the test set. The test set data is input into the model, and evaluation metrics are calculated.

[0077] In addition to existing evaluation metrics (such as accuracy and mean squared error), more evaluation metrics related to the actual effectiveness of teaching and learning should be introduced. For example, for student profile models, indicators such as learning progress rate (measuring the extent to which students' academic performance improves over a period of time) and changes in knowledge mastery can be added; for teacher profile models, indicators such as continuous improvement rate of teaching effectiveness (observing the trend of changes in teachers' teaching evaluation scores over multiple semesters) can be added to comprehensively evaluate the performance of the model from different perspectives.

[0078] After model deployment, a real-time monitoring mechanism is established to continuously track the model's performance in practical applications. Based on the monitoring results, any performance degradation or deviations in the model are promptly detected, and the model is automatically retrained or adjusted to ensure it maintains good performance. For example, if the prediction accuracy of the student profile model for a certain type of student is found to have significantly decreased, more up-to-date data on that type of student is automatically collected for model updates. Step S6. Generate a multi-dimensional profile based on the multi-dimensional index. Specifically: Based on S5 tracking, a multi-dimensional profile system is constructed for teachers, students, courses, and majors. As time progresses and teaching activities continue, data changes, and the profiles need timely updates. A mechanism for regular automatic profile updates is established to ensure that the profiles always reflect the latest teaching and learning situations. Simultaneously, during the update process, attention must be paid to the continuity and comparability of the data to avoid significant fluctuations or deviations in the profiles due to data updates. A commonly used method is K-Fold Cross-Validation, a classic method in machine learning for validating the generalization ability of a model. Its core principle is: 1. Divide the complete dataset into K mutually exclusive subsets (called "folds").

[0079] 2. Use K-1 folds as the training set each time, and use the remaining 1 fold as the validation set.

[0080] 3. Repeat the training-validation process K times, and finally take the average performance index as the model evaluation result.

[0081] This article will update K-Fold Cross Validation, using alternating K-fold cross-validation to verify the generalization ability of the model. The core principle is: 1. Divide the complete dataset evenly into K*K mutually exclusive subsets (called "folds").

[0082] 2. Use K*K-2 folds as the training set each time, and use the remaining 1 fold as the validation set.

[0083] 3. Use K*K-2 folds as the training set each time, and use the remaining 1 fold as the validation set.

[0084] 3. Repeat the training-validation process K*K times, and finally take the average performance index as the model evaluation result.

[0085] Example 2 This embodiment provides a multi-dimensional profile construction system based on teaching data index, including: The data acquisition module is used to extract raw data from systems related to teaching data and preprocess it to form a structured dataset, which includes teacher subjective rating data and objective behavioral data. The data correction module is used to construct a rational teacher model, correct the teacher's subjective rating to generate an objective rating, and merge the objective rating with the objective behavioral data to form a hybrid dataset. The feature processing module is used to perform feature dimensionality reduction on the data in the mixed dataset and extract key feature vectors; the index generation module is used to train a multilayer perceptron model based on the key feature vectors, configure dynamic batch processing, learning rate scheduling and regularization strategies, and train to generate a multidimensional index. Evaluation module. It is used to evaluate the generalization ability of the trained multilayer perceptron model, introduces the learning progress rate and the continuous improvement rate of teaching effect as indicators, and establishes a real-time monitoring and automatic update mechanism to update the multidimensional index synchronously. The portrait generation module is used to generate a multidimensional portrait based on the multidimensional index. It should be understood that any parts not described in detail in this specification belong to the prior art.

[0086] It should be understood that the above description of the preferred embodiments is quite detailed, but this should not be construed as limiting the scope of protection of this invention. It is neither necessary nor possible to exhaustively describe all possible implementations. Those skilled in the art, guided by this invention, can make substitutions or modifications without departing from the scope of the claims, all of which fall within the scope of protection of this invention. The scope of protection of this invention should be determined by the appended claims.

Claims

1. A method for constructing a multidimensional profile based on teaching data indices, characterized in that, Includes the following steps: Step S1. Extract raw data from the system related to teaching data and preprocess it to form a structured dataset, which includes teacher subjective rating data and objective behavior data; Step S2. Construct a rational teacher model, correct the teacher's subjective rating to generate an objective rating, and merge the objective rating with the objective behavioral data to form a hybrid dataset; Step S3. Perform feature dimensionality reduction on the data in the hybrid dataset and extract key feature vectors; Step S4. Based on the key feature vectors, a multilayer perceptron model is used for training, with dynamic batch processing, learning rate scheduling and regularization strategies configured, and a multidimensional exponent is generated during training. Step S5. Evaluate the generalization ability of the trained multilayer perceptron model, introduce the learning progress rate and the continuous improvement rate of teaching effect indicators, and establish a real-time monitoring and automatic update mechanism to update the multidimensional index synchronously; Step S6. Generate a multidimensional profile based on the multidimensional index.

2. The method for constructing a multi-dimensional profile based on teaching data index according to claim 1, characterized in that, The systems related to teaching data include teaching evaluation systems, academic affairs systems, online course learning systems, OA systems, library systems, student affairs management systems, enrollment and employment systems, and teaching and research systems.

3. The method for constructing a multi-dimensional profile based on teaching data index according to claim 1, characterized in that, The preprocessing in step S1 includes: cleaning, standardizing and uniformly identifying the raw data, wherein the uniform identification process includes: using data warehouse technology to achieve the fusion of multi-source heterogeneous data and establish a unified identification system; performing semantic annotation and structure transformation on unstructured data to generate standardized fields.

4. The method for constructing a multi-dimensional profile based on teaching data index according to claim 1, characterized in that, The construction of the rational teacher model in S2 includes: Based on the SVM double-blind peak-shaving comparison method, the teacher's subjective ratings were standardized and key features were extracted. The weight of each key feature is determined using the analytic hierarchy process (AHP) or expert scoring. Based on the SVM double-blind peak-shaving control method, the classification boundary is generated by utilizing teaching outcomes, teaching process, educational background and training experience, academic research results, teaching attitude, and work pressure coping ability. Cluster analysis was performed on the generated objective scores to remove extreme value data.

5. The method for constructing a multi-dimensional profile based on teaching data index according to claim 1, characterized in that, The feature dimensionality reduction in step S3 further includes: Dimensionality reduction of high-dimensional data in mixed datasets: Principal component analysis is used for initial dimensionality compression; t-SNE algorithm is combined to further reduce the dimensionality of nonlinear features, and cosine distance is used to optimize the similarity measure; sliding window statistics are performed on time series data to extract periodic features; Extracting key feature vectors: Ranking the importance of features in structured numerical data and retaining the Top-N features; using convolutional neural networks to extract teaching behavior features from unstructured data.

6. The method for constructing a multi-dimensional profile based on teaching data index according to claim 1, characterized in that, The generation of multidimensional indices in step S4 includes: Constructing a multilayer perceptron model based on key feature vectors: The input layer dimension matches the number of key features, and the hidden layer uses the Swish activation function. The output layer generates multidimensional indices, including teaching competence index, teaching digitization index, teacher-student interaction index, and learning challenge index. Configure dynamic training strategies, including dynamic batch processing, learning rate scheduling, and regularization strategies.

7. The method for constructing a multi-dimensional profile based on teaching data index according to claim 6, characterized in that, Step S4, training the multilayer perceptron model, further includes: A multi-task learning architecture is adopted to simultaneously optimize sub-tasks related to teaching ability, digital literacy, and scientific research level. Recurrent neural networks are used to extract temporal features from unstructured data.

8. The method for constructing a multi-dimensional profile based on teaching data index according to claim 7, characterized in that, Step S5 employs alternating K-fold cross-validation to evaluate the generalization ability of the trained multilayer perceptron model, specifically including: The key feature vectors are divided into K×K subsets. K×K-2 subsets are used for each training session, and the remaining 1 subset is used as the validation set. Introduce learning progress rate and continuous improvement rate of teaching effectiveness as supplementary evaluation indicators; The sensitivity of the model to the generation of different multidimensional indices was verified by dynamically adjusting the weights.

9. The method according to claim 1, characterized in that, The dynamic profile generation in step S6 further includes: dynamically configuring the weight ratios based on the generated multidimensional indices to generate a comprehensive multidimensional profile of teachers, students, courses and majors, and outputting the results through a visualization interface.

10. A multi-dimensional profile construction system based on teaching data index, characterized in that, include: The data acquisition module is used to extract raw data from systems related to teaching data and preprocess it to form a structured dataset, which includes teacher subjective rating data and objective behavioral data. The data correction module is used to construct a rational teacher model, correct the teacher's subjective rating to generate an objective rating, and merge the objective rating with the objective behavioral data to form a hybrid dataset. The feature processing module is used to perform feature dimensionality reduction on the data in the mixed dataset and extract key feature vectors. index The generation module is used to train a multilayer perceptron model based on the key feature vectors, configure dynamic batch processing, learning rate scheduling and regularization strategies, and train to generate multidimensional exponents. Evaluation module. It is used to evaluate the generalization ability of the trained multilayer perceptron model, introduces the learning progress rate and the continuous improvement rate of teaching effect as indicators, and establishes a real-time monitoring and automatic update mechanism to update the multidimensional index synchronously. The portrait generation module is used to generate a multidimensional portrait based on the multidimensional index. The multidimensional profile construction system based on teaching data index is used to perform the steps in the multidimensional profile construction method based on teaching data index as described in any one of claims 1-9.

Images