Classroom content evaluation system and method based on student questions and answers

Abstract

The invention discloses a classroom content evaluation system based on student questions and answers, and the system comprises a classroom information collection module, a data processing module, a classroom question generation module, a student answer interaction module, a semantic analysis module, a classroom evaluation generation module, a classified storage module, and a teaching strategy optimization module. The classroom information acquisition module monitors and acquires videos, audios, blackboard-writing texts and courseware images in real time, the data processing module converts multi-modal data into structured texts through voice recognition and OCR technologies, and the classroom question generation module generates a question set based on teaching content. The student answering interaction module receives answering data in a text, voice or image form, and the semantic analysis module analyzes teaching texts and answering contents and marks knowledge blind areas; the system has the advantages that multi-modal teaching content is integrated in real time, adaptability problems are dynamically generated, knowledge blind areas are accurately recognized, and the classroom effect is quantitatively evaluated in a multi-dimensional mode.

Description

Technical Field

[0001] This invention relates to the field of educational information technology, and in particular to a classroom content evaluation system and method based on student question-and-answer sessions. Background Technology

[0002] Current classroom teaching evaluation systems face numerous technical bottlenecks in practical applications: most systems rely on manual feedback after class through assignments, tests, or subjective observation, which is not only time-consuming and labor-intensive but also highly subjective, making it difficult to cover real-time dynamic data in the classroom. Existing systems can only collect single data such as student answers or class attendance, lacking the ability to integrate and structure multimodal teaching content such as video, audio, blackboard writing, and courseware in real time, resulting in fragmented teaching content analysis.

[0003] Traditional systems generate fixed question types, failing to dynamically adjust question difficulty and format based on teaching subjects and grade levels, thus hindering targeted teaching assessment. Semantic analysis of student descriptive answers remains at the level of simple keyword matching, lacking deep learning-based semantic understanding capabilities and failing to accurately identify logical errors and knowledge gaps in student responses. Regarding evaluation dimensions, existing systems primarily focus on the single indicator of knowledge point accuracy, lacking comprehensive quantitative assessments of multiple dimensions such as teaching pace, teacher-student interaction effectiveness, and content comprehension. System interaction design flaws prevent dynamic switching between open-ended and closed-ended question types based on student response characteristics, resulting in insufficient matching between questions and student learning. Furthermore, traditional teaching optimization schemes rely excessively on teacher experience and judgment, lacking intelligent analysis models based on multi-dimensional data such as student error patterns and interaction response times, making it difficult to provide teachers with quantitative suggestions for adjusting teaching content and methods, thus limiting the closed-loop optimization effect of the teaching process. Summary of the Invention

[0004] This invention addresses the shortcomings of existing technologies that make it difficult to provide teachers with quantitative suggestions for adjusting teaching content and methods by providing a classroom content evaluation system and method based on student questions and answers.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a classroom content evaluation system based on student question-and-answer sessions, comprising: The classroom information collection module is used to monitor and collect classroom teaching content in real time, including video, audio, blackboard text, and courseware images. The data processing module is connected to the classroom information collection module via a data interface. It is used to convert the collected audio into text through speech recognition, recognize the blackboard text through OCR technology, and extract the text content from the courseware images. The classroom question generation module is connected to the data processing module and is used to generate a collection of questions based on the processed classroom teaching content. The student response interaction module is used to receive students' text, voice, or image responses to a collection of questions; The semantic analysis module is used to analyze the semantics of classroom teaching content, and to provide semantic interpretation of students' descriptive answers in the question set. It also marks knowledge blind spots for knowledge points that students did not answer correctly by marking database fields or highlighting them in a visualization interface. The classroom evaluation generation module outputs a classroom evaluation report that includes knowledge blind spot markers, teaching pace scores, content comprehension indices, and classroom interaction scores. The student knowledge point mastery report generated based on the knowledge blind spot markers is sent to the teacher's terminal.

[0006] By adopting the above technical solutions, a fully automated classroom evaluation closed loop is achieved. Through multimodal acquisition of video / audio / blackboard writing / courseware and data processing using speech recognition / OCR, questions are automatically generated and student answers are analyzed. Semantic analysis is used to mark knowledge gaps, and evaluation reports with dimensions such as knowledge gap markings and teaching pace scores, as well as personalized knowledge point mastery reports, are output. This solves the problems of low efficiency and strong subjectivity of traditional manual evaluation, and improves the real-time and objectivity of teaching feedback.

[0007] The present invention is further configured such that: the set of questions includes a set of closed-ended questions and a set of open-ended questions; The student response interaction module prioritizes pushing open-ended question sets. When the number of characters in a student's answer is lower than a preset character threshold based on the teaching subject and grade level, it automatically switches to a closed-ended question set. The semantic analysis module extracts keywords from open-ended answers and dynamically adjusts the content of the closed-ended question set based on the TF-IDF weights of the keywords and the cosine similarity calculation results of the closed-ended question set. When the cosine similarity is lower than the preset threshold of 0.6, a question with higher similarity is selected from the closed-ended question set to replace the current question.

[0008] By adopting the above technical solutions, the question interaction logic is optimized, open-ended questions are prioritized and closed-ended questions are automatically switched according to the number of characters in the answer. The question content is dynamically adjusted by combining TF-IDF keyword weights and cosine similarity of the closed-ended question library, which solves the problems of low student participation and poor question suitability, and improves the quality of answers and the accuracy of assessment.

[0009] The present invention is further configured such that the classroom question generation module includes: The teaching attribute recognition unit is used to automatically identify the teaching subject and grade level based on the teaching content in the classroom. The knowledge point identification unit is used to identify knowledge points in the teaching content. The question generation unit is used to generate questions based on teaching objectives and knowledge points.

[0010] By adopting the above technical solutions, the relevance of question generation is enhanced. By identifying teaching subjects, grade levels, and knowledge points, questions are automatically generated in conjunction with teaching objectives, avoiding biases from manual design and ensuring a strong correlation between questions and teaching content, thus supporting personalized assessment and customized learning paths.

[0011] The present invention is further configured to include a classification storage module, which is connected to the data processing module and the classroom question generation module, and is configured with a classification storage repository for classifying and storing the question collection according to teaching subjects, grade levels and knowledge points.

[0012] By adopting the above technical solutions, a structured question storage mechanism is established, which stores question sets according to subject / grade / knowledge point, improves data reuse rate and retrieval efficiency, reduces the overhead of repeated generation, and provides a data foundation for long-term teaching analysis.

[0013] The present invention is further configured such that the semantic analysis module includes: The text preprocessing unit is used to perform word segmentation and part-of-speech tagging on classroom teaching texts and student responses; The semantic understanding unit is used to analyze the logical structure of text through a semantic model based on the BERT architecture; The key information extraction unit is used to generate keyword vectors from the answers to a collection of open-ended questions.

[0014] By adopting the above technical solutions, semantic analysis capabilities are enhanced. By parsing the logical structure of text through word segmentation, part-of-speech tagging, and the BERT architecture model, keyword vectors are extracted from open-ended answers, solving the problem of superficiality in traditional text analysis and improving the accuracy and depth of knowledge blind spot labeling.

[0015] The present invention is further configured such that: the semantic analysis module also includes a student answer analysis unit, connected to the classification and storage module, used to calculate the error rate of question answers generated based on knowledge points, wherein: Error rate = Number of knowledge-based incorrect answers / (Total number of answers - Number of unanswered questions). The number of knowledge-based incorrect answers refers to the number of answers that are judged to be incorrect based on the standard answer for the knowledge point.

[0016] By adopting the above technical solutions, the degree of mastery of knowledge points is quantified, and standardized evaluation indicators are generated through error rate formulas to eliminate subjective scoring bias and provide objective basis for identifying individual and group weaknesses.

[0017] The present invention is further configured such that the classroom evaluation generation module includes: The summary unit, connected to the student answer analysis unit, is used to generate statistical data on the mastery of knowledge points in the whole class based on the error rate of each student's knowledge points. The statistical data includes the class average error rate, standard deviation of the error rate, and proportion of students who have mastered each knowledge point. The report generation unit is used to associate the statistical data on the mastery of knowledge points of the whole class with the knowledge point mastery report of individual students, mark students whose error rate is higher than the average of the whole class and common knowledge blind spots whose error rate of the whole class exceeds a preset ratio, and generate a visual student knowledge point mastery report including a knowledge point mastery heatmap.

[0018] By adopting the above technical solutions, data-driven visualization reports can be generated, summarizing statistical data such as the average error rate and standard deviation of knowledge points for the whole class, linking individual reports and marking students with high error rates and common blind spots, and presenting teaching weaknesses intuitively through heat maps to assist teachers in precise intervention.

[0019] The present invention is further configured such that the classroom evaluation generation module also includes: The feature extraction unit extracts problem feature vectors containing common knowledge gaps, student error rates, and interaction response times; The evaluation rule matching unit is used to match the problem feature vector with the teaching syllabus evaluation rule library pre-stored in the system memory; The report generation unit generates a classroom evaluation report based on the results of the feature extraction unit and the evaluation rule matching unit.

[0020] By adopting the above technical solutions, we can ensure the compliance of evaluations, extract feature vectors such as common blind spots and error rates to match the teaching syllabus rule base, generate standardized reports, solve the problem of the disconnect between evaluation and teaching objectives, and improve the authority of reports and their alignment with teaching.

[0021] The present invention is further configured such that: the system also includes a teaching strategy optimization module, connected to the classroom evaluation generation module, comprising: The data receiving unit is used to receive the classroom evaluation report and the student knowledge point mastery report; The analysis and optimization unit analyzes classroom evaluation reports and student knowledge mastery reports based on a neural network model trained with supervised learning, and generates improvement plans, including suggestions for adjusting teaching content and optimization plans for explanation speed. The strategy feedback unit automatically feeds back the improvement plan to the question generation unit of the classroom question generation module through the data interface; The teacher interaction unit pushes improvement plans and knowledge point mastery reports to the teacher's terminal.

[0022] By adopting the above technical solutions, a closed-loop optimization mechanism is constructed. Based on the analysis report of the supervised learning neural network, improvement plans such as content adjustment suggestions are generated and automatically fed back to the problem generation module and pushed to the teacher's terminal, realizing automatic iteration from evaluation to optimization and continuously improving teaching quality.

[0023] A classroom content assessment method based on student question-and-answer sessions includes the following steps: S1: The classroom teaching content is monitored and collected in real time through the classroom information collection module. The teaching content includes video, audio, blackboard text and courseware images. S2: The data processing module is used to convert the collected audio into text through speech recognition, and the OCR technology is used to recognize the text written on the blackboard and extract the text content in the courseware images to form structured teaching text data. S3: The classroom question generation module identifies the teaching subject, grade level, and knowledge point based on the teaching content. It combines the structured teaching objective data pre-stored in the system database or input by the teacher's terminal to generate a question set containing closed-ended question sets and open-ended question sets. The questions are then stored in a classified storage repository according to the teaching subject, grade level, and knowledge point. S4: The student response interaction module prioritizes pushing open-ended question sets. When the number of characters in a student's response is lower than a preset value, it automatically switches to closed-ended question sets to receive the student's text, voice, or image responses. S5: The semantic analysis module performs word segmentation and part-of-speech tagging on teaching texts and student answers. It analyzes the logical structure of the text using a pre-trained model for the education domain based on the BERT-base architecture and extracts keyword vectors from open-ended answers. S6: The student answer analysis unit generates questions and answers based on knowledge points, calculates the error rate of individual students' knowledge points, and summarizes and generates the knowledge point statistics of the whole class; the classroom evaluation generation module links the knowledge point statistics of the whole class with the individual student reports, marks students whose error rate is higher than the average of the whole class and common knowledge blind spots whose error rate of the whole class exceeds the preset ratio, and generates a visual student knowledge point mastery report that includes a knowledge point mastery heatmap. S7: Extract problem feature vectors containing knowledge point coverage, student error rate, and interaction response time, match them with the pre-stored teaching syllabus evaluation rule base, and generate a classroom evaluation report containing knowledge blind spot markers, teaching pace scores, content comprehension index, and classroom interaction scores. S8: The teaching strategy optimization module analyzes classroom evaluation reports and student knowledge mastery reports based on a supervised learning neural network model, and generates improvement plans such as suggestions for adjusting teaching content and optimizing explanation speed. S9: The teaching strategy optimization module automatically feeds back the improvement plan to the classroom question generation module through the data interface and pushes it to the teacher's terminal.

[0024] By adopting the above technical solutions, an end-to-end automated process is defined, from content collection, question generation, and answer analysis to report generation and strategy optimization. By integrating system functions through standardized steps, the inefficiency of manual processes is solved, maximizing the real-time performance and scalability of classroom assessment.

[0025] The present invention, by adopting the above technical solutions, has significant technical effects: The classroom content evaluation system and method based on student question and answer provided in this application includes a classroom information collection module that collects multimodal teaching data in real time, a data processing module that performs structured processing, a classroom question generation module that dynamically generates appropriate questions, a semantic analysis module that accurately identifies knowledge gaps, and a classroom evaluation generation module that outputs a multidimensional quantitative evaluation report. It has the advantages of real-time integration of multimodal teaching content, dynamic generation of appropriate questions, accurate identification of knowledge gaps, and multidimensional quantitative evaluation of classroom effectiveness. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the classroom content evaluation system. Detailed Implementation

[0027] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0028] Example: Existing classroom teaching evaluation systems generally adopt manual after-class feedback and rely on homework correction or subjective observation to obtain teaching effectiveness data. Traditional systems have insufficient multimodal data integration capabilities and cannot process heterogeneous teaching content such as videos, audios, blackboard writing, and courseware in real time. Due to the lack of a structured processing mechanism, the analysis of teaching content is fragmented, resulting in low matching degree between problem generation and grade level. Existing semantic analysis methods are limited to keyword matching and have difficulty identifying knowledge gaps. The evaluation dimensions are single and lack visualization, making it difficult for teachers to identify common learning problems in a timely manner.

[0029] To address the aforementioned issues, it is necessary to construct a processing mechanism capable of integrating multi-source teaching data in real time, establish a dynamic question generation and intelligent response analysis system, consider setting up multimodal data acquisition devices to address the problem that traditional systems cannot capture dynamic classroom information, design text processing units with logical structure parsing capabilities to overcome the deficiency of insufficient semantic analysis depth, and construct a comprehensive evaluation model covering knowledge mastery, teaching pace, and interaction effectiveness to improve the completeness of evaluation dimensions.

[0030] This application proposes a system comprising a classroom information acquisition module, a data processing module, a classroom question generation module, a student response interaction module, a semantic analysis module, and a classroom evaluation generation module. The classroom information acquisition module monitors and acquires video, audio, blackboard text, and courseware images in real time. The data processing module converts multimodal data into structured text using speech recognition and OCR technologies. The classroom question generation module generates a set of questions based on the teaching content. The student response interaction module receives response data in text, voice, or image format. The semantic analysis module parses the teaching text and response content, identifies knowledge gaps, and the classroom evaluation generation module outputs a visual report containing multidimensional indicators.

[0031] The classroom information acquisition module refers to a hardware combination capable of simultaneously capturing multiple teaching media. Specifically, it can be implemented using a multi-channel video capture card in conjunction with an array microphone to ensure complete recording of the teaching process. The data processing module refers to a heterogeneous data conversion unit, which can be implemented using a speech recognition engine and an image OCR processing chip to convert unstructured data into analyzable text. The classroom question generation module refers to an automated test question generation system, which can be implemented using a natural language processing model combined with a knowledge point graph to dynamically generate a set of questions that match the teaching content. The student response interaction module refers to a multi-channel response receiving device, which can be implemented using a touch screen and a voice input component to support data acquisition through multiple interaction methods. The semantic analysis module refers to a text logic parsing unit, which can be implemented using a deep learning model combined with a knowledge base matching algorithm to identify conceptual biases in responses. The classroom evaluation generation module refers to a data analysis and visualization component, which can be implemented using a data statistics engine in conjunction with a chart generation tool to integrate multi-dimensional indicators to form an evaluation report.

[0032] The video stream generated during the teaching process is used to extract the blackboard images through frame capture technology. The audio stream is converted into text records after noise reduction. Key content is extracted from the courseware images through feature recognition. The structured teaching text is input into the question generation engine, which generates a combination of open-ended and closed-ended questions based on knowledge point association rules. Students input text answers or provide voice feedback through touch terminals. The response data is preprocessed and input into a semantic analysis model. This model identifies logical breaks in the answers through context association algorithms and maps incorrect knowledge points to corresponding items in the teaching syllabus. The evaluation report automatically integrates the class error rate distribution, interactive response time series, and knowledge point coverage indicators to generate a visual heat map and optimization suggestions.

[0033] This solution breaks through the processing limitations of a single data source, realizing the real-time synchronous acquisition of multimodal teaching data. Traditional systems cannot dynamically adjust question types, while this solution achieves intelligent switching between open-ended and closed-ended questions through response feature analysis. Existing technologies lack deep semantic parsing capabilities, while this solution uses logical structure analysis methods to accurately locate knowledge blind spots. Traditional evaluation reports only reflect accuracy indicators, while this solution constructs a multi-dimensional evaluation system covering teaching pace and interaction effects.

[0034] The comprehensive collection and structured processing of classroom teaching data solves the technical problem of integrating multimodal information. Through dynamic question generation and intelligent response analysis mechanisms, the real-time performance and accuracy of teaching feedback are improved. The application of semantic parsing models shifts the identification of knowledge blind spots from surface keyword matching to in-depth logical analysis, helping teachers quickly locate weak links in teaching. The introduction of multidimensional evaluation indicators constructs a three-dimensional classroom evaluation system, providing data support for optimizing teaching strategies.

[0035] The question set includes a closed-ended question set and an open-ended question set. The student answer interaction module prioritizes pushing open-ended question sets. When the number of characters in a student's answer is lower than a preset character threshold based on the teaching subject and grade level, it automatically switches to the closed-ended question set. The semantic analysis module extracts keywords from the open-ended answers and dynamically adjusts the content of the closed-ended question set based on the TF-IDF weights of the keywords and the cosine similarity calculation results of the closed-ended question set. When the cosine similarity is lower than a preset threshold, for example, a preset threshold of 0.6, a question with higher similarity is selected from the closed-ended question set to replace the current question.

[0036] The closed-ended question set refers to a collection of questions with fixed answers, such as multiple-choice and true / false questions. This can be implemented using pre-set answer templates and automatic scoring algorithms to quickly assess students' mastery of basic knowledge points. The open-ended question set refers to a collection of open-ended questions requiring students to elaborate or explain. This can be generated using natural language generation models combined with the curriculum syllabus to deeply examine students' logical thinking and knowledge application abilities. The character threshold refers to the minimum answer length standard set according to the complexity of different subjects and the age group of students. This can be set through historical answer data statistics and teaching expert experience values ​​to determine whether students have sufficient expression ability. TF-IDF weight refers to calculating the importance of keywords in the answer using the term frequency-inverse document frequency algorithm. This can be achieved by using text mining tools to extract features from open-ended answers to quantify the core content of students' answers. Cosine similarity refers to calculating the degree of matching between students' answers and the closed-ended question database using a vector space model. This can be achieved by using numerical calculation methods to compare the similarity of keyword vectors to dynamically optimize the question recommendation strategy.

[0037] When the system detects that the number of characters in a student's open-ended answer is lower than the preset threshold for the corresponding subject, it automatically switches the interactive interface to a collection of closed-ended questions to avoid invalid answers due to insufficient expression ability. After extracting keywords from the open-ended answer, the semantic analysis module calculates its cosine similarity with each question in the closed-ended question library. If the highest similarity is lower than 0.6, it is determined that the current question is not well matched, and questions with higher similarity are automatically selected from the question library for replacement. This process dynamically adjusts the difficulty and relevance of questions by analyzing the quality of student answers in real time to ensure that subsequent questions can accurately cover the student's knowledge gaps.

[0038] Traditional systems cannot automatically switch question types based on the quality of students' real-time responses, resulting in a large amount of invalid data generated by students with weaker expressive abilities in open-ended question sections. Existing technologies use a fixed question database push mode, which lacks a dynamic optimization mechanism based on semantic similarity, making it difficult to guarantee the matching accuracy between questions and knowledge gaps. This solution triggers question type switching through character thresholds and combines TF-IDF and cosine similarity analysis to achieve adaptive adjustment of the question push strategy.

[0039] This application addresses the shortcomings of traditional systems that cannot dynamically adjust question types based on the quality of student responses. It avoids data noise caused by insufficient student expression in open-ended question sections. Through semantic similarity calculation and dynamic question database updates, it ensures that subsequent questions accurately target knowledge points that students have not mastered, improving the efficiency of identifying knowledge gaps. Based on character thresholds set for teaching subjects and grade levels, it effectively distinguishes differences in students' cognitive levels and provides suitable question interaction modes for students of different ability levels.

[0040] The classroom question generation module includes a teaching attribute identification unit, a knowledge point identification unit, and a question generation unit. The teaching attribute identification unit is used to automatically identify the teaching subject and grade level based on the classroom teaching content; the knowledge point identification unit is used to identify the knowledge points in the teaching content; and the question generation unit is used to generate questions based on the teaching objectives and knowledge points.

[0041] The teaching attribute identification unit refers to analyzing subject-specific vocabulary and grade-level-appropriate terms in teaching texts using natural language processing technology. Specifically, it can be implemented using a rule-based text classifier, such as matching the teaching content with a pre-set subject keyword database to automatically determine the subject category and applicable grade level of the current class. The knowledge point identification unit refers to extracting core concepts and relationships from structured teaching texts. Specifically, it can be implemented using a deep learning-based keyword extraction model, such as using the BERT model to encode teaching texts and identify high-frequency terms and their contextual relationships to form a knowledge point graph. The question generation unit refers to constructing a set of questions based on the correlation between teaching objectives and knowledge points. Specifically, it can be implemented using template-based automatic question-and-answer generation technology, such as calling a pre-set question template based on the knowledge point type and combining the characteristics of the teaching subject to generate different forms of questions, such as multiple-choice questions and short-answer questions.

[0042] The teaching attribute identification unit analyzes the frequency distribution of professional terms in teaching texts, such as detecting subject-specific words like "equation" and "geometric shape" in mathematics classes, and combines them with grade-level appropriate words such as "beginner" and "advanced" to determine the subject attributes and applicable grade level of the current teaching content. The knowledge point identification unit performs semantic analysis on the pre-processed text data to identify core concept nodes with teaching value and their related paths. For example, it extracts knowledge points such as "Newton's laws" and "action and reaction forces" and their logical relationships in physics courses. The question generation unit calls the corresponding question template library according to the identified knowledge point type. For example, it generates concept discrimination questions for definition-type knowledge points and calculation application questions for formula-type knowledge points. At the same time, it adjusts the question expression method according to the characteristics of the teaching subject to ensure that the generated questions are highly consistent with the teaching objectives.

[0043] Existing systems rely on manually pre-set question banks for question generation and cannot adapt to dynamic teaching content. This solution automatically identifies teaching attributes and knowledge points, and can match the current teaching progress and subject characteristics in real time. In existing technologies, fixed question banks are difficult to cover the differences in knowledge depth across different learning stages. This solution dynamically adjusts the difficulty gradient of questions based on the learning stage identification results. For example, it generates questions of different complexity for the same knowledge points in junior high school physics and senior high school physics.

[0044] The dynamic adaptation of problem generation to teaching objectives solves the technical shortcomings of traditional systems in terms of the relevance of problems. By automatically identifying the characteristics of teaching subjects and grade levels, it ensures that the generated problems are consistent with the requirements of the curriculum outline. The problem generation mechanism based on knowledge point graphs effectively improves the accuracy of the correlation between problems and teaching focus, avoiding the problem of incomplete knowledge point coverage in manually pre-set question banks.

[0045] The classification storage module, connected to the data processing module and the classroom question generation module, is configured with a classification repository for classifying and storing question sets according to teaching subjects, grade levels, and knowledge points. This module is responsible for the structured storage of the generated question sets, and can be implemented using a distributed database system or a relational database. It logically isolates teaching data with different attributes. The classification repository is a container for storing the classified question sets, and can be implemented using MySQL or MongoDB. It enables fast data retrieval through preset storage paths and indexing rules. The classification method based on teaching subjects, grade levels, and knowledge points refers to dividing the question set storage hierarchy according to the attributes of the teaching content. This can be achieved through preset tags or metadata fields, such as setting the tag "MATH" for mathematics and "JUNIOR" for junior high school.

[0046] After the data processing module converts classroom teaching content into structured text data, the classroom question generation module generates corresponding question sets based on the identified teaching subjects, grade levels, and knowledge points. The classification and storage module receives the question set data and establishes a multi-level directory structure through preset subject tags, grade level tags, and knowledge point tags. The question sets are then stored hierarchically in the classification and storage repository. When the semantic analysis module needs to call historical question data for error rate statistics, the classification and storage module quickly locates the corresponding storage location based on the subject, grade level, and knowledge point tags in the query conditions and extracts the relevant data for analysis.

[0047] Traditional systems do not categorize question collections according to teaching attributes, resulting in low data retrieval efficiency and an inability to dynamically adjust question types. This solution establishes a categorized storage mechanism, enabling question collections to match subject knowledge systems with teaching subjects, differentiate difficulty levels by grade level, and associate teaching content with knowledge points. This improves the efficiency of question data retrieval and supports the subsequent semantic analysis module in accurately locating common knowledge gaps.

[0048] This application further addresses the technical problems of low retrieval efficiency and inability to dynamically adjust question types caused by the chaotic storage of question collections in existing technologies. It achieves structured storage and rapid retrieval of question data, providing reliable data support for the subsequent generation of student knowledge mastery reports.

[0049] The semantic analysis module includes a text preprocessing unit, a semantic understanding unit, and a key information extraction unit. The text preprocessing unit performs word segmentation and part-of-speech tagging on classroom teaching texts and student answers. The semantic understanding unit analyzes the logical structure of the text using a semantic model based on the BERT architecture. The key information extraction unit generates keyword vectors from the answers to the set of open-ended questions.

[0050] The text preprocessing unit refers to the component that performs basic language processing on the teaching text. Specifically, it can use Chinese word segmentation tools such as Jieba to perform word segmentation operations and part-of-speech tagging tools such as LTP to perform grammatical tagging, providing structured input for subsequent semantic analysis. The semantic understanding unit refers to the module that parses the deep logical relationships in the text. Specifically, it can use a pre-trained model based on the BERT-base architecture, and by fine-tuning and adapting it to the text features of the education field, it can realize the semantic association mapping between teaching knowledge points and student answers. The key information extraction unit refers to the device that captures core concepts from unstructured answers. Specifically, it can use the TF-IDF algorithm combined with word vector clustering technology to generate a set of keyword vectors to represent the core semantic content of student answers.

[0051] The text preprocessing unit performs unified word segmentation on the transcribed classroom teaching text and student response text, eliminates stop word interference, and completes part-of-speech tagging to form a structured text data stream. The semantic understanding unit receives the preprocessed text and captures the logical connections between teaching knowledge points through the multi-layer attention mechanism of the BERT model, establishing a semantic network of knowledge points. The key information extraction unit calculates the TF-IDF weights of keywords for open-ended responses and generates high-dimensional vector representations. Through vector space mapping, it identifies the core concepts in the responses. Traditional systems only use simple keyword matching for semantic analysis, which cannot capture the logical relationships between teaching knowledge points. This solution uses the BERT model to parse the deep semantic structure of the text and combines it with keyword vector extraction technology to accurately identify semantic fragments in student responses that deviate from standard knowledge points.

[0052] The deep semantic analysis of teaching texts and student responses, through structured processing and vectorized representation, effectively solves the problem of fragmented semantic analysis in traditional systems. This solution can accurately locate knowledge comprehension deviations in student responses, providing reliable data support for subsequent knowledge blind spot labeling.

[0053] The semantic analysis module also includes a student answer analysis unit, which is connected to the classification and storage module. It is used to calculate the error rate of questions and answers generated based on knowledge points. The error rate is equal to the number of knowledge-based incorrect answers divided by the total number of answers minus the number of unanswered questions. The number of knowledge-based incorrect answers is the number of answers judged as incorrect based on the standard answer of the knowledge point.

[0054] The student answer analysis unit is a module that automatically counts student answers using algorithms. Specifically, it can interface with a database query interface and a classification storage module to obtain answer data for each knowledge point in real time. The number of knowledge-based incorrect answers refers to the number of answers determined to be incorrect after comparison with the standard answer database for each knowledge point using a semantic matching algorithm. This can be achieved using semantic similarity calculation methods from natural language processing. The total number of answers refers to the total number of valid answers submitted by all students for a specific knowledge point. The number of unanswered answers refers to the number of answers not submitted within the specified time or submitted blank answers. The student answer analysis unit obtains student answer data for each knowledge point through the classification storage module. First, it filters out invalid data from unanswered answers. Then, it performs semantic matching between valid answers and pre-stored standard answers. For closed-ended questions, it directly determines correctness by comparing answer options; for open-ended answers, it uses semantic vector similarity calculation to determine whether the answer deviates from the core content of the knowledge point. Error rate calculation eliminates the interference of unanswered data on the statistical results, reflecting only the actual answer performance and the degree of knowledge mastery. The analysis results are updated in real time to the classification storage module, which can then be used by the classroom evaluation generation module to generate visual reports.

[0055] Traditional systems only count the correctness of answers without distinguishing between unanswered and incorrect answers, making it difficult for teachers to accurately identify students' true knowledge gaps. This solution uses an error rate calculation model to separate unanswered data, focusing on knowledge-based errors in actual answers. Combined with a standard answer database of knowledge points, it automates error detection, avoiding the subjective bias of manual grading, and providing accurate data support for the subsequent generation of common knowledge gap markers.

[0056] This application can dynamically quantify the distribution of students' errors for each knowledge point, accurately identify high-frequency error knowledge points, help teachers quickly locate weak links in teaching, and by eliminating the interference of unanswered data, the error rate calculation results more realistically reflect the students' actual knowledge mastery level, providing a reliable basis for adjusting personalized teaching strategies.

[0057] The classroom assessment generation module includes a summary unit and a report generation unit. The summary unit is connected to the student answer analysis unit. It generates statistical data on the mastery of knowledge points for the whole class based on the error rate of each student's knowledge points. The statistical data includes the class average error rate, the standard deviation of the error rate, and the proportion of students who have mastered each knowledge point. The report generation unit links the statistical data on the mastery of knowledge points for the whole class with the individual student's knowledge point mastery report. It marks students whose error rate is higher than the class average and common knowledge blind spots where the class error rate exceeds a preset ratio. For example, when common knowledge blind spots where the class error rate exceeds 30% are used, a visual student knowledge point mastery report with a knowledge point mastery heatmap is generated.

[0058] The statistical data on the mastery of knowledge points in the whole class refers to the quantitative indicator formed by calculating the error rate distribution of the student group for each knowledge point. Specifically, the average error rate can be calculated using a weighted average algorithm, such as multiplying the error rate of each student by the weight coefficient of the corresponding knowledge point and then summing them up to obtain the average. The standard deviation can be used to reflect the dispersion of the error rate distribution through variance calculation. The proportion of students who have mastered the knowledge can be automatically calculated by setting an error rate threshold. These indicators are used to help teachers quickly identify the weak points of the class as a whole.

[0059] Association refers to mapping and comparing the error rate data of an individual student with the statistical data of the class. Specifically, it can be done by using data table association or API interface calls to bind the student ID with the error rate data of knowledge points. Through database queries, the data of individuals and groups can be displayed synchronously. This operation allows teachers to observe the difference in knowledge mastery between specific students and the class as a whole at the same time.

[0060] Common knowledge blind spots refer to the set of knowledge points where the error rate of the whole class exceeds a preset threshold, such as 30%. By traversing the error rate data of all knowledge points, the results that exceed the threshold can be filtered out and marked in the visualization interface through color marking or pop-up prompts. This marking method can intuitively present the common weak knowledge points that need to be emphasized.

[0061] A visualized student knowledge mastery report refers to a data analysis result that presents knowledge mastery in a graphical form. Specifically, it can be in the form of a heatmap, which maps knowledge points to different colored blocks according to their error rates. For example, red represents a high error rate and green represents a low error rate. The heatmap can be generated using Python's Matplotlib library or JavaScript's D3.js library, and supports dynamic interaction and multi-dimensional data switching display.

[0062] Specifically, after receiving error rate data from the student answer analysis unit, the aggregation unit first stores it in the database according to knowledge points. For each knowledge point, it calculates the arithmetic mean of all students' error rates as the class average error rate, and calculates the standard deviation to reflect the dispersion of the error distribution. The percentage of students with mastery is obtained by statistically analyzing the percentage of students with error rates below a preset passing threshold, for example, a preset passing threshold of 20%. After the report generation unit calls these statistical data, it compares each student's individual error rate with the class average for the corresponding knowledge point. When a student's error rate exceeds the average, a special marker is added to the report. At the same time, it iterates through all knowledge points and filters out items with a class error rate exceeding 30% as common blind spots, which are displayed as highlighted blocks on the heatmap. The final generated visualization report includes interactive charts, allowing teachers to click and view specific error distribution details.

[0063] Traditional systems only count students' answer accuracy and lack comparative analysis at the class level, failing to distinguish between individual differences and common group problems. This solution establishes a class statistical indicator system and combines it with a dynamic threshold judgment mechanism to automatically identify common teaching blind spots that need to be addressed first. At the same time, through data association technology, it simultaneously presents the difference between each student's report and the class average, solving the problems of isolated individual data and lack of reference benchmarks in traditional reports. The application of visual heatmaps replaces the original tabular data display method, significantly improving the efficiency of identifying weak teaching links.

[0064] The automated statistical analysis of class knowledge mastery helps teachers quickly locate frequently missed knowledge points. The marking function of common blind spots provides data support for adjusting teaching focus, avoiding reliance on subjective experience. The visual report presents the error distribution intuitively through heatmaps, reducing the time cost for teachers to analyze teaching effectiveness. The correlation comparison between individual and group data provides a basis for implementing differentiated teaching and supports teachers in developing personalized tutoring plans for students with error rates higher than the average.

[0065] The classroom evaluation generation module also includes a feature extraction unit and an evaluation rule matching unit. The feature extraction unit is used to extract problem feature vectors containing common knowledge gaps, student error rates, and interactive response times. The evaluation rule matching unit is used to match the problem feature vectors with the teaching syllabus evaluation rule library pre-stored in the system memory. The report generation unit generates a classroom evaluation report based on the results of the feature extraction unit and the evaluation rule matching unit.

[0066] The problem feature vector refers to a data structure that transforms multi-dimensional classroom evaluation indicators into numerical representations. Specifically, it can be implemented using a vector space model, combining data such as the frequency of occurrence of common knowledge gaps, the percentage of student error rates, and millisecond-level interaction response times into multi-dimensional vectors. This is used to quantify key features that characterize classroom teaching effectiveness. The teaching syllabus evaluation rule base refers to a database of evaluation indicators established according to the subject curriculum standards. Specifically, it can be implemented by parsing the teaching syllabus document to extract evaluation rules such as knowledge point weights and interaction frequency requirements, and using a relational database to store the mapping relationship between knowledge points and evaluation indicators. This provides a standardized reference for feature vector matching. Evaluation rule matching refers to the process of associating the real-time extracted feature vectors with preset rules. Specifically, it can be implemented by using a cosine similarity algorithm to calculate the matching degree between the feature vectors and the standard vectors in the rule base, or by executing preset logical judgment conditions through a rule engine. This is used to determine whether the classroom teaching effectiveness meets the requirements of the teaching syllabus.

[0067] During classroom teaching, the feature extraction unit continuously receives common knowledge blind spot marker data output by the semantic analysis module, error rate data calculated by the student answer analysis unit, and interactive response time data recorded by the student answer interaction module. After normalization, these data are arranged and combined according to preset vector dimensions to form a problem feature vector. The evaluation rule matching unit calls the subject knowledge point attainment standards, interactive response time thresholds, and other evaluation rules stored in the teaching syllabus evaluation rule base. It calculates the similarity between the real-time generated problem feature vector and the standard vector in the rule base. When the student error rate dimension in the feature vector exceeds the threshold set by the rule base, the report generation unit automatically triggers the early warning mechanism and marks the teaching effect of the knowledge point as not meeting the standard in the classroom evaluation report. At the same time, the interactive response time dimension data is compared with the classroom rhythm standard in the rule base to generate a teaching rhythm scoring index.

[0068] Existing classroom evaluation systems can only assess a single dimension based on the accuracy of answers, lacking comprehensive analysis of composite indicators such as the timeliness of interaction and the distribution of knowledge gaps. This solution constructs a multi-dimensional question feature vector to quantitatively integrate indicators such as teaching pace, knowledge mastery, and interaction efficiency. Combined with the teaching syllabus rule base, it achieves multi-dimensional cross-validation. In existing technologies, manual evaluation rule matching requires teachers to manually compare with the teaching syllabus, while this solution, through an automated vector matching mechanism, can detect deviations between teaching effectiveness and curriculum standards in real time.

[0069] The multi-dimensional quantitative evaluation and standardized rule matching of classroom teaching effectiveness, by extracting composite feature vectors containing timeliness indicators and knowledge mastery levels, solves the technical shortcomings of traditional systems with single evaluation dimensions. Combined with the automatic matching mechanism of the teaching syllabus rule base, it can accurately identify the degree of deviation between teaching effectiveness and curriculum standards, providing data support for generating comprehensive evaluation reports that include pace scores and comprehension indices. The dynamic matching mechanism between feature vectors and the rule base enables the system to adapt to the differences in teaching standards across different subjects and grade levels, improving the objectivity and relevance of the evaluation reports.

[0070] The system includes a teaching strategy optimization module, which is connected to the classroom evaluation generation module. This module comprises a data receiving unit, an analysis and optimization unit, a strategy feedback unit, and a teacher interaction unit. The data receiving unit receives classroom evaluation reports and student knowledge mastery reports. The analysis and optimization unit analyzes the reports based on a supervised learning-trained neural network model and generates improvement plans. The strategy feedback unit feeds the improvement plans back to the problem generation unit through a data interface. The teacher interaction unit pushes the improvement plans and reports to the teacher's terminal.

[0071] The neural network model for supervised learning training refers to a model trained by mapping labeled teaching effectiveness data to optimization strategies. Specifically, it can be implemented using a multilayer perceptron or convolutional neural network architecture to establish a mapping relationship between classroom evaluation indicators and teaching strategy adjustments. The data interface refers to the communication protocol that enables data transmission between modules. Specifically, it can be implemented using an HTTP interface or WebSocket protocol to ensure the automatic flow of improvement plans within the system. The teacher terminal refers to the electronic device used by the teacher. Specifically, it can be implemented using a computer or mobile terminal with a dedicated client installed to receive optimization suggestions generated by the system in real time.

[0072] After the data receiving unit obtains the classroom evaluation report, the analysis and optimization unit inputs the data such as knowledge blind spot markings and interaction scores from the report into the neural network model. The model calculates the teaching pace adjustment coefficient through the pre-trained weight matrix and generates an optimization plan for the explanation speed based on the error rate distribution. The strategy feedback unit transmits the generated adjustment suggestions to the question generation module through the API interface, triggering the weight update mechanism of the question library. Simultaneously, the teacher interaction unit converts the key knowledge point reinforcement suggestions in the optimization plan into visual charts and pushes them to the message queue on the teacher's terminal.

[0073] Compared with existing technologies, traditional teaching optimization relies on teachers manually collecting error rate data and adjusting lesson plans based on experience, which cannot link the dynamic relationship between classroom interaction pace and knowledge mastery. This solution automatically mines multi-dimensional features in the evaluation report through a supervised learning model. For example, it incorporates the correlation between frequently erroneous knowledge points and corresponding teaching time into the analysis, achieving accurate matching between teaching strategies and real-time learning. Compared with static optimization systems based on rule bases, neural network models can capture the non-linear relationship between teaching speed and average class comprehension.

[0074] The system automatically transforms classroom evaluation data into actionable improvement plans through dynamic closed-loop optimization of teaching strategies. For example, it generates specialized training question sets for knowledge points with an error rate exceeding 30%, and recommends speech rate adjustment ranges based on the distribution of interactive response time. Teachers receive optimization suggestions in real time, avoiding the response lag problem of traditional manual analysis, and enabling teaching adjustments to accurately match the learning characteristics of the class.

[0075] The student question-and-answer-based classroom content evaluation method includes the following steps: A classroom information acquisition module monitors and collects classroom teaching content in real time, including video, audio, blackboard text, and courseware images; a data processing module converts the collected audio into text through speech recognition, uses OCR technology to recognize the blackboard text, and extracts text content from courseware images to form structured teaching text data; a classroom question generation module identifies the teaching subject, grade level, and knowledge points based on the teaching content, and combines this with pre-stored structured teaching objective data from the system database or input by the teacher's terminal to generate a question set containing both closed-ended and open-ended questions, which are then categorized and stored in a classified repository according to teaching subject, grade level, and knowledge points; a student response interaction module prioritizes pushing open-ended questions, automatically switching to closed-ended questions when the number of characters in a student's response falls below a preset value, and receiving students' text, voice, or image responses; a semantic analysis module performs word segmentation and part-of-speech tagging on the teaching text and student responses, using a BERT-based pre-training architecture for the education domain. The model analyzes the text's logical structure and extracts keyword vectors from open-ended answers. The student answer analysis unit calculates the error rate of individual students based on the questions and answers generated from the knowledge points, and summarizes and generates statistical data on the knowledge points of the whole class. The classroom evaluation generation module associates the statistical data of the whole class with the individual student reports, marks students with error rates higher than the class average and common knowledge blind spots where the class error rate exceeds 30%, and generates a visual student knowledge point mastery report that includes a heatmap of knowledge point mastery. It extracts question feature vectors containing knowledge point coverage, student error rate, and interaction response time, matches them with a pre-stored teaching syllabus evaluation rule base, and generates a classroom evaluation report that includes knowledge blind spot marking, teaching pace score, content comprehension index, and classroom interaction score. The teaching strategy optimization module analyzes the classroom evaluation report and student knowledge point mastery report based on a supervised learning neural network model and generates improvement plans such as teaching content adjustment suggestions and explanation speed optimization schemes. The teaching strategy optimization module automatically feeds back the improvement schemes to the classroom question generation module through a data interface and pushes them to the teacher's terminal.

[0076] Real-time monitoring and acquisition of classroom teaching content refers to the simultaneous acquisition of video streams, audio streams, blackboard images, and courseware images through multimodal sensors. This can be achieved using distributed camera arrays and microphone arrays to address the fragmented data acquisition problem of traditional systems. Speech recognition to text conversion refers to converting the teacher's audio into editable text data. This can be achieved using a deep learning-based speech recognition engine to build a structured teaching data foundation. OCR technology for recognizing blackboard text refers to optical character recognition of blackboard images. This can be achieved using a convolutional neural network model to extract knowledge point information from handwritten blackboard text. Structured teaching objective data refers to a set of teaching objective parameters constructed according to the subject knowledge system. Specifically, it can be stored in XML or JSON format to guide the logical coherence of question generation. The BERT-based pre-trained model for the education domain refers to a semantic understanding model optimized for educational scenarios. Specifically, it can be fine-tuned on the textbook corpus using transfer learning methods to improve the semantic parsing accuracy of students' descriptive answers. The knowledge point mastery heatmap is a visual chart that reflects the degree of mastery of knowledge points in a class using color gradients. Specifically, it can be generated using data visualization tools to intuitively present the distribution of common knowledge blind spots. The supervised learning neural network model refers to a strategy optimization algorithm trained based on historical teaching data. Specifically, it can be implemented using a multilayer perceptron architecture to automatically generate teaching improvement plans.

[0077] After classroom teaching content is captured in real time by multimodal acquisition devices, audio data is converted into text through speech recognition, and text information is extracted from blackboard writing and courseware images through OCR, forming a unified structured teaching database. The question generation module combines teaching objectives and knowledge point tags to dynamically generate combinations of open-ended and closed-ended questions, and automatically switches question types based on the number of characters in student answers. During semantic analysis, the pre-trained model in the education field analyzes the semantic logic of student answers, extracts keyword vectors, and calculates similarity with standard answers. The error rate statistics unit calculates the degree of mastery of individuals and groups based on knowledge point dimensions, and marks common knowledge blind spots through heat maps. When the evaluation report is generated, multidimensional feature vectors are matched with the teaching syllabus rule base to quantitatively evaluate the teaching pace and interaction effect. The strategy optimization module analyzes class and individual data, automatically generates teaching adjustment suggestions, and feeds them back to the question generation unit, forming a closed-loop optimization mechanism for teaching.

[0078] Traditional methods rely on manual analysis of homework data, which cannot process multimodal teaching information in real time. The question types are fixed and lack dynamic adjustment mechanisms. Semantic analysis is limited to keyword matching. This method realizes real-time structured processing of all teaching elements. It triggers a question type switching mechanism through a character count threshold, combines a pre-trained model in the education field to deeply analyze semantic logic, constructs a multi-dimensional quantitative assessment system for the mastery of knowledge points, and automatically generates executable optimization strategies based on a supervised learning model.

[0079] The automated collection and structured processing of classroom teaching data solved the problem of multimodal information integration; a dynamic question generation and switching mechanism was established to improve the adaptability of the question-and-answer process; deep semantic analysis accurately located knowledge gaps, overcoming the limitations of traditional keyword matching; a multi-dimensional evaluation system covering teaching pace and interaction effects was constructed, breaking through the defects of single accuracy rate assessment; and a closed-loop optimization path for teaching strategies based on data analysis was formed, replacing the subjective decision-making model that relies on teacher experience.

Claims

1. A classroom content evaluation system based on student question-and-answer sessions, characterized in that, include: The classroom information collection module is used to monitor and collect classroom teaching content in real time, including video, audio, blackboard text, and courseware images. The data processing module is connected to the classroom information collection module via a data interface. It is used to convert the collected audio into text through speech recognition, recognize the blackboard text through OCR technology, and extract the text content from the courseware images. The classroom question generation module is connected to the data processing module and is used to generate a collection of questions based on the processed classroom teaching content. The student response interaction module is used to receive students' text, voice, or image responses to a collection of questions; The semantic analysis module is used to analyze the semantics of classroom teaching content, and to provide semantic interpretation of students' descriptive answers in the question set. It also marks knowledge blind spots for knowledge points that students did not answer correctly by marking database fields or highlighting them in a visualization interface. The classroom evaluation generation module outputs a classroom evaluation report that includes knowledge blind spot markers, teaching pace scores, content comprehension indices, and classroom interaction scores. The student knowledge point mastery report generated based on the knowledge blind spot markers is sent to the teacher's terminal.

2. The classroom content evaluation system based on student question-and-answer as described in claim 1, characterized in that, The set of questions includes a set of closed-ended questions and a set of open-ended questions; The student response interaction module prioritizes pushing open-ended question sets. When the number of characters in a student's answer is lower than a preset character threshold based on the teaching subject and grade level, it automatically switches to a closed-ended question set. The semantic analysis module extracts keywords from open-ended answers and dynamically adjusts the content of the closed-ended question set based on the TF-IDF weights of the keywords and the cosine similarity calculation results of the closed-ended question set. When the cosine similarity is lower than a preset threshold, a question with higher similarity is selected from the closed-ended question set to replace the current question.

3. The classroom content evaluation system based on student question-and-answer as described in claim 1, characterized in that, The classroom question generation module includes: The teaching attribute recognition unit is used to automatically identify the teaching subject and grade level based on the teaching content in the classroom. The knowledge point identification unit is used to identify knowledge points in the teaching content. The question generation unit is used to generate questions based on teaching objectives and knowledge points.

4. A classroom content evaluation system based on student question-and-answer as described in claim 3, characterized in that, It also includes a classification storage module, which is connected to the data processing module and the classroom question generation module. It is configured with a classification storage repository for classifying and storing the question collection according to teaching subjects, grade levels and knowledge points.

5. A classroom content evaluation system based on student question-and-answer as described in claim 4, characterized in that, The semantic analysis module includes: The text preprocessing unit is used to perform word segmentation and part-of-speech tagging on classroom teaching texts and student responses; The semantic understanding unit is used to analyze the logical structure of text through a semantic model based on the BERT architecture; The key information extraction unit is used to generate keyword vectors from the answers to a collection of open-ended questions.

6. A classroom content evaluation system based on student question-and-answer as described in claim 5, characterized in that, The semantic analysis module also includes a student answer analysis unit, connected to the classification and storage module, used to calculate the error rate of questions and answers generated based on knowledge points, wherein: Error rate = Number of knowledge-based incorrect answers / (Total number of answers - Number of unanswered questions). The number of knowledge-based incorrect answers refers to the number of answers that are judged to be incorrect based on the standard answer for the knowledge point.

7. A classroom content evaluation system based on student question-and-answer as described in claim 6, characterized in that, The classroom assessment generation module includes: The summary unit, connected to the student answer analysis unit, is used to generate statistical data on the mastery of knowledge points in the whole class based on the error rate of each student's knowledge points. The statistical data includes the class average error rate, standard deviation of the error rate, and proportion of students who have mastered each knowledge point. The report generation unit is used to associate the statistical data on the mastery of knowledge points of the whole class with the knowledge point mastery report of individual students, mark students whose error rate is higher than the average of the whole class and common knowledge blind spots whose error rate of the whole class exceeds a preset ratio, and generate a visual student knowledge point mastery report including a knowledge point mastery heatmap.

8. A classroom content evaluation system based on student question-and-answer as described in claim 7, characterized in that, The classroom assessment generation module also includes: The feature extraction unit extracts problem feature vectors containing common knowledge gaps, student error rates, and interaction response times; The evaluation rule matching unit is used to match the problem feature vector with the teaching syllabus evaluation rule library pre-stored in the system memory; The report generation unit generates a classroom evaluation report based on the results of the feature extraction unit and the evaluation rule matching unit.

9. A classroom content evaluation system based on student question-and-answer as described in claim 8, characterized in that, The system also includes a teaching strategy optimization module, connected to the classroom evaluation generation module, including: The data receiving unit is used to receive the classroom evaluation report and the student knowledge point mastery report; The analysis and optimization unit analyzes classroom evaluation reports and student knowledge mastery reports based on a neural network model trained with supervised learning, and generates improvement plans, including suggestions for adjusting teaching content and optimization plans for explanation speed. The strategy feedback unit automatically feeds back the improvement plan to the question generation unit of the classroom question generation module through the data interface; The teacher interaction unit pushes improvement plans and knowledge point mastery reports to the teacher's terminal.

10. A classroom content evaluation method based on student question-and-answer, applied to the classroom content evaluation system according to any one of claims 1-9, characterized in that, Includes the following steps: S1: The classroom teaching content is monitored and collected in real time through the classroom information collection module. The teaching content includes video, audio, blackboard text and courseware images. S2: The data processing module is used to convert the collected audio into text through speech recognition, and the OCR technology is used to recognize the text written on the blackboard and extract the text content in the courseware images to form structured teaching text data. S3: The classroom question generation module identifies the teaching subject, grade level, and knowledge point based on the teaching content. It combines the structured teaching objective data pre-stored in the system database or input by the teacher's terminal to generate a question set containing closed-ended question sets and open-ended question sets. The questions are then stored in a classified storage repository according to the teaching subject, grade level, and knowledge point. S4: The student response interaction module prioritizes pushing open-ended question sets. When the number of characters in a student's response is lower than a preset value, it automatically switches to closed-ended question sets to receive the student's text, voice, or image responses. S5: The semantic analysis module performs word segmentation and part-of-speech tagging on teaching texts and student answers. It analyzes the logical structure of the text using a pre-trained model for the education domain based on the BERT-base architecture and extracts keyword vectors from open-ended answers. S6: The student answer analysis unit generates questions and answers based on knowledge points, calculates the error rate of individual students' knowledge points, and summarizes and generates the knowledge point statistics of the whole class; the classroom evaluation generation module links the knowledge point statistics of the whole class with the individual student reports, marks students whose error rate is higher than the average of the whole class and common knowledge blind spots whose error rate of the whole class exceeds the preset ratio, and generates a visual student knowledge point mastery report that includes a knowledge point mastery heatmap. S7: Extract problem feature vectors containing knowledge point coverage, student error rate, and interaction response time, match them with the pre-stored teaching syllabus evaluation rule base, and generate a classroom evaluation report containing knowledge blind spot markers, teaching pace scores, content comprehension index, and classroom interaction scores. S8: The teaching strategy optimization module analyzes classroom evaluation reports and student knowledge mastery reports based on a supervised learning neural network model, and generates improvement plans such as suggestions for adjusting teaching content and optimizing explanation speed. S9: The teaching strategy optimization module automatically feeds back the improvement plan to the classroom question generation module through the data interface and pushes it to the teacher's terminal.

Images