Classroom knowledge point coverage analysis method based on keyword and semantic matching

By using keyword and semantic matching analysis methods, the correlation between course content and teaching syllabus is automatically assessed, solving the subjectivity and efficiency problems of traditional assessment methods and achieving efficient and accurate improvement in teaching quality.

CN119047481AInactive Publication Date: 2024-11-29BEIJING NORMAL UNIVERSITY

Patent Information

Application Number
CN202411166314.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-23
Publication Date
2024-11-29
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Traditional classroom teaching assessment methods rely on teachers' subjective judgment, which has problems such as strong subjectivity, high time and manpower costs, difficulty in real-time feedback, and inability to accurately assess the matching degree between the teaching content and the teaching syllabus.

Method used

This study employs a keyword and semantic matching-based analysis method. By combining speech recognition and text analysis techniques with rule-based keyword matching and semantic analysis, it automatically assesses the relevance of teachers' course content to the teaching syllabus. This includes speech recognition, keyword extraction, semantic recognition, and similarity calculation.

Benefits of technology

It improves the accuracy and efficiency of teaching assessment, reduces manpower consumption, provides detailed teaching feedback, helps teachers optimize teaching content, and ensures that key knowledge points are fully imparted.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119047481A_ABST
    Figure CN119047481A_ABST
Patent Text Reader

Abstract

The invention discloses a classroom knowledge point coverage analysis method based on keyword and semantic matching. The method comprises the steps of obtaining teaching audio data and a teaching outline text in a teaching process; performing voice recognition on the teaching audio data to obtain a teaching text, and segmenting the teaching text to obtain teaching text segments; performing keyword extraction on the teaching outline text to obtain teaching outline keywords; performing keyword matching on each teaching text fragment based on the teaching outline keywords to obtain knowledge point mention frequency so as to realize teaching outline knowledge point coverage analysis on the keyword level; performing semantic recognition on the teaching text fragments to generate a teaching text summary language; and performing similarity calculation on the teaching text summary language and the teaching outline text to obtain semantic similarity between the teaching text and the teaching outline text so as to realize teaching outline knowledge point coverage analysis on the semantic level.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of teaching evaluation, and in particular relates to a classroom knowledge point coverage analysis method based on keyword and semantic matching. Background Art

[0002] Classroom teaching evaluation is one of the important means to improve teaching quality and promote educational reform and innovation. Through evaluation and feedback, teachers can understand the strengths and weaknesses of their teaching, so as to improve teaching content and methods in a targeted manner and promote the continuous improvement of teaching quality. In this process, the analysis and evaluation of teaching content is of great significance. Judging whether the teaching content of teachers covers the difficulties and key points pre-defined in the syllabus can help teachers find knowledge points that are not mentioned or not fully explained, thereby improving the teaching quality and ensuring that every key knowledge point is effectively taught. However, traditional evaluation methods mainly rely on teachers' subjective judgment and classroom observation, which usually have the problems of strong subjectivity, high time and labor costs, and difficulty in real-time feedback, which may lead to the inability to adjust teaching strategies in time and meet students' learning needs. With the rapid development of artificial intelligence technology, innovative tools are provided for the analysis and evaluation of teaching content. Using artificial intelligence technology to analyze the content of teachers' classroom teaching and judge whether each knowledge point is fully covered can significantly improve the efficiency and accuracy of teaching evaluation.

[0003] In summary, the existing technology has the problem that traditional teaching content analysis methods usually only extract keywords from teachers' classroom speech without considering their relevance to the teaching syllabus, and are unable to quantify and explicitly reflect the specific situation and dynamic changes of the teaching content knowledge point coverage. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a classroom knowledge point coverage analysis method based on keyword and semantic matching, which uses speech recognition and text analysis technology to parse and understand the teaching content, and automatically analyzes teaching keywords and deep semantics to evaluate the knowledge point coverage of the teacher's teaching content. The present invention combines rule-based keyword matching and semantic analysis to comprehensively evaluate whether the teaching content is closely related to the syllabus from two perspectives, and can deeply analyze the correlation between the teacher's teaching content and the teaching syllabus, in order to improve the evaluation accuracy of the matching degree between the teaching content and the syllabus, provide feedback to teachers, and then improve the teaching quality, which can effectively solve the above technical problems.

[0005] To achieve the above object, the present invention provides a classroom knowledge point coverage analysis method based on keyword and semantic matching, comprising:

[0006] Acquire the teaching audio data and the teaching outline text during the teaching process; perform speech recognition on the teaching audio data to obtain the teaching text, and segment the teaching text to obtain teaching text segments;

[0007] Extract keywords from the syllabus text to obtain syllabus keywords; based on the syllabus keywords, perform keyword matching on each teaching text segment to obtain the frequency of knowledge point mentions to achieve syllabus knowledge point coverage analysis at the keyword level;

[0008] Perform semantic recognition on lecture text fragments to generate a lecture text summary; perform semantic recognition conversion on the lecture text summary and the syllabus text to obtain a lecture semantic vector and a syllabus semantic vector; perform similarity calculation on the lecture semantic vector and the syllabus semantic vector to obtain the semantic similarity between the lecture text and the syllabus text to achieve syllabus knowledge point coverage analysis at the semantic level.

[0009] Optionally, before performing speech recognition on the teaching audio data, the method further includes:

[0010] The teaching audio data is format converted and normalized, the normalized audio signal is Fourier transformed to obtain a spectrum diagram, the spectrum diagram is noise-removed through a high-pass filter and time-domain transformed to obtain preprocessed teaching audio data.

[0011] Optionally, the process of performing speech recognition on the teaching audio data includes:

[0012] The teaching audio data is subjected to speech recognition by using the Whisper algorithm to obtain the teaching text.

[0013] Optionally, the process of extracting keywords from the syllabus text includes:

[0014] The teaching outline text is segmented by the Jieba word segmentation method, and the word segmentation result is screened by the Chinese stop word list to obtain the teaching outline keywords.

[0015] Optionally, the process of keyword matching each lecture text segment includes:

[0016] In each lecture text segment, the number of occurrences of the syllabus keywords is counted using regular expressions to obtain the frequency of knowledge point mentions in each lecture text segment.

[0017] Optionally, the process of semantically identifying the lecture text segment includes:

[0018] Each lecture text segment is summarized through the Llama model, and each lecture text segment has a lecture text summary.

[0019] Optionally, the process of performing semantic recognition conversion on the teaching text summary and the teaching syllabus text includes:

[0020] The sentence embedding representation model is used to perform semantic recognition conversion on each lecture text summary and syllabus text sentence in the syllabus text, and several lecture semantic vectors and several syllabus semantic vectors are obtained.

[0021] Optionally, similarity calculation is performed on the teaching semantic vector and the syllabus semantic vector by using cosine similarity:

[0022] vec(b i )=SemtenceBert(b i )

[0023] vec(d j )=SentenceBert(d j )

[0024]

[0025] Among them, SentenceBert() means that it has been processed by the SentenceBert model, b i represents the summary of the i-th lecture text, d j represents the jth syllabus text sentence in the syllabus text, vec(b i ) represents the teaching semantic vector corresponding to the summary of the i-th teaching text, vec(d j ) represents the outline semantic vector corresponding to the jth outline text sentence, ||x|| represents the Euclidean norm of vector x, and Similarity represents the similarity;

[0026] Based on the syllabus semantic vectors of different syllabus text sentences, the similarity of the teaching semantic vectors corresponding to each teaching text summary is calculated respectively to obtain the similarity sequence of each teaching text summary to different syllabus text sentences. The maximum similarity value in each similarity sequence is taken as the similarity value of the corresponding teaching text summary to the teaching syllabus text, and the semantic similarity sequence, i.e., the semantic similarity between the teaching text and the teaching syllabus text, is obtained.

[0027] Optionally, the frequency of mentioning the knowledge points and the semantic similarity are visualized, wherein the timestamp of the teaching text segment is obtained, the timestamp is used as the horizontal axis, the frequency of mentioning the knowledge points and the semantic similarity are used as the vertical axis, the value of the timestamp is marked on the horizontal axis, and the value of the similarity value in the frequency of mentioning the knowledge points and the semantic similarity corresponding to the teaching text segment is marked on the vertical axis, to form a visualization chart for visualization display.

[0028] Compared with the prior art, the present invention has the following advantages and technical effects:

[0029] 1. Improve teaching quality and enhance teaching effectiveness:

[0030] Through detailed data analysis, it can reveal deviations and deficiencies in teaching knowledge points, provide feedback to teachers, and enable them to optimize teaching content in a targeted manner, improve teaching quality, and ensure that students can comprehensively learn and master the necessary knowledge and skills.

[0031] 2. Improve efficiency and reduce manpower consumption:

[0032] Automated teaching content analysis can significantly reduce the workload of teachers in the teaching evaluation process and reduce dependence on professionals. The original manual analysis that may take several hours can be shortened to automatic completion in a few minutes, greatly reducing the cost of teaching evaluation.

[0033] 3. Improve the accuracy of teaching content analysis:

[0034] Different from the teaching content analysis technology that only relies on keyword extraction, this method combines keyword matching technology with semantic understanding technology to calculate the matching degree between teaching content and teaching syllabus from the two dimensions of vocabulary and semantics, providing a comprehensive and precise evaluation and improving the evaluation accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0036] Figure 1 This is a flowchart of audio data acquisition and preprocessing according to an embodiment of the present invention;

[0037] Figure 2 This is a flow chart of the classroom knowledge point coverage analysis method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0039] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0040] The present invention uses speech recognition and text analysis technology to propose a classroom knowledge point coverage analysis method based on keyword and semantic matching. This method comprehensively evaluates the teaching content from two different dimensions: semantic relevance and keyword mention frequency. Based on a large amount of text data generated by long-term classroom teaching, including teachers' explanations and other teaching-related content, models such as Llama and SentenceBert are used to automatically analyze the teacher's classroom lecture content, accurately match the teaching content with the keywords and semantics of the syllabus, so as to more accurately evaluate the knowledge point coverage of the teacher's teaching content, reveal possible teaching blind spots or areas that need to be strengthened, and provide strong data support for teachers.

[0041] This paper focuses on the analysis of classroom knowledge point coverage based on keyword and semantic matching, aiming to provide a comprehensive and automated solution for evaluating the matching degree between teaching content and teaching syllabus. This method first collects and processes the classroom teaching audio and teaching syllabus, as shown in the attached figure. Figure 1 Then, we compared the teaching content with the syllabus from the perspectives of keyword mention frequency and semantic relevance to analyze whether the teacher’s teaching covered the syllabus knowledge points, as shown in the attached figure. Figure 2 The following is a detailed description of the technical solution of the present invention.

[0042] Step 1: Data collection. Data collection includes two parts: lecture audio recording and syllabus text collection. A microphone is used to connect to the computer and is specially installed in the best position above the middle of the classroom podium to ensure that the teacher's lecture audio data can be clearly collected. The collected lecture audio data will be transmitted to the connected computer in real time and stored on the local server. The lecture audio data of each class will be marked by date and time for subsequent classification and retrieval. The syllabus text data is generally uploaded by the teacher.

[0043] Step 2: Audio processing. The lecture audio data is first preprocessed, including normalization and denoising. The audio signal is converted to WAV format, and the spectrum is analyzed by Fourier transform to identify and remove noise components. After processing by a high-pass filter, the signals above a specific cutoff frequency in the lecture audio data are retained (mainly the teacher's voice), thereby clearly capturing the teaching content. Then, speech recognition is performed through Whisper, an automatic speech recognition system developed by OpenAI, and the lecture voice, i.e., the lecture audio data, is converted into lecture text.

[0044] Step 3: Extract keywords from the syllabus. Use Jieba word segmentation technology to automatically segment the syllabus text, and use the Chinese stop word list to remove redundant words. The phrase set obtained after word segmentation and stop word removal is used as the syllabus keywords.

[0045] Step 4: Keyword matching. First, divide the teaching text into multiple two-minute paragraphs, and each paragraph is distinguished by a timestamp. Dividing the teaching text by time units not only simplifies subsequent data processing, but also facilitates the statistics of knowledge point coverage in different time periods to show the dynamic changes in the classroom. Then, regular expression technology is used to identify and calculate the frequency of occurrence of syllabus keywords in the segmented teaching text fragments. In theory, the larger the frequency value, the more times the teacher mentions the syllabus keywords during the lecture, the higher the match between the teaching content and the syllabus, and the better the knowledge point coverage.

[0046] Step 5: Summary of lecture text. Keyword matching only focuses on the number of times a word is mentioned, while ignoring semantic relevance, resulting in incomplete and in-depth analysis results. Therefore, this solution also matches the syllabus and lecture content at the semantic level to improve the evaluation effect. Before semantic matching, the Llama model based on the Transformer architecture is used to summarize the segmented lecture text fragments obtained in step 4, generate a short summary sentence, and extract the central theme of the lecture content in each time period for better semantic matching.

[0047] Step 6: Semantic matching. The syllabus text and the lecture text summary obtained in step 5 are converted into semantic vectors using the SentenceBert model. The cosine similarity between the two semantic vectors is calculated to evaluate the semantic relevance between the lecture content and the syllabus points in each time period. This process not only accurately measures the coverage and depth of the lecture content, but also reveals possible teaching blind spots or areas that need to be strengthened.

[0048] Step 7: Visualize the results. After obtaining the frequency of keyword occurrence and the semantic relevance between teaching content and knowledge points, a visualization chart is generated to present these data to reflect the coverage of classroom knowledge points in different time periods, helping teachers understand their own teaching characteristics and omissions, and helping teachers improve teaching quality.

[0049] The above technical solution is described in detail below:

[0050] Step 1: Data collection:

[0051] Data collection consists of two parts: lecture audio recording and outline text collection

[0052] 1. Lecture audio recording

[0053] The present invention uses Hikvision's microphone to connect to a computer to collect audio data when teachers are teaching in the classroom. The data collected is the real teaching audio data of all classes in the classroom. The teacher turns on the microphone after class and turns off the microphone after class. Therefore, the data collected in this way are all classroom audio scenes, excluding the noise between classes.

[0054] The microphone is installed in the best position in the classroom to collect the teacher's audio, that is, above the middle of the podium, to ensure that the teacher's teaching audio data can be clearly collected. The collected teaching audio data is transmitted to the connected computer in real time and stored on the local server to ensure data security and accessibility. The data is stored in MP3 format. The date and time of each class's teaching audio data are marked for subsequent classification and retrieval.

[0055] 2. Outline text collection

[0056] The collection of teaching outline texts should include teaching objectives and teaching content, clarify the knowledge points and concepts that students need to master in this class, and decompose the teaching content in the teaching outline text D into multiple long sentences, namely, outline text sentences, recorded as D = {d1, d2, ..., d m}. i represents the i-th outline text sentence, m represents the total number of outline text sentences, and each long sentence d i Corresponding to specific knowledge points, in order to match knowledge points and track teaching progress, such as "deep understanding and implementation of arrays, linked lists, stacks and queues", "graph representation methods (adjacency lists, adjacency matrices), traversal algorithms (depth-first, breadth-first)". This part of information is generally clearly stated before the teaching activities begin, and the teacher should upload the teaching objectives and content of the syllabus according to the course.

[0057] Step 2: Audio Processing

[0058] 1. Preprocessing of classroom audio data.

[0059] The lecture audio data obtained in step 1 needs to be preprocessed by normalization and denoising. First, convert the audio signal into WAV format and normalize it so that the amplitude of the audio signal is within a fixed range. Then perform Fourier transform on the audio signal to obtain a spectrum, identify the noise components in the spectrum, and remove these noise frequency bands through filters. The specific formula for Fourier transform of data is as follows:

[0060]

[0061] Among them, X(f) represents the frequency domain signal, x(t) represents the time domain signal, f represents frequency, t represents time, and j is an imaginary unit.

[0062] In a classroom environment, common noise types and their frequency ranges in the collected audio data include: background noise: low-frequency noise such as air conditioners and computer fans (usually in the range of 0-300Hz); environmental noise: medium- and low-frequency noise such as door opening and closing, footsteps (usually in the range of 300-1000Hz); class noise: medium-frequency noise such as students talking and chair moving (usually in the range of 1000-3000Hz); high-frequency noise: high-frequency noise such as electronic equipment (usually above 3000Hz). Therefore, for classroom audio data, a high-pass filter is usually a more appropriate choice, which can remove most of the low-frequency background noise and retain the teacher's speech information. The formula for the high-pass filter is:

[0063]

[0064] Where H(f) is the frequency response of the filter. f is the frequency of the signal. c is the filter cutoff frequency. Its effect is to c Attenuation occurs at frequencies above f c The signal is retained in the place.

[0065] 2. Extract the lecture text

[0066] The Whisper algorithm is used to convert the lecture audio data into lecture text. Whisper is a model framework based on deep learning and sequence-to-sequence (seq2seq). It uses the Transformer architecture. Its core idea is to capture long-term dependencies through the attention mechanism when processing the input audio sequence and generate the corresponding text sequence. The Whisper algorithm can not only generate text data, but also provide the timestamp of each text. Using this model, you can get the lecture text of the classroom and the timestamp of the text, which are stored in txt format.

[0067] Step 3: Extract syllabus keywords

[0068] The syllabus text obtained in Step 1 is generally quite rigorous, with each sentence being concise and clear. However, the content taught by teachers tends to be more colloquial. Directly matching and analyzing the syllabus text with the teacher's teaching text results in poor matching and recognition effects. Therefore, in this implementation method, we first extract the syllabus keywords through the following methods and then perform keyword matching on the syllabus text and the teaching text. First, use jieba to segment the entire sentence content of the syllabus text D into multiple Chinese words. Then, use a publicly available Chinese stop word list to remove some unimportant and insignificant words. The words included in the stop word list are those that appear frequently but contribute little to the text theme and content. For example, "de", "le", "zai", "shi" are very common in Chinese texts, but they do not actually contribute to the text theme. The set of phrases obtained after word segmentation and stop word removal is used as the syllabus keywords.

[0069] Step 4: Keyword Matching

[0070] To facilitate the matching of the syllabus keywords obtained in Step 3 with the teaching text obtained in Step 2, first segment the teaching text. For example, the teaching time for a general university course is 90 minutes, which can be divided into 45 two-minute text segments for more detailed analysis. The teaching text obtained in Step 2 has corresponding timestamp information, and the teaching text is segmented every two minutes based on the timestamp. For convenience of representation, the teaching text is denoted as S here, and the segmented teaching text segments are s i ∈{s1, s2,..., s n}, s i represents the i-th teaching text segment, and n represents the total number of teaching text segments.

[0071] Then, by matching the segmented teaching text segment s i with the syllabus keywords obtained in Step 3, calculate the number of occurrences f i of the syllabus keywords in s i . f i reflects the matching degree between the teacher's teaching content and the syllabus keywords during the time period where s i is located. The larger this frequency value, the higher the matching degree between the teaching content and the syllabus, and the better the knowledge point coverage; vice versa. This implementation method uses a rule-based matching algorithm, namely regular expressions, for text matching. Regular expressions are a tool for performing pattern matching and text processing tasks in text. It identifies specific character sequences by defining a search pattern, which consists of a series of characters and special symbols. Regular expressions are widely used in scenarios such as searching, editing, and data verification that require pattern recognition. Thus, a sequence F = {f1, f2,... f n}, which includes the frequency of mention of knowledge points in different teaching text segments in corresponding time periods.

[0072] Step 5: Summary of the Lecture Text

[0073] Keyword matching only focuses on the number of times a word is mentioned, but ignores semantic relevance, which results in incomplete and in-depth analysis results. Therefore, this method also matches the syllabus and teaching content at the semantic level to improve the evaluation effect.

[0074] Before semantic matching, the segmented teaching text fragments s i Summarize and output the corresponding summary as the summary of the teaching text, so as to extract the central theme of the teaching content for better semantic matching. This implementation method uses the Llama model to respectively analyze the teaching text segments s1, s2, ...s n Summarize. Llama is an open source large model based on the Transformer architecture, which is used to handle a variety of natural language processing tasks, such as text generation, summary generation, translation, and question answering. However, the Llama model is large and has high hardware requirements. In this implementation method, the API of the Llama model in Alibaba Cloud is called to summarize the text. Specifically, the prompt word "Summarize the following paragraph, no more than 30 words: s i " is input into the Llama model to obtain the lecture text segment s i Corresponding summary b i Thus, we can get a sequence B consisting of the summary of the teaching text = {b1, b2, ...b n}, contains the lecture text summary corresponding to each lecture text segment.

[0075] Step 6: Semantic Matching

[0076] Use SentenceBert and cosine similarity to convert the summary of the lecture text into i With all outline texts collected in step 2 D = {d1, d2, ..., d m} are matched one by one and the coverage of knowledge points in the syllabus is evaluated from a semantic level.

[0077] SentenceBert is a sentence embedding representation model, which is often used to match text semantic similarity. SentenceBert can generate a fixed-size vector representation of the input sentence to represent the semantics of the entire sentence. Specifically, the summary of the lecture text b i and outline text sentence d i Input SentenceBert to obtain semantic vectors respectively, and then calculate the cosine similarity between the two vectors to measure the semantic relevance. The formula is as follows:

[0078] vec(b i )=SentenceBert(b i )

[0079] vec(d j )=SentenceBert(d j )

[0080]

[0081] Among them, SentenceBert() means that it has been processed by the SentenceBert model, b i represents the summary of the i-th lecture text, d j represents the jth syllabus text sentence in the syllabus text, vec(b i ) represents the teaching semantic vector corresponding to the summary of the i-th teaching text, vec(d j ) represents the outline semantic vector corresponding to the j-th outline text sentence, Similarity represents the similarity; · represents the vector dot product, and ||x|| represents the Euclidean norm of vector x.

[0082] By fixing a single summary statement b i By traversing and replacing the outline text sentences one by one and calculating the similarity, we can get the summary sentence b i With all outline texts D = {d1, d2, ..., d m} in the similarity sequence in, Indicates the summary of the lecture text b i Sentence d with outline text j The similarity of Select the outline sentence with the largest similarity in the sequence as b i The knowledge point is matched, and the similarity value is used as b i The semantic matching degree between the teaching content and the knowledge points in the syllabus during the time period:

[0083]

[0084] Then, the summary words are traversed and replaced one by one, and after a single summary word replacement, the related subsequent operations of obtaining the similarity of a fixed single summary word are performed to traverse and obtain the semantic matching degree of each summary word. In this way, the summary words {b1, b2, ... b n} and the semantic relevance of the outline text D {L1, L1, ..., L n}, that is, the semantic matching degree between the teacher's teaching content and the syllabus knowledge points in each time period of the entire teaching process.

[0085] Step 7: Visualize the results

[0086] Through steps 4 and 6, we can get two sequences, namely, the frequency of knowledge point mentions {f1, f2, ..., f n} and semantic relevance {L1,L1,…,L n {f1,f2,…f n} represents the number of times the outline keywords appear in the teaching content in each time period of the teaching process. n} indicates the semantic relevance between the teaching content and the syllabus knowledge points in each time period during the teaching process.

[0087] Through the two sets of sequences, a visualization graph that changes over time can be made. The horizontal axis of the graph represents time in minutes, and the interval between each horizontal axis point is 2 minutes, reflecting the time series of classroom lectures. The value on the left side of the vertical axis is the frequency, that is, the frequency of matching between the teaching content and the predetermined knowledge points in the syllabus in a specific time period, and the value on the right side of the vertical axis is the semantic relevance between the teaching content and the syllabus knowledge points, which is a percentage value. The matching frequency is plotted as a bar chart, and the closeness is plotted as a line chart, so that the two screenshots can be presented on the same graph at the same time. Through this graph, teachers can intuitively identify which time periods have better covered the requirements of the syllabus, and which time periods may need to further strengthen or adjust the teaching strategy to ensure that all predetermined teaching objectives are fully explained and valued.

[0088] The present invention proposes a method for analyzing the coverage of classroom knowledge points. This method comprehensively evaluates the teaching content from two different dimensions: semantic relevance and keyword mention frequency. By processing and analyzing the voice and text data of teachers in the classroom in real time, the matching degree between the teaching content and the syllabus is automatically evaluated, revealing possible teaching blind spots or areas that need to be strengthened, thereby providing strong data support for teachers.

[0089] The present invention proposes to use Whisper algorithm to recognize classroom speech, and adopts Jieba word segmentation technology to accurately segment the outline text to obtain outline keywords, and finally applies regular expressions to match the two, automatically calculates the frequency of occurrence of outline keywords in the teaching content, and evaluates the matching degree between the teaching content and the teaching outline from the keyword level.

[0090] The present invention proposes to use the open source large model Llama to semantically summarize the teaching content to obtain the central theme, and adopts the SentenceBert model to extract the high-dimensional semantic vectors of the teaching text summary and the teaching outline. Finally, the semantic matching is achieved by calculating the correlation between the two through cosine similarity, and the matching degree between the teaching content and the teaching outline is evaluated from the semantic level.

[0091] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. Classroom knowledge point coverage analysis method based on keyword and semantic matching, It is characterized in that include: Obtain the lecture audio data and syllabus text during the lecture; Performing speech recognition on the teaching audio data to obtain teaching text, and segmenting the teaching text to obtain teaching text segments; Extract keywords from the syllabus text to obtain syllabus keywords; based on the syllabus keywords, perform keyword matching on each teaching text segment to obtain the frequency of knowledge point mentions to achieve syllabus knowledge point coverage analysis at the keyword level; Perform semantic recognition on lecture text fragments to generate a lecture text summary; perform semantic recognition conversion on the lecture text summary and the syllabus text to obtain a lecture semantic vector and a syllabus semantic vector; perform similarity calculation on the lecture semantic vector and the syllabus semantic vector to obtain the semantic similarity between the lecture text and the syllabus text to achieve syllabus knowledge point coverage analysis at the semantic level.

2. The method according to claim 1, It is characterized in that Before performing speech recognition on the teaching audio data, the method further includes: The teaching audio data is format converted and normalized, the normalized audio signal is Fourier transformed to obtain a spectrum diagram, the spectrum diagram is noise-removed through a high-pass filter and time-domain transformed to obtain preprocessed teaching audio data.

3. The method according to claim 1, It is characterized in that The process of performing speech recognition on the teaching audio data includes: The teaching audio data is subjected to speech recognition by using the Whisper algorithm to obtain the teaching text.

4. The method according to claim 1, It is characterized in that The process of extracting keywords from the syllabus text includes: The teaching outline text is segmented by the Jieba word segmentation method, and the word segmentation result is screened by the Chinese stop word list to obtain the teaching outline keywords.

5. The method according to claim 1, It is characterized in that The process of keyword matching for each lecture text segment includes: In each lecture text segment, the number of occurrences of the syllabus keywords is counted using regular expressions to obtain the frequency of knowledge point mentions in each lecture text segment.

6. The method according to claim 1, It is characterized in that The process of semantic recognition of lecture text fragments includes: Each lecture text segment is summarized through the Llama model, and each lecture text segment has a lecture text summary.

7. The method according to claim 1, It is characterized in that The process of semantic recognition conversion of the teaching text summary and the teaching syllabus text includes: The sentence embedding representation model is used to perform semantic recognition conversion on each lecture text summary and syllabus text sentence in the syllabus text, and several lecture semantic vectors and several syllabus semantic vectors are obtained.

8. The method according to claim 7, It is characterized in that The similarity between the teaching semantic vector and the outline semantic vector is calculated by cosine similarity: vec(b i )=SentenceBert(b i ) vec(d j )=SentenceBert(d j ) Among them, SentenceBert() means that it has been processed by the SentenceBert model, b j represents the summary of the i-th lecture text, d j represents the jth syllabus text sentence in the syllabus text, vec(b i ) represents the teaching semantic vector corresponding to the summary of the i-th teaching text, vec(d j ) represents the outline semantic vector corresponding to the jth outline text sentence, ||x|| represents the Euclidean norm of vector x, and Similarity represents the similarity; Based on the syllabus semantic vectors of different syllabus text sentences, the similarity of the teaching semantic vectors corresponding to each teaching text summary is calculated respectively to obtain the similarity sequence of each teaching text summary to different syllabus text sentences. The maximum similarity value in each similarity sequence is taken as the similarity value of the corresponding teaching text summary to the teaching syllabus text, and the semantic similarity sequence, i.e., the semantic similarity between the teaching text and the teaching syllabus text, is obtained.

9. The method according to claim 1, It is characterized in that Also includes: The frequency of mentioning the knowledge points and the semantic similarity are visualized, wherein the timestamp of the teaching text segment is obtained, the timestamp is used as the horizontal axis, the frequency of mentioning the knowledge points and the semantic similarity are used as the vertical axis respectively, the value of the timestamp is marked on the horizontal axis, and the value of the similarity value in the frequency of mentioning the knowledge points and the semantic similarity corresponding to the teaching text segment is marked on the vertical axis, forming a visualization chart for visualization display.

Citation Information

Patent Citations

  • Question-answer quality determination method and device, server and storage medium

    CN108960574A

  • A classroom teaching content evaluation system based on an intelligent analysis technology and a method thereof

    CN109710931A

  • Online course video resource content identification and evaluation method and intelligent system

    CN111898441A

  • Multi-dimensional intelligent teaching quality evaluation method and system

    CN115600922A

  • Teacher teaching ability evaluation method and system for curriculum heavy and difficult point knowledge

    CN117933787A

Cited By

  • Voice intention recognition method, electronic equipment and storage medium

    CN120279912A

  • Teaching quality determination system and method based on multi-modal data

    CN120782325A

  • Micro-course extraction method, related method and related device

    CN121982695A