Chinese reading intelligent teaching method
By introducing natural language processing and speech recognition technology into the Chinese teaching system, the reading speed is dynamically adjusted, and the problem of neglected reading training in the existing system and the lack of targeted teaching solutions is solved, and a personalized learning experience and better teaching effect is achieved.
Patent Information
- Application Number
- CN202510433283.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-08
- Publication Date
- 2025-06-24
AI Technical Summary
The existing Chinese teaching system neglects the reading training, and the teaching plan is not targeted, resulting in poor teaching results.
The intelligent teaching method of Chinese reading is adopted, and through natural language processing and speech recognition technology, the reading speed is dynamically adjusted according to the difficulty of the text and the students' learning status to achieve a personalized learning experience.
It effectively improves the pertinence and teaching effect of reading training, ensuring that every student can conduct personalized learning based on their own comprehension abilities.
Smart Images
Figure CN120199250A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the technical field of information processing, and particularly relates to an intelligent Chinese reading teaching method. Background Art
[0002] Language ability plays an important role in people's daily life, work and social interaction. Language teaching is an important teaching method to improve users' language ability. The main ways of language teaching include listening, speaking, reading and writing. Although the purpose of Chinese teaching is to guide students to correctly understand and use the language and characters of their motherland, in teaching, there is still a phenomenon of emphasizing writing training and neglecting other aspects of training. Among them, reading training is the most easily neglected, but it is the basic skill of language ability. Therefore, in teaching, we must pay attention to strengthening the guidance and training of reading. The existing language teaching systems adopt a unified teaching plan for all students, that is, unified courses and unified teaching, without formulating a teaching plan targeted at the actual situation of students, nor differentiating the difficulty level of texts, resulting in poor teaching effects. Summary of the Invention
[0003] In view of this, the purpose of this application is to provide an intelligent Chinese reading teaching method, which can automatically adjust the reading speed according to the difficulty level of the text, and at the same time judge the understanding ability of students by monitoring the learning status of students, so as to dynamically adjust the reading speed and achieve a personalized learning experience.
[0004] This application provides an intelligent Chinese reading teaching method, including: S1. Obtain the target text, perform semantic analysis on the target text using natural language processing technology to obtain the semantic feature vector of the target text, and use the pre-established text difficulty grading model to evaluate the difficulty level of the target text to generate the initial reading speed; S2. Evaluate the difficulty of each sentence in the target text, and adjust the initial reading speed of difficult sentences to obtain the standard reading speed of the target text; S3. Real-time obtain the behavior data of the target student to generate the learning status feature vector of the target student, and obtain the current understanding ability level of the target student; wherein, the behavior data includes: initial behavior data and learning behavior data; S4. Adjust the standard reading speed based on the current understanding ability level of the target student; S5. Perform speech synthesis on the target text according to the adjusted standard reading speed to obtain the teaching audio of the target text for students to follow and read, and repeat steps S3 - S5.
[0005] Further, obtaining the target text in step S1, and performing semantic analysis on the target text using natural language processing technology to obtain the semantic feature vector of the target text, including: Performing word segmentation on the target text to obtain a corresponding sequence of basic language units; Judging the part-of-speech of each basic language unit, and using the dependency parsing method to determine the dependency relationship between each basic language unit to obtain the syntactic structure tree of the target text; wherein, the part-of-speech is used to reflect the role of the basic language unit in the syntactic structure; Based on the syntactic structure tree of the target text, using the word embedding technology to map each language unit into a multi-dimensional real number vector to obtain the semantic feature vector of the target text.
[0006] Further, the difficulty assessment for each sentence in the target text in step S2 includes: For each sentence in the target text, using the word segmentation technology to obtain the corresponding word sequence of the sentence, and using the word embedding technology to map each word into the corresponding semantic feature vector; Judging the difficulty weight of each word according to the pre-set vocabulary difficulty grading model; Based on the semantic feature vector and difficulty weight of each word, using the weighted average method to calculate the semantic difficulty feature vector of the sentence, and using the pre-trained sentence difficulty grading model to obtain the comprehensive difficulty score of the sentence; Comparing the comprehensive difficulty score of the sentence with the pre-set difficult sentence threshold; When the comprehensive difficulty score of the sentence is greater than the difficult sentence threshold, it is determined that the sentence is a difficult sentence.
[0007] Further, adjusting the initial reading speed of the difficult sentence in step S2 to obtain the standard reading speed of the target text, including: According to the comprehensive difficulty score of the difficult sentence, using the dynamic adjustment algorithm to calculate the reduction ratio of the reading speed, and adjusting the initial reading speed of the difficult sentence; Based on the adjusted reading speed of the difficult sentence, using the smoothing algorithm to eliminate the sudden change in the reading speed between the difficult sentence and the adjacent sentences to obtain the standard reading speed of the target text, and at the same time marking the sentence to attract the attention of students.
[0008] Further, when the behavior data is the initial behavior data, the real-time acquisition of the behavior data of the target student in step S3 to generate the learning state feature vector of the target student, including: Obtain the audio data of the target student when completing a specific reading task and the text data during the answering process as the initial behavior data; Perform speech recognition on the obtained audio data to obtain the corresponding text content, and compare it with the standard reading text to count the pause frequency and repetition times of the target student during the reading process; Calculate the answering accuracy rate of the target student based on the obtained text data; Generate the learning status feature vector of the target student based on the pause frequency, repetition times, and answering accuracy rate;
[0009] Further, when the behavior data is learning behavior data, the step of obtaining the behavior data of the target student in real time in step S3 to generate the learning status feature vector of the target student includes: Obtain the learning behavior data of the target student during the follow-up reading process, including reading speed and eye movement trajectory; Preprocess the learning behavior data to remove noise and outliers, and extract key feature parameters from the preprocessed learning behavior data to generate the learning status feature vector of the target student.
[0010] Further, the adjustment of the standard reading speed in step S4 based on the current comprehension ability level of the target student includes: Compare the current comprehension ability level of the target student with the historical comprehension ability level to judge the change trend of the comprehension ability level of the target student; When the comprehension ability level of the target student decreases, use a dynamic adjustment strategy to adjust the standard reading speed.
[0011] Further, the standard reading speed is adjusted by the following method: Based on the current comprehension ability level of the target student, match the corresponding adjustment coefficient from the pre-set reading speed adjustment coefficient mapping table; Obtain the current reading speed of the target student and multiply it by the matched adjustment coefficient to obtain the adjusted standard reading speed.
[0012] Further, the step of performing speech synthesis on the target text according to the adjusted standard reading speed to obtain the teaching audio of the target text includes: Obtain the adjusted standard reading speed to determine the parameters related to the reading speed; Input the parameters related to the reading speed and the target text into the pre-trained speech synthesis model, and generate the corresponding speech feature sequence through the forward propagation process; Convert the voice feature sequence into an audio signal using a vocoder and perform waveform reconstruction to obtain teaching audio corresponding to the target text.
[0013] Further, inputting the parameter related to the reading speed and the target text into a pre-trained speech synthesis model, and generating a corresponding voice feature sequence through a forward propagation process, includes: Perform semantic and syntactic analysis on the target text to determine the structure and pause positions of each sentence; Based on the structure, pause positions of each sentence, and the parameter related to the reading speed, use a Markov model for modeling to obtain initial voice parameters related to the speech prosody and pause positions of the target text; Use the backpropagation algorithm to dynamically adjust the initial voice parameters, obtain optimized voice parameters, and generate a corresponding voice feature sequence; Judge whether the voice feature sequence meets the requirements of naturalness and accuracy; When the voice feature sequence meets the requirements of naturalness and accuracy, output the voice feature sequence; otherwise, continue to optimize the voice parameters until the voice feature sequence meets the requirements of naturalness and accuracy.
[0014] The intelligent Chinese reading teaching method provided by this application can automatically adjust the reading speed according to the difficulty of the text, and at the same time, by monitoring the learning status of students, judge the understanding ability of students, so as to dynamically adjust the reading speed and achieve a personalized learning experience. Brief Description of the Drawings
[0015] Figure 1 Shows the flowchart of the intelligent Chinese reading teaching method provided by the embodiment of this application. Detailed Description of the Embodiment
[0016] To make the purpose, technical solution and advantages of this technical solution clearer, the following further details this technical solution in combination with specific embodiments. It should be understood that these descriptions are exemplary and not intended to limit the scope of this technical solution.
[0017] First, introduce the application scenario of this application. The technical solution of this application can be applied to the server of a learning device, and the application end and the server of this learning device have an interaction function.
[0018] Please refer to Figure 1 the flowchart of the intelligent Chinese reading teaching method as shown. As Figure 1 shown, the method includes: S101. Obtain the target text, perform semantic analysis on the target text using natural language processing technology to obtain the semantic feature vector of the target text, and use the pre-established text difficulty grading model to evaluate the difficulty level of the target text to generate the initial reading speed.
[0019] In this step, first obtain the target text from the teaching resource library or input it by the user, then convert the target text into a standard text format (such as UTF-8 encoding format), and filter out irrelevant symbols including page numbers and illustrations while preserving the integrity of the standard text; then perform semantic analysis on the target text using natural language processing technology to obtain the semantic feature vector of the target text.
[0020] In specific implementation, the semantic feature vector of the target text can be obtained through the following method: Step 1011. Perform word segmentation on the target text to obtain the corresponding basic language unit sequence.
[0021] Step 1012. Determine the part of speech of each basic language unit, and use the dependency parsing method to determine the dependency relationship between each basic language unit to obtain the syntactic structure tree of the target text; where the part of speech is used to reflect the role of the basic language unit in the syntactic structure.
[0022] Step 1013. Based on the syntactic structure tree of the target text, use the word embedding technology to map each language unit into a multi-dimensional real number vector to obtain the semantic feature vector of the target text.
[0023] In addition, the text difficulty grading model is pre-established through the following method: 1) Obtain multiple texts, and obtain the semantic feature vector of each text according to the methods in steps 1011 - 1013; 2) Mark the new words, sentence length, and metaphorical sentences in each text, and count the new word density, sentence length, and the proportion of metaphorical sentences as the first difficulty evaluation indicators; 3) Set the weights corresponding to each first difficulty evaluation indicator; 4) For each text, obtain the difficulty level of the text according to the first difficulty evaluation indicators of the text and the corresponding weights; 5) Use the semantic feature vector of each text as the input and the corresponding difficulty level as the output to train the XGBoost classifier to obtain the text difficulty grading model.
[0024] S102. Evaluate the difficulty of each sentence in the target text, and adjust the initial reading speed of difficult sentences to obtain the standard reading speed of the target text.
[0025] In specific implementation, the difficulty of each sentence in the target text can be evaluated through the following method: Step 1021: For each sentence in the target text, use the word segmentation technology to obtain the corresponding word sequence of the sentence, and use the word embedding technology to map each word to the corresponding semantic feature vector.
[0026] Step 1022: According to the pre-set vocabulary difficulty grading model, judge the difficulty weights of each word.
[0027] In this step, the vocabulary difficulty grading model includes words of different difficulty levels and the corresponding difficulty weights.
[0028] Step 1023: Based on the semantic feature vectors and difficulty weights of each word, use the weighted average method to calculate the semantic difficulty feature vector of the sentence, and use the pre-trained sentence difficulty grading model to obtain the comprehensive difficulty score of the sentence.
[0029] In this step, the sentence difficulty grading model is established in advance in the following way: 1) Obtain multiple sentences, and obtain the semantic difficulty feature vector of each sentence according to the methods in Steps 1021 - 1023; 2) Mark the word meanings and parts of speech of each word in each sentence as the second difficulty evaluation index, and obtain the difficulty weights of each word; 3) For each sentence, obtain the comprehensive difficulty score based on the second difficulty evaluation index and the corresponding weights of each word in the sentence; 4) Use the semantic difficulty feature vector of each sentence as the input and the corresponding comprehensive difficulty score as the output to train the support vector machine to obtain the sentence difficulty grading model.
[0030] Step 1024: Compare the comprehensive difficulty score of the sentence with the pre-set difficult sentence threshold.
[0031] When the comprehensive difficulty score of the sentence is greater than the difficult sentence threshold, then execute Step 1025: Judge that the sentence is a difficult sentence.
[0032] In specific implementation, the initial reading speed of the difficult sentence can be adjusted in the following way to obtain the standard reading speed of the target text: Step 1026: According to the comprehensive difficulty score of the difficult sentence, use the dynamic adjustment algorithm to calculate the reduction ratio of the reading speed, and adjust the initial reading speed of the difficult sentence.
[0033] Step 1027: Based on the adjusted reading speed of the difficult sentence, use the smoothing algorithm to eliminate the sudden change in the reading speed between the difficult sentence and the adjacent sentences to obtain the standard reading speed of the target text, and at the same time mark the sentence to attract the attention of students.
[0034] S103: Real-time obtain the behavior data of the target student to generate the learning state feature vector of the target student, and obtain the current comprehension ability level of the target student.
[0035] Among them, the behavior data includes: initial behavior data and learning behavior data.
[0036] When the behavior data is initial behavior data, in specific implementation, the learning status feature vector of the target student can be generated in the following way: Step 1031: Obtain the audio data when the target student completes a specific reading task and the text data during the answering process as the initial behavior data.
[0037] Step 1032: Perform speech recognition on the obtained audio data to obtain the corresponding text content, and compare it with the standard reading text to count the pause frequency and repetition times during the target student's reading process.
[0038] Step 1033: Calculate the answering accuracy rate of the target student based on the obtained text data.
[0039] Step 1034: Generate the learning status feature vector of the target student based on the pause frequency, repetition times, and answering accuracy rate.
[0040] When the behavior data is learning behavior data, in specific implementation, the learning status feature vector of the target student can be generated in the following way: Step 1035: Obtain the learning behavior data including reading speed and eye movement trajectory of the target student during the shadowing process.
[0041] In this step, the learning behavior data of the target student is obtained through the camera and microphone installed on the learning device; specifically, the facial expressions and eye directions of the target student are captured through the camera, and based on the change trend of the eye directions in the time series, the eye movement trajectory is obtained using the background server, and the reading audio of the target student is collected through the microphone, and the reading speed of the target student is calculated using the background server.
[0042] Step 1036: Perform preprocessing of denoising and outliers on the learning behavior data, and extract key feature parameters from the preprocessed learning behavior data to generate the learning status feature vector of the target student.
[0043] S104: Adjust the standard reading speed based on the current comprehension ability level of the target student.
[0044] In specific implementation, the standard reading speed can be adjusted based on the current comprehension ability level of the target student in the following way: Step 1041: Compare the current comprehension ability level of the target student with the historical comprehension ability level to determine the change trend of the target student's comprehension ability level.
[0045] Step 1042: When the comprehension ability level of the target student decreases, adjust the standard reading speed.
[0046] Specifically, adjust the standard reading speed through the following method: Step 201: Based on the current comprehension ability level of the target student, match the corresponding adjustment coefficient from the pre-set mapping table of reading speed adjustment coefficients.
[0047] Step 202: Obtain the current reading speed of the target student and multiply it by the matched adjustment coefficient to obtain the adjusted standard reading speed.
[0048] S105: Perform speech synthesis on the target text according to the personalized adjusted standard reading speed to obtain the teaching audio of the target text for the student to follow and repeat steps S103 - S105.
[0049] In specific implementation, the teaching audio of the target text can be obtained through the following method: Step 1051: Obtain the adjusted standard reading speed to determine the parameters related to the reading speed.
[0050] Step 1052: Input the parameters related to the reading speed and the target text into the pre-trained speech synthesis model, and generate the corresponding speech feature sequence through the forward propagation process.
[0051] Specifically, the input of the parameters related to the reading speed and the target text into the pre-trained speech synthesis model to generate the corresponding speech feature sequence through the forward propagation process includes: Step 301: Perform semantic and syntactic analysis on the target text to determine the structure and pause positions of each sentence.
[0052] Step 302: Based on the structure, pause positions of each sentence and the parameters related to the reading speed, use the Markov model for modeling to obtain the initial speech parameters related to the speech prosody and pause positions of the target text.
[0053] Step 303: Dynamically adjust the initial speech parameters using the backpropagation algorithm to obtain the optimized speech parameters and generate the corresponding speech feature sequence.
[0054] Step 304: Determine whether the speech feature sequence meets the requirements of naturalness and accuracy.
[0055] When the speech feature sequence meets the requirements of naturalness and accuracy, step 305 is executed to output the speech feature sequence.
[0056] Otherwise, step 306 is executed to continue optimizing the speech parameters until the speech feature sequence meets the requirements of naturalness and accuracy.
[0057] Step 1053: Use a vocoder to convert the speech feature sequence into an audio signal and perform waveform reconstruction to obtain a teaching audio corresponding to the target text.
[0058] In addition, the intelligent Chinese reading teaching method further includes: Obtain the basic information of the target student including the mother tongue background and learning purpose, and determine the teaching key points and difficulties according to the basic information to match the corresponding teaching text; Conduct pronunciation demonstrations for the teaching content, collect the speech information of the target student following the pronunciation, and judge whether the pronunciation of the target student is accurate through speech comparison. When the pronunciation of the target student is inaccurate, provide a correction plan.
[0059] The above content is only the preferred embodiment of the present invention. For those of ordinary skill in the art, many changes can be made in the specific implementation manners and application scopes according to the idea of the technical content of the present application. As long as these changes do not deviate from the concept of the present invention, they all fall within the protection scope of this patent.
Claims
1. A Chinese reading intelligent teaching method, characterized in that: The method comprises: S1. Obtain a target text, perform semantic analysis on the target text using natural language processing technology to obtain a semantic feature vector of the target text, and use a pre-established text difficulty grading model to evaluate the difficulty level of the target text to generate an initial reading speed; S2, evaluating the difficulty of each sentence in the target text, and adjusting the initial reading speed of the difficult sentences to obtain a standard reading speed of the target text; S3, acquiring the behavior data of the target student in real time to generate the learning state feature vector of the target student and obtain the current comprehension ability level of the target student; wherein the behavior data includes: initial behavior data and learning behavior data; S4, adjusting the standard reading speed based on the current comprehension ability level of the target student; S5. Perform speech synthesis on the target text according to the adjusted standard reading speed to obtain the teaching audio of the target text for students to follow the reading, and repeat steps S3-S5.
2. The method according to claim 1, characterized in that The step S1 of obtaining a target text and performing semantic analysis on the target text using natural language processing technology to obtain a semantic feature vector of the target text includes: Performing word segmentation processing on the target text to obtain a corresponding basic language unit sequence; Determine the part of speech of each basic language unit, and use the dependency syntactic analysis method to determine the dependency relationship between the basic language units, so as to obtain the grammatical structure tree of the target text; wherein the part of speech is used to reflect the role of the basic language unit in the grammatical structure; Based on the grammatical structure tree of the target text, each language unit is mapped into a multi-dimensional real number vector by using word embedding technology to obtain the semantic feature vector of the target text.
3. The method according to claim 1, characterized in that The step S2 of performing difficulty assessment on each sentence in the target text includes: For each sentence in the target text, a word sequence corresponding to the sentence is obtained by using word segmentation processing technology, and each word is mapped into a corresponding semantic feature vector by using word embedding technology; Determine the difficulty weight of each word according to the pre-set vocabulary difficulty grading model; Based on the semantic feature vector and difficulty weight of each word, the semantic difficulty feature vector of the sentence is calculated using the weighted average method, and the comprehensive difficulty score of the sentence is obtained using the pre-trained sentence difficulty grading model; Compare the comprehensive difficulty score of the sentence with a pre-set difficult sentence threshold; When the comprehensive difficulty score of the sentence is greater than the difficult sentence threshold, the sentence is judged to be a difficult sentence.
4. The method according to claim 1, characterized in that The initial reading speed for the difficult sentence in step S2 is adjusted to obtain the standard reading speed of the target text, including: According to the comprehensive difficulty score of the difficult sentence, a reduction ratio of the reading speed is calculated using a dynamic adjustment algorithm, and the initial reading speed of the difficult sentence is adjusted; Based on the adjusted reading speed of the difficult sentence, a sudden change in reading speed between the difficult sentence and adjacent sentences is eliminated through a smoothing algorithm to obtain a standard reading speed of the target text, and the sentence is marked to attract students' attention.
5. The method according to claim 1, characterized in that When the behavior data is initial behavior data, the step S3 of acquiring the behavior data of the target student in real time to generate the learning state feature vector of the target student includes: Acquire the audio data when the target student completes the specific reading task, and the text data when answering questions, as the initial behavior data; Performing speech recognition on the acquired audio data to obtain corresponding text content, and comparing it with the standard reading text to count the pause frequency and repetition number of the target student during the reading process; Calculating the correctness of the target student's answers based on the acquired text data; Based on the pause frequency, the number of repetitions and the correct answer rate, a learning status feature vector of the target student is generated.
6. The method according to claim 1, characterized in that When the behavior data is learning behavior data, the step S3 of acquiring the behavior data of the target student in real time to generate the learning state feature vector of the target student includes: Acquire learning behavior data including reading speed and eye movement trajectory of the target student during the shadow reading process; The learning behavior data is preprocessed to remove noise and outliers, and key feature parameters are extracted from the preprocessed learning behavior data to generate a learning state feature vector of the target student.
7. The method according to claim 1, characterized in that The step S4 of adjusting the standard reading speed based on the current comprehension ability level of the target student includes: Comparing the current comprehension ability level of the target student with the historical comprehension ability level to determine the changing trend of the comprehension ability level of the target student; When the target student's comprehension ability level decreases, the standard reading speed is adjusted.
8. The method according to claim 7, characterized in that The standard reading speed is adjusted by the following method: Based on the current comprehension ability level of the target student, matching a corresponding adjustment coefficient from a preset reading speed adjustment coefficient mapping table; The current reading speed of the target student is obtained, and multiplied by the matched adjustment coefficient to obtain the adjusted standard reading speed.
9. The method according to claim 1, characterized in that The step S5 of performing speech synthesis on the target text according to the adjusted standard reading speed to obtain the teaching audio of the target text includes: Obtaining the adjusted standard reading speed to determine parameters related to the reading speed; Inputting the parameters related to the reading speed and the target text into a pre-trained speech synthesis model, and generating a corresponding speech feature sequence through a forward propagation process; The speech feature sequence is converted into an audio signal by using a vocoder, and waveform reconstruction is performed to obtain teaching audio corresponding to the target text.
10. The method according to claim 9, characterized in that The parameters related to the reading speed and the target text are input into a pre-trained speech synthesis model, and a corresponding speech feature sequence is generated through a forward propagation process, including: Conduct semantic and grammatical analysis on the target text to determine the structure and pause position of each sentence; Based on the structure, pause position and parameters related to the reading speed of each sentence, a Markov model is used to perform modeling to obtain initial speech parameters related to the speech rhythm and pause position of the target text; Dynamically adjusting the initial speech parameters using a back propagation algorithm to obtain optimized speech parameters and generate a corresponding speech feature sequence; Determining whether the speech feature sequence meets the requirements of naturalness and accuracy; When the speech feature sequence meets the requirements of naturalness and accuracy, the speech feature sequence is output; otherwise, the speech parameters are continuously optimized until the speech feature sequence meets the requirements of naturalness and accuracy.