English word learning tool and use method thereof
By integrating Q-learning algorithm, collaborative filtering algorithm and DTW technology, real-time analysis of user pronunciation and learning data, dynamically adjusting the exercise content and difficulty of English word learning tools, solving the problem of lack of personalization and dynamic adaptation of existing tools, and realizing the Personalized and intelligent English word learning experience.
Patent Information
- Application Number
- CN202411046903.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2024-11-19
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing English word learning tools lack personalized and dynamic adaptability, and are difficult to meet the personalized learning needs of different users, especially in identifying and adjusting learning strategies for users’ weak links.
Using Q-learning algorithm, collaborative filtering algorithm and DTW technology means, users' pronunciations are analyzed in real time through speech acquisition and deep learning models, dynamically adjusting the content and difficulty of the exercises, and providing a personalized learning experience.
It realizes a personalized, intelligent and efficient English word learning experience, significantly improves the learning effect and satisfaction of users, and can dynamically adjust the content and difficulty of exercises based on the user's real-time feedback to meet the learning needs of different users.
Smart Images

Figure CN118982947A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of intelligent learning tools, and more specifically, to an English word learning tool and a method for using the tool. Background Art
[0002] In English learning, the accumulation and application of vocabulary is particularly important. Existing English word learning tools mostly focus on the learning and practice of words and their expanded vocabulary, but ignore the practice of oral pronunciation, which makes English learning mostly stay on paper exams and difficult to apply to real life and work.
[0003] Patent document CN210895920U discloses an English word auxiliary learning device, which facilitates word query through the English auxiliary learning machine, and reviews the learned words to learn new words. Accurate pronunciation can also deepen the user's impression of the word, which is conducive to improving learning efficiency. However, it only has a single function of word query pronunciation, and cannot evaluate the user's pronunciation, and thus cannot improve the user's pronunciation accuracy. Prior art CN115545980A discloses a word auxiliary learning system for English education, which uses an English word test unit of a learning achievement test unit to test and judge the spoken words of college students, and judges the college students' word accuracy through the calculation and analysis of voice big data. Whether the pronunciation is accurate, a virtual person is generated through virtual reality technology, and a randomly learned English word is given to pronounce the word. Through the calculation and analysis of voice big data, the mouth shape of college students' pronunciation is displayed in a virtual scene. Through comparative analysis of mouth shape and pronunciation form, college students can simulate learning and increase the interest of English word learning. Although the word-assisted learning system evaluates the user's oral pronunciation and provides measures to improve learning, it still provides fixed learning content and exercises in a static way, lacks personalization and dynamic adaptability, and is difficult to meet the personalized learning needs of different users, especially in identifying and adjusting learning strategies based on users' weak links. Summary of the invention
[0004] 1. Technical problems to be solved
[0005] The purpose of this application is to provide an English word learning tool and a method of using the same, which solves the technical problem raised in the above background technology that the existing English word learning tools lack personalization and dynamic adaptability, and are difficult to meet the personalized learning needs of different users. By integrating multiple technical means such as Q-learning algorithm, collaborative filtering algorithm and DTW, the user's learning needs and weak links are analyzed in real time, and the generated exercise content is dynamically adjusted according to the user's real-time feedback, thereby providing more targeted learning content.
[0006] 2. Technical solution
[0007] Based on the above purpose, the present application provides an English word learning tool, including a voice collector, a speaker, a controller and a computer-readable storage medium, wherein an English word learning system is stored in the computer-readable storage medium, and the English word learning system includes a pronunciation analysis module, an automatic pronunciation scoring module and an adaptive exercise generation module, wherein:
[0008] The pronunciation analysis module obtains the standard pronunciation of the learning target, uses the deep learning model to perform real-time analysis on the characteristic parameters of the user's pronunciation collected by the voice collector, and obtains the pronunciation analysis result by comparing the difference between the characteristic parameters of the user's pronunciation and the standard pronunciation;
[0009] The automatic pronunciation scoring module scores the characteristic parameters of the user's pronunciation individually based on the pronunciation analysis results, generates a characteristic scoring vector, and calculates the comprehensive score of the user's pronunciation based on the characteristic scoring vector;
[0010] The adaptive exercise generation module analyzes the user's learning data, identifies the user's weaknesses, and uses an adaptive algorithm to dynamically generate personalized exercise content based on the user's current learning level and weaknesses.
[0011] Furthermore, the English word learning system also includes a dynamic difficulty adjustment module and an exercise optimization module, wherein the dynamic difficulty adjustment module adjusts the difficulty and complexity of the personalized exercise content in real time by analyzing the user's learning data, so that the user always learns within the difficulty range suitable for him / her;
[0012] The exercise optimization module collects and analyzes a large amount of user learning data, dynamically optimizes the difficulty of exercises through data mining technology, and regularly updates and expands the question bank content to optimize questions.
[0013] Furthermore, the pronunciation analysis module includes a feature extraction unit and a standard comparison unit, wherein the feature extraction unit extracts key feature parameters from the input user pronunciation audio, the feature parameters including but not limited to pitch, speaking speed, stress position and fluency, and generates a feature vector; the standard comparison unit compares the user pronunciation with the annotated pronunciation based on the feature parameters extracted by the feature extraction unit, and calculates the difference between the two.
[0014] Furthermore, the automatic pronunciation scoring module includes a feature scoring unit and a comprehensive scoring unit, wherein the feature scoring unit scores each feature parameter individually through a dynamic time warping model based on the difference comparison data between the feature parameters of the user pronunciation and the standard pronunciation, and generates a feature scoring vector;
[0015] The comprehensive scoring unit calculates the user's comprehensive pronunciation score based on the feature scoring vector:
[0016]
[0017] Among them, T, P, R, and F are the pitch score, stress position score, speaking speed score, and fluency score respectively, w1, w2, w3, and w4 are the weights of the corresponding factors, and the sum of the weights is 1. α is a nonlinear adjustment parameter.
[0018] Based on the above technical solution, the dynamic time warping model is used to calculate the DTW distance between the user pronunciation features and the standard pronunciation features, and the distance calculated by DTW is mapped to a score to convert it into a similarity score.
[0019] Furthermore, the pronunciation analysis module also includes a feedback generation unit and a learning tracking unit, wherein the feedback generation unit generates a feedback report based on the feature score and the comprehensive score result; the learning tracking unit records the user's learning data, analyzes the progress trend, and generates a user learning progress report.
[0020] Furthermore, the adaptive exercise generation module includes a user data analysis unit and an exercise generation unit, wherein the user data analysis unit uses a collaborative filtering algorithm to analyze user learning data, determines similar learning tasks based on the target learning tasks, and infers and identifies the weaknesses of the target user based on the target user's performance and / or scores in similar learning tasks; the exercise generation unit is based on a reinforcement learning model, and dynamically selects and generates exercise content according to the user's current learning level and weaknesses.
[0021] Based on the above technical solution, the Q-learning algorithm is used to build a reinforcement learning model. The structure of the Q-learning algorithm is:
[0022] The state space S is the user's learning data, and the learning data used includes but is not limited to historical and current practice scores, historical and current pronunciation score data, and completion time;
[0023] Action space A is the selection of exercises, which include but are not limited to vocabulary exercises, pronunciation exercises, grammar exercises, listening exercises and speaking exercises;
[0024] The reward function R is the correct rate reward R acc , Pronunciation score reward R score , Completion time reward R time and weak points focus on rewarding R ws The comprehensive reward function is:
[0025] R=W acc ·R acc +W score ·R score +W time ·R time +w ws ·R ws
[0026] Among them, w acc 、w score 、w time 、w ws are the weights of the corresponding factors, and the sum of the weights is 1;
[0027] The Q value is the expected cumulative reward after taking action a in state s.
[0028] The present application also provides a method for using an English word learning tool, comprising the following steps:
[0029] Step 1: Pronunciation practice and analysis: Users practice pronunciation through the pronunciation analysis module. Based on the standard pronunciation of the learning target, the system analyzes the pronunciation tone, speaking speed, stress position and fluency in real time;
[0030] Step 2: Comprehensive pronunciation scoring: After completing the pronunciation practice, the automatic pronunciation scoring module will perform feature scoring and comprehensive scoring on the user's pronunciation based on the pronunciation analysis results;
[0031] Step 3 Pronunciation feedback improvement: Generate a feedback report based on the feature score and comprehensive score results, provide improvement suggestions, and users improve their pronunciation based on the suggestions;
[0032] Step 4: Personalized exercise generation: Analyze the user's learning data based on the collaborative filtering algorithm, determine the user's weaknesses, and use the adaptive algorithm to dynamically generate personalized exercise content based on the user's current learning level and weaknesses, focusing on breaking through the user's weaknesses;
[0033] Step 5: Dynamic difficulty adjustment: By analyzing the user's learning data, the difficulty and complexity of the exercises are adjusted in real time to ensure that the user always learns within the most suitable difficulty range.
[0034] 3. Beneficial effects
[0035] The English word learning tool and method of use of the present application have at least the following technical effects or advantages:
[0036] The English word learning tool successfully solves the problems of personalized learning and dynamic adjustment in existing technologies by integrating multiple technical means such as Q-learning algorithm, collaborative filtering algorithm and DTW. The tool can analyze users' learning needs and weak links in real time, and dynamically adjust the generated exercise content according to users' real-time feedback, thereby providing more targeted learning content.
[0037] Through detailed multi-dimensional reward function design, this tool can enhance user engagement while significantly improving user learning outcomes in terms of accuracy, pronunciation, learning efficiency and weaknesses, thereby achieving a personalized, intelligent and efficient English word learning experience and greatly improving user learning outcomes and satisfaction.
[0038] In the process of real-time adjustment of the complexity of exercise content, the difficulty of exercises required by users is predicted based on their learning data. According to the predicted results and current learning data, the difficulty of exercises can be dynamically adjusted to provide users with the most appropriate exercise difficulty, so that users can always practice in the best learning state to ensure that the user's learning experience remains between the optimal level of challenge and understanding, improve learning effects and user satisfaction, and achieve the goal of personalized education. It is suitable for different users and learning goals and is highly flexible and scalable. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 It is a schematic diagram of the structural framework of the English word learning tool of the present application;
[0040] Figure 2 It is a structural diagram of a pronunciation analysis module in the English word learning system of the present application;
[0041] Figure 3 It is a structural diagram of the automatic pronunciation scoring module in the English word learning system of the present application;
[0042] Figure 4 It is a structural diagram of a dynamic difficulty adjustment module in the English word learning system of the present application;
[0043] Figure 5 It is a flow chart of the method of using the English word learning tool of the present application. DETAILED DESCRIPTION
[0044] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.
[0045] Example 1
[0046] See also Figure 1 The embodiment of the present invention provides an English word learning tool, including a voice collector, a speaker, a controller and a computer-readable storage medium, wherein an English word learning system is stored in the computer-readable storage medium, and the English word learning system includes:
[0047] The pronunciation analysis module obtains the standard pronunciation of the learning target, uses the deep learning model to perform real-time analysis on the feature parameters of the user pronunciation collected by the voice collector, and obtains the pronunciation analysis results by comparing the differences between the feature parameters of the user pronunciation and the standard pronunciation.
[0048] The automatic pronunciation scoring module scores the characteristic parameters of the user's pronunciation individually based on the pronunciation analysis results, generates a characteristic scoring vector, and calculates the comprehensive score of the user's pronunciation based on the characteristic scoring vector.
[0049] The adaptive exercise generation module analyzes the user's learning data, identifies the user's weaknesses, and uses an adaptive algorithm to dynamically generate personalized exercise content based on the user's current learning level and weaknesses.
[0050] The dynamic difficulty adjustment module analyzes the user's learning data and adjusts the difficulty and complexity of personalized exercise content in real time, so that users can always learn within the difficulty range that suits them.
[0051] In this embodiment, the user can take the initial test, check the learning progress, perform pronunciation practice and complete exercises through the user interface of the English word learning tool. After the user registers and takes the initial test, the system evaluates the user's English vocabulary level and pronunciation ability, and can generate a personalized learning path and goals based on the test results. The user can learn English words according to the learning path and goals.
[0052] Among them, for the English pronunciation of the user's learning process, the pronunciation analysis module and the automatic pronunciation scoring module are used to perform pronunciation analysis and scoring. For details, see Figure 2 and Figure 3 The pronunciation analysis module includes a feature extraction unit and a standard comparison unit. The feature extraction unit extracts key feature parameters from the input user pronunciation audio and generates a feature vector, wherein the feature parameters include tone, pitch, speaking speed, stress position, fluency and clarity, etc. The standard comparison unit compares the user pronunciation with the marked pronunciation based on the feature parameters extracted by the feature extraction unit, and calculates the similarity and difference between the two. The automatic pronunciation scoring module includes a feature scoring unit and a comprehensive scoring unit. The feature scoring unit compares the difference data between the feature parameters of the user pronunciation and the standard pronunciation, scores each feature parameter separately through the dynamic time warping model, and generates a feature scoring vector; the comprehensive scoring unit calculates the user's comprehensive pronunciation score based on the feature scoring vector and the weights of each factor.
[0053] In this embodiment, the pronunciation analysis module receives the user's pronunciation audio and performs preliminary processing to convert it into a standard format, such as a 16kHz mono WAV file, and performs noise reduction processing to improve the signal quality. Subsequently, the feature extraction unit uses the MFCC algorithm to extract the spectral features of the audio signal to form time series data. Specifically, a feature extraction model is constructed using a combination of Mel spectrum cepstral coefficients (MFCC) and a convolutional neural network. The implementation process is: using short-time Fourier transform (STFT) to convert the audio signal into a frequency domain signal, calculating the Mel spectrum graph, and converting it into Mel spectrum cepstral coefficients. The input layer of the convolutional neural network receives the MFCC features, the convolution layer + pooling layer extracts local features and reduces the dimension, the fully connected layer performs feature fusion and classification, and the output layer provides the feature extraction results.
[0054] After completing feature extraction, the standard comparison unit uses the dynamic time warping (DTW) model to calculate the difference between the user's pronunciation and the standard pronunciation to obtain a similarity score. Specifically, the dynamic time warping model is used to calculate the DTW distance between the user's pronunciation features and the standard pronunciation features, and the distance calculated by DTW is mapped to the score to convert it into a similarity score. DTW compares time series through time alignment, allowing time series to be nonlinearly stretched and compressed in the time dimension. DTW calculates the minimum distance between the feature parameter time series data of the user's pronunciation and the standard pronunciation. The DTW distance calculation formula is:
[0055] DTW(X,Y)=D(N,M)
[0056] Where X = (x1, X2, ..., x N ) and Y = (y1, y2, ..., y N ) are two time series, representing the user pronunciation features and the standard pronunciation features respectively. The lengths of the time series are N and M respectively. The DTW distance matrix D is defined as N×M, where each element D(i, j) represents the DTW distance between the first i elements of sequence X and the first j elements of sequence Y;
[0057] Initialize the DTW distance matrix D: D(0,0)=0, D(i,0)=D(0,j)=∞(i>0,j>0),
[0058] And perform recursive calculation:
[0059] D((i, j) = d(x i ,y j )+min(D(i-1,j),D(i,j-1),D(i-1,j-1))
[0060] Where d(x i ,y j) represents the local distance between sequences X and Y at points i and j, and the Euclidean distance is usually used to calculate d(x i ,y j )=(x i -y j ) 2 , thus obtaining the final DTW distance.
[0061] In this embodiment, the similarity scores of the four characteristic parameters of pitch, speaking speed, stress position and fluency are calculated as the corresponding characteristic scores, which are respectively recorded as pitch score T, speaking speed score R, stress position score P and fluency score F, where:
[0062] The pitch score f can evaluate the degree of match between the pitch of the user's pronunciation and the standard pitch. It is generally determined by comparing the pitch curve of the user's pronunciation with the standard pitch curve. The score range is 0 (complete mismatch) to 1 (complete match);
[0063] The speech rate score R can assess whether the user's pronunciation speed is moderate, that is, whether it meets the standard speaking speed requirements. It is determined by calculating the difference between the number of words per minute (WPM) of the user's pronunciation and the target speaking speed. The score range is 0 (much lower or much higher than the standard speaking speed) to 1 (standard speaking speed);
[0064] The stress placement score P evaluates whether the user's stress placement in words and sentences is correct. It is determined by analyzing the stress placement and the stress placement of the standard pronunciation. The score ranges from 0 (completely wrong stress placement) to 1 (completely correct stress placement).
[0065] The fluency score F assesses the fluency of the user's pronunciation, checking for pauses, repetitions, or hesitations. It is determined by analyzing the pause time and fluency index in the speech, with a score range of 0 (very unfluent) to 1 (very fluent).
[0066] Substitute each feature score into the comprehensive scoring formula to calculate the user's comprehensive pronunciation score S:
[0067]
[0068] Among them, T, P, R, and F are the pitch score, stress position score, speaking speed score, and fluency score, respectively; w1, w2, w3, and w4 are the weights of the corresponding factors, and the sum of the weights is 1; α is a nonlinear adjustment parameter, which is usually a positive number greater than 1 (such as 2). The formula uses nonlinear adjustment to perform nonlinear weighted processing on pitch and stress position, reflecting the combined influence of pitch score and stress position score. The speed and fluency parts are unified to between 0 and 1 through exponential adjustment, and then nonlinear weighted processing is performed. The final score is processed through a unified nonlinear adjustment parameter α to ensure the rationality of the comprehensive score.
[0069] For example, if the pitch score of a user's pronunciation audio is T=0.7, the speaking speed score is R=0.8, the stress position score is P=0.9, the fluency score is F=0.85, and the weights w1=0.3, w2=0.2, w3=0.25, and w4=0.25, then the comprehensive score S is calculated as follows:
[0070]
[0071] Through the collaborative work of the pronunciation analysis module and the automatic pronunciation scoring module, a comprehensive analysis and scoring of user pronunciation can be achieved. The pronunciation analysis module is responsible for audio feature extraction, standard comparison and feedback generation, and provides personalized improvement suggestions. The automatic pronunciation scoring module focuses on scoring multiple pronunciation features and comprehensively considers the mutual influence between multiple pronunciation features to ensure the accuracy and logic of the scoring.
[0072] After the automatic pronunciation scoring module completes the pronunciation scoring, the pronunciation analysis module can analyze and provide feedback to the user based on the learning records and pronunciation scoring data. Specifically, the pronunciation analysis module also includes a feedback generation unit and a learning tracking unit. The feedback generation unit generates a feedback report based on the feature scoring and comprehensive scoring results, and provides specific improvement suggestions and guidance. The learning tracking unit records the user's learning data, provides historical data analysis and trend charts, analyzes progress trends, and generates a user learning progress report to help users continue to improve.
[0073] Based on the above embodiments, see Figure 3 , use the adaptive exercise generation module to determine the user's weaknesses, and dynamically generate personalized exercise content based on the user's current learning level and weaknesses, so as to focus on breaking through the user's weaknesses. Specifically, the adaptive exercise generation module includes a user data analysis unit and an exercise generation unit. The user data analysis unit uses a collaborative filtering algorithm to analyze user learning data, determines similar learning tasks based on the target learning tasks, and infers and identifies the target user's weaknesses based on the target user's performance and / or score in similar learning tasks; the exercise generation unit is based on a reinforcement learning model, and dynamically selects and generates exercise content based on the user's current learning level and weaknesses.
[0074] In this embodiment, the user data analysis unit performs data cleaning and preprocessing on the collected user learning data, and uses a collaborative filtering algorithm to analyze the user learning data to identify the weaknesses of the target user, wherein the learning data includes:
[0075] (1) Learning records: including all learning tasks completed by the user and their corresponding completion time and number of times.
[0076] (2) Practice results: the user's score and incorrect answers in each practice.
[0077] (3) Completion time: the time it takes the user to complete the exercise.
[0078] (4) Accuracy: The accuracy of the user’s answers during practice, including the accuracy of single subjects and the comprehensive accuracy of multiple subjects for various exercise types.
[0079] (5) Pronunciation score: Pronunciation feature score and comprehensive score obtained by the pronunciation analysis module and the automatic scoring module.
[0080] (6) Learning time: the time users spend on each learning task.
[0081] (7) Progress tracking: The user’s progress throughout the learning process, such as completed chapters, achieved learning goals, etc.
[0082] (8) Improvement: The user’s progress after repeated practice.
[0083] (9) User feedback: users’ subjective evaluation of the difficulty and content of exercises.
[0084] …
[0085] The learning data used in the collaborative filtering algorithm are pronunciation scores, practice scores, accuracy, and learning time. Based on the user's performance in different learning tasks, the user's learning data is compared with the overall average level to determine the user's weak points. Specifically, project-based collaborative filtering is used to first find other learning tasks similar to the target learning task, summarize the target user's scores and / or scores in similar learning tasks, infer and identify the target user's weaknesses, and use the Pearson correlation coefficient to calculate the similarity between different learning tasks:
[0086]
[0087] Among them, Sim(i,j) is the similarity between learning tasks i and learning tasks j, U ij is the set of users who have completed learning tasks i and j, r ui and r uj are the scores or ratings of user u in learning task i and learning task j, respectively. and are the average scores or ratings of learning task i and learning task j respectively.
[0088] Predict user u’s score for target learning task i based on learning task similarity:
[0089]
[0090] in, is the predicted score, N(i) is the set of learning tasks similar to learning task i, and Sim(i,j) is the similarity between item i and item j.
[0091] By comparing the user's actual scores and predicted scores for different learning tasks, we can identify the user's weak links in certain similar learning tasks. The inference formula is:
[0092]
[0093] Among them, Δr u,i represents the difference between the predicted score and the actual score of user u for learning task i. u,i If it is negative and its absolute value exceeds the set threshold, it means that the user’s actual performance on learning task i is lower than expected, and it is judged as a weak link.
[0094] It should be noted that different standards can be used to judge weak links. In addition to the above-mentioned score gap, one or more of the following can be used to comprehensively evaluate the user's weak links in each learning task: error rate, whether the score is lower than the average level, the learning time of a certain task is significantly higher than other tasks but the score does not meet expectations, historical progress analysis, etc.
[0095] The exercise generation unit in the adaptive exercise generation module uses the Q-learning algorithm in the reinforcement learning model to dynamically select and generate exercise content based on the user's current learning level and weaknesses. The structure of the Q-learning algorithm is:
[0096] (1) The state space S is the user's learning data, which includes but is not limited to historical and current practice scores, historical and current pronunciation score data, and completion time;
[0097] (2) Action space A is the selection of exercises, which include but are not limited to vocabulary exercises, pronunciation exercises, grammar exercises, listening exercises, and speaking exercises;
[0098] (3) Use the exercise accuracy, pronunciation score, and completion time to characterize the user's current learning level, and construct a reward function based on the user's current learning level and weaknesses, that is, the reward function R is the exercise accuracy reward R acc , Pronunciation score reward R score , Completion time reward R time and weak points focus on rewarding R ws The comprehensive reward function is:
[0099] R=w acc ·R acc +w score ·R score +w time ·R time +wws ·R ws
[0100] Among them, w acc 、w score 、w time 、w ws are the weights of the corresponding factors, and the sum of the weights is 1;
[0101] Exercise Correct Rate Reward R acc : Rewards are given based on the user's accuracy in the exercises. The higher the accuracy, the greater the reward. Exercises refer to exercises excluding pronunciation exercises and oral exercises. This avoids overlap between exercise accuracy rewards and pronunciation score rewards. The formula is:
[0102]
[0103] Among them, R max It is the maximum value of the correctness reward for the exercise.
[0104] Pronunciation score reward R score : Rewards are given based on the user's pronunciation score. The higher the pronunciation score, the greater the reward.
[0105]
[0106] Among them, R max is the maximum value of the pronunciation score reward.
[0107] Completion time reward R time : Rewards are given based on the time it takes users to complete the exercises. The shorter the completion time, the greater the reward.
[0108]
[0109] Among them, max_time is the maximum completion time, R max is the maximum value of the completion time reward.
[0110] Weakness Focus Reward R ws :
[0111]
[0112] Among them, C min and C max is the minimum and maximum ratios set, G is the reward value set, ω and is the reward value coefficient, generally ω>1, It can be selected between 0 and 1, or a value greater than 1.
[0113] (4) The Q value is the expected cumulative reward after taking action a in state s.
[0114] After completing the construction of the reinforcement learning model based on the Q-learning algorithm, input the training data, train the model through the Q-learning algorithm, continuously update the Q-table, use the ε-greedy strategy to optimize the strategy, and adjust the content of the exercises. Use the validation data set to evaluate the model and observe the performance of the model under different exercise content adjustment strategies. According to the evaluation results, adjust the model parameters and reward function to further optimize the model performance. The specific process of implementing the Q-learning algorithm is as follows:
[0115] (1) Initialize Q value: Initialize the Q table, where each element Q(s,a) represents the Q value of taking action a in state s. The initial value can be set to zero or a random value.
[0116] (2) Action selection: Use the ε-greedy strategy to select actions, select a random action with a probability of ε, and select the action with the largest current Q value with a probability of 1-ε.
[0117] (3) Execute action: Execute action a and obtain the new state s' and immediate reward r based on the user's feedback.
[0118] (4) Update Q value: Use Q-learning to update the formula:
[0119]
[0120] Among them, α is the learning rate and γ is the discount factor.
[0121] (5) Repeat the above steps: Repeat the process of selecting actions, executing actions, and updating Q values until the set number of iterations or cumulative reward value is reached.
[0122] By combining the user's real-time feedback, such as current learning level and weaknesses, and using the Q-learning algorithm to dynamically adjust the selected and generated exercises, personalized learning path optimization can be achieved. The system can respond to the user's learning performance in real time, and by continuously optimizing the Q value table, improve the accuracy and adaptability of exercise generation, and ultimately achieve the purpose of improving the user's learning effect.
[0123] Furthermore, considering that the adaptive exercise generation module focuses on the personalized selection of exercises, in order to make the generated personalized exercise content consistent with the student's knowledge level, the difficulty of the exercises can be adjusted in real time to adapt to the user's learning status, so that the user can always learn within the difficulty range that suits him / her. Therefore, the English word learning system also includes a dynamic difficulty adjustment module, such as Figure 4As shown, by analyzing the user's learning data, the difficulty and complexity of the personalized exercise content are adjusted in real time, so that the dynamic difficulty adjustment module extends the function based on the adaptive exercise generation module.
[0124] The dynamic difficulty adjustment module includes a dynamic difficulty prediction unit and an exercise difficulty adjustment unit. The dynamic difficulty prediction unit uses a random forest regression model to predict the difficulty of exercises required by users based on the user's learning data, such as pronunciation scores, exercise scores, accuracy, and learning time, and outputs the predicted difficulty level of the exercises, such as 1 to 5, with 1 being easy and 5 being difficult. The implementation process is as follows: prepare the user's historical learning data set, mark the target difficulty level of each record, train a random forest regression model based on the data set, map the input features to the output target, and after training, input the user's current learning features into the model to predict the difficulty level that is suitable for the user in the current situation. The model's input data can then be dynamically updated based on the user's real-time learning performance to ensure that the model always reflects the latest learning status.
[0125] The exercise difficulty adjustment unit sets the initial difficulty based on the predicted difficulty level obtained by the dynamic difficulty prediction unit and the user's historical learning data, and adjusts it according to the user's feedback and performance after completing each exercise. The dynamic adjustment formula is:
[0126] New Difficulty=Predicted Difficulty+β×(UserPerformance-TargetPerformance)
[0127] Among them, Predicted Difficulty is the difficulty predicted by the model, β is the adjustment coefficient that controls the adjustment step, User Performance is the user's current performance score, and Target Performance is the target performance, such as 80% accuracy.
[0128] When the user performs well at the current difficulty level, higher difficulty exercises are automatically selected, and when the user performs poorly, the user is relegated to lower difficulty exercises to strengthen the basics.
[0129] Among them, the adjustment of difficulty level may include the following parameters:
[0130] (1) Question type: Choose different types of questions according to the difficulty level, such as multiple-choice questions, fill-in-the-blank questions, and writing questions.
[0131] (2) Question length: Adjust the length and complexity of the question content, for example, increase or decrease the length of the question stem, and adjust the number and content of options.
[0132] (3) Grammatical difficulty: For language learning, the grammatical structures used can be adjusted, such as from simple sentences to complex sentences, and from common vocabulary to advanced vocabulary.
[0133] (4) Topic relevance: Select exercises that are highly relevant to the user’s current learning topic to improve learning outcomes.
[0134] In the process of adjusting the complexity of exercise content in real time, the predicted difficulty of the exercise that is most suitable for the user at the moment plays a key role. Based on the user's learning data, the difficulty of the exercise required by the user is predicted. According to the predicted results and the current learning data, an intelligent dynamic exercise difficulty adjustment system is designed, which can dynamically adjust the difficulty of the exercise to provide the user with the most suitable exercise difficulty, so that the user can always practice in the best learning state to ensure that the user's learning experience remains between the best challenge and understanding level, improve learning effects and user satisfaction, and achieve the goal of personalized education. It is suitable for different users and learning goals and has high flexibility and scalability.
[0135] In addition, the English word learning system can also include an exercise optimization module, which collects and analyzes a large amount of user learning data, uses data mining technology to dynamically optimize the difficulty of exercises, and regularly updates and expands the question bank content to optimize questions.
[0136] Example 2
[0137] The following describes the method for using the English word learning tool provided by the present invention, which can be used in correspondence with the English word learning tool described above. Figure 5 As shown, the embodiment of the present invention discloses a method for using an English word learning tool, comprising the following steps:
[0138] Step 1: Pronunciation practice and analysis: Users practice pronunciation through the pronunciation analysis module. Based on the standard pronunciation of the learning target, the system analyzes the pronunciation tone, speaking speed, stress position and fluency in real time;
[0139] Step 2: Comprehensive pronunciation scoring: After completing the pronunciation practice, the automatic pronunciation scoring module will perform feature scoring and comprehensive scoring on the user's pronunciation based on the pronunciation analysis results;
[0140] Step 3 Pronunciation feedback improvement: Generate a feedback report based on the feature score and comprehensive score results, provide improvement suggestions, and users improve their pronunciation based on the suggestions;
[0141] Step 4: Personalized exercise generation: Analyze the user's learning data based on the collaborative filtering algorithm, determine the user's weaknesses, and use the adaptive algorithm to dynamically generate personalized exercise content based on the user's current learning level and weaknesses, focusing on breaking through the user's weaknesses;
[0142] Step 5: Dynamic difficulty adjustment: By analyzing the user's learning data, the difficulty and complexity of the exercises are adjusted in real time to ensure that the user always learns within the most suitable difficulty range.
[0143] Through the above technical solutions, the user's pronunciation level can be comprehensively evaluated, and targeted improvement suggestions can be provided to help users gradually improve their pronunciation quality and English speaking ability. Through the cooperation of collaborative filtering algorithm and adaptive algorithm, personalized exercise content can be dynamically generated. As users continue to practice pronunciation and complete exercises, the system continuously adjusts the learning path and exercise difficulty based on the latest learning data, forming a dynamic, personalized learning and feedback loop to ensure that users continue to improve.
[0144] It should be noted that the system provided in the above embodiment only uses the division of the above functional modules as an example to implement its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.
[0145] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.
Claims
1. An English word learning tool, comprising a voice collector, a speaker, a controller and a computer-readable storage medium, wherein an English word learning system is stored in the computer-readable storage medium, characterized in that: The English word learning system includes: The pronunciation analysis module obtains the standard pronunciation of the learning target, uses the deep learning model to perform real-time analysis on the characteristic parameters of the user's pronunciation collected by the voice collector, and obtains the pronunciation analysis result by comparing the difference between the characteristic parameters of the user's pronunciation and the standard pronunciation; The automatic pronunciation scoring module scores the characteristic parameters of the user's pronunciation individually based on the pronunciation analysis results, generates a characteristic scoring vector, and calculates the comprehensive score of the user's pronunciation based on the characteristic scoring vector; The adaptive exercise generation module analyzes the user's learning data, identifies the user's weaknesses, and uses an adaptive algorithm to dynamically generate personalized exercise content based on the user's current learning level and weaknesses.
2. The English word learning tool according to claim 1, characterized in that: The English word learning system also includes: a dynamic difficulty adjustment module, which adjusts the difficulty and complexity of personalized exercise content in real time by analyzing the user's learning data, so that users can always learn within the difficulty range that suits them.
3. The English word learning tool according to claim 1, characterized in that: The English word learning system also includes: an exercise optimization module, which collects and analyzes a large amount of user learning data, dynamically optimizes the difficulty of exercises through data mining technology, and regularly updates and expands the question bank content to optimize questions.
4. The English word learning tool according to claim 1, characterized in that: The pronunciation analysis module includes a feature extraction unit and a standard comparison unit, wherein the feature extraction unit extracts key feature parameters from the input user pronunciation audio, the feature parameters including but not limited to pitch, speech speed, stress position and fluency, and generates a feature vector; The standard comparison unit compares the user pronunciation with the marked pronunciation based on the feature parameters extracted by the feature extraction unit, and calculates the difference between the two.
5. The English word learning tool according to claim 4, characterized in that: The automatic pronunciation scoring module includes a feature scoring unit and a comprehensive scoring unit, wherein the feature scoring unit scores each feature parameter individually through a dynamic time warping model based on the difference comparison data between the feature parameters of the user's pronunciation and the standard pronunciation, and generates a feature scoring vector; The comprehensive scoring unit calculates the user's comprehensive pronunciation score based on the feature scoring vector: Among them, T, P, R, and F are the pitch score, stress position score, speaking speed score, and fluency score respectively, w1, w2, w3, and w4 are the weights of the corresponding factors, and the sum of the weights is 1. α is a nonlinear adjustment parameter.
6. The English word learning tool according to claim 5, characterized in that: The dynamic time warping model is used to calculate the DTW distance between the user's pronunciation features and the standard pronunciation features, and the distance calculated by DTW is mapped to a score to convert it into a similarity score.
7. The English word learning tool according to claim 5, characterized in that: The pronunciation analysis module also includes a feedback generation unit and a learning tracking unit, wherein the feedback generation unit generates a feedback report according to the feature scoring and the comprehensive scoring results; The learning tracking unit records the user's learning data, analyzes progress trends, and generates user learning progress reports.
8. The English word learning tool according to claim 1, characterized in that: The adaptive exercise generation module includes a user data analysis unit and an exercise generation unit, wherein the user data analysis unit uses a collaborative filtering algorithm to analyze user learning data, determines similar learning tasks based on the target learning task, and infers and identifies the weaknesses of the target user based on the target user's performance and / or score in similar learning tasks; The exercise generation unit is based on the reinforcement learning model and dynamically selects and generates exercise content according to the user's current learning level and weaknesses.
9. The English word learning tool according to claim 8, characterized in that: Use the Q-learning algorithm to build a reinforcement learning model. The structure of the Q-learning algorithm is: The state space S is the user's learning data, and the learning data used includes but is not limited to historical and current practice scores, historical and current pronunciation score data, and completion time; Action space A is the selection of exercises, which include but are not limited to vocabulary exercises, pronunciation exercises, grammar exercises, listening exercises and speaking exercises; The reward function R is the correct rate reward R acc , Pronunciation score reward R score , Completion time reward R time and weak points focus on rewarding R ws The comprehensive reward function is: R=w acc ·R acc +w score ·R score +w time ·R time +w ws ·R ws Among them, w acc 、w score 、w time 、w ws are the weights of the corresponding factors, and the sum of the weights is 1; The Q value is the expected cumulative reward after taking action a in state s.
10. A method for using an English word learning tool, characterized in that: The following steps are involved: Step 1: Pronunciation practice and analysis: Users practice pronunciation through the pronunciation analysis module. Based on the standard pronunciation of the learning target, the system analyzes the pronunciation tone, speaking speed, stress position and fluency in real time; Step 2: Comprehensive pronunciation scoring: After completing the pronunciation practice, the automatic pronunciation scoring module will perform feature scoring and comprehensive scoring on the user's pronunciation based on the pronunciation analysis results; Step 3 Pronunciation feedback improvement: Generate a feedback report based on the feature score and comprehensive score results, provide improvement suggestions, and users improve their pronunciation based on the suggestions; Step 4: Personalized exercise generation: Analyze the user's learning data based on the collaborative filtering algorithm, determine the user's weaknesses, and use the adaptive algorithm to dynamically generate personalized exercise content based on the user's current learning level and weaknesses, focusing on breaking through the user's weaknesses; Step 5: Dynamic difficulty adjustment: By analyzing the user's learning data, the difficulty and complexity of the exercises are adjusted in real time to ensure that the user always learns within the most suitable difficulty range.
Citation Information
Cited By
Multi-dimensional oral language ability quantitative evaluation and weakness tracing system
CN120071693A
Emotion accompanying type children education interaction method and device, medium and program product
CN121365130A
