English vocabulary cognitive training method and system based on interval repetition memory method

Abstract

The invention provides an English vocabulary cognition training method and system based on an interval repetition memory method, and relates to the technical field of language cognition training, abnormal value elimination and normalization preprocessing are performed on collected data, a memory intensity coefficient is calculated, a memory intensity level is judged, and a final review interval is generated in combination with a vocabulary basic interval and a special adjustment rule; each layer is matched with a differentiated multi-modal training task, and the vocabulary adding proportion, cognitive layer dynamic switching and fatigue relieving links in the training process are controlled. According to the method, multi-dimensional data such as learning duration, memory performance and vocabulary difficulty and dynamic review intervals are fused, compared with fixed period review, the one-week memory retention rate of vocabularies is increased to 85%, the two-week memory retention rate reaches 72%, the long-term memory effect is remarkably enhanced, and the problem that multi-dimensional data such as intervals, non-fused reaction time and cognitive states are adjusted only based on correct rate is solved. And personalized and efficient vocabulary cognition training requirements cannot be met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of language cognitive training, in particular to an English vocabulary cognitive training method and system based on spaced repetition. BACKGROUND

[0002] English vocabulary cognitive training refers to a process of improving learners' recognition, understanding, memory and application of English vocabulary through a series of purposeful and structured exercises and activities.

[0003] The traditional method of English vocabulary cognitive training in the prior art, such as Ebbinghaus fixed curve, uses preset intervals of 1 day, 3 days and 7 days for training, but only adjusts the interval based on accuracy, without integrating multi-dimensional data such as reaction time and cognitive state, and some multi-modal training schemes only add voice function without combining with interval repetition depth, which cannot meet the needs of individualized and efficient vocabulary cognitive training. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides an English vocabulary cognitive training method and system based on spaced repetition, to solve the problems mentioned in the background art.

[0005] To solve the above technical problems, the present application is realized by the following technical scheme: The present application is an English vocabulary cognitive training method and system based on spaced repetition, which specifically includes the following steps: One, system initialization and individual memory parameter configuration: the user inputs the English level, daily training time and memory achievement target into the parameter initialization module system, the system loads the pre-constructed vocabulary difficulty-initial interval mapping table, and initializes the core algorithm, including: LSTM memory decay prediction model, multi-modal training rule engine, error type classification model parameters, and simultaneously matches the vocabulary library containing multi-modal resources according to the user level; Two, the multi-dimensional data acquisition module acquires real-time data from multiple dimensions of memory performance, and synchronously acquires core memory data such as accuracy R, reaction time T and error type E at a preset frequency, as well as cognitive state data such as concentration C and fatigue F, and the acquired data is transmitted to the cloud server for storage through the HTTPS protocol; Three, the AI dynamic interval calculation module performs outlier rejection and normalization preprocessing on the collected data, calculates the memory strength coefficient and determines the memory strength level, generates the final review interval in combination with the vocabulary basic interval and special adjustment rules; Four, the multi-modal hierarchical training module performs cognitive training execution, divides the vocabulary into basic, advanced and proficient cognitive layers according to the memory strength coefficient, matches differential multi-modal training tasks for each layer, controls the vocabulary addition proportion in the training process, dynamically switches the cognitive layers and relieves fatigue. Five, the feedback optimization module pushes multi-dimensional feedback reports to the user end, teacher or parent end, updates the LSTM memory decay prediction model parameters daily, and optimizes the vocabulary difficulty-initial interval mapping table and training task weight monthly.

[0006] Further, the English level in step one is A1-C2, A1-C2 corresponds to the European Common Reference Framework, and the daily training time is 30-90 minutes; In the vocabulary difficulty-initial interval mapping table, the A1 level vocabulary has an initial interval of 1 day, the B2 level has an initial interval of 3 days, and the C1 level has an initial interval of 5 days; The error type classification model is based on a CNN text classification algorithm and can identify 8 types of errors such as semantic confusion, spelling errors, and pronunciation deviation; The vocabulary library multi-modal resource includes: a British or American dual pronunciation audio with a sampling rate of 44.1 kHz and a bit rate of 128 kbps, 3-5 oral or written example sentences with an adaptation degree of ≥90% selected based on a BERT semantic model, and spelling resources with root and affix disassembly.

[0007] Further, the core memory data collection frequency in step two is once every 1 vocabulary training task completed, wherein the reaction time timeout is 30000 ms, and the error type is automatically identified by the error type classification model; The cognitive state data collection frequency is once every 5 minutes, the concentration is based on eye movement recognition of the training terminal, for example, a gaze screen duration ratio of ≥80% is recorded as 100%, and a ratio of <50% is recorded as 0%, and the fatigue degree is calculated based on reaction time fluctuation, a fluctuation amplitude of >50% is recorded as 100%, and a fluctuation amplitude of <10% is recorded as 0%; The data is stored in a MySQL distributed database, each data is associated with a user ID-vocabulary ID-training timestamp-data type label, and the retention period is ≥1 year.

[0008] Further, the data preprocessing in step three adopts the box plot method to eliminate extreme data with a reaction time of <100 ms or >30000 ms, and normalizes the accuracy, reaction time, concentration, and fatigue to the [0,1] interval to obtain a standardized data matrix; The calculation formula of the memory strength coefficient M is M=0.4×R'+0.3×(1-T')+0.2×C'-0.1×F', wherein R' is the standardized accuracy, T' is the standardized reaction time, C' is the standardized concentration, and F' is the standardized fatigue, and the weight is calibrated by 10000+ user training data; according to the LSTM memory decay model, the future 7-day memory decay value is output by inputting the 4 data R', T', C', and F'; According to the memory strength grade division standard: M≥0.8 is high memory strength, 0.5≤M<0.8 is medium memory strength, and M<0.5 is low memory strength.

[0009] Further, the basic interval in step three is retrieved from the vocabulary difficulty-initial interval mapping table according to the vocabulary difficulty level to calculate the final review interval D, and the interval adjustment formula D=D0×(1+0.6×M-0.3×(1-M)) is used. The special adjustment rules include: shortening the interval by 20% when the error type is semantic confusion, lengthening the interval by 50% when M≥0.9 for three times in a row, rounding the interval calculation result, and generating a dynamic interval table.

[0010] Further, the multi-modal hierarchical training task in step four meets: Basic cognitive layer, including word plus explanation pairing, voice follow-up verification task; Advanced cognitive layer, including context fill-in-the-blank, spelling dictation task; Master cognitive layer, including scene sentence making, vocabulary association task.

[0011] Further, in step four, the training process control meets that in daily training, the proportion of new words is ≤40%, the words to be reviewed and new words are selected according to the dynamic interval table, the proportion of new words is ≤40%, and memory overload is avoided; Cognitive layer dynamic switching rule: automatically downgrade to basic cognitive layer for two consecutive task errors, and automatically upgrade to master cognitive layer for three consecutive task corrects; Every 10 vocabulary training inserts ≤3 minutes of vocabulary game to relieve fatigue and synchronously collect cognitive state data.

[0012] Further, in step five, the user end feedback includes: real-time feedback for each task, and daily personal memory report; The teacher or parent end views the 7-day memory trend chart and weak cognitive layer distribution, supports pushing targeted training tasks; Model optimization includes: adding training data to the training set of the LSTM memory decay prediction model every day, updating model parameters, and optimizing the vocabulary difficulty-initial interval mapping table and training task weight based on 1 month of data every month.

[0013] The English vocabulary cognitive training system based on interval repetition memory method comprises a parameter initialization module, a multi-dimensional data acquisition module, an AI dynamic interval calculation module, a multi-modal hierarchical training module and a feedback optimization module.

[0014] Further, the training terminal comprises a tablet computer and earphones, the tablet computer is an iPadPro with a screen size of 11 inches, and the earphones are Huawei FreeBudsPro2; The training set of the LSTM memory decay model in the AI dynamic interval calculation module comprises 100,000 pieces of data of 5,000 users, the prediction error is less than or equal to 3%, and the error type set comprises 50,000 samples of 8 types of errors. In the multi-modal hierarchical training module, the voice recognition of the data acquisition in daily training adopts an MFCC plus DTW algorithm, and the semantic analysis adopts a BERT or GPT-3.5 fine-tuning model.

[0015] The English vocabulary cognitive training method and system based on the interval repetition memory method have the following beneficial effects: In use, the present application fuses multi-dimensional data such as learning duration, memory performance and vocabulary difficulty, dynamically reviews intervals, and compared with fixed cycle review, the vocabulary 1-week memory retention rate is improved to 85%, the 2-week retention rate is 72%, and the long-term memory effect is significantly enhanced.

[0016] The method covers the complete cognitive path of memory-understanding-application-innovation, combines voice recognition, spelling practice and real context application, realizes multi-sensory collaborative input, and the vocabulary application ability of student users in practical language tasks such as sentence making and filling in the blanks is improved by 60%.

[0017] The system automatically identifies and subdivides 8 typical errors such as spelling errors, misuse of word forms and improper collocation, provides personalized error correction suggestions and reinforcement training schemes, helps users to achieve targeted breakthrough, and the average monthly error rate is reduced by 35%.

[0018] Through intelligent task design and attention monitoring, the proportion of daily effective training time is improved to 75%, the vocabulary mastery amount in unit time is increased by 2 times, and the learning input-output ratio is significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments will be briefly introduced below.

[0020] The accompanying drawings, which are described below, relate to some embodiments of the present application and are not limiting of the present application.

[0021] In the drawings: Figure 1 The overall architecture diagram of the practice system of the present application is shown, including parameter initialization, multi-dimensional data acquisition, AI dynamic interval calculation, multi-modal hierarchical training, feedback optimization of 5 big modules, and data interaction between modules through HTTPS protocol; Figure 2 The parameter initialization input flowchart of the present application is shown; Figure 3 The multi-dimensional data acquisition flowchart of the present application is shown; Figure 4 The multi-modal hierarchical training task schematic diagram of the present application is shown; Figure 5 The memory strength change curve diagram of the present application is shown.

[0022] List of reference signs 1, parameter initialization module; 2, multi-dimensional data acquisition module; 3, AI dynamic interval calculation module; 301, cloud server; 4, multi-modal hierarchical training module; 5, feedback optimization module; 6, training terminal; 601, tablet computer; 602, earphone. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be described clearly and completely below in conjunction with the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0024] Please refer to Figures 1 to 5 : Embodiment one: the present application proposes an English vocabulary cognitive training method and system based on interval repetition memory method, including the following steps: One, system initialization and individual memory parameter configuration: the user inputs the English level grade, daily training time and memory achievement target into the parameter initialization module 1 system, the system loads the pre-constructed vocabulary difficulty-initial interval mapping table, and initializes the core algorithm, including: LSTM memory decay prediction model, multi-modal training rule engine, error type classification model parameter, and simultaneously matches the vocabulary library containing multi-modal resources according to the user level grade; Secondly, the multi-dimensional data acquisition module 2 collects real-time data from multiple dimensions of memory performance, and synchronously collects core memory data such as accuracy R and reaction time T at a preset frequency, and cognitive state data such as concentration C and fatigue F. The collected data is transmitted to the cloud server 301 for storage through the HTTPS protocol; Thirdly, the AI dynamic interval calculation module 3 performs outlier rejection and normalization preprocessing on the collected data, calculates the memory strength coefficient and determines the memory strength level, generates the final review interval in combination with the vocabulary basic interval and special adjustment rules; Fourthly, the multi-modal hierarchical training module 4 performs cognitive training, divides the vocabulary into basic, advanced, and proficient cognitive layers according to the memory strength coefficient, matches differentiated multi-modal training tasks for each layer, and controls the vocabulary addition proportion, cognitive layer dynamic switching, and fatigue relief in the training process. Fifthly, the feedback optimization module 5 pushes multi-dimensional feedback reports to the user end, teacher or parent end, updates the LSTM memory decay prediction model parameters daily, and optimizes the vocabulary difficulty-initial interval mapping table and training task weight every month. The English level in step one is A1-C2, and A1-C2 corresponds to the European Common Reference Framework. The daily training duration is 30-90 minutes. In the vocabulary difficulty-initial interval mapping table, the initial interval of A1-level vocabulary is 1 day, B2-level is 3 days, and C1-level is 5 days. The error type classification model is based on the CNN text classification algorithm, which can identify 8 types of errors such as semantic confusion, spelling errors, and pronunciation deviation. The multi-modal resources of the vocabulary library include: English or American dual pronunciation audio with a sampling rate of 44.1 kHz and a bit rate of 128 kbps, 3-5 spoken or written examples with an adaptation degree of ≥90% based on the BERT semantic model, and spelling resources with root and affix disassembly; The core memory data acquisition frequency in step two is once per completed vocabulary training task, where the reaction time timeout is 30000 ms, and the error type is automatically identified by the error type classification model. The cognitive state data acquisition frequency is once every 5 minutes, the concentration is based on the training terminal 6 eye movement recognition, for example, the fixation screen duration proportion ≥80% is recorded as 100%, and the proportion <50% is recorded as 0%. The fatigue is calculated based on the reaction time fluctuation, and the fluctuation amplitude >50% is recorded as 100%, and the fluctuation amplitude <10% is recorded as 0%. The data is stored in a MySQL distributed database, and each data is associated with user ID-vocabulary ID-training timestamp-data type label, with a retention period of ≥1 year; The data preprocessing in step three adopts the box plot method to eliminate extreme data with reaction time <100 ms or >30000 ms, and normalizes the accuracy, reaction time, concentration, and fatigue to the interval [0, 1] to obtain a standardized data matrix; the calculation formula of the memory strength coefficient M is M = 0.4 x R' + 0.3 x (1-T') + 0.2 x C' - 0.1 x F', wherein R' is the standardized accuracy, T' is the standardized reaction time, C' is the standardized concentration, and F' is the standardized fatigue, and the weight is calibrated through 10000+ user training data; according to the LSTM memory decay model, the future 7-day memory decay value is output by inputting the four data R', T', C', and F'; according to the memory strength grade division standard, M≥0.8 is high memory strength, 0.5≤M<0.8 is medium memory strength, and M<0.5 is low memory strength; The basic interval in step three is obtained from the vocabulary difficulty-initial interval mapping table according to the vocabulary difficulty level, the final review interval D is calculated, and the interval adjustment formula is D = D0 x (1 + 0.6 x M - 0.3 x (1-M)); special adjustment rules include: shortening the interval by 20% when the error type is semantic confusion, lengthening the interval by 50% when M≥0.9 for 3 consecutive times, rounding the interval calculation result, and generating a dynamic interval table; The multi-modal hierarchical training task in step four meets the following requirements: basic cognitive layer, including word plus definition pairing and voice follow-up verification task; advanced cognitive layer, including context fill-in-the-blank and spelling dictation task; expert cognitive layer, including scene sentence making and vocabulary association task; in step four, the training process control meets the following requirements: in daily training, the proportion of new words is ≤40%, the words to be reviewed and new words are selected according to the dynamic interval table, the proportion of new words is ≤40%, and memory overload is avoided; cognitive layer dynamic switching rules: automatically downgrade to basic cognitive layer for 2 consecutive task errors, and automatically upgrade to expert cognitive layer for 3 consecutive task corrects; insert ≤3 minutes of vocabulary games after completing 10 vocabulary training, which is used to relieve fatigue and simultaneously collect cognitive state data; The user end feedback in step five includes: real-time feedback for each task, and daily personal memory report; the teacher or parent end views the 7-day memory trend chart and weak cognitive layer distribution, supports pushing targeted training tasks; model optimization includes: adding training data to the training set of the LSTM memory decay prediction model every day, updating model parameters, and optimizing the vocabulary difficulty-initial interval mapping table and training task weight based on 1 month of data every month.

[0025] Example two, on the basis of example one, the English vocabulary cognitive training system based on spaced repetition method, including: parameter initialization module 1: for receiving user input parameters, load mapping table and AI model parameters, match multi-modal vocabulary library; Multidimensional data acquisition module 2 contains training terminal 6 and data transmission unit, acquisition core memory data and cognitive state data; AI dynamic interval calculation module 3 is deployed in cloud server 301, executes data preprocessing, memory strength calculation and dynamic interval generation; Multi-modal hierarchical training module 4 contains task generation unit and process control unit; Feedback optimization module 5 contains feedback generation unit and model iteration unit; Training terminal 6 includes tablet computer 601 and earphone 602, tablet computer 601 adopts 11 inch iPadPro, earphone 602 adopts Huawei FreeBudsPro2; The training set of LSTM memory decay model in AI dynamic interval calculation module 3 contains 5000+ user 100000+ data, the prediction error is less than or equal to 3%, and the error type set contains 80000+ samples of 8 types of errors; In multi-modal hierarchical training module 4, the speech recognition of data acquisition in daily training adopts MFCC plus DTW algorithm, and the semantic analysis adopts BERT or GPT-3.5 fine-tuning model.

[0026] Example three, on the basis of example one, high school students in the second grade use the training system and training method, English level B1, daily training 40 minutes, target 300 high school entrance examination core words in 30 days, the system matches B1 high school entrance examination vocabulary library, the initial interval is set to 2 days, and the LSTM memory decay prediction model and error type classification model parameters are loaded; The data acquisition of the first day training, 20 new words, such as abandon, achieve, complete the word-semantic matching and voice reading task, collect data: R=75%, T=1200ms, C=85%, F=15%, E=2 times, pronunciation deviation, such as ch pronunciation into sh; According to dynamic interval calculation: the standardized data R'=0.75, T'=0.041200 / 30000, C'=0.85, F'=0.15 are obtained; Memory strength coefficient M=0.4×0.75+0.3×(1-0.04)+0.2×0.85-0.1×0.15=0.743; Basic interval D0=2 days, final interval D=2×(1+0.6×0.743-0.3×(1-0.743))≈3 days; After the hierarchical training, review on the 4th day, interval expiration, memory strength coefficient M = 0.743, enter the advanced cognitive layer, complete the context fill-in-the-blank, such as "After years of hard work, she finally ____ her goal, spelling and dictation tasks, correct rate 90%, memory strength coefficient M is updated to 0.82, interval adjustment is 3x(1+0.6x0.82-0.3x0.18)≈4 days; After 30 days, feedback and optimization are carried out, and the user end report shows that 292 words are mastered in 30 days, the 1-week retention rate is 88%, and the high-frequency error is spelling error-double consonant omission; model optimization: the prediction error of the LSTM memory decay prediction model is reduced from the initial 5% to 2.8%, and the B1-level vocabulary initial interval is adjusted to 1.5 days for this user.

[0027] The working principle of the embodiment is as follows: after the user inputs the English level, the daily training time and the learning goal, the system automatically matches the corresponding difficulty vocabulary library, and loads the preset vocabulary difficulty-initial interval mapping table and prediction model, multi-modal training rule engine, CNN error classification model, in the training process, the system continuously collects data through the tablet 601 and the earphone 602, records the correct rate, reaction time and error type every time a task is completed, collects the concentration and fatigue degree every 5 minutes, all the obtained data is transmitted to the cloud MySQL database for storage through HTTPS encryption, and is stored for more than 1 year, the AI dynamic interval calculation module 3 processes the collected data, calculates the individualized memory strength coefficient M, and dynamically adjusts the review interval of each vocabulary according to the memory strength, divides the vocabulary into three cognitive levels according to the memory strength, matches different training tasks, and adjusts the level according to the performance, and inserts a small game to relieve fatigue, the user pushes real-time feedback and daily memory report, and the teacher or parent end can view the trend chart and weak link, improve the prediction accuracy, and the system realizes self-iterative closed loop.

[0028] In this paper, the following points need attention: 1. The drawings of the embodiment of the present application only involve the structures involved in the embodiment of the present application, and other structures can refer to the usual design.

[0029] 2. In the case of no conflict, the embodiments of the present application and the features in the embodiments can be combined to obtain new embodiments.

[0030] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited to this, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for English vocabulary recognition training based on spaced repetition, characterized in that, Includes the following steps: First, the user inputs their English proficiency level, daily training duration, and memory achievement goals into the parameter initialization module (1). The system loads the pre-built vocabulary difficulty-initial interval mapping table and initializes the core algorithm, including: LSTM memory decay prediction model, multimodal training rule engine, error type classification model parameters, and matches a vocabulary library containing multimodal resources according to the user's proficiency level. Second, the multi-dimensional data acquisition module (2) collects memory performance data from multiple dimensions in real time, and collects core memory data and cognitive state data synchronously at a preset frequency. The collected data is transmitted to the cloud server (301) for storage via HTTPS protocol. Third, the AI ​​dynamic interval calculation module (3) performs outlier removal and normalization preprocessing on the collected data, calculates the memory strength coefficient and determines the memory strength level, and generates the final review interval by combining the vocabulary basic interval and special adjustment rules. Fourth, the multimodal hierarchical training module (4) performs cognitive training. Based on the memory intensity coefficient, the vocabulary is divided into three cognitive layers: basic, intermediate, and advanced. Each layer is matched with a differentiated multimodal training task. The proportion of new vocabulary, dynamic switching of cognitive layers, and fatigue relief are controlled in the training process. Fifth, the feedback optimization module (5) pushes multi-dimensional feedback reports to users, teachers or parents, updates the LSTM memory decay prediction model parameters daily, and optimizes the vocabulary difficulty-initial interval mapping table and training task weights monthly.

2. The English vocabulary recognition training method based on spaced repetition as described in claim 1, characterized in that, The English proficiency level in Step 1 is A1-C2, which corresponds to the Common European Framework of Reference for Languages ​​(CEFR). The daily training time is 30-90 minutes. In the vocabulary difficulty-initial interval mapping table, the initial interval for A1 level vocabulary is 1 day, for B2 level it is 3 days, and for C1 level it is 5 days. The error type classification model is based on the CNN text classification algorithm and can identify semantic confusion, spelling errors, and pronunciation errors. The vocabulary database includes multimodal resources such as: dual-speech audio in English or American English with a sampling rate of 44.1kHz and a bit rate of 128kbps; 3-5 spoken or written example sentences with a fit of ≥90% selected based on the BERT semantic model; and spelling resources containing root and affix breakdowns.

3. The English vocabulary recognition training method based on spaced repetition as described in claim 1, characterized in that, In step two, the core memory data collection frequency is once for each completed vocabulary training task, with reaction time exceeding 30,000 ms and error type automatically identified by an error type classification model. The cognitive state data was collected every 5 minutes. Attention level was based on eye movement recognition of the training terminal (6). The percentage of time spent looking at the screen was ≥80% and recorded as 100%, and <50% and recorded as 0%. Fatigue level was calculated based on the fluctuation of reaction time. The fluctuation range was >50% and recorded as 100%, and the fluctuation range was <10% and recorded as 0%. Data is stored in a distributed MySQL database. Each data entry is associated with a user ID, a vocabulary ID, a training timestamp, and a data type label. The data is retained for at least one year.

4. The English vocabulary recognition training method based on spaced repetition as described in claim 3, characterized in that, In step three, the data preprocessing method uses box plots to remove extreme data with reaction times <100ms or >30000ms, and normalizes the accuracy, reaction time, focus, and fatigue to the [0,1] interval to obtain a standardized data matrix. The formula for calculating the memory strength coefficient M is M=0.4×R'+0.3×(1-T')+0.2×C'-0.1×F', where R' is the standardized accuracy, T' is the standardized reaction time, C' is the standardized focus, and F' is the standardized fatigue. The weights are calibrated using training data from 10,000+ users. Based on the LSTM memory decay model, the model analyzes and outputs the memory decay value for the next 7 days using the four input data points R', T', C', and F'. The standard for classifying memory strength is as follows: M≥0.8 is high memory strength, 0.5≤M<0.8 is medium memory strength, and M<0.5 is low memory strength.

5. The English vocabulary recognition training method based on spaced repetition as described in claim 1, characterized in that, In step three, the basic interval is retrieved from the vocabulary difficulty-initial interval mapping table based on the vocabulary difficulty level, and the final review interval D is calculated. The interval adjustment formula is D=D0×(1+0.6×M-0.3×(1-M)). Special adjustment rules include: shortening the interval by 20% when the error type is semantic confusion, extending the interval by 50% when M≥0.9 for 3 consecutive errors, and executing the interval calculation after rounding the result to generate a dynamic interval table.

6. The English vocabulary recognition training method based on spaced repetition as described in claim 5, characterized in that, The multimodal hierarchical training task in step four satisfies: Basic cognitive level, including word and definition matching, and voice reading verification task; Advanced cognitive level, including contextual fill-in-the-blank and spelling / dictation tasks; Proficient in cognitive tasks, including scenario-based sentence construction and vocabulary association.

7. The English vocabulary recognition training method based on spaced repetition as described in claim 1, characterized in that, In step four, the training process control ensures that the proportion of newly added vocabulary is ≤40% in daily training. Based on the dynamic interval table, the vocabulary to be reviewed and the newly added vocabulary are selected for the day, with the proportion of new words being ≤40% to avoid memory overload. Cognitive level dynamic switching rules: Two consecutive incorrect tasks will automatically downgrade to the basic cognitive level, and three consecutive correct tasks will automatically upgrade to the advanced cognitive level. After every 10 vocabulary words are learned, a vocabulary mini-game of ≤3 minutes is inserted, and cognitive status data is collected simultaneously.

8. The English vocabulary recognition training method based on spaced repetition as described in claim 7, characterized in that, Step five includes user feedback: real-time feedback for each task and daily personal memory reports; Teachers or parents can view a 7-day memory trend chart, the distribution of weak cognitive layers, and push targeted training tasks. Model optimization includes: adding training data to the training set of the LSTM memory decay prediction model daily, updating model parameters, and optimizing the vocabulary difficulty-initial interval mapping table and training task weights monthly based on one month's data.

9. An English vocabulary recognition training system based on spaced repetition, implementing the method of any one of claims 1-8, characterized in that, include: Parameter initialization module (1): used to receive user input parameters, load mapping table and AI model parameters, and match multimodal vocabulary; The multi-dimensional data acquisition module (2) includes a training terminal (6) and a data transmission unit, which collect core memory data and cognitive state data; the AI ​​dynamic interval calculation module (3) is deployed on a cloud server (301) to perform data preprocessing, memory strength calculation and dynamic interval generation; the multimodal hierarchical training module (4) includes a task generation unit and a process control unit; the feedback optimization module (5) includes a feedback generation unit and a model iteration unit.

10. The English vocabulary recognition training system based on spaced repetition as described in claim 9, characterized in that, The training terminal (6) includes a tablet computer (601) and headphones (602). The training set of the LSTM memory decay model in the AI ​​dynamic interval calculation module (3) contains 5,000+ users and 100,000+ data points, with a prediction error ≤3% and an error type set containing 8 types of errors and 50,000+ samples. The speech recognition data collected in the daily training of the multimodal hierarchical training module (4) adopts the MFCC plus DTW algorithm.

Images