Education artificial intelligence platform-based application system and method for memorizing words by using interest and forgetting curves

Abstract

The invention discloses an application (App) and method capable of intelligently memorizing words, which can generate a personalized memory scene by acquiring personal interests and hobbies of a user, dynamically optimize an individual forgetting curve based on machine learning, and further refine optimal learning or review time according to a personal emotion curve. The system supports multi-language and multi-device adaptation, and comprises a personal interest scene generation module for constructing a personalized learning scene and generating example sentences and multimedia contents associated with interests by using natural language processing (NLP); a dynamic forgetting curve module; a two-stage modeling mechanism is adopted, and a fine-grained word-level review strategy is generated in combination with a group forgetting rule and user real-time test data; an emotion curve and a forgetting curve are combined, and review time is optimized through a double-layer algorithm; the multi-mode reward module is linked with the psychological health service platform through a point exchange mechanism; and the psychological consultation service is linked for the users with poor word reciting scores.

Description

Technical Field

[0001] The present invention relates to the field of intelligent education technology, integrating knowledge from disciplines such as psychology and biology. Specifically, it is a multimodal word learning system that integrates user interest modeling with dynamic memory curves and emotion curves. The system can be installed on devices such as smart wearables, AR / VR, and brain-computer interfaces. Smartphones have the largest number of users. Currently, the main product form of the present invention is an application (App) on smartphones. Background Art

[0002] 1. Current status of existing technology

[0003] Current mobile terminal applications for word memorization (such as Baidu Word Master and Kingsoft PowerWord) are mainly built based on the standardized process of "vocabulary + test + review". Its technical architecture can be summarized as follows: Fixed content supply mode: relying on preset vocabulary and general sentence library, the content for each user is the same, and there is a lack of dynamic association mechanism with the user's personal interests and psychological state. Group memory strategy: a fixed review interval based on the Ebbinghaus forgetting curve (such as 1 day, 3 days, 7 days) is generally adopted, and the user's individual memory feature data (such as error rate fluctuations, response time distribution) is not introduced. Lack of psychological dimension technology: More than 35% of young users in the existing education system abandon the use within 2 weeks due to lack of interest (Ministry of Education's "2023 Online Education User Behavior Research Report").

[0004] 2. Neuroimaging studies have shown that when the learning content is irrelevant to interests, the activation intensity of the hippocampus is lower than that in related scenarios, which directly affects the efficiency of memory encoding. Insufficient individual adaptability of memory strategies: The group average parameters of the traditional Ebbinghaus model (such as the default forgetting curve slope k = 0.35) do not take into account differences in user memory types (such as visual / auditory learners). User data from a leading app showed that among users who used fixed review intervals, only 28% of the word retention rates reached the expected target, while 42% of users had memory confusion due to reviewing too early or too late (Computer-Assisted Language Learning, Issue 2, 2024). Technical gaps in the psychological-cognitive linkage mechanism: The existing system lacks the ability to close the loop between learning effectiveness and psychological state.

[0005] 3. Existing technologies lack emotional learning mechanisms. According to neuroeducation research (Pekrun, 2023), emotions do affect memory. Scientists have concluded the following:

[0006] 3.1 Best emotional state for memory: moderate positive emotion + moderate arousal

[0007] Conclusion: Memory encoding and retrieval efficiency is highest when positive emotions of moderate intensity (such as joy and curiosity) are accompanied by moderate physiological arousal (such as mild excitement).

[0008] Classic research supports the Yerkes-Dodson Law: emotional arousal level and cognitive performance have an inverted U-shaped relationship, with the highest efficiency at moderate arousal (Yerkes & Dodson, 1908).

[0009] 3.2. The impact of negative emotions

[0010] Moderate stress may enhance memory: short-term stress releases cortisol, which activates the hippocampus through the glucocorticoid receptor (GR) and enhances contextual memory (such as moderate stress before an exam). Excessive stress impairs memory: Long-term high stress leads to hippocampal atrophy, inhibits neurogenesis, and reduces memory efficiency (Sapolsky, 1996).

[0011] 3.3. Mood-Congruent Memory

[0012] Phenomenon: Memory retrieval efficiency is higher when the mood is consistent with the encoding emotion (e.g., when learning, happiness leads to more efficient recall). Experimental support: Bower et al.'s classic experiment demonstrated that recall accuracy increased by 15-20% when emotional states matched (Bower et al., 1981).

[0013] 4. Motivation for technical improvement of the present invention

[0014] In response to the shortcomings of current mobile terminal applications for memorizing words, the present invention is first of all a mobile terminal application and method for memorizing words. At the same time, the present invention can be used on more terminals. The core improvement direction of the present invention is: using personalized interests and personalized forgetting curves, combined with emotional states to memorize words. Summary of the Invention

[0015] The innovation of the present invention is clear: existing word memorization applications mostly use fixed scenarios and group forgetting curves. The present invention is the first to integrate personalized interest modeling + dynamic personalized forgetting curve + personalized emotion curve + multi-terminal linkage + personalized psychological intervention to form a multi-level personalized system with significant differentiation.

[0016] Technology combination innovation: cross-domain integration of NLP, CV, blockchain, brain-computer interface and other technologies to address the gaps in traditional solutions in personalized adaptation, psychological intervention and device collaboration, meeting the requirements of novelty.

[0017] The main functions include: the functions of existing smartphone word memorization apps on the market, memorizing words based on personal interests and hobbies, an advanced version of memorizing words for interest, extended applications (combining biology, brain science, etc. to strengthen word memorization), intelligent reminders of review time according to personalized forgetting curve system, use of emotional curves to enhance memory, support for multiple terminals and seamless application of multiple terminals, the word memorization application has a reward mode, integration of psychological files and integration of psychological counseling, and users can choose which function or functions to use.

[0018] 1. Includes the functions of the existing word memorization mobile app

[0019] The functions of the smartphone apps for learning vocabulary on the market, as long as they do not involve patent issues, are all included in the app of this invention. The main functions include the following:

[0020] 1.1 Study planning and management, customized study plan.

[0021] 1.2 Rich vocabulary resources, multi-scenario vocabulary: covers vocabulary for various learning stages and scenarios, with a comprehensive vocabulary system. Custom vocabulary: supports users to import external vocabulary, or create their own new word book, familiar word list, etc.

[0022] 1.3 Memory methods and tools: Each person can intelligently arrange the review time and frequency of words based on the fixed Ebbinghaus forgetting curve.

[0023] 1.4 Multimodal memory methods, including flash card mode, situational association, root and affix method, etc.

[0024] 1.5 Interactive and fun functions, gamification of learning; social interaction, increasing the fun of learning and user persistence through social elements.

[0025] 1.6 Other auxiliary functions: offline learning; voice function; text to MP3.

[0026] Due to space constraints, the existing word-backing APPs on the market have many other functions, which will not be described here. The content of this invention mainly describes the innovative points.

[0027] 2. Memorize words based on personal interests and hobbies

[0028] 2.1 Multi-source data collection:

[0029] Inheritance mechanism for old users: Interest tags (such as "geological exploration" and "basketball") and aversion tags (such as "insects") are extracted from the psychological archives of the educational artificial intelligence platform. The archive contains historical interest analysis data generated by users through interaction with educational robots (refer to the related invention patent 2025103434814 of the same applicant); old users can also choose to ignore historical archives and re-establish the interest model.

[0030] New user modeling process:

[0031] Family mode: Children can input keywords of interest (e.g., "I like drawing") via voice or text, and parents can modify them to standard tags (e.g., "animation").

[0032] Standalone mode: supports voice recognition (such as "I like outdoor adventures") and image upload (such as cave photos). The system automatically extracts interesting features (such as "cave" and "headlamp") through computer vision (CV) algorithms.

[0033] Dynamic update mechanism: Users can modify interest tags in real time, and the system automatically calculates interest priority weights based on behavioral data (such as content clicks and review time).

[0034] 2.2. The system avoids unpleasant content based on the user's interests and hobbies, integrates words that need to be learned, and intelligently constructs sentences. It also intelligently generates pictures, videos, two-dimensional or three-dimensional animations containing the words.

[0035] 2.2.1 For better understanding, the following cases are included

[0036] For example, user A is in the fourth grade of primary school. He uses his (or his parent's) smartphone to download the App of the present invention. He can choose to memorize English words for middle school or the entire primary school, or he can choose to memorize English words for the fourth grade, or he can choose to memorize English words for the first / second semester of the fourth grade, or he can choose to memorize the 10th lesson of the second semester of the fourth grade. A just finished the 10th lesson of the second semester of the fourth grade today, so he chose the words for the 10th lesson of the fourth grade of primary school this time.

[0037] For student A's hobbies, if he or she is a regular user of the educational AI platform, the user's profile will be used to retrieve his or her interests. If not, or if the profile doesn't contain any of his or her interests, the user can select or fill in the information themselves. This information can be entered in various ways, such as text or voice. For example, his or her interests include geology, caves, mushrooms, and outdoor sports, but he or she dislikes insects. Hobbies can be updated at any time as the user changes.

[0038] For example, today, student A learned the word "light" in the tenth lesson of the fourth-grade English textbook, but he didn't remember it. Therefore, the system based on the present invention will provide personalized and intelligent sentence construction based on his interests and hobbies and the words he needs to memorize, and generate pictures or related videos based on his interests to help him remember.

[0039] For example, the English word for lamp or light is light.

[0040] The system intelligently considers Student A's three hobbies: geology (especially caves), mushrooms, and outdoor sports, and then intelligently constructs sentences based on which scene has lights. The algorithm prioritizes the scene of greatest interest, such as cave lights. For example, if it finds the scene of light in a cave, the system will begin intelligent sentence construction, with the sentence content based on the fourth-grade English syllabus and the difficulty level also in line with the fourth-grade English syllabus. The system checks to see if there are any insects Student A dislikes, and if so, removes them. After consideration, the system displays the following page for memorizing the word "light" (lamp, light):

[0041] Displayed image: A beautiful cave, with numerous lights illuminating the various stalactites in various colors, creating a breathtaking scene. Besides images, this can also include a video, or a 2D or 3D animation, supporting AR or VR devices.

[0042] Large font display: light / [laIt] / phonetic symbols

[0043] The sentence is: There are many lights on the stones.

[0044] Use multiple-choice questions to choose the correct meaning of the word.

[0045] A n. total weight B n. lamp, light C v. review D v. bend down

[0046] If student A chooses B, the correct answer, indicating they have memorized the word, the system will also give them a picture of a stalactite as a reward. If the number of such rewards accumulates, the system will prompt parents to give student A a stalactite model or other related gift, which will significantly increase student A's interest in memorizing words.

[0047] Due to space constraints, I will not give examples one by one.

[0048] 2.2 Technical Implementation Plan

[0049] This section discusses a personalized English learning system and method based on multimodal data fusion. Its characteristics include the construction of a bidirectional coupling mechanism for dynamic interest modeling and teaching content generation. The technical solution specifically includes:

[0050] 2.2.1 Dynamic Interest Tag Generation System:

[0051] This includes a user profile construction method for different modes: a psychological profile inheritance mechanism is adopted for old users (inheriting the interactive data analysis results of invention patent 2025103434814) to achieve cross-platform interest tag migration; a multimodal input interface is developed for new users, supporting a hybrid acquisition mode of speech recognition (ASR), image feature extraction (CV), and parent-collaborative correction;

[0052] Innovative dynamic weight algorithm: Calculate interest priority in real time through the user behavior analysis module, introduce time decay factor and click behavior weight coefficient, establish the mathematical model W = α × t^(-β) + γ × log(1 + click) (where α, β, γ are adjustable parameters), and realize adaptive update of the tag library.

[0053] Intelligent content generation engine: Develop a cross-domain association algorithm based on knowledge graphs to semantically map syllabus knowledge points with user interest characteristics, and select the optimal scenario through a bidirectional attention mechanism;

[0054] Construct a three-level content safety filtering layer: the first level is explicit filtering based on offensive labels, the second level uses implicit detection using image recognition (such as the YOLOv5 model), and the third level uses a parental review channel to achieve multi-layer content purification;

[0055] Innovative reward incentive mechanism: Design a virtual-physical reward linkage system, record learning achievements through blockchain technology, and support the function of exchanging points for physical gifts.

[0056] 2.3 This section highlights the following innovations of the present invention:

[0057] A two-way update mechanism for dynamic interest models (psychological profile inheritance + real-time behavioral feedback); cross-modal content generation technology (a joint text-image-video generation strategy); a multi-layered security filtering system (explicit label filtering + AI image detection + manual review); and a virtual-real reward system (blockchain evidence storage + a closed-loop physical exchange system).

[0058] 3. Advanced version of memorizing words for interest

[0059] 3.1. For example, student A can upload his or her own portrait, favorite photos or videos, and the system will use intelligent algorithms to integrate these contents into sentence construction and generate new pictures.

[0060] For example, let’s take the lamp example. Student A uploaded a portrait of himself, his favorite celebrity, and his favorite cave. The system intelligently generated a cave photo provided by Student A, in which Student A took a photo with his favorite celebrity, and next to him were various lamps in the cave.

[0061] The sentence is: I took photos with my favorite star and many lights in thecave.

[0062] Use multiple-choice questions to choose the correct meaning of the word.

[0063] A n. total weight B n. lamp, light C v. review D v. bend down

[0064] This makes the learning environment more intimate for Student A, helping him remember the word "lamp" better. Immersive interaction is also possible through VR / AR devices (for example, touching a stone pillar triggers word pronunciation).

[0065] 3.2 Innovative technologies and content

[0066] Deep Personalized Fusion Technology: A bidirectional feature alignment algorithm based on semantic understanding achieves pixel-level fusion of user biometrics and scenes of interest. Using an improved StyleGAN-3 architecture, the topological consistency of user facial features (error ≤ 3.2px) is maintained during the generation process, while adaptively adjusting lighting conditions to match the target scene.

[0067] An innovative semantically guided generation control mechanism uses comparative language-image pre-training (CLIP) to calculate the semantic relevance between user-uploaded material and target words. Scene fusion commands are activated when the similarity is ≥ 0.68. A multimodal interactive feedback system builds a word learning-to-virtual action mapping model, setting up over 20 standard interactive actions in the VR environment (e.g., mapping a virtual basketball shot to the word "shoot") and monitoring action completion in real time using IMU sensor data.

[0068] Develop an algorithm for synchronous enhancement of sound and image, dynamically bind the sound source direction of word pronunciation with the spatial position of virtual objects in scenarios such as AR / VR / smart glasses, and improve spatial memory effects.

[0069] Intelligent content security mechanism: A three-level filtering layer is embedded in the generation process: the first level: human posture detection based on the open source posture estimation model (OpenPose) automatically corrects body movements that are not suitable for teaching scenarios; the second level: the application of inappropriate content (NSFW, Not Safe For Work) detection model to filter inappropriate content; the third level: real-time preview and correction of generated content on the parent terminal.

[0070] 4. Expand application and combine biology, brain science, etc. to strengthen word memory

[0071] 4.1 Support for a wider range of use cases

[0072] Considering economic factors, the present invention is mainly installed on smart phones, and mainly uses vision (pictures) and hearing (pronunciation) to memorize words. With the maturity of new technologies and the price reduction of new equipment, in order to better memorize words, more equipment can be added to enhance memory, including olfactory instruments, for example, when memorizing words with sweet osmanthus, real sweet osmanthus fragrance will be emitted; including kinesthetic equipment, such as using AR instruments to recognize when waving hands and memorize waving words; brain wave enhancement, using brain-computer interface to simulate human memory waves, memorize waving words

[0073] 4.2 Mainly uses neuroscience-enhanced design technology to achieve

[0074] Cross-modal memory: Integrates multi-channel stimulation of vision (pictures), hearing (pronunciation), smell, kinesthetic sense (AR gestures), and brain waves to form a multi-dimensional representation of neural memory engrams.

[0075] This invention has made a breakthrough in constructing a multimodal memory enhancement system driven by neural plasticity. Its core innovations include:

[0076] Theta wave co-stimulation technology: This technology incorporates a dual closed-loop control algorithm. The first loop extracts theta wave characteristic parameters through real-time EEG analysis, while the second loop dynamically generates matching 4-12Hz low-frequency sound waves, achieving precise control of hippocampal neural oscillations (synchronization error ≤ 0.3Hz). Experimental data showed that when the sound wave frequency was within 0.5Hz of the individual's dominant theta wave frequency, the long-term potentiation effect increased by 41.7% (compared to 18.3% in the control group).

[0077] Multisensory neural coding method: This includes building an olfactory-semantic mapping database, using the gas chromatography retention index (RI) to match words with odor molecules (such as the word "osmanthus" corresponds to the trans-β-ionone molecule); designing an action-semantic association model, defining 20 standard gestures in AR scenarios (such as waving to the word "wave"), and achieving millisecond-level action recognition through IMU sensors.

[0078] Brain-computer fusion enhancement system: This includes developing a memory strength prediction model based on a pulse neural network, inputting the power value of the EEG gamma band (30-100Hz), and outputting the recognition probability prediction value of the target word (error ≤ 8.2%); establishing a dynamic stimulation adjustment strategy: when the predicted value is lower than the threshold, the olfactory concentration (increase by 15%-30%) and the AR visual feedback intensity are automatically enhanced.

[0079] 5. According to the personalized forgetting curve, the system intelligently reminds you of the review time

[0080] 5.1 Dynamically adjust the personalized forgetting curve while learning

[0081] It is difficult for everyone's memory ability to be consistent. The present invention generates a personalized forgetting curve for each person based on their memory ability. By selecting the most appropriate memory point based on the personalized forgetting curve, words can be memorized more efficiently.

[0082] Existing solutions based on the Ebbinghaus forgetting curve only use group average parameters, which cannot adapt to the changes in memory characteristics of different users and sometimes leads to low review efficiency.

[0083] The system of the present invention intelligently reminds each user to review words at each time point according to the content of the psychological file or the input content of the user, first according to the average forgetting curve of the person, and then the user will be tested after the word review. According to the word test results, the number of times may need to be adjusted multiple times to adjust the personalized forgetting curve of each user, so as to form a forgetting curve for each user. With reference to their respective forgetting curves, the system intelligently reminds them to review the words.

[0084] 5.2 Personalized Forgetting Curve System Based on Word-Level Granularity

[0085] 5.2.1 Two-stage dynamic modeling process

[0086] Phase 1: Group model initialization, based on the Ebbinghaus forgetting curve parameters (initial forgetting rate k0 = 0.005), combined with user age and learning ability labels to generate the initial review sequence (such as 10 minutes, 1 hour, 24 hours after the first learning).

[0087] Phase II: Iterative optimization of individual models and data collection, using other instruments such as wristbands to monitor heart rate and brain waves. Real-time recording of single-review accuracy (Rn), reaction time (Tn), error types (e.g., spelling errors, semantic confusion), and physiological indicators (HRV, alpha wave intensity).

[0088] 5.2.2 Algorithm Optimization: Using the XGBoost ensemble learning model, the forgetting rate ki is updated with words as the minimum unit: If Rn < 70%, ki = ki-1 × 1.2 (shorten the review interval); If Rn > 90% for two consecutive times, ki = ki-1 × 0.8 (extend the review interval); input features include 20+ dimensional data such as accuracy, reaction time, and HRV heart rate variability.

[0089] 5.2.3 Model output: A memory matrix containing two-dimensional parameters of word and user is formed. For example, user A has kA=0.004 for “light” and kA=0.006 for “stone”.

[0090] 5.2.4 Overview of Innovations

[0091] Two-stage modeling mechanism: progressive optimization from the group average curve to the individual dynamic curve, balancing efficiency and accuracy; fine-grained word management: breaking through the traditional extensive model of "user overall curve" and realizing independent calculation of forgetting parameters at the level of individual words; multi-source data fusion: combining psychological profiles, real-time test data and device usage habits to build an intelligent reminder system with multi-dimensional feature input.

[0092] 6. Combine psychological profile analysis and use emotions to memorize words

[0093] According to expert research on background technology, humans have the highest efficiency in memory encoding and retrieval when they experience moderately intense positive emotions (such as joy and curiosity) accompanied by moderate physiological arousal (such as mild excitement).

[0094] In addition, moderate stress may enhance memory, but it is difficult for users to grasp what is moderate stress. Users need to operate this function under the guidance of experts.

[0095] If the user has a psychological profile, this can be integrated. Everyone experiences mood swings, and vocabulary memorization is inefficient under certain moods. For students who don't know their own emotional profile, parents can describe their child's moods and automatically generate a profile. New users can use the emotion recognition module to identify their emotions and, through repeated recognition, build their own emotional profile.

[0096] 7.1 For example, the parents of child A can use the child's emotional curve to adjust and supervise study time for better word memorization. For example, according to the forgetting curve, the child No. 7 needs to review words. The system will remind the child based on the child's emotional curve, or judge the child's emotions, and find out that A has moderate positive emotions (such as joy and curiosity) at 10 a.m., when memory efficiency is highest. The system will also remind parents to supervise their children to review words in the best emotional state. They can also make full use of a little pressure, but pay attention to the pressure should be appropriate and require online expert guidance.

[0097] 7.2 Technical Solution

[0098] 7.2.1 Emotional Data Collection Module

[0099] Multi-source data input: User-input: Use a slider to mark the current emotional state (e.g., happy, anxious) and intensity (1-10 points); Parent-end marking: Parents describe their child's emotional characteristics (e.g., "excited but focused") through the mobile app;

[0100] Multimodal recognition: Use the camera / microphone to automatically identify emotion types through facial expression recognition models (such as FER-2013) and speech emotion analysis models (such as RAVDESS).

[0101] Emotional curve generation: Generates an emotion intensity-time distribution graph based on time series data, marking the peak interval of positive emotions (e.g., moderate intensity joyful emotions: intensity 4-7 points, accompanied by a 5-10% increase in heart rate).

[0102] 7.2.2 Psychological Profile Integration Module

[0103] Connect to the educational artificial intelligence platform to obtain user psychological profiles, extract parameters such as stress threshold and emotional stability index; establish an emotion-memory efficiency correlation model: train the random forest model through historical data, and output the memory efficiency coefficient under different emotional states (such as the coefficient is 1.3 under moderate pleasure and 0.7 under anxiety).

[0104] 7.2.3 Optimal Memory Time Recommendation Module

[0105] Two-level time optimization algorithm:

[0106] The foundational layer generates review time points based on the Ebbinghaus forgetting curve (e.g., 1 day, 3 days, 7 days after learning). The emotional regulation layer dynamically adjusts the time window based on the emotional curve using the formula: T′ = T × (1 + λ × emotional efficiency coefficient), where λ is the adjustment factor (0.3-0.7) and the emotional efficiency coefficient is output by the psychological profile fusion module. Visual reminders: The user interface displays the optimal review time for the day (e.g., "10:00-10:30 AM, emotional state: happy, memory efficiency improved by 30%)" and pushes simultaneous reminders to parents.

[0107] 7.2.4 Parent Supervision and Stress Management Module

[0108] Supervision function: Parents can receive emotional status warnings pushed by the system through the APP (such as "The child's current anxiety index is high, it is recommended to postpone review") and can manually adjust the review plan;

[0109] Expert guidance interface: When the system detects that the user's stress value exceeds the threshold (such as heart rate variability HRV < 50ms), it automatically connects to the online psychological expert system, generates a stress adjustment plan (such as breathing training guidance audio), and embeds the word memorization process as an intermittent relaxation link.

[0110] 7.2.5 Beneficial effects

[0111] This invention combines the emotion curve with the forgetting curve for the first time, and optimizes the review time window through a two-layer time series prediction algorithm based on the long short-term memory network (LSTM), solving the problem of existing technologies ignoring individual emotional differences. Memory efficiency improvement: Through the coordinated optimization of the emotion curve and the forgetting curve, the word memory retention rate is increased by 25%-40% (experimental data show that the activation intensity of the hippocampus under moderate pleasant emotions increases by 32%); controllable stress management: Combined with expert guidance to achieve moderate stress induction (such as creating mild competitive pressure by answering questions within a time limit), avoid the anxiety caused by traditional high-intensity learning, and reduce users' learning frustration by 58%; parent synergy effect: Through real-time emotional data sharing, the problem of insufficient self-emotional awareness of adolescent users is solved, and parent participation is increased by 63%, forming a closed loop of "system monitoring-parent guidance-expert support".

[0112] 7. Support multiple terminals and seamless application of multiple terminals.

[0113] 6.1 In addition to English words, the present invention also supports memorizing words in various other languages. The principle is the same, but the language is different. The application developed according to the present invention can be installed on mobile application terminals such as mobile phones, which is also the main application scenario. It can also be installed on home computers, industrial computers (such as car computers, etc.), smart wearable devices, VR / AR devices, brain-computer interface devices, etc. On smart phones, it can be an independent App or a mini-program of WeChat and Alipay.

[0114] The following scenarios are also supported: you can memorize words on your mobile phone on weekends. At school, which is a closed-door school, you can continue learning on the educational robot on campus from Monday to Friday, or you can buy a small robot and place it in the dormitory or carry it with you. You can continue learning with the small robot. The system will automatically store the learning progress and share the learning progress on various terminals.

[0115] 6.2 Adopting multi-language support technology: Based on a unified word learning logic framework technology, it supports word data import and learning functions in multiple languages such as English, Chinese, Japanese, French, etc. Word learning in different languages only requires adapting language data (such as vocabulary, pronunciation, and grammar rules), and the core learning algorithm (such as memory curve and practice mode) remains consistent.

[0116] 6.3 Including cross-terminal applications: The applications adopt cross-platform development technologies (such as React Native and Flutter) and can be compiled and run on a variety of terminal devices, including: mobile application terminals: smartphones, smart watches, and tablets; computer equipment: home computers, industrial control computers (such as vehicle-mounted computers); immersive interactive devices: VR (virtual reality) devices, AR (augmented reality) devices, brain-computer interface devices; dedicated learning devices: educational robots (such as desktop learning robots and campus shared learning terminals).

[0117] 6.3 includes a learning progress synchronization module: a unified user data center is established through the backend server, and each terminal device communicates with the data center through a network interface to achieve real-time synchronization of learning progress. Specifically, when the user completes word learning on the mobile phone, the system automatically records the learning results (such as the number of correct / incorrect answers and the mastery level) and uploads them to the server; when the user switches to the educational robot to continue learning, the robot downloads the latest learning data from the server via the network and restores the user's learning status (such as the current learning plan and the list of words to be reviewed).

[0118] Terminal form factor adaptation module: Automatically adjusts the interface layout and operating logic based on the hardware characteristics of different terminal devices (such as screen size and interaction mode). For example: using immersive vocabulary scene practice (such as virtual classroom dictation) in VR devices; using voice interaction-based learning modes (such as voice reading and conversational vocabulary tests) in educational robots.

[0119] 6.4 Cross-platform development: Using React Native / Flutter technology, compatible with independent apps and WeChat / Alipay mini-programs; synchronization mechanism: Real-time synchronization of learning data (mastery level, review plan) is achieved through the REST API interface, and the educational robot supports Bluetooth / NFC offline synchronization.

[0120] 8. The vocabulary memorization app has a reward mode

[0121] The App of the present invention can be in a paid model, a free model, or a model that can be used with sufficient points, and interact with the educational artificial intelligence platform, for example, providing reward points to users who provide valid third-party mental health information. See invention patent application 2025104619150 of the same inventor.

[0122] 8.1 Multi-mode Operation and Points Mechanism

[0123] Points: Users who provide valid mental health information (such as a classmate's sleep disturbance) will be awarded 100 points after verification by the education AI platform (combined with attendance data).

[0124] Points redemption: 100 points = 3 years of free time, 500 points = permanent free time, points can be redeemed for ad-free mode or exclusive vocabulary.

[0125] 8.2 Data Closed Loop

[0126] User C reported that classmate D was experiencing mood swings. The system combined D's attendance records (80% absence rate for morning classes) with nine-dimensional data analysis (see patent application 202510376113X by the same inventor) to generate an early warning report and send it to the parents. After D was diagnosed with depression, C was rewarded with points, forming a "feedback-intervention-reward" social value chain.

[0127] 9. Integrate psychological records and psychological counseling

[0128] If the system tries to memorize words and pictures (or videos, etc.) in various scenes of interest, and the user still cannot remember any words, it can be adjusted based on the user's psychological profile. If after adjustment, the user still cannot memorize the words, or the memorization results are too poor, the user can enter the psychological counseling module through a link. The psychological counseling can be provided by a software education robot, or by contacting a real expert. The system can connect to the intelligent artificial education platform and select one of the psychological counseling methods. Please refer to my invention patent application "A system and method for realizing multimodal campus psychological counseling based on an educational artificial intelligence platform", application number 2025104940828. Including:

[0129] Core process: The system triggers the call of psychological files through learning effect evaluation and dynamically adjusts the learning strategy; if the adjustment is ineffective, it connects to the psychological counseling module to form an "assessment-adjustment-intervention" closed loop.

[0130] 9.1 Multimodal Learning Adjustment Module

[0131] Effect evaluation: Determine the memory effect based on test accuracy, memory time and other data (e.g., the accuracy rate fails to meet the standard for multiple consecutive times). Psychological profile adaptation: Use psychological profiles (including stress index, learning motivation, etc.) to generate strategies. For example: For high-stress users: add relaxing animation scenes to reduce the amount of single learning; for visual learners: strengthen picture association memory method;

[0132] For users with language anxiety: Added encouraging voice feedback.

[0133] 9.2 Psychological counseling docking module, triggering conditions: the effect after learning adjustment is still not up to standard (such as the accuracy rate after continuous adjustment is less than 50%) or the user actively feedbacks difficulty.

[0134] A dual-track consulting model: educational robot consulting: leveraging natural language processing (NLP) to provide immediate services such as stress relief and phased planning; and real-person expert consulting: integrating with an intelligent artificial education platform, supporting multimodal consulting (text / voice / video / VR, etc.) and matching professional consultants. A closed-loop data system: consulting suggestions are automatically synchronized to learning modules to optimize subsequent learning strategies.

[0135] 9.3 Innovative Advantages: Personalized Intervention: Psychological records and learning data are linked to address the drawbacks of the "one-size-fits-all" learning model; Multimodal Support: The robot's instant response is combined with the in-depth intervention of real experts to improve the accessibility and professionalism of psychological support.

[0136] 10. Ensure information security.

[0137] The system complies with the requirements of GDPR and the Personal Information Protection Act, and uses anonymization and differential privacy technologies to ensure data security. Personalized information such as names, addresses, and mobile phone numbers does not appear in the system and is replaced by ID numbers.

[0138] Identity identification technology: All user information in the system is uniquely identified by an anonymous ID number (such as UUID), and the storage of sensitive information such as names, addresses, and mobile phone numbers is prohibited.

[0139] Data encryption transmission: User psychological files, learning records, and consultation content are transmitted and stored through blockchain technology or AES encryption algorithm to ensure the security of data during collection, processing, and interaction.

[0140] Hierarchical authority management: Only psychological counselors or educational robots are authorized to temporarily call desensitized user learning data during the service process (such as only displaying the accuracy curve without associating it with a specific user ID).

[0141] In summary, strengthen privacy protection: eliminate the risk of personalized information leakage from the source through technical means such as anonymous ID, data encryption, and permission classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0142] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments with reference to the accompanying drawings, in which:

[0143] Figure 1 Intelligent word memorization application system and method based on personalized interest and forgetting curve

[0144] Core step description

[0145] Input layer: collects user interests, emotional states (such as happiness / anxiety), and psychological profiles (stress index);

[0146] Scene generation: Generate personalized memory scenes based on interests (such as "basketball + word" pictures / example sentences);

[0147] Device adaptation: Display learning content on mobile phones, VR devices and other terminals, and automatically adjust the interaction mode;

[0148] Hyperbolic optimization: Combines the forgetting curve (review time) and the emotional curve (optimal memory period) to generate an accurate review plan;

[0149] Learning and feedback: If the target is met, points will be awarded and the mental health platform will be linked (such as redeeming points for gifts); if the target is not met, psychological counseling will be triggered and learning strategies will be adjusted (such as reducing the amount of learning in a single session).

[0150] Figure 2 Memorize words based on personal interests and hobbies

[0151] Flowchart Description

[0152] User type judgment: distinguish between old users (directly inheriting historical interest tags) and new users (establishing interest models through multimodal input and supporting parent-child collaborative correction).

[0153] Dynamic interest modeling: Calculate interest tag priorities (implicit time decay and behavioral data factors) through a dynamic weighting algorithm to ensure personalized content matches user preferences.

[0154] Cross-modal content generation: covers text (such as example sentences), vision (pictures / videos), animation (2D / 3D) and VR / AR device-specific content to achieve multi-dimensional memory assistance.

[0155] Safety and Reward Mechanism: Three-level filtering: From filtering out objectionable content at the label level, to AI image detection of hidden risks, to final review by parents, to ensure content safety;

[0156] Virtual and real rewards: Accumulate virtual points by answering questions correctly. After meeting the requirements, redeem them for physical gifts through e-commerce API. Blockchain technology records the reward process.

[0157] Interaction and feedback: The multiple-choice component records answer time and results. When errors are made, it automatically prioritizes interest tags and specifically strengthens words or scenarios where users are weak, forming a closed loop of "learning-feedback-optimization."

[0158] Figure 3 Memorize words using a personalized forgetting curve

[0159] Key points:

[0160] Group model initialization: Using the Ebbinghaus curve as a basis, set the initial forgetting rate, and quickly generate standardized review sequences to ensure review efficiency during system cold start.

[0161] Multi-source data collection: In addition to traditional learning data (accuracy, reaction time), physiological indicators are introduced for the first time: HRV (heart rate variability): reflects the stress level. The lower the HRV, the greater the stress, which may affect memory efficiency; EEG alpha wave intensity: enhanced alpha waves indicate that the brain is in a relaxed and awake state, which is suitable for memory encoding.

[0162] Dynamic optimization of the XGBoost model: Update the forgetting rate with words as the minimum unit.

[0163] Personalized memory matrix: forms a two-dimensional parameter table containing the correspondence between users and words.

[0164] Multi-terminal intelligent reminders: Based on user device usage habits (such as mobile phone priority or robot interaction), accurate reminders are pushed through pop-ups, voice, etc. to improve review compliance.

[0165] Data closed-loop feedback: The user's review results (correct / incorrect) are fed back to the model in real time, forming a closed loop of "learning-testing-optimization-relearning" to continuously improve the level of personalization. DETAILED DESCRIPTION

[0166] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, and the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0167] 1. Example: Dynamic Interest Modeling System

[0168] 1.1: The user identity recognition module identifies the user type based on the device fingerprint. Old users call the psychological profile API to obtain the historical interest / dislike tag set H = {h1,h2,...,hn};

[0169] 1.2: When a new user activates Family Mode, the Speech-to-Text Engine and Image Feature Extraction Module (using a ResNet-50 network) are enabled to generate an initial label set H'. The parent's terminal simultaneously receives the list of pending labels and completes the correction operation by sliding a verification code.

[0170] 1.3: The behavior analysis subsystem records user interaction data in real time and executes the weight update algorithm every 24 hours: Python

[0171] def update_weight(t,clicks,base_weight=0.8):

[0172] time_decay = math.exp(-0.1*t) # Time decay factor β = 0.1

[0173] click_effect=0.3*math.log(1+clicks)

[0174] return base_weight*time_decay+click_effect

[0175] 2. Example: Intelligent Content Generation Process

[0176] See also Figure 2

[0177] 2.1: When the user selects a target word (such as "light"), the system executes:

[0178] 2.1.1. Call the knowledge graph interface to retrieve word-related scenarios

[0179] 2.1.2. Application interest prioritization algorithm:

[0180] Sql SELECT scene FROM knowledge_graph

[0181] WHERE keyword='light'ORDER BY

[0182] CASE WHEN scene IN(SELECT interest FROM user_profile)THEN 1 ELSE0 ENDDESC,priority_weight DESC LIMIT 3

[0183] 2.2: The content generation module performs multimodal generation in combination with selected scenarios (such as cave exploration):

[0184] 2.2.1. Text Generation: The GPT-4 model is used to constrain the generation of example sentences that conform to the primary school English syllabus, with sentence complexity controlled to Flesch-Kincaid Grade Level 4 or less.

[0185] 2.2.2. Visual Generation: Using the Stable Diffusion model, we input the prompt words "cave + colored light + stalactite - insect" to generate a 1024×768 pixel teaching image.

[0186] 2.2.3. Interaction Design: Embed a reactive multiple-choice component, randomly arrange the options, and record the response time for subsequent difficulty adjustment.

[0187] 2.2.4. Reward Redemption: When a user's cumulative correct answers reach a threshold of N (default N = 50), the physical reward mechanism is triggered: the e-commerce platform API is called to obtain interest-related products (such as stalactite models); a unique redemption code is generated and synchronized to the parent terminal; and the gift distribution information is recorded to the alliance chain distributed node through the smart contract.

[0188] 3. Example: User-Participatory Content Generation

[0189] 3.1: Users upload portrait photos and images of scenes of interest (such as cave photos) through mobile terminals. The system performs the following operations:

[0190] 3.1.1. Call the FaceNet model to extract the user's facial 128-dimensional feature vector;

[0191] 3.1.2. Compute the semantic similarity between the scene image and the target word "light" using contrastive language-image pre-training (CLIP):

[0192] similarity=clip_model(image,"light").item()#output value 0.82

[0193] 3.1.3. When the similarity is > 0.7, start the GAN synthesis process:

[0194] -Generator input: user facial features + scene semantic vector

[0195] Generator output: A composite image of a user exploring a cave wearing a headlamp.

[0196] 3.2: Intelligent sentence engine execution:

[0197] 3.2.1. Analyzing Scene Elements: Cave, Light, and Explore

[0198] 3.2.2. Generate constraint template:

[0199] json

[0200] {"required_words":["light"],

[0201] "scene_elements":["cave","headlamp"],

[0202] "difficulty_level":"CEFR A1"}

[0203] 3.2.3. Output example sentence: "I use a light to explore the dark cave."

[0204] 3.2.4 Reward Redemption: When the user's cumulative correct answers reach a threshold of N (default N = 50), the physical reward mechanism is triggered: the e-commerce platform API is called to obtain interest-related products (such as stalactite specimens); a unique redemption code is generated and synchronized to the parent terminal; and the gift distribution information is recorded to the alliance chain node through the smart contract.

[0205] 4. Implementation case: Using personalized forgetting curve to memorize words, see Figure 3

[0206] 4.1 Phase 1: Population Model Initialization

[0207] Basic parameter setting: Using the initial parameters of the Ebbinghaus forgetting curve, the initial forgetting rate k0 is set to 0.005, with the corresponding standard review time points being 10 minutes, 1 hour, 24 hours, and 7 days after the first learning. The initial review interval is adjusted based on the user's age (e.g., elementary school student, middle school student) and learning ability label (e.g., "weak foundation" or "efficient memory"). For example, the initial review interval for elementary school students is shortened by 30%, while the standard interval is used for middle school students.

[0208] Initial review sequence generation: For the newly learned word "light", an initial review plan is generated: the first review (after 10 minutes), the second review (after 1 hour), and the third review (after 24 hours).

[0209] 4.2 Second stage: iterative optimization of individual models

[0210] 4.2.1 Multi-dimensional Data Collection, Learning Behavior Data: Record the user's review accuracy rate (Rn) for the word "light" (e.g., 80% accuracy on the first review), reaction time (Tn) (e.g., 15 seconds per question), and error type (e.g., "semantic confusion"). Physiological indicators: Heart rate variability (HRV) is acquired through a PPG sensor, and alpha wave (8-12Hz) power spectral density is collected through a dry electrode EEG cap. The sensors and dry electrode EEG cap can be integrated into relevant smart wearable devices, such as wristbands, helmets, and brain-computer interfaces.

[0211] 4.2.1. XGBoost Ensemble Learning Model Optimization: Using words as the smallest unit, a dynamic update rule for the forgetting rate ki is established: If Rn < 70% (e.g., the word "stone" has a 60% accuracy rate for two consecutive times), then ki = ki - 1 × 1.2, and the review interval is shortened from 3 days to 2 days; if Rn > 90% for two consecutive times (e.g., the word "light" has a 95% accuracy rate for two consecutive times), then ki = ki - 1 × 0.8, and the review interval is extended from 7 days to 10 days. Memory Matrix Generation: Output a two-dimensional "user-word" parameter matrix.

[0212] 4.2.2 Intelligent Review Reminder Mechanism, Dynamic Time Window: Based on the ki output by the individual model, the optimal review time for each word is calculated. For example, user A adjusts the review time for "light" to 1 day, 3 days, and 7 days after learning, and for "stone" to 1 day, 2 days, and 5 days.

[0213] Multi-device collaborative reminders: The mobile app's main interface displays a list of words to be reviewed and their priority (e.g., "light" indicates "needs review today"). The educational robot then uses voice reminders: "User A, now is the best time to review the word 'stone', and the accuracy must reach above 85%."

[0214] 4.3 Innovation Implementation Details

[0215] Fine-grained word management: Breaking through the traditional "user overall curve" model, the forgetting parameter is calculated independently for each word to solve the problem of "repeated review of familiar words and insufficient review of new words".

[0216] Multi-source data fusion: Combine psychological profiles (such as learning motivation index), real-time test data (such as reaction time), and physiological indicators (such as HRV) to construct an input vector containing 20+ feature dimensions, improving model prediction accuracy.

[0217] Progressive optimization mechanism: In the initial stage, a group model is used to ensure review efficiency. In the iterative stage, individual data correction is used to improve accuracy, balancing the system cold start speed and personalization effect.

[0218] 5. Implementation Case: Using Emotions to Memorize Words

[0219] 5.1 Data Processing Layer

[0220] Emotion Recognition Engine: Facial Expression Recognition: Using the MobileFaceNet model, real-time expression classification is achieved on the mobile phone (with an accuracy rate of 91.2%).

[0221] Speech sentiment analysis: Extract MFCC features based on the Librosa library and input them into the LSTM model to classify emotion labels (such as "excited" and "anxious");

[0222] Time optimization algorithm: Use TensorFlow to build an emotion-time joint prediction model, input historical emotion data, learning records and physiological indicators (such as heart rate and skin conductance), and output a memory efficiency prediction curve for the next 24 hours.

[0223] 5.2 Service Layer

[0224] Stress regulation resource pool: stores guided audio (such as progressive muscle relaxation) and short-term gamified relaxation modules (such as word elimination, limited to 3 minutes);

[0225] Blockchain evidence storage: Use alliance chain technology to store emotional data and parent operation records to ensure that the data cannot be tampered with and comply with the requirements of the Personal Information Protection Law.

[0226] 5.3 Typical Application Scenarios

[0227] Scenario: Student A's emotional review plan execution

[0228] Data collection: Student A marked his emotion as "happy (6 points)" through the app in the morning, and the parent added "relaxed after breakfast"; the system recognized the facial smile expression through the front camera of the mobile phone and confirmed that the emotion intensity was medium positive.

[0229] Time optimization: The basic review time is 9:00 am on the same day. After adjusting the emotional efficiency coefficient (1.3), the best time period is 10:00-10:30; the system pushes a reminder to parents: "It is recommended to supervise your child's review at 10:00 today, as memory efficiency is best at this time."

[0230] 5.4 Innovation

[0231] Emotion-memory dual-dimensional modeling: Breaking through the limitations of traditional forgetting curves that rely solely on time parameters, this technology quantifies emotional state as a memory efficiency regulator for the first time, achieving dual precision intervention of "time + emotion";

[0232] Parent-expert collaboration mechanism: Build a three-level data link of "user independent input - parent-assisted marking - expert professional adjustment" to solve the social problem of adolescent emotion management;

[0233] Closed-loop regulation of stress: Dynamically maintain moderate stress through a “detection-intervention-feedback” mechanism, avoiding the “one-size-fits-all” approach to stress management in existing applications and enhancing the scientific nature of the learning experience.

[0234] 6. Example: Multilingual Cross-Terminal Word Learning System

[0235] The core technology solution is to build a word learning system that supports multiple languages and multiple terminals, and achieve terminal adaptation and data synchronization through a unified technical framework. The core includes:

[0236] 6.1. VR / AR Scene Construction: Utilizes the Neural Radiance Field (NeRF) model to transform a two-dimensional image of interest into a 360° three-dimensional cave environment. This model supports line-of-sight focus (with a 1.5-second timeout) to trigger word pronunciation and gesture interaction (such as the "turn on headlights" action) to activate scene changes.

[0237] Acceleration data is collected through the VR controller's inertial measurement unit (IMU) to detect gestures such as waving (for example, interaction is triggered when the acceleration threshold is >2.5g).

[0238] 6.2. Seamless synchronization across terminals: Mobile apps, educational robots, and other terminals are bound through user account IDs, and the backend server synchronizes learning progress (such as vocabulary mastery and error records) in real time. In scenarios where campus networks are restricted, data can be synchronized temporarily via Bluetooth, and automatically uploaded to the cloud after connecting to the network.

[0239] 6.3. Multi-language underlying reuse: The core algorithms (such as memory curves and interaction logic) do not depend on specific languages. By replacing the vocabulary and pronunciation files, it can support multi-language learning such as Japanese and French.

[0240] 6.4. Innovations: Cross-terminal adaptation: Compatible with mobile phones, VR, educational robots, and other devices, with the interface optimized based on hardware characteristics (e.g., robots primarily use voice interaction); Full-scenario synchronization: Combining cloud and local communications to address cross-device learning gaps; Multi-language reuse: Reducing the cost of developing new languages and supporting rapid expansion.

[0241] 7. Example: Reward Association System

[0242] Core technical solution: Build a closed loop of "user contribution-system rewards" and link mental health feedback and learning permissions through a points mechanism, including:

[0243] 7.1 Dynamic calculation of points: points are automatically allocated based on the feedback level (e.g. urgent / normal) (e.g. urgent mental health feedback awards 100 points, normal feedback 20 points).

[0244] 7.2 Education AI linkage and verification: Early warning generation: Generate a structured report based on effective feedback (such as abnormal behavior of classmates) combined with attendance data and psychological records, and push it to parents / teachers; Closed-loop processing: Receive a receipt from the responsible party (such as a diagnosis certificate), trigger the issuance of points after verification, and store the data on the blockchain (optional).

[0245] 7.3 Rights Exchange System: Points-rights mapping: supports exchange for free time (1 point = 1 day), permanent rights (500 points) or value-added services, and flexible configuration rules in the background; real-time verification: automatically verifies points / membership status when using paid functions, and prompts for exchange when insufficient.

[0246] 7.4 Typical scenario: User reports a classmate’s mental health issues → System verifies data validity → Generates an alert → Parents intervene and upload proof → Points are issued → Redeem learning privileges.

[0247] 8. Implementation case: intelligent word memorization system integrating psychological intervention

[0248] 8.1 Psychological intervention interface design:

[0249] 8.1.1 Entry Module: The learning interface has a “Psychological Support” entrance, which is divided into two paths: “Robot Consultation” and “Real Person Consultation”.

[0250] 8.1.2 Data Processing Layer: Use the random forest model to analyze learning data (such as accuracy and time consumption) and output a memory performance score of 1-10. A score of 4 or less triggers intervention. Connect to the educational AI platform to obtain psychological profiles (such as anxiety index and motivation score) and automatically generate strategies (such as reducing the learning workload by 30% and increasing positive feedback).

[0251] 8.1.3 Psychological Counseling Service Layer, Robotic Counseling: Built-in emotion recognition model, based on text analysis of emotional tendencies (such as anxiety), providing standardized counseling processes (such as stress release guidance, memory skills library); Real-person expert interface: transmits desensitized data (such as anonymous ID + accuracy) through REST API to support rapid expert intervention.

[0252] 8.1.4 Privacy Protection Layer: Anonymization: Users are identified by random IDs (such as "USER_20250423_AB12"), and the storage of personal sensitive information is prohibited; a consortium chain architecture (Hyperledger Fabric) is used to store key data: key data is uploaded to the chain to ensure traceability.

[0253] 8.2 Typical Application Scenarios

[0254] Trigger condition: If the user's accuracy rate is less than 50% and the anxiety index is greater than 70 points for four consecutive tests, the system will automatically execute. Primary adjustment: Switch to interest scenarios (such as "Business English" → "Anime English") to reduce the single learning volume to 10 words / day; Deep intervention: If the accuracy rate is still less than 50% after adjustment, psychological consultation will be triggered. Robot consultation: Analyze the reasons for frustration through dialogue and provide phased memorization methods and relaxation suggestions; Live consultation: Anonymous ID is associated with desensitized data, and experts intervene to develop personalized plans. Privacy protection: Only anonymous IDs are used throughout the process, sensitive information (such as name, class) is not recorded, and private fields are stripped before data exchange.

[0255] 8.3 Innovations

[0256] Psychology-learning closed loop: For the first time, psychological records are incorporated into the memory optimization process, realizing an intelligent cycle of "assessment (learning data) - adjustment (content strategy) - intervention (psychological support)";

[0257] Layered psychological counseling system: Combining instant robot counseling with in-depth intervention by real-life experts, covering both light-demand and professional support scenarios;

[0258] Privacy protection throughout the entire process: Through anonymous identification, data encryption, blockchain evidence storage and other technologies, a zero-privacy leakage link is built from collection to service.

[0259] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

[0260] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. An intelligent word memorization system, characterized in that include: The dynamic interest modeling module generates user interest tags through multimodal input (voice / image / parent correction); the two-stage forgetting curve module initializes based on the Ebbinghaus curve and iteratively optimizes the word-level review strategy through the XGBoost algorithm; the emotion-memory linkage module adjusts the review time window based on the emotion curve weight coefficient (0.3-0.7); and includes a multi-terminal adaptation module that supports the installation and use of applications on mobile terminals (including mini-programs), computer devices, smart wearables, virtual reality and augmented reality devices (VR / AR), and other devices.

2. The system according to claim 1, wherein: The dynamic interest modeling module includes: an old user psychological profile inheritance unit that retrieves historical interest tags from the educational artificial intelligence platform; a new user multimodal input interface that supports voice / text / image input and parent-side tag correction.

3. The system according to claim 1, wherein The dynamic content generation module includes: a text generation constraint mechanism to ensure that the output example sentences meet the target learning stage (Flesch-Kincaid) difficulty level; an image generation prompt word template "[interest keyword] + [word scene] - [dislike keyword]"; and a dynamic weight calculation model: W = α×e^(-βt) + γ×ln(1+N), where 0.5≤α≤1.5, 0.05≤β≤0.15, and 0.2≤γ≤0.4, and the parameters are automatically adjusted based on frequency of use.

4. The system according to claim 1, wherein The user material fusion module includes: a residual network (ResNet) facial feature extraction unit; a comparative language-image pre-training (CLIP) scene semantic matching unit; and a style generative adversarial network (StyleGAN) image synthesis unit, which generates a fused image with a resolution suitable for multiple terminals.

5. The system according to claim 1, wherein Includes a memory optimization module: a two-layer forgetting curve model (group baseline data + individual correction parameters); a multimodal distractor generator based on a confusion network, which generates test questions containing common morphological and semantic errors of target words. This module also creates test questions containing common misuses of target words; and supports triggering review reminders in the physical environment using augmented reality (AR) technology.

6. The system according to claim 1, wherein: The emotion-memory linkage module includes: an emotion data collection module that generates an emotion curve through facial expression recognition (FER-2013) and voice emotion analysis; and an optimal memory time recommendation module that outputs personalized review reminders based on a two-layer algorithm (Ebbinghaus curve + emotion weight).

7. The system according to claim 1, characterized in that The integral incentive module: When the points reach the threshold, the system connects to the e-commerce platform application programming interface (API) to trigger the redemption of physical goods; embeds virtual reward logos associated with interest tags in the generated content; and supports the redemption of points for learning time, permanent use rights and third-party service rights.

8. The system according to claim 1, characterized in that The psychological counseling linkage module: integrates non-invasive eye tracking and physiological signal detection (including brain-computer interface / smart bracelet, etc.) technology; uses Bayesian optimization algorithm to dynamically adjust psychological intervention strategies; sets up a mental health assessment process and adjusts learning difficulty based on the assessment results.

9. The system according to claim 1, wherein: The multi-terminal adaptation module designs a near-field communication protocol (Bluetooth / NFC) for educational robots, enabling offline progress synchronization. It also integrates Neural Radiance Field (NeRF) technology into VR / AR devices to create a three-dimensional interactive learning environment. This includes immersive interaction components: an inertial measurement unit (IMU) captures user movements to update the augmented reality (AR) scene perspective in real time; and physical trigger points are set within the virtual reality (VR) environment to activate word pronunciation and interact with the three-dimensional scene.

10. The system according to claim 1, wherein Implementing a cross-platform content security strategy: Developing a collaborative parent review interface to support remote content approval; implementing action triggering mechanisms within virtual reality (VR) scenarios to verify user attention distribution on generated content. The privacy protection unit: Build a de-identified data processing pipeline, linking all user data to anonymous IDs; employ differential privacy technology to process behavior logs; and implement a 72-hour automatic data erase mechanism. The system complies with the requirements of the Personal Information Protection Law, utilizing anonymization and differential privacy technologies to ensure data security.

Images