Globalized intelligent English learning method and system based on multi-dimensional user portraits

Abstract

The invention discloses a globalized intelligent English learning method and system based on a multi-dimensional user portrait, and relates to the technical field of English learning, and the method comprises the steps: constructing the multi-dimensional user portrait; performing intelligent level evaluation and dynamic updating; personalized example sentences and articles are dynamically generated based on the user portrait; carrying out deep analysis and visualization on roots and affixes; intelligent association learning is carried out through a learning method of finding similarities and differences; carrying out multi-modal auxiliary memory by utilizing an AI image generation technology; multiple intelligent word importing modes are supported; international architecture design is adopted to support globalized multi-language service. And an embedded real-time translation assisting function is provided. The invention further provides a multilayer content safety filtering method throughout the whole process, and safety, health and compliance of the generated content are ensured through active defense of the input end, process control of the generation end and reaction disposal of the output end. According to the invention, personalized learning experience, learning efficiency and globalized service capability are improved, and a safe and reliable intelligent learning environment is provided for the user.

Description

Technical Field

[0001] This invention relates to the field of English learning technology, and more specifically to a global intelligent English learning method and system based on multi-dimensional user profiles. Background Technology

[0002] Currently, English vocabulary learning mainly relies on traditional dictionaries, standardized textbooks, and various digital learning applications. These tools assist users by providing word definitions, fixed example sentences, graded word databases, and basic exercises.

[0003] However, existing technologies generally have significant limitations: learning content is mostly static and pre-set, making it difficult to dynamically generate personalized example sentences and materials based on users' personal interests, professional backgrounds, or real-time levels; the difficulty adjustment mechanism is crude and cannot precisely adapt to subtle changes in users' abilities; the analysis of word roots and affixes only stays at the surface level, lacking systematic explanations of meanings and association management; there is a lack of intelligent knowledge transfer and association display between new and old words, resulting in an isolated learning process.

[0004] More seriously, when using artificial intelligence (AI) technology to generate learning content, existing technologies often lack proactive, multi-layered filtering and review mechanisms, posing a risk of generating inappropriate or harmful information and creating a potential threat to users, especially minors.

[0005] Therefore, how to construct an intelligent English learning system and method that can deeply integrate multi-dimensional user profiles, achieve dynamic personalization of content, adaptive adjustment of difficulty, and have a sound content security mechanism is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a global intelligent English learning method and system based on multi-dimensional user profiles, which overcomes the above-mentioned defects.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: This application discloses a global intelligent English learning method based on multi-dimensional user profiles, the specific steps of which are as follows: Collect and store multidimensional information about users, and construct a user profile data model; the multidimensional information includes basic information, common scenarios, and personal preferences; An initial language ability assessment is performed on the user to generate an initial ability model, and the initial ability model is dynamically updated based on the user's learning behavior data to obtain the current ability model; Based on the user profile data model and the current capability model, personalized learning content is generated for the user; and content security filtering is performed on the generation process of the personalized learning content.

[0008] Another aspect of this application discloses a global intelligent English learning system based on multi-dimensional user profiles, characterized by comprising: The user profiling module is used to collect and store multidimensional information about users and to build user profile data models. The proficiency assessment module is used to perform an initial language proficiency assessment on the user to generate an initial proficiency model, and to dynamically update the initial proficiency model based on the user's learning behavior data to obtain the current proficiency model; The personalized content generation module is used to generate personalized learning content for the user based on the user profile data model and the current capability model. The content security filtering module is used to perform content security filtering during the generation process of personalized learning content.

[0009] As can be seen from the above technical solution, the present invention discloses a global intelligent English learning method and system based on multi-dimensional user profiles, which has the following beneficial effects compared with the prior art: 1. By constructing user profiles that encompass multi-dimensional information such as interests, professional backgrounds, common scenarios, content style preferences, and learning preferences, and by dynamically generating example sentences and articles closely related to users' actual lives and preferences based on these profiles and a large language model, the relevance and relevance of learning content are enhanced, thereby effectively increasing learning interest and long-term memory.

[0010] 2. By establishing an intelligent level assessment and dynamic update mechanism, the system accurately quantifies users' multidimensional abilities and adjusts the content difficulty in real time accordingly. This ensures that the learning content remains within an appropriate challenge range, avoiding low learning efficiency caused by difficulty mismatch. Furthermore, by combining this with the "find similarities and differences" intelligent comparison method, the system promotes positive transfer and clear differentiation between new and old knowledge, reducing cognitive load and improving knowledge mastery efficiency.

[0011] 3. By deeply analyzing and visualizing word roots and affixes and their meanings, and combining this with statistics on mastery levels, users can systematically build a vocabulary knowledge network. This network is further strengthened through a word root and affix association learning mechanism, which actively pushes words containing the same word roots and affixes, promoting the ability to apply knowledge to new situations and efficiently expanding vocabulary.

[0012] 4. Utilize AI image generation technology to generate highly personalized visual materials based on user profiles, combining abstract words with concrete and relevant visual scenes to enhance the memorability and retrieval of information.

[0013] 5. It integrates multiple word import methods such as OCR photo taking, document upload, text pasting, and API interface, comprehensively covering different input scenarios, reducing the operational cost of users initializing learning data, improving overall usage efficiency, and also improving word import efficiency.

[0014] 6. The platform adopts an international architecture design that supports multilingual interfaces and definitions, enabling it to flexibly adapt to the localization needs of users with different native languages. This effectively overcomes language barriers and lays a solid technical foundation for serving users worldwide.

[0015] 7. The modular design clarifies system responsibilities and enhances the flexibility of function expansion and iterative maintenance. Simultaneously, relying on an AI-driven content generation mechanism reduces reliance on a pre-built, massive static content library, fundamentally lowering the R&D and maintenance costs of content updates.

[0016] 8. Embedded real-time translation assistance eliminates learning interruptions caused by switching between different tools. The hybrid translation strategy optimizes cost and response speed while ensuring translation quality. Continuously updating user profiles based on translation history allows the system to accurately pinpoint knowledge gaps and provide feedback to the content generation stage. Optimized translation caching, personalized translation settings, and multi-granularity translation support collectively achieve fast, flexible, and scenario-based translation assistance. Ultimately, deep integration of translation and learning data greatly ensures the continuity and autonomy of the learning process. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0018] Figure 1 is a schematic diagram of the method flow provided by the present invention; Figure 2 is a flowchart of the multi-dimensional user profile construction process provided by the present invention; Figure 3 is a flowchart of the intelligent level assessment and dynamic update provided by the present invention; Figure 4 is a flowchart of the personalized example sentence generation process provided by the present invention; Figure 5 is a schematic diagram of the deep analysis and visualization of word roots and affixes provided by the present invention; Figure 6 is a flowchart of the "finding similarities and differences" learning method provided by the present invention; Figure 7 is a flowchart of the multimodal personalized auxiliary memory provided by the present invention; Figure 8 is a flowchart of the intelligent word import process provided by the present invention; Figure 9 is a diagram of the internationalization architecture provided by this invention; Figure 10 is a flowchart of the embedded real-time translation assistance provided by the present invention; Figure 11 is a schematic diagram of the content security filtering method provided by the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] This invention discloses a global intelligent English learning system based on multi-dimensional user profiles, such as... Figure 1 As shown, it includes: Collect and store multidimensional information about users, and build user profile data models; the multidimensional information includes basic information, common scenarios, and personal preferences; An initial language proficiency assessment is performed on the user to generate an initial proficiency model, and the initial proficiency model is dynamically updated based on the user's learning behavior data to obtain the current proficiency model; Based on the user profile data model and the current capability model, personalized learning content is generated for the user; and content security filtering is performed on the generation process of the personalized learning content.

[0021] In one embodiment, the multidimensional information includes the user's basic information, interests, occupation / educational background, common scenarios, content style preferences, learning preferences, etc.

[0022] Furthermore, basic information includes name / nickname, age / grade, identity (K12 student / university student / working professional / other), native language, learning goals, etc.; interests include multiple dimensions such as sports and fitness, technology and digital products, arts and culture, lifestyle and leisure, business and finance, etc., supporting multiple selections and custom input; professional / learning background includes students' grade and major, and working professionals' industry and position, etc.; common scenarios include multiple dimensions such as daily life, study and work, social entertainment, travel, professional scenarios, etc., supporting multiple selections and custom scenarios; content style preferences include humorous and lighthearted, serious and professional, heartwarming and healing, adventurous and exciting, science fiction and futuristic, artistic and refreshing, practical and useful, positive and inspiring, etc.; learning preferences include how much time they are willing to spend learning English each day and their preferred learning methods. The collected information is stored as structured data to build a user profile data model, and the construction process is as follows: Figure 2 As shown.

[0023] In one embodiment, the step of obtaining the current capability model is as follows: Obtain users' basic information and generate personalized language proficiency test questions based on the basic information; analyze and evaluate users' answers to the language proficiency test questions and generate an initial ability model containing multi-dimensional ability scores; The system acquires the user's learning progress, mastery of the learning content, and difficulty trends in real time. When preset trigger conditions are met, the system updates the initial ability model using a smoothing algorithm based on the real-time acquired data to obtain the current ability model.

[0024] Furthermore, during the ability model construction, an initial assessment is conducted, including a questionnaire survey, AI-generated customized tests, AI analysis and evaluation, and the ability model itself. The questionnaire survey collects basic user information; the AI-generated customized tests, based on the questionnaire results, utilize a large language model to generate customized reading comprehension and writing tests. The AI ​​analysis and evaluation uses natural language processing technology to analyze users' answers, assessing vocabulary, grammar level, reading comprehension ability, writing expression ability, and overall level score; the ability model includes overall level, scores for each dimension of ability, and weakness analysis, and its construction process is as follows: Figure 3 As shown.

[0025] Furthermore, a dynamic update mechanism is established, including update trigger conditions, update criteria, and update strategies. Update trigger conditions include completing a certain number of words learned, completing a certain number of articles read, periodic automatic assessments, and user-initiated reassessments. Update criteria include learning speed, learning progress, mastery level, trends in the difficulty of unlearned words, and trends in the difficulty of reading articles. The update strategy employs a smooth update algorithm to avoid drastic fluctuations in scores and ensure a continuous learning experience.

[0026] In one embodiment, the step of generating personalized learning content includes: Receive target vocabulary input; Based on the user profile data model and the current capability model, construct and generate prompts; The system utilizes an AI-powered content generation model to generate English learning content that matches the target vocabulary and the user's personalized characteristics and language proficiency, based on generation prompts.

[0027] In one embodiment, personalized learning content includes personalized example sentence generation, the specific steps of which are as follows: Construct example sentence generation parameters, which include target words, user profile data model, and difficulty target; Example sentence generation prompts are constructed based on example sentence generation parameters; Input the example sentences into a large language model to generate personalized example sentences; We conduct quality checks on personalized example sentences and then display the qualified personalized example sentences along with their corresponding translations to users.

[0028] In one embodiment, the quality check includes grammar checking, difficulty assessment, originality checking, and relevance checking.

[0029] Furthermore, such as Figure 4 As shown, personalized example sentence generation includes: First, determining the generation timing, adopting a strategy of generation upon viewing plus scrolling pre-generation. Pre-generation upon login: After a user logs in, the background asynchronously generates example sentences for the first few words in the vocabulary list; Scrolling pre-generation: As the user progresses through their learning, example sentences for subsequent words are pre-generated, maintaining a certain number of buffers; Real-time generation: If there are no example sentences for a word the user is viewing, they are immediately generated and displayed; Regeneration: As the user's level improves, the system prompts the user to generate more suitable example sentences based on their new level.

[0030] Secondly, the generation parameters are constructed, including the target word, user profile (level, vocabulary, grammar level, interests, scenarios, and style), and difficulty target. Example sentences are generated using a large language model to construct AI prompts. These example sentences must contain the target word, their scenarios and content should align with the user's interests and common situations, their difficulty should match the user's level, and they should be natural, idiomatic, and native-speaker-like. The style of the example sentences should also match the user's preferences, and a Chinese translation is provided. Finally, the large language model API is called, with the prompts input, to obtain the generated results.

[0031] Next, quality checks are performed, including grammar checks, difficulty assessments, authenticity checks, and relevance checks. Grammar checks use a grammar checking API to ensure example sentences are grammatically correct; difficulty assessments calculate the lexical difficulty and grammatical complexity of sentences to ensure the difficulty matches the user's level; authenticity checks use multiple large language models to score example sentences, and if the score is below a threshold, a new sentence is generated; relevance checks ensure that example sentences contain target words and are relevant to the user's interests and the context.

[0032] Finally, the information is stored in the database and displayed to the user, showing personalized example sentences in word cards. Each example sentence includes an English sentence, a Chinese translation, a scene label, and a pronunciation button.

[0033] In one embodiment, personalized learning content further includes personalized article generation, the specific steps of which are as follows: Obtain a list of target words, and then obtain the correlation between the part of speech, semantics, and difficulty of each target word in the list; Based on the target word list, user profile data model, and current ability model, construct article generation suggestions; Input the article generation prompts into a large language model to obtain personalized articles; Perform quality checks on personalized articles and display only those that pass the checks to users.

[0034] In one embodiment, the quality checks include: target word coverage check, difficulty assessment, grammar check, idiomaticity check, and plot coherence check.

[0035] Furthermore, personalized article generation includes: First, identifying target words. Users can select several words from a vocabulary list as target words for article generation. The system intelligently recommends words to be reviewed based on the user's learning progress and mastery level. Next, the system analyzes the part of speech, difficulty, and semantic relationships of the target words, identifying the connections between them.

[0036] Secondly, construct the article generation parameters, including the target word list, user profile, and article requirements (word count, difficulty target, genre, and theme). Utilize a large language model to generate the article, constructing AI-generated prompts. The article must contain all target words, its theme and plot should align with user interests, its difficulty level should match the user's skill level, it should have a plot and story to attract user interest, its style should match user preferences, and it should be natural, idiomatic, and native-speaker-like. Provide a Chinese translation and indicate the position of each target word in the article. Call the large language model API, inputting the prompts, to obtain the generated results.

[0037] Next, quality checks are performed, including target word coverage checks, difficulty assessment, grammar checks, authenticity checks, and plot coherence checks. Specifically: the target word coverage check ensures that the article contains all target words; if any target words are missing, the sentence is regenerated or supplemented; the difficulty assessment calculates the vocabulary difficulty and grammatical complexity of the article to ensure that the difficulty matches the user's level; the grammar check uses a grammar checking API to ensure that the article is free of grammatical errors; the authenticity check uses multiple large language models to score the article; if the score is below a threshold, it is regenerated; and the plot coherence check ensures that the article has a clear plot and storyline, and that the content is relevant to the user's interests and the context.

[0038] Finally, the data is stored in the database and displayed to users. Personalized articles are shown on the article reading page, with target words in the article highlighted. Clicking on these words allows users to view definitions and example sentences. The system also provides article reading and translation functions.

[0039] In one embodiment, it also includes in-depth analysis and visualization of root words and affixes, such as... Figure 5 As shown, specifically: This method performs word root and affix identification and decomposition, utilizing a word root and affix database to construct a database containing common prefixes, roots, suffixes, and their meanings. AI-driven word root and affix decomposition involves using an AI model to decompose an input word into its roots and affixes, obtaining prefixes, roots, suffixes, and their meanings.

[0040] The meanings of word roots and affixes are marked using color-coded visual annotations, with prefixes, roots, and suffixes each using different color gradients. The meanings of each word root and affix are explained below, showcasing the complete derivation process of their meanings.

[0041] This feature provides a root and affix list function, which counts the root and affixes the user has mastered. Each time a user learns a word, the system automatically extracts its root and affixes, and counts the number and types of root and affixes the user has mastered. The root and affix list is displayed, sorted by frequency of use, and shows the meaning, example words, and mastery level of each root and affix.

[0042] The system performs word root and affix association learning, recommending related words. When a user learns a word, the system suggests other words containing the same root, helping the user to learn by analogy and quickly expand their vocabulary. It also generates word root and affix exercises using AI.

[0043] In one embodiment, when learning personalized learning content, an intelligent associative learning method can be used, the specific steps of which are as follows: Get any unlearned word and calculate the multi-dimensional similarity between the unlearned word and each word in the learned word bank; the multi-dimensional similarity includes: root and affix similarity, substring similarity and pronunciation similarity; Select a preset number of related words from the learned vocabulary based on similarity. The related words are compared side by side with the unlearned words, and the similarities and differences are highlighted.

[0044] Furthermore, such as Figure 6 As shown, when performing intelligent word association recommendations, it is necessary to identify the most relevant learned words to the unlearned words. Specifically, the user inputs an unlearned word, and related words are retrieved from the user's learned word database. Similarity calculation is performed using a multi-dimensional approach, including root and affix similarity (highest priority), substring similarity, and pronunciation similarity. Based on the similarity calculation results, intelligent word association recommendations suggest the most relevant learned words to the unlearned word, up to a maximum of several.

[0045] The learning method employs a two-step approach. Step one involves identifying commonalities to facilitate knowledge transfer. This involves extracting the roots and affixes of both unlearned and learned words, highlighting the commonalities, and generating knowledge transfer hints. Step two involves identifying differences to avoid confusion. This involves identifying the different parts of both unlearned and learned words, highlighting the differences, and generating hints to prevent confusion.

[0046] Generate visual comparison displays, including highlighting, comparison cards, and educational tips. Highlighting uses one color to highlight identical parts and another color to highlight different parts. Comparison cards show a comparison between unlearned and learned words, including spelling, pronunciation, roots and affixes, meaning, and example sentences. Educational tips emphasize knowledge transfer and avoiding confusion, such as "You already know most of this word! Just remember their combinations!" and "Pay attention to the different meanings of the roots! Although they may share some parts, the different parts determine the specific meaning!"

[0047] Record learning behavior, including which related words the user viewed, which similar and different parts they viewed, the duration of their study, and the learning outcomes.

[0048] In one embodiment, when learning personalized learning content, the method further includes memory-aiding techniques, specifically: Based on the target words and user profile data model, image-generated prompts are constructed. Based on image generation prompts, generate personalized memory images related to the meaning of the target word; Display personalized memory images to users.

[0049] Furthermore, such as Figure 7 As shown, the memory-aiding methods include: determining when to generate images, such as user-initiated requests, system-automatic recommendations, or regeneration; constructing image generation parameters, including target words, user profiles (interests, scenarios, styles), and image requirements (style, scenarios, elements); using AI image generation technology to generate personalized images and constructing image generation prompts, requiring that the image be relevant to the meaning of the word, the image's scenario and elements be combined with the user's interests and scenario, the image style be consistent with the user's preferences, and the image be clear, aesthetically pleasing, and attractive; and calling the AI ​​image generation API, passing in the prompts, and obtaining the generated results.

[0050] Quality checks are conducted, including relevance checks, aesthetic checks, and sharpness checks. Specifically: the relevance check ensures that the image is relevant to the meaning of the words and to the user's interests and the context; the aesthetic check uses an image quality assessment model to score the image, and if the score is below a threshold, it is regenerated; the sharpness check ensures that the image is clear, without blur or noise.

[0051] Store the images in a file and display them to the user. Show personalized images in word cards and provide an image zoom-in function.

[0052] In one embodiment, the method disclosed in this embodiment also includes various intelligent word import methods, such as Figure 8 As shown, specifically: Supports OCR photo import: Users can take photos containing English words, and OCR technology will recognize the English words in the photos, extract a word list, and add it to the vocabulary list in batches after user confirmation. Supports document upload import: Users can upload documents containing English words, and the system will parse the document content, extract the English words, and add them to the vocabulary list in batches after user confirmation. Supports text paste import: Users can paste text containing English words, and the system will parse the text content, extract the English words, and add them to the vocabulary list in batches after user confirmation. Supports API import: The system provides an API interface, allowing third-party applications to import words in batches via the API.

[0053] Perform word deduplication and validation. Deduplication checks if the imported words already exist in the vocabulary list; if so, prompt the user. Validation checks if the imported words are valid English words; if invalid, prompt the user.

[0054] In one embodiment, such as Figure 9 As shown, the system using the method disclosed in this embodiment adopts an internationalized architecture design, supports multilingual interfaces, and allows users to select the interface language. The system provides interface translations in multiple languages. It also supports multilingual definitions, allowing users to select the definition language, and the system provides word definitions, example sentence translations, and article translations in multiple languages. Furthermore, it supports multilingual content generation, generating definitions, translations, and educational prompts in the corresponding language based on the user's native language.

[0055] Establish a multilingual data management system, construct a multilingual database, and store interface translations, word definitions, example sentence translations, and article translations in multiple languages. Support dynamic switching; users can switch interface and definition languages ​​at any time, and the system updates the interface and content in real time.

[0056] In one embodiment, such as Figure 10 As shown, the method disclosed in this embodiment supports multi-granularity text selection and type recognition. When reading English content, users can select words, phrases, sentences, paragraphs, and articles for translation. The system automatically identifies the type of the selected text and selects an appropriate translation strategy based on the type.

[0057] A hybrid translation strategy and intelligent routing are employed. For word-level translation, dictionary APIs are prioritized for their fast response, low cost, and high accuracy. For phrase, sentence, paragraph, and article-level translation, a large language model is used to understand the context and provide more accurate and natural translations. Intelligent routing automatically selects the most appropriate translation strategy based on text type, length, and complexity.

[0058] Generate and display personalized translations. For word-level translations, display dictionary definitions, phonetic symbols, parts of speech, and example sentences. For phrase, sentence, paragraph, and article-level translations, display translation results generated by a large language model, combined with the user's native language and proficiency to provide more understandable translations. Provide a translation history, allowing users to view and review previously translated content.

[0059] Record translation history and update user profiles, noting what content users translated, the frequency of translation, and the difficulties they encountered. Based on the translation history, analyze user weaknesses, update user profiles, and optimize subsequent content generation and recommendations.

[0060] To optimize translation caching and control costs, a caching mechanism is used for frequently used words and phrases to avoid repeated API calls, thereby reducing costs and improving response speed. For long text translation, a segmented translation strategy is used to avoid excessively long texts in a single API call, further reducing costs.

[0061] It offers personalized translation settings and user controls. Users can configure translation preferences, including whether to enable real-time translation, the default translation language, and the translation display method. Users can also control translation behavior, including pausing translation, clearing translation history, and re-translating.

[0062] This embodiment also discloses a global intelligent English learning system based on multi-dimensional user profiles, which performs the above method, including: The user profiling module is used to collect and store multidimensional information about users and to build user profile data models. The proficiency assessment module is used to conduct an initial language proficiency assessment of users to generate an initial proficiency model, and dynamically update the initial proficiency model based on users' learning behavior data to obtain the current proficiency model; The personalized content generation module is used to generate personalized learning content for users based on user profile data models and current capability models. The content security filtering module is used to perform content security filtering during the generation process of personalized learning content.

[0063] In one embodiment, the system further includes: The user management module is used for user registration, login, authentication, authorization, user profile management, and performance evaluation.

[0064] In one embodiment, the level assessment module includes: The learning management unit is used for learning progress tracking, ability assessment, learning records, and learning behavior analysis.

[0065] In one embodiment, the personalized content generation module includes: The content management unit is used for vocabulary management, example sentence management, article management, and image management.

[0066] The AI ​​generation engine unit includes sub-units for vocabulary analysis, example sentence generation, article generation, and image generation. The vocabulary analysis sub-unit is used for root and affix decomposition, phonetic transcription generation, definition lookup, and similar word analysis. The example sentence generation sub-unit is used for personalized example sentence generation, difficulty control, and quality checking. The article generation sub-unit is used for customized article generation, scene fusion, and length control. The image generation sub-unit is used for personalized image generation, style control, and quality checking.

[0067] In one embodiment, the personalized content generation module further includes an intelligent recommendation unit for word recommendation, article recommendation, association learning recommendation, and root and affix recommendation.

[0068] In one embodiment, the system further includes an intelligent real-time translation assistance module for multi-granularity text selection and type recognition, hybrid translation strategies and intelligent routing, personalized translation content generation and display, translation history and user profile updates, translation cache optimization and cost control, and personalized translation settings and user control.

[0069] In one embodiment, the system further includes a data storage module, comprising a database submodule, a cache submodule, and a file storage submodule. The database submodule is used to store user data, learning data, and content data; the cache submodule is used to store session cache, generation cache, and hot data; the file storage submodule is used to store audio files, image files, and exported files; and the internationalization module is used for multilingual interface support, multilingual interpretation support, multilingual content generation, and multilingual data management.

[0070] In one embodiment, the AI ​​generation engine unit uses large language model technology to dynamically generate personalized example sentences and articles based on user profiles and learning levels; the image generation subunit uses AI image generation technology to generate personalized images based on user profiles and word meanings.

[0071] In one embodiment, the intelligent recommendation unit employs a multi-dimensional similarity calculation algorithm, including root and affix similarity, substring similarity, and pronunciation similarity, to identify the learned words most relevant to the new word, thus realizing the "find similarities and differences" learning method.

[0072] In one embodiment, the intelligent real-time translation assistance module adopts a hybrid translation strategy. For word-level translation, it prioritizes the use of dictionary APIs, while for phrase, sentence, paragraph, and article-level translation, it uses large language models and automatically selects the most suitable translation strategy based on text type, length, and complexity.

[0073] In one embodiment, the data storage module adopts a multi-level caching architecture, including an L1 cache for storing user-level frequently used word translations, an L2 cache for storing globally frequently used word translations, and an L3 cache for storing phrase and sentence translations, thereby improving translation response speed and reducing translation costs.

[0074] This embodiment further elaborates on this application based on a specific application scenario, and the specific steps are as follows: like Figure 2 As shown in the figure, this embodiment provides a method for constructing a multi-dimensional user profile, and the specific steps are as follows: After user Zhang San registers and logs into the system, the system guides him to complete the user profile collection process. The steps are as follows: Collect basic information: Zhang San fills in the following basic information: Name "Zhang San", Age 15, Status "K12 student", Grade "Junior High School Grade 9", Mother tongue "Chinese", Learning goal "Junior High School Entrance Examination"; Collect interests and hobbies: Zhang San selects the following interests and hobbies: football, programming, music, and travel; Collect occupational / academic background: Zhang San selects student type, grade "Junior High 3", and major "Science"; Collect common scenarios: Zhang San selects common scenarios: coffee shop, school, football field, airport; Content style preferences: Zhang San selected the following content styles: humorous and lighthearted, practical and useful information; Learning preferences were collected: Zhang San chose to spend "30-60 minutes" a day learning English, and his preferred learning methods were "reading articles" and "memorizing words".

[0075] Building a user profile data model: The system stores the collected information as structured data, builds a user profile data model, and stores it in the database.

[0076] like Figure 3 As shown in the figure, this embodiment provides a method for intelligent level assessment and dynamic updating, and the specific steps are as follows: Initial Assessment: Based on Zhang San's basic information, the system generates a customized reading comprehension test using a large language model. The test includes an article and several questions. After Zhang San completes the test, the system analyzes his answers using natural language processing technology, assessing his vocabulary (approximately 2000 words), grammar level (intermediate), reading comprehension ability (intermediate), and writing expression ability (intermediate), resulting in an overall score of 65. The system then generates an ability model and stores it in the database.

[0077] Dynamic Update: After Zhang San learns 20 words, the system triggers a dynamic update. The system retrieves Zhang San's learning data from the past 7 days, analyzes his learning performance, and calculates a new ability score. The system uses a smooth update algorithm to calculate a weighted average of the new and old scores to update the ability model. The updated ability model is then stored in the database.

[0078] like Figure 4 As shown in the figure, this embodiment provides a method for generating personalized example sentences based on user profiles. The specific steps are as follows: Timing of word generation: After Zhang San logs in, the system backend asynchronously generates example sentences for the first 10 words in the vocabulary list. When Zhang San learns the 5th word, the system pre-generates example sentences for words 11-20.

[0079] Parameters are generated: Zhang San looks up the word "achieve". The system generates parameters, including the target word "achieve", user profile (level intermediate, vocabulary 2000, grammar level intermediate, interests: football, programming, music, scenarios: coffee shop, school, football field, style: humorous and lighthearted, practical and useful information), and target difficulty 65.

[0080] The system utilizes a large language model to generate example sentences. It constructs AI prompts and generates three example sentences, each containing the word "achieve." The scenario and content of the example sentences should be relevant to Zhang San's interests and common scenarios. The difficulty level should match Zhang San's proficiency, and the sentences should be natural, idiomatic, and native-speaker-like. The style should suit Zhang San's preferences, and each sentence should not exceed 20 words. A Chinese translation is provided. The system calls the large language model API, inputs the prompts, and obtains the generated results. Generated example sentences include: "After months of practice, our team finally achieved victory in the championship." (After several months of practice, our team finally achieved victory in the championship.) "The AI ​​model achieved high accuracy after fine-tuning." (After fine-tuning, this AI model achieved high accuracy.) "She achieved her dream of becoming a musician by performing at coffee shops." (She achieved her dream of becoming a musician by performing at coffee shops.)

[0081] Quality checks: The system uses a grammar checker API to ensure example sentences are grammatically correct. The system calculates the lexical difficulty and grammatical complexity of the sentences to ensure the difficulty matches Zhang San's level. The system uses multiple large language models to score the example sentences, and all scores are above a threshold. The system ensures that the example sentences contain the target words and are relevant to Zhang San's interests and the context.

[0082] Storage and Display: The system stores example sentences in the database and displays them to Zhang San in word cards. Each example sentence includes an English sentence, a Chinese translation, a scene label, and a pronunciation button.

[0083] This embodiment provides a personalized article generation method based on user profiles, the specific steps of which are as follows: Target words are identified: Zhang San selects 5 words from the vocabulary list: achieve, challenge, practice, victory, and teamwork. The system analyzes the part of speech, difficulty, and semantic relationships of the target words, and identifies the connections between them.

[0084] Article generation parameters are constructed: The system constructs article generation parameters, including a target word list, user profile (level intermediate, vocabulary 2000, grammar level intermediate, interests: football, programming, music, scene: coffee shop, school, football field, style: humorous and lighthearted, practical and informative), and article requirements (word count 200, difficulty target 65, genre: narrative, theme: football match).

[0085] The system generates articles using a large language model: It constructs AI-generated prompts and generates an English article of approximately 200 words. The article must contain all target words, its theme and plot should align with Zhang San's interests (e.g., football), its difficulty level should match Zhang San's skill level, it should have a plot and story to engage Zhang San, its style should suit Zhang San's preferences (humorous and lighthearted, practical and informative), and it should be natural, idiomatic, and native-like. A Chinese translation is provided, along with annotations of the position of each target word within the article. The system calls the large language model API, inputs the prompts, and retrieves the generated result. The generated article is titled "The Championship Game" and contains the following content: "Our school football team faced a huge challenge last Saturday. We had been practicing for months to achieve our goal of winning the championship. The game was tough, but our teamwork made the difference. In the final minutes, we scored the winning goal and achieved victory. It was an unforgettable moment!"

[0086] Quality checks: The system ensures the article contains all target words. The system calculates the article's lexical difficulty and grammatical complexity to ensure the difficulty matches Zhang San's level. The system uses a grammar checker API to ensure the article is free of grammatical errors. The system uses multiple large language models to score the article, and all scores are above the threshold. The system ensures the article has a clear plot and storyline, and that the content is relevant to Zhang San's interests and the given context.

[0087] Storage and Display: The system stores articles in a database and displays them to Zhang San on the article reading page. Target words in the article are highlighted, and clicking on them displays definitions and example sentences. The system provides article reading and translation functions.

[0088] like Figure 5 As shown, this embodiment provides a method for in-depth analysis and visualization of word roots and affixes. The specific steps are as follows: Word root and affix identification and decomposition: Zhang San learns the word "unbelievable". The system uses an AI model to decompose the word root and affix, and obtains the prefix "un-" (not, not), the word root "-believe-" (believe), and the suffix "-able" (able).

[0089] Word root and affix meaning annotation: The system uses color-coded visual annotation, with prefixes using a purple gradient, roots using a pink gradient, and suffixes using a blue gradient. The system annotates the meaning of each word root and affix below it, showing the derivation process of the complete meaning: "un-believe-able → not + believe + able → cannot believe → unbelievable".

[0090] The root and affix list function automatically extracts the root and affixes of "unbelievable" words and counts the number and types of root and affixes that Zhang San knows. The system displays the root and affix list, sorted by frequency of use, and shows the meaning, example words, and level of mastery for each root and affix.

[0091] Word root and affix association learning: The system recommends other words containing the same word root, such as "believe", "believable", and "disbelieve". The system uses AI to generate word root and affix practice questions, such as "Given the word root '-believe-' (believe), please write 3 words containing this word root".

[0092] like Figure 6 As shown, this embodiment provides a "finding similarities and differences" learning method, the specific steps of which are as follows: Intelligent word association recommendation: Zhang San inputs the new word "uncomfortable," and the system retrieves related words from Zhang San's existing vocabulary. The system calculates similarity, including root and affix similarity, substring similarity, and pronunciation similarity. Based on the similarity calculation results, the system recommends the most relevant words from the existing vocabulary: "comfortable" (similarity 0.85), "unable" (similarity 0.45), and "unhappy" (similarity 0.40).

[0093] The two-step learning method: The system first identifies common elements. It extracts the root and affixes of the new word "uncomfortable" ([un-, comfort, -able]) and the previously learned word "comfortable" ([comfort, -able]). The system then identifies the common root and affixes ([comfort, -able]). The system highlights these common elements and generates a knowledge transfer prompt: "You already know most of this word! You just need to remember their combinations!" The system then performs the second step: finding the different parts. It identifies the different parts of the new word "uncomfortable": [un-], and the different parts of the already learned word "comfortable": []. The system highlights the different parts and generates a prompt to avoid confusion: "Pay attention to the meanings of different word roots! Although they share some parts, the different parts determine the specific meaning!"

[0094] Generate a visual comparison display: the system highlights similar parts in blue and different parts in red. The system displays comparison cards, including the spelling, pronunciation, roots and affixes, meaning, and example sentences of new and learned words. The system also displays educational tips.

[0095] Learning behavior recorded: The system recorded that Zhang San viewed the related word "comfortable", viewed the same part [comfort, -able] and different part [un-], the learning time was 2 minutes, and the learning effect was good.

[0096] like Figure 7 As shown, this embodiment provides a multimodal personalized memory assistance method, the specific steps of which are as follows: Determine when to generate an image: Zhang San learns the word "achieve" and actively requests the generation of a personalized image.

[0097] Image generation parameters are constructed: The system constructs image generation parameters, including the target word "achieve", user profile (interests: football, programming, music, scenes: coffee shop, school, football field, style: humorous and lighthearted, practical and useful information), and image requirements (style: cartoon, scene: football field, elements: football, trophy, player).

[0098] Personalized images are generated using AI image generation technology: The system constructs image generation prompts, requiring the image to be related to the meaning of the word "achieve" (to realize, to reach), the image's scene and elements to reflect Zhang San's interests (football) and the setting (football field), the image style to match Zhang San's preferences (humorous and lighthearted), and the image to be clear, aesthetically pleasing, and attractive. The system calls the AI ​​image generation API, inputs the prompts, and obtains the generated result. The generated image shows a football player raising a trophy to celebrate victory.

[0099] Quality checks: The system ensures that the image is relevant to the meaning of the word "achieve" and to Zhang San's interests and the context. The system uses an image quality assessment model to score the image, with scores exceeding a threshold. The system ensures the image is clear, without blur or noise.

[0100] Storage and Display: The system stores the images in a file and displays them to Zhang San in the word cards. The system provides an image zoom-in view function.

[0101] like Figure 8 As shown, this embodiment provides multiple intelligent word import methods, and the specific steps are as follows: OCR Image Import: Zhang San takes a photo containing the English words "achievement," "challenge," and "practice." The system uses OCR technology to recognize the English words in the photo and extracts a word list. The system displays the extracted word list, and after Zhang San confirms it, it is added to the vocabulary list in batches.

[0102] Document Upload and Import: Zhang San uploads a Word document containing the English words "victory," "teamwork," and "success." The system parses the document content, extracts the English words, and creates a word list. The system displays the extracted word list, which Zhang San then adds to his vocabulary list in batches after confirmation.

[0103] Text Paste Import: Zhang San pastes a text containing the English words "effort," "goal," and "improve." The system parses the text, extracts the English words, and creates a word list. The system displays the extracted word list, which Zhang San then confirms and adds to his vocabulary list in batches.

[0104] API Import: A third-party application imports the words "confidence", "motivation", and "perseverance" in batches via API. The system receives the API request, extracts the word list, and adds it to Zhang San's vocabulary.

[0105] Word deduplication and validation: The system checks if the imported words already exist in the vocabulary list. If they do, a message is displayed to Zhang San. The system also checks if the imported words are valid English words. If they are invalid, a message is displayed to Zhang San.

[0106] like Figure 9 As shown, this embodiment provides global service capabilities, and the specific steps are as follows: Multilingual interface support: Zhang San can select the interface language, and the system provides interface translations for multiple languages ​​including Chinese, English, Japanese, Korean, French, German, and Spanish. If Zhang San selects the Chinese interface, the system will update the interface to Chinese in real time.

[0107] Supports multilingual definitions: Zhang San can select the definition language, and the system provides word definitions, example sentence translations, and article translations in multiple languages. If Zhang San selects Chinese definitions, the system displays Chinese word definitions, example sentence translations, and article translations.

[0108] Supports multilingual content generation: The system generates definitions, translations, and educational tips in the corresponding language based on Zhang San's native language (Chinese).

[0109] Multilingual Data Management: The system constructs a multilingual database, storing interface translations, word definitions, example sentence translations, and article translations in multiple languages. The system supports dynamic switching; Zhang San can switch the interface language and definition language at any time, and the system updates the interface and content in real time.

[0110] like Figure 10 As shown, this embodiment provides an embedded real-time translation assistance method, the specific steps of which are as follows: Multi-granular text selection and type recognition: When Zhang San is reading an English article, he selects the word "achievement". The system automatically recognizes the type of the selected text as word level.

[0111] Hybrid translation strategy and intelligent routing: The system selects a dictionary API for translation based on the text type (word level). The system calls the dictionary API to obtain the definition, phonetic transcription, part of speech, and example sentences for the word "achievement".

[0112] Personalized translation content generation and display: The system displays the dictionary definition of "achievement, accomplishment", phonetic transcription " / əˈtʃiːvmənt / ", part of speech "noun", and example sentence: "Winning the championship was a great achievement."

[0113] Translation history and user profile updates: The system records Zhang San's translation of the word "achievement," including the translation time and frequency. Based on this history, the system analyzes Zhang San's weaknesses and updates his user profile accordingly.

[0114] Translation caching optimization and cost control: The system caches the translation result of the word "achievement". The next time Zhang San translates the word, he can directly retrieve it from the cache without repeatedly calling the API.

[0115] Personalized translation settings and user controls: Zhang San can set translation preferences, including whether to enable real-time translation, the default translation language, and the translation display method. Zhang San can control translation behavior, including pausing translation, clearing translation history, and re-translating.

[0116] In one embodiment, content security filtering is integrated throughout the entire content generation process, specifically as follows: Before generating personalized learning content, keyword filtering, semantic analysis, and intent recognition are performed on the user's input to intercept inappropriate requests; During the personalized learning content generation process, the generation behavior is constrained based on preset security instructions and security parameters. After personalized learning content is generated, it is reviewed and then classified and categorized according to the review results.

[0117] Furthermore, such as Figure 11 As shown, this embodiment also includes a content security filtering method that runs through the entire content generation process, aiming to ensure that all content generated by AI or input by users complies with laws, regulations, and ethical standards. This method, as a fundamental security safeguard, is applied to all stages involving content creation, such as personalized example sentence generation, article generation, and image generation. Specifically: Input-side proactive defense steps: triggered when the system receives any form of user input (e.g., learning requests, search queries, uploaded text or images), with the aim of intercepting inappropriate or malicious requests at the source; The processing methods include: Keyword filtering: The system first quickly matches user input against a pre-set, dynamically updated sensitive word database, which includes words in various categories such as violence, pornography, drugs, gambling, and politically sensitive terms. Once a match is found, the request will be directly rejected or marked as high-risk.

[0118] Semantic analysis: To prevent bypassing keyword filtering through homophones, variations, metaphors, etc., the system will call one or more professional content security big language models to perform deep semantic understanding and vectorization analysis on the text entered by the user to determine whether its true intent is illegal.

[0119] Intent recognition: The system also analyzes patterns in user input to identify whether there are attempts at "prompt injection" or leading questions, such as trying to induce the system to generate inappropriate content through role-playing or fictional scenarios.

[0120] The generation process control steps are activated when the system calls the Large Language Model (LLM) or image generation model to perform the generation task. These steps aim to internally constrain the model's behavior and involve the following steps: System-level security instruction injection: When constructing the final prompt word sent to the large language model, the system automatically injects a high-priority security instruction at the top level that cannot be overridden by user input. This instruction explicitly and mandatorily requires the model not to generate any text involving prohibited content categories.

[0121] Security parameter configuration: The system will configure specific security parameters for the generation task, such as disabling or restricting certain creative features of the model that may lead to the generation of extreme or inappropriate content, thereby reducing the risk technically.

[0122] Output response processing steps: After content is generated and before it is presented to the user, this step is performed as the last line of defense, and its processing steps are as follows: Multi-model cross-verification: To avoid misjudgments or vulnerabilities in a single verification model, the system adopts a multi-model cross-verification mechanism, which sends the generated content to at least two independent content verification models from different vendors or architectures for analysis. Only when all models determine that the content is safe will the content be directly approved.

[0123] Three-level classification and handling: The review results are divided into three levels: (1) Pass: The content is safe and is presented directly to the user; (2) Review pending: The content has uncertainty or potential risks, so it is temporarily suspended and not presented to the user, and is immediately submitted to the manual review platform; (3) Reject: The content is clearly in violation of regulations, so it is directly discarded, and the user can be warned or their behavior recorded according to the policy.

[0124] Human arbitration and feedback loop: For content "awaiting human review," a professionally trained human reviewer makes the final judgment. The reviewer's judgment not only determines whether the content is visible to users, but also serves as high-quality annotation data, used to fine-tune the content security model, thereby continuously improving the accuracy of automated review and forming a continuously optimized data feedback loop.

[0125] This embodiment constructs a content security system with defense in depth and multi-layered redundancy, which maximizes the protection of the positive, healthy and compliant nature of learning content, and creates a safe and reliable intelligent learning environment for users, especially minors.

[0126] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0127] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A global intelligent English learning method based on multi-dimensional user portrait, characterized in that, The specific steps are: Collect and store the multi-dimensional information of the user, and build a user portrait data model; the multi-dimensional information includes basic information, common scenarios, and personal preferences; Perform initial language ability assessment on the user to generate an initial ability model, and dynamically update the initial ability model based on the user's learning behavior data to obtain a current ability model; According to the user portrait data model and the current ability model, personalized learning content is generated for the user; and the generation process of the personalized learning content is subjected to content security filtering. 2.The globalized intelligent English learning method based on multi-dimensional user portrait according to claim 1, characterized in that, The steps for obtaining the current ability model are: Obtain the basic information of the user, generate personalized language ability test questions based on the basic information, analyze and evaluate the user's answers to the language ability test questions, and generate an initial ability model containing multi-dimensional ability scores; Real-time acquisition of the user's learning progress, learning content mastery and difficulty trend, when the preset triggering condition is met, the initial ability model is updated by using a smoothing algorithm according to the real-time acquisition data to obtain the current ability model. 3.The globalized intelligent English learning method based on multi-dimensional user portrait according to claim 1, characterized in that, The personalized learning content includes personalized example sentence generation, and the specific steps are: Build an example sentence generation parameter, which includes a target word, a user portrait data model, and a difficulty target; Build an example sentence generation prompt word based on the example sentence generation parameter; Input the example sentence generation prompt word into a large language model to generate a personalized example sentence; Quality check the personalized example sentence, and display the qualified personalized example sentence and the corresponding translation to the user.

4. The method according to claim 3, wherein, The quality check includes syntax check, difficulty assessment, nativeness check and relevance check.

5. The method of claim 1, wherein the method further comprises: The personalized learning content also includes personalized article generation, and the specific steps are: Obtain a target word list and the relevance of the part of speech, semantics and difficulty of each target word in the target word list; Based on the target word list, the user portrait data model and the current ability model, build an article generation prompt; Input the article generation prompt into a large language model to obtain a personalized article; Quality check the personalized article, and display the qualified personalized article to the user.

6. The method of claim 5, wherein the method further comprises: The quality check includes target word coverage check, difficulty assessment, syntax check, nativeness check and plot coherence check.

7. The method of claim 1, wherein the method further comprises: Content security filtering runs through the whole process of content generation, specifically: Before generating personalized learning content, filter the user's input content by keywords, analyze the semantics and identify the intent to intercept inappropriate requests; During the generation of personalized learning content, based on the preset security instructions and security parameters, the generation behavior is constrained; After generating personalized learning content, the personalized learning content is audited, and the classification processing is performed according to the audit result. 8.The globalized intelligent English learning method based on multi-dimensional user portrait according to claim 7, characterized in that, Before generating personalized learning content, the specific steps of content security filtering are: Match and filter the user's input content with the preset sensitive word library; Call the content security model to analyze the semantics of the user's input content and identify abnormal expressions; According to the mode of the input content of the user, it is judged whether there is a prompt word attack or an attempt of induced questioning. 9.The globalized intelligent English learning method based on multi-dimensional user profiling according to claim 8, wherein, After the personalized learning content is generated, the specific steps of content security filtering are: Cross-verification of the generated personalized learning content is performed by using at least two independent audit models; According to the audit result, the personalized learning content is executed for classification disposal, and the disposal result at least includes passing, waiting for manual audit and rejecting; For the personalized learning content waiting for manual audit, it is submitted to a manual audit platform for arbitration, and the arbitration result is fed back to the content security model.

10. The method of claim 1, wherein the method further comprises: When learning the personalized learning content, an intelligent association learning method can be used, and the specific steps are: An unlearned word is obtained, and multi-dimensional similarity of the unlearned word and each word in a learned word library is calculated; The multi-dimensional similarity includes root and affix similarity, substring similarity and pronunciation similarity; According to the similarity, a preset number of associated words are selected from the learned word library; The associated words and the unlearned word are compared, and the same part and the different part are both prompted.

11. The method of claim 3, wherein the method further comprises: When learning the personalized learning content, an auxiliary memory method is also included, and the specific steps are: Based on the target word and the user portrait data model, a picture generation prompt is constructed; According to the picture generation prompt, an individualized memory picture related to the meaning of the target word is generated; The individualized memory picture is displayed to the user.

12. The method of claim 1, wherein the method further comprises: When generating the personalized learning content, a viewing-time generation and rolling pre-generation strategy is used.

13. A globalized intelligent English learning system based on multi-dimensional user profiling, performing a globalized intelligent English learning method based on multi-dimensional user profiling according to any one of claims 1-12, characterized in that, It includes: A user portrait module is used to collect and store multi-dimensional information of a user, and a user portrait data model is constructed; A horizontal evaluation module is used to perform initial language ability evaluation on the user to generate an initial ability model, and based on learning behavior data of the user, the initial ability model is dynamically updated to obtain a current ability model; A personalized content generation module is used to generate personalized learning content for the user according to the user portrait data model and the current ability model; A content security filtering module is used to perform content security filtering on the generation process of the personalized learning content.

Images