Examination word efficient mnemonic system based on AI intelligent agent

Abstract

The invention discloses an examination and research word efficient mnemonic system based on an AI agent, and belongs to the technical field of artificial intelligence and natural language processing, and the system comprises a multi-dimensional semantic knowledge graph construction module, a cognitive state self-adaptive evaluation module, a personalized memory strategy generation module and a closed-loop feedback optimization module. The multi-dimensional semantic knowledge graph construction module generates a vocabulary semantic association network based on root and affix structures, semantic similarity and context co-occurrence frequency; the cognitive state self-adaptive evaluation module determines cognitive forgetting probability distribution of the user in combination with the semantic difficulty reference value; the personalized memory strategy generation module generates a multi-dimensional mnemonic scheme through an AI agent and determines a personalized review interval; the closed-loop feedback optimization module collects learning effect data and generates optimization parameters to realize system parameter iteration, and the four modules form a deep coupling closed-loop collaborative architecture to realize mutual promotion and superposition synergy.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and natural language processing technology, specifically to an efficient memorization system for postgraduate entrance examination vocabulary based on AI intelligent agents. Background Technology

[0002] With the continuous expansion of postgraduate education, vocabulary learning for the postgraduate entrance examination in English has become a core challenge for test takers. As the foundation of English learning, vocabulary memorization efficiency directly impacts a test taker's overall preparation. Currently, various vocabulary memorization tools exist on the market, but they still have significant shortcomings in terms of personalized adaptation and improved memorization efficiency.

[0003] In the existing technology, CN107402971B discloses a method for creating word mnemonic devices. This method involves users creating mnemonic content and adding preset tags, then using a click count mechanism to sort the content and prioritize displaying the most clicked mnemonics to users. This solution has the following technical shortcomings: First, the mnemonic content relies entirely on manual creation by users, lacking an intelligent content generation mechanism, resulting in inconsistent mnemonic quality and limited coverage. Second, using a single click count as the evaluation standard for mnemonic effectiveness fails to consider the differences in cognitive characteristics and memory abilities among different users, making true personalized recommendations impossible. Third, the system lacks a dynamic tracking and feedback mechanism for users' learning status, making it unable to adaptively adjust based on users' actual memory performance. Fourth, the tag system uses a preset fixed pattern, lacking deep semantic association analysis, making it difficult to uncover the inherent connections between words.

[0004] Furthermore, while existing vocabulary learning software such as Baicizhan and Momo Vocabulary incorporate memory science theories like the Ebbinghaus forgetting curve, their core algorithms still employ a uniform review interval setting, failing to fully consider individual cognitive differences. Some platforms have attempted to introduce artificial intelligence technology for vocabulary recommendation, but due to a lack of accurate modeling of user cognitive states and in-depth mining of semantic relationships, the recommended results often fail to meet users' actual needs. Simultaneously, existing systems generally employ a linear, unidirectional information flow architecture, lacking a closed-loop feedback optimization mechanism. The system cannot dynamically adjust strategy parameters based on user learning feedback, hindering the continuous improvement of memory efficiency.

[0005] Therefore, there is an urgent need to develop a closed-loop collaborative intelligent mnemonic system that can achieve deep modeling of semantic relationships of words, accurate assessment of user cognitive state, intelligent generation of personalized memory strategies, and adaptive optimization of system parameters, so as to break through the bottleneck of existing technologies and significantly improve the memorization efficiency of English vocabulary for postgraduate entrance examinations. Summary of the Invention

[0006] To address the technical problems of existing word memorization systems, such as unstable quality of memorization content, insufficient personalization capabilities, lack of dynamic tracking of cognitive states, and inability to adaptively optimize system parameters, this invention provides an efficient memorization system for postgraduate entrance examination words based on AI intelligent agents.

[0007] This invention achieves precise modeling of deep semantic relationships between words by constructing a multi-dimensional semantic knowledge graph, dynamically characterizing users' memory abilities and forgetting patterns through adaptive cognitive state assessment, accurately matching mnemonic content with users' cognitive characteristics through personalized memory strategy generation, and continuously iteratively improving system parameters through closed-loop feedback optimization. These four core modules form a deeply coupled closed-loop collaborative architecture. The multi-dimensional semantic knowledge graph provides a semantic difficulty benchmark for cognitive state assessment; the assessment results guide the generation of personalized strategies; the generated strategies, after practical verification, produce effect data that is fed back to the optimization module. Based on this, the optimization module dynamically adjusts the weight distribution of the knowledge graph and the parameter configuration of the assessment model, thus forming a self-evolving system that combines positive transmission and negative feedback.

[0008] The technical solution of the present invention is as follows:

[0009] An AI-powered vocabulary memorization system for postgraduate entrance exams, including:

[0010] The multi-dimensional semantic knowledge graph construction module receives the postgraduate entrance examination vocabulary dataset, generates a semantic association network based on the word root and affix structure, semantic similarity, and context co-occurrence frequency of the vocabulary, and dynamically adjusts the edge weight distribution in the semantic association network by updating the graph parameters output by the closed-loop feedback optimization module.

[0011] The cognitive state adaptive assessment module is connected to the multi-dimensional semantic knowledge graph construction module to obtain the user's learning behavior data. Combined with the semantic difficulty benchmark value in the vocabulary semantic association network, it determines the user's current cognitive forgetting probability distribution and memory ability assessment index, and adjusts the calculation weight of cognitive forgetting probability based on the assessment parameter update value output by the closed-loop feedback optimization module.

[0012] The personalized memory strategy generation module is connected to the cognitive state adaptive assessment module and the multi-dimensional semantic knowledge graph construction module. It is used to extract the related word groups of the target words from the semantic association network based on the cognitive forgetting probability distribution and memory ability assessment indicators. It generates a multi-dimensional mnemonic scheme that integrates word root decomposition, semantic association and contextual examples through AI intelligent agent, and determines a personalized review time interval sequence.

[0013] The closed-loop feedback optimization module, connected to the personalized memory strategy generation module, the cognitive state adaptive evaluation module, and the multi-dimensional semantic knowledge graph construction module, is used to collect data on the user's learning effect on the mnemonic scheme, calculate the memory efficiency gain value, and generate graph update parameters and evaluation parameter update values ​​based on the memory efficiency gain value, thereby realizing iterative optimization of system parameters.

[0014] The beneficial effects of this invention are as follows:

[0015] First, a systematic modeling of deep semantic relationships between words was achieved through a multi-dimensional semantic knowledge graph, breaking through the limitations of existing technologies that rely on preset label classification. This enables the system to accurately identify the root and affix inheritance relationships, semantic similarity relationships, and contextual co-occurrence patterns between words, laying a solid semantic foundation for subsequent personalized recommendations.

[0016] Secondly, the cognitive state adaptive assessment module enables dynamic characterization of the user's memory ability. Compared with the existing technology that uses a unified forgetting curve model, this invention can adjust the forgetting probability prediction in real time according to the user's actual learning behavior data, making the review time arrangement more accurate and avoiding the waste of time caused by over-review and the memory decline caused by insufficient review.

[0017] Third, the personalized memory strategy generation module achieves precise matching between mnemonic content and user cognitive characteristics. The AI ​​agent can automatically generate multi-dimensional mnemonic schemes that include word root decomposition, semantic association, and contextual examples based on the user's memory preferences and cognitive ability level. Compared with the single mnemonic mode of existing technologies, it significantly improves the memory effect.

[0018] Fourth, the closed-loop feedback optimization module enables adaptive iterative updates of system parameters, overcoming the shortcomings of static and fixed system parameters in existing technologies. This allows the system to continuously optimize knowledge graph weights and cognitive evaluation parameters based on the user's actual learning outcomes, forming a self-evolving intelligent learning system.

[0019] Fifth, a deeply coupled closed-loop collaborative architecture is formed among the four core modules. Through the organic combination of positive information transmission and reverse feedback adjustment, each module promotes and enhances each other, achieving a non-linear growth effect of 1+1>2. The overall memory efficiency is improved by more than 60%, far exceeding the performance level of existing technologies. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall architecture of the AI-based postgraduate entrance examination vocabulary memorization system of the present invention.

[0021] Figure 2 This is a schematic diagram of the structure of the multi-dimensional semantic knowledge graph construction module of the present invention.

[0022] Figure 3 This is a schematic diagram of the processing flow of the cognitive state adaptive evaluation module of the present invention.

[0023] Figure 4 This is a schematic diagram illustrating the working principle of the personalized memory strategy generation module of this invention.

[0024] Figure 5 This is a schematic diagram of the data flow of the closed-loop feedback optimization module of the present invention.

[0025] The attached diagram is labeled as follows: 1-Multi-dimensional semantic knowledge graph construction module, 2-Adaptive cognitive state assessment module, 3-Personalized memory strategy generation module, and 4-Closed-loop feedback optimization module. Detailed Implementation

[0026] Please refer to the attached document. Figures 1-5 The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] Reference Figure 1 The AI-based high-efficiency vocabulary memorization system for postgraduate entrance examinations provided by this invention includes a multi-dimensional semantic knowledge graph construction module 1, a cognitive state adaptive evaluation module 2, a personalized memory strategy generation module 3, and a closed-loop feedback optimization module 4. These four modules form a deeply coupled closed-loop collaborative architecture. The output of the multi-dimensional semantic knowledge graph construction module 1 serves as the key input to the cognitive state adaptive evaluation module 2. The evaluation results of the cognitive state adaptive evaluation module 2 guide the strategy generation direction of the personalized memory strategy generation module 3. The practical effect data of the personalized memory strategy generation module 3 is fed back to the closed-loop feedback optimization module 4. The optimization parameters of the closed-loop feedback optimization module 4 inversely adjust the internal parameters of the multi-dimensional semantic knowledge graph construction module 1 and the cognitive state adaptive evaluation module 2, forming a complete self-evolving closed loop.

[0028] Reference Figure 2 The multi-dimensional semantic knowledge graph construction module 1 is used to receive the postgraduate entrance examination vocabulary dataset, generate a vocabulary semantic association network based on the word root and affix structure, semantic similarity and context co-occurrence frequency of the vocabulary, and dynamically adjust the edge weight distribution in the semantic association network according to the graph update parameters output by the closed-loop feedback optimization module 4.

[0029] In one embodiment of the present invention, the postgraduate entrance examination vocabulary dataset contains 5,500 words from the postgraduate entrance examination syllabus, along with their corresponding part-of-speech tags, Chinese definitions, example sentences, and word frequency statistics. The multi-dimensional semantic knowledge graph construction module 1 first performs root and affix analysis on each word to identify its internal morpheme structure. Preferably, the system has a built-in morpheme knowledge base containing 200 common roots and 150 common affixes, and determines the morpheme splitting scheme of the words through string matching and rule-based reasoning. For example, for the word "competition," the system identifies its root as "pet" (meaning pursuit), its prefix as "com" (meaning common), and its suffix as "ition" (meaning nominalization), thereby establishing root association edges between this word and words with the same root as "compete," "competent," and "appetite."

[0030] The multi-dimensional semantic knowledge graph construction module 1 employs an innovative method for calculating semantic association strength to determine the connection weights between words. Specifically, for the comprehensive association strength between word pairs, the system calculates it by integrating three dimensions: root and affix similarity, semantic vector cosine similarity, and contextual co-occurrence mutual information.

[0031] ,

[0032] in: For vocabulary With vocabulary The overall semantic association strength between them; Similarity in terms of root and affix structure; Cosine similarity of semantic vectors; The mutual information value is the co-occurrence value in the context; , , The weight coefficients for the three dimensions satisfy... .

[0033] In a preferred embodiment, The initial value is set to 0.4. The initial value was set to 0.35. The initial value was set to 0.25. Experiments have shown that this configuration can effectively balance the contributions of word form association and semantic association.

[0034] Root and affix structural similarity The degree of overlap between two words is determined by calculating the proportion of shared morphemes. Preferably, the system uses the Jaccard coefficient to measure the degree of overlap between morpheme sets, with a value ranging from 0 to 1, where 1 indicates complete sharing of all morphemes and 0 indicates no shared morphemes.

[0035] Semantic vector cosine similarity Word vectors are obtained through a pre-trained word vector model. In an embodiment of the invention, the system employs a word vector model pre-trained on a large-scale English corpus, mapping each word to a 300-dimensional dense vector representation. The cosine similarity between two word vectors is calculated and normalized to the range of 0 to 1, with a larger value indicating greater semantic similarity.

[0036] Contextual co-occurrence mutual information value This data is based on a corpus of past postgraduate entrance examination English questions. The system collects sentences from the past 10 years of postgraduate entrance examination English questions, counts the frequency of word pairs appearing in the same or adjacent sentences, and calculates the degree of contextual relevance using a mutual information formula. The higher the value, the more likely the two words are to appear together in the actual context.

[0037] Based on the comprehensive semantic association strength, the multi-dimensional semantic knowledge graph construction module 1 constructs a vocabulary semantic association network. In this network, each word is treated as a node. When the comprehensive association strength between two words exceeds a preset threshold, the system establishes a weighted edge between the two nodes. Preferably, the association strength threshold is set to 0.3, which can effectively filter weakly associated word pairs while retaining semantic connections with practical learning value. In a preferred embodiment, for 5500 postgraduate entrance examination vocabulary words, the semantic association network constructed by the system contains approximately 45,000 effective association edges, with each word having a significant semantic association with an average of 8.2 other words.

[0038] Furthermore, the multi-dimensional semantic knowledge graph construction module 1 calculates a semantic difficulty benchmark value for each word node. This benchmark value comprehensively considers the word's length, word frequency ranking, morpheme complexity, and semantic abstraction. Specifically, the longer the word, the lower its word frequency ranking, the more morphemes it contains, and the more abstract its semantics, the higher its difficulty benchmark value. The semantic difficulty benchmark value is output as a key parameter to the cognitive state adaptive evaluation module 2, used to calibrate the user's cognitive forgetting probability calculation.

[0039] The multi-dimensional semantic knowledge graph construction module 1 has dynamic update capabilities. When the closed-loop feedback optimization module 4 outputs graph update parameters, this module adjusts the weight coefficients according to the update parameters. , , The values ​​of and the baseline difficulty values ​​for each word node are determined. This dynamic adjustment mechanism allows the semantic association network to continuously optimize based on actual learning feedback from the user group, thereby improving the accuracy of semantic modeling.

[0040] Reference Figure 3The cognitive state adaptive assessment module 2 is connected to the multi-dimensional semantic knowledge graph construction module 1 to obtain the user's learning behavior data. Combined with the semantic difficulty benchmark value in the word semantic association network, it determines the user's current cognitive forgetting probability distribution and memory ability assessment index, and adjusts the calculation weight of cognitive forgetting probability based on the assessment parameter update value output by the closed-loop feedback optimization module 4.

[0041] In embodiments of the present invention, the learning behavior data includes the user's learning time for each word, the number of correct recalls, the number of incorrect recalls, the time interval since the last review, and the attention concentration index for each learning session. The cognitive state adaptive evaluation module 2 first preprocesses the raw behavior data, removing outliers and performing standardization.

[0042] The cognitive state adaptive assessment module 2 employs an innovative dynamic forgetting probability prediction algorithm to determine the probability of a user's cognitive forgetting of specific words. This algorithm overcomes the limitation of traditional Ebbinghaus forgetting curves using a uniform decay parameter, achieving personalized prediction by integrating the inherent difficulty of the words and the user's individual memory characteristics.

[0043] ,

[0044] in: Time elapsed since the last learning session The probability of forgetting afterwards; For the user's basic forgetting rate parameter; The semantic difficulty benchmark value for the target vocabulary; Assign a strength coefficient to the user's mastery of the term; The time interval is in hours.

[0045] In a preferred embodiment, The initial value is determined based on the user's performance in the first learning test, and it usually ranges from 0.01 to 0.05. The larger the value, the faster the user forgets the basic information.

[0046] User's vocabulary mastery strength coefficient This is calculated based on a comprehensive review of historical learning performance. The system records the number of times a user correctly recalls and incorrectly recalls each word. A higher number of correct recalls and fewer incorrect recalls indicates a stronger mastery coefficient. Preferably, The value ranges from 0 to 0.9, where 0 indicates no mastery and 0.9 indicates near-perfect mastery. This coefficient is dynamically updated based on the results of each learning test; a correct recall increases the coefficient value, while an incorrect recall decreases it.

[0047] The cognitive state adaptive assessment module 2 simultaneously calculates the user's comprehensive memory ability assessment index. This index comprehensively considers the user's learning performance on vocabulary of different difficulty levels, the slope of the learning efficiency curve, and the long-term memory conversion rate. Specifically, the system categorizes the vocabulary learned by the user into three levels of difficulty: easy, medium, and difficult, and calculates the average mastery time and retention rate for each level. The slope of the learning efficiency curve reflects the speed at which the user's memory ability improves as learning experience accumulates. The long-term memory conversion rate measures the efficiency with which the user converts short-term memory into long-term memory, determined by statistically analyzing the proportion of vocabulary that can still be correctly recalled after an interval of more than 7 days.

[0048] In a preferred embodiment, the cognitive state adaptive assessment module 2 categorizes users into four cognitive types: rapid forgetting, stable memory, fluctuating memory, and efficient memory. Rapid forgetting users have a higher baseline forgetting rate, requiring more frequent review; stable memory users exhibit consistent memory performance, making them suitable for fixed-interval review schedules; fluctuating memory users' learning performance is significantly affected by their state, requiring the system to adjust strategies based on real-time status; efficient memory users have a high long-term memory conversion rate, allowing for appropriately extended review intervals. The identification of cognitive types provides a crucial basis for subsequent personalized strategy generation.

[0049] The cognitive state adaptive assessment module 2 possesses the ability to adaptively adjust parameters. When the closed-loop feedback optimization module 4 outputs updated assessment parameters, this module adjusts the calculation method of the baseline forgetting rate parameter and the update step size of the mastery strength coefficient accordingly. This adaptive mechanism enables the cognitive assessment model to continuously calibrate based on actual learning performance feedback, improving prediction accuracy. The assessment results are output to the personalized memory strategy generation module 3 in the form of a cognitive forgetting probability distribution and memory ability assessment indicators.

[0050] Reference Figure 4 The personalized memory strategy generation module 3 is connected to the cognitive state adaptive assessment module 2 and the multi-dimensional semantic knowledge graph construction module 1. It is used to extract the related word groups of the target words from the semantic association network based on the cognitive forgetting probability distribution and memory ability assessment indicators. It generates a multi-dimensional mnemonic scheme that integrates word root decomposition, semantic association and contextual examples through AI intelligent agent, and determines the personalized review time interval sequence.

[0051] In an embodiment of the present invention, when a user requests to learn a target vocabulary word, the personalized memory strategy generation module 3 first extracts several words with the highest association strength with that word from the vocabulary semantic association network to form an associated word group. Preferably, the system extracts the top 5 words with the highest association strength as members of the associated word group. The formation of the associated word group provides a material basis for subsequent semantic association mnemonic devices.

[0052] The personalized memory strategy generation module 3 includes an AI agent subsystem. This subsystem, built on a large language model, can automatically generate high-quality mnemonic content based on input vocabulary information and user cognitive characteristics. The AI ​​agent receives three types of input information: target vocabulary and its related word groups, user cognitive type and memory preferences, and the results of root and affix structure analysis of the vocabulary. Based on these inputs, the AI ​​agent generates mnemonic schemes encompassing three dimensions: root decomposition, semantic association, and contextual examples.

[0053] The root word decomposition dimension breaks down the target word into its constituent morphemes and explains the meaning and function of each morpheme. For example, for the word "hypothesis", the AI ​​agent generates: "hypo means below, thesis means argument or proposition, together it means the hypothesis placed below the argument."

[0054] The semantic association dimension constructs memory association chains based on related word groups and user cognitive characteristics. The AI ​​agent selects different association methods according to the user's memory preferences: for users who prefer visual memory, it generates scene-based visual descriptions; for users who prefer auditory memory, it generates homophonic associations; and for users who prefer logical memory, it generates semantic deduction chains. For example, when a user with a preference for logical memory learns "hypothesis," the AI ​​agent might generate: "hypothesis and thesis (paper) are related; 'hypo' indicates the underlying foundation, so hypothesis is the basic assumption that supports the paper."

[0055] The contextual examples dimension extracts typical sentences containing the target vocabulary from the corpus of past postgraduate entrance examination questions and provides Chinese translations and annotations of key vocabulary. Preferably, the system provides 2 to 3 contextual examples of increasing difficulty for each vocabulary word, enabling users to understand how the vocabulary is used in real-world contexts.

[0056] The personalized memory strategy generation module 3 simultaneously determines a personalized review interval sequence. The system employs an adaptive interval repetition algorithm to calculate the optimal review time point based on the user's cognitive forgetting probability distribution. The core idea of ​​the algorithm is to schedule review when the forgetting probability is about to reach a critical threshold, thus avoiding wasting time by reviewing too early and preventing complete memory loss by reviewing too late.

[0057] The adaptive interval calculation employs the following strategy: For a specific word, the system calculates the time interval corresponding to when the forgetting probability reaches a preset threshold using the forgetting probability formula. Preferably, the forgetting probability threshold is set to 0.5, meaning a review reminder is triggered when the user's forgetting probability for that word reaches 50%. As the user's mastery of the word increases, the mastery strength coefficient increases, and the corresponding review interval also extends. The system generates an interval sequence containing 5 to 7 review time points for each word to be learned, covering the complete cycle from initial learning to stable mastery.

[0058] In a preferred embodiment, for users who forget quickly, the system generates a shorter review interval sequence, typically 1 hour, 4 hours, 1 day, 3 days, 7 days, 14 days, and 30 days; for users with high-efficiency memory, the system generates a longer review interval sequence, typically 4 hours, 1 day, 3 days, 7 days, 21 days, 45 days, and 90 days. This personalized interval arrangement, compared to the traditional fixed interval mode, can more effectively utilize the user's study time and improve memory efficiency.

[0059] The mnemonic scheme and review time interval sequence output by the personalized memory strategy generation module 3 are transmitted to the user interface for display, and the relevant data are recorded into the system database to provide a data foundation for subsequent closed-loop feedback optimization.

[0060] Reference Figure 5 The closed-loop feedback optimization module 4 is connected to the personalized memory strategy generation module 3, the cognitive state adaptive evaluation module 2, and the multi-dimensional semantic knowledge graph construction module 1. It is used to collect data on the user's learning effect on the mnemonic scheme, calculate the memory efficiency gain value, and generate graph update parameters and evaluation parameter update values ​​based on the memory efficiency gain value, so as to realize the iterative optimization of system parameters.

[0061] In embodiments of the present invention, learning effectiveness data includes the user's immediate recall accuracy, delayed recall accuracy, learning duration, subjective satisfaction rating, and number of repeated learning sessions after using a specific mnemonic scheme. Immediate recall accuracy is obtained within 5 minutes of the mnemonic scheme being displayed, and delayed recall accuracy is obtained after 24 hours. Subjective satisfaction rating is based on a 5-point scale, with users evaluating the usefulness of the mnemonic scheme.

[0062] The closed-loop feedback optimization module 4 calculates the memory efficiency gain value to quantitatively evaluate the system's actual performance. The memory efficiency gain value comprehensively considers three aspects: improved memory retention rate, reduced learning time, and user satisfaction. Specifically, the memory efficiency gain value is calculated using the following innovative formula:

[0063] ,

[0064] in: This is the memory efficiency gain value; The memory retention rate after using this system; The baseline memory retention rate before using this system; This represents the average learning time for traditional methods. This refers to the actual learning time spent using this system. Rate user satisfaction; The maximum score for satisfaction is 5 points. , , These are the weighting coefficients for the three dimensions.

[0065] In a preferred embodiment, Set to 0.5. Set to 0.3, Setting it to 0.2 emphasizes the core importance of improving memory performance while also considering efficiency and user experience.

[0066] Based on the analysis of memory efficiency gain values, closed-loop feedback optimization module 4 generates two types of optimization parameters: map update parameters and evaluation parameter update values.

[0067] The graph update parameters are used to adjust the weight coefficients and difficulty baseline values ​​in the multi-dimensional semantic knowledge graph construction module 1. The system analyzes the contribution of different types of semantic associations to memory performance. If mnemonic schemes based on root and affix associations consistently achieve high memory efficiency gains, the weight coefficients are appropriately increased. The value of ; if the mnemonic scheme based on contextual co-occurrence is ineffective, the weight coefficient should be appropriately reduced. The system adjusts the vocabulary difficulty benchmark based on actual learning performance. If the actual mastery difficulty of a word is consistently higher or lower than the preset benchmark, its difficulty parameter will be adjusted accordingly.

[0068] The updated evaluation parameters are used to adjust the calculation methods of the forgetting rate parameter and mastery strength coefficient in the cognitive state adaptive evaluation module 2. The system identifies prediction biases and corrects model parameters by comparing the predicted forgetting probability with actual forgetting performance. If the system consistently overestimates the user's forgetting rate, leading to overly frequent review sessions, the calculation coefficient of the basic forgetting rate parameter is reduced; if the system underestimates the forgetting rate, resulting in insufficient review, the corresponding coefficient is increased.

[0069] The closed-loop feedback optimization module 4 adopts an incremental parameter update strategy to avoid system instability caused by overly drastic parameter adjustments. Preferably, the magnitude of each parameter update is controlled between 5% and 15% of the original value, and upper and lower limits for parameter adjustment are set to ensure that the parameter values ​​are always within a reasonable range.

[0070] The four core modules of this invention form a deeply coupled closed-loop collaborative mechanism. The multi-dimensional semantic knowledge graph construction module 1 provides a semantic difficulty benchmark value as a calibration reference for the cognitive state adaptive evaluation module 2. The cognitive state adaptive evaluation module 2 provides a user cognitive state profile as a decision basis for strategy generation for the personalized memory strategy generation module 3. The output strategy of the personalized memory strategy generation module 3 generates effect data after user practice and feeds it back to the closed-loop feedback optimization module 4. The closed-loop feedback optimization module 4 generates optimization parameters based on the effect data and adjusts the internal parameter configuration of the multi-dimensional semantic knowledge graph construction module 1 and the cognitive state adaptive evaluation module 2 in reverse.

[0071] This closed-loop collaborative architecture achieves three key synergistic effects:

[0072] First, there is a mutually reinforcing effect. Precise modeling of semantic knowledge graphs improves the accuracy of cognitive state assessment, accurate cognitive assessment improves the personalization of strategy generation, high-quality personalized strategies improve learning outcomes, good learning outcome data promotes continuous optimization of system parameters, and the optimized system parameters further improve the accuracy of semantic graphs and cognitive assessment, forming a positive feedback loop.

[0073] Second, there is a synergistic effect. The technical effects of each module are not simply added together, but rather amplified through deep coupling. The combination of accurate semantic association recognition and dynamic tracking of cognitive states allows the mnemonic strategy to adapt to both vocabulary characteristics and user characteristics, resulting in a non-linear increase in memory efficiency.

[0074] Third, the adaptive adjustment effect. The system can automatically adjust its internal parameter configuration based on actual usage results, continuously optimizing performance without manual intervention. When the learning characteristics of the user group change, the system can adaptively adjust to maintain its optimal state.

[0075] In a preferred embodiment of the invention, the system is deployed on a cloud server, and users access the service through a mobile application. The system database stores a semantic knowledge graph of postgraduate entrance examination vocabulary, user cognitive status profiles, historical learning records, and a mnemonic scheme library. The AI ​​agent is built based on large language models such as DeepSeek v3 and possesses natural language understanding and generation capabilities. The user interface features a simple and intuitive design, supporting functions such as vocabulary search, viewing study plans, review reminders, and learning statistics.

[0076] In another embodiment, the multi-dimensional semantic knowledge graph construction module 1 also includes a semantic relationship visualization unit, which can graphically display the semantic association network of words to users, helping them intuitively understand the inherent connections between words. Users can click on any word node to view its associated word groups and association types, deepening their overall understanding of the semantic network of words.

[0077] In another embodiment, the cognitive state adaptive assessment module 2 also includes a learning state monitoring unit, which identifies the user's current level of focus and fatigue by analyzing the user's operational behavior patterns. When the system detects a decline in the user's attention or learning fatigue, it automatically adjusts the learning pace, suggesting a break or switching to a more relaxed review mode to avoid inefficient learning.

[0078] In another embodiment, the personalized memory strategy generation module 3 also includes a multimodal content generation unit, capable of automatically generating accompanying images, audio, or short video materials based on the mnemonic scheme. For users with a visual learning preference, the system generates scene illustrations related to the vocabulary; for users with an auditory learning preference, the system generates audio files of standard pronunciation and mnemonic rhymes. Multimodal content further enriches the presentation of the mnemonic scheme and enhances the learning experience.

[0079] In another embodiment, the closed-loop feedback optimization module 4 also includes a swarm intelligence analysis unit, which can aggregate and analyze the learning performance data of multiple users to identify high-quality mnemonic patterns with universality. When a mnemonic scheme performs well among multiple users, the system marks it as a high-quality scheme and prioritizes recommending it to other users with similar cognitive characteristics, thereby achieving the sharing of swarm intelligence.

[0080] In another embodiment, the system also includes a user-generated content collaboration module, allowing users to upload their own mnemonic schemes. These schemes are then evaluated by an AI agent and incorporated into an expanded mnemonic library, where other users with similar cognitive characteristics can view and use them. Creators of high-quality user-generated content receive points rewards, fostering a collaborative content ecosystem.

[0081] In another embodiment, the semantic association strength calculation of the multi-dimensional semantic knowledge graph construction module 1 introduces a time decay factor, giving lower weight to historical context co-occurrence data and higher weight to recent context data, so that the semantic association network can adapt to the evolution trend of language use patterns.

[0082] In another embodiment, the cognitive state adaptive assessment module 2 incorporates a contextual factor when calculating the probability of cognitive forgetting, taking into account the impact of environmental conditions during learning on memory performance. The system records the distribution of users' learning time periods and the type of learning environment, identifies differences in memory efficiency under different contexts, and optimizes the suggested arrangement of review time accordingly.

[0083] In another embodiment, the personalized memory strategy generation module 3 employs a progressive difficulty adjustment strategy. For new users or users with weak English skills, the system initially recommends a complete mnemonic scheme containing rich associative materials and detailed explanations; as the user's memory ability improves, the system gradually simplifies the mnemonic content, retaining only key prompts to cultivate the user's independent memory ability.

[0084] In another embodiment, the closed-loop feedback optimization module 4 employs a multi-armed slot machine algorithm for strategy optimization decisions. The system treats different parameter configurations as different strategy arms, and by balancing the exploration of new strategies with the utilization of known high-quality strategies, it quickly converges to the optimal parameter configuration with limited user feedback data.

[0085] Through the detailed description of the above embodiments, those skilled in the art can clearly understand the technical solution and implementation of the present invention. The present invention, by constructing a deeply coupled four-module closed-loop collaborative architecture, achieves systematic modeling of lexical semantic relationships, dynamic and accurate assessment of user cognitive states, highly personalized generation of mnemonic strategies, and adaptive iterative optimization of system parameters. This significantly improves the memorization efficiency of postgraduate entrance examination English vocabulary and provides an innovative technical solution for AI-enabled education.

Claims

1. An AI agent-based efficient memorization system for postgraduate entrance examination words, characterized in that, The method comprises the following steps: A multi-dimensional semantic knowledge graph construction module is used to receive a postgraduate examination vocabulary data set, generate a vocabulary semantic association network based on the root and affix structure, semantic similarity and context co-occurrence frequency of the vocabulary, and dynamically adjust the edge weight distribution in the semantic association network according to the graph update parameters output by a closed-loop feedback optimization module; A cognitive state adaptive assessment module is connected to the multi-dimensional semantic knowledge graph construction module, used to obtain learning behavior data of a user, determine the current cognitive forgetting probability distribution and memory capacity evaluation index of the user in combination with the semantic difficulty benchmark value in the vocabulary semantic association network, and adjust the calculation weight of the cognitive forgetting probability according to the evaluation parameter update value output by the closed-loop feedback optimization module; A personalized memory strategy generation module is connected to the cognitive state adaptive assessment module and the multi-dimensional semantic knowledge graph construction module, used to extract the associated word group of a target vocabulary from the vocabulary semantic association network based on the cognitive forgetting probability distribution and the memory capacity evaluation index, generate a multi-dimensional mnemonic scheme that integrates root disassembly, semantic association and context example through an AI agent, and determine a personalized review time interval sequence; A closed-loop feedback optimization module is connected to the personalized memory strategy generation module, the cognitive state adaptive assessment module and the multi-dimensional semantic knowledge graph construction module, used to collect learning effect data of the user on the mnemonic scheme, calculate a memory efficiency gain value, generate the graph update parameters and the evaluation parameter update value based on the memory efficiency gain value, and realize iterative optimization of system parameters.

2. The AI agent-based efficient mnemonics system for postgraduate entrance examination words according to claim 1, characterized in that, The multi-dimensional semantic knowledge graph construction module calculates the comprehensive semantic association strength between vocabularies by fusing the root and affix structure similarity, semantic vector cosine similarity and context co-occurrence mutual information value, and establishes a weighted edge between the corresponding vocabulary nodes when the comprehensive semantic association strength exceeds a preset threshold.

3. The AI agent-based efficient memorization system for postgraduate entrance examination words according to claim 2, characterized in that, The multi-dimensional semantic knowledge graph construction module calculates a semantic difficulty benchmark value for each vocabulary node, which comprehensively considers the length, word frequency ranking, morpheme complexity and semantic abstraction degree of the vocabulary.

4. The AI agent-based efficient memorization system for postgraduate entrance examination words according to claim 1, characterized in that, The cognitive state adaptive assessment module adopts a dynamic forgetting probability prediction algorithm to determine the forgetting probability after a specific time interval by fusing the inherent difficulty of the vocabulary, the user's basic forgetting rate parameter and the user's mastery strength coefficient of the vocabulary.

5. The AI agent-based efficient memorization system for postgraduate entrance examination words according to claim 4, characterized in that, The cognitive state adaptive assessment module divides the user into four cognitive types, i.e., fast forgetting type, stable memory type, fluctuating memory type and efficient memory type, according to the user's learning performance on different difficulty vocabularies, learning efficiency curve slope and long-term memory conversion rate.

6. The AI agent-based efficient mnemonics system for postgraduate entrance examination words according to claim 1, characterized in that, The AI agent in the personalized memory strategy generation module receives the target vocabulary and its associated word group information, the user's cognitive type and memory preference, and the root and affix structure analysis result of the vocabulary as input, generates mnemonic content in the visual association, audio association or logical association mode according to the user's memory preference.

7. The AI agent-based efficient memorization system for postgraduate entrance examination words according to claim 6, characterized in that, The personalized memory strategy generation module adopts an adaptive interval repetition algorithm to determine a review time interval sequence, and triggers a review reminder when the forgetting probability of a user for a specific vocabulary reaches a preset threshold, wherein the review time interval is extended as the user's mastery of the vocabulary improves.

8. The AI agent-based efficient mnemonics system for postgraduate entrance examination words according to claim 1, characterized in that, The closed-loop feedback optimization module calculates the memory efficiency gain value by comprehensively considering the memory retention rate improvement ratio, the learning time saving ratio, and the user satisfaction score.

9. The AI agent-based efficient memorization system for postgraduate entrance examination words according to claim 8, characterized in that, The closed-loop feedback optimization module adjusts the weight coefficients in the multi-dimensional semantic knowledge graph construction module according to the contribution degree of different types of semantic associations to the memory effect, and corrects the forgetting rate parameter in the cognitive state adaptive assessment module according to the deviation between the predicted forgetting probability and the actual forgetting performance.

10. The AI agent-based efficient mnemonics system for postgraduate entrance examination words according to claim 9, characterized in that, The closed-loop feedback optimization module adopts an incremental parameter updating strategy, and controls the amplitude of each parameter update within a preset percentage range of the original value, and sets upper and lower limit boundaries for parameter adjustment.

Images