Personalized early education content recommendation system and method based on big data analysis

Abstract

The invention relates to the technical field of artificial intelligence and education, in particular to a personalized early education content recommendation system and method based on big data analysis. The personalized early education content recommendation system comprises a multi-modal data acquisition module, an early education big data processing module, a dynamic user portrait construction module, an intelligent recommendation engine module and an interactive feedback optimization module. The personalized early education content recommendation method comprises the following steps: collecting multi-modal data; preprocessing the big data; constructing a dynamic portrait; personalized recommendation is carried out; and feedback optimization. According to the early education content recommendation system and method, multi-dimensional data can be integrated, the requirements of children can be dynamically met, and the continuous optimization capability is achieved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and education technology, and in particular to a personalized early childhood education content recommendation system and method based on big data analysis. Background Technology

[0002] The period from 0 to 6 years old is a critical time for children's growth and development, and the suitability of early childhood education content has a significant impact on children's cognitive development and interest cultivation. With the rapid development of big data technology, artificial intelligence technology, and wearable devices, the early childhood education field is gradually transforming from traditional standardized content supply to personalized services. Parents' demand for early childhood education content that matches their children's individual characteristics, developmental stages, and family educational needs is becoming increasingly urgent. Against this backdrop, collecting and analyzing multi-source data to uncover children's developmental patterns, constructing accurate user profiles, and achieving intelligent content recommendation have become the core directions for improving the effectiveness of early childhood education.

[0003] Current early childhood education content recommendation technologies generally suffer from a lack of comprehensive data collection dimensions. They often focus solely on children's behavioral interaction data, failing to fully integrate key influencing factors such as physiological characteristics and family environment, resulting in an incomplete assessment of each child's individual state. In data processing, there is a lack of refined processing mechanisms tailored to early childhood education scenarios. The accuracy of outlier removal is insufficient, and data security measures are inadequate, making it difficult to protect the privacy of children and parents. Furthermore, user profiles are often built statically, failing to update in real-time based on data changes during a child's development, leading to a disconnect between profiles and actual needs. Regarding recommendation algorithms, they often rely on single algorithm models, failing to effectively integrate group preferences and individual characteristics, resulting in low accuracy and suitability of recommended content. In addition, most systems lack a complete feedback and optimization loop, unable to adjust recommendation strategies promptly based on children's interaction effects and parental feedback, hindering the continuous improvement of recommendation effectiveness. Summary of the Invention

[0004] This invention provides an early childhood education content recommendation system and method that can integrate multi-dimensional data, dynamically adapt to children's needs, and have continuous optimization capabilities.

[0005] The technical solution adopted in this invention is: a personalized early childhood education content recommendation system based on big data analysis, comprising: a multimodal data acquisition module, an early childhood education big data processing module, a dynamic user profile construction module, an intelligent recommendation engine module, and an interactive feedback optimization module.

[0006] The multimodal data acquisition module is used to collect physiological characteristic data, behavioral interaction data, environmental scene data, and parental education needs data of children aged 0-6. The physiological characteristic data includes heart rate, attention concentration, and sleep quality data collected through wearable devices; the behavioral interaction data includes touch operation trajectory, content dwell time, and answer accuracy data; the environmental scene data includes family living area, lighting conditions, and noise level data; and the parental education needs data includes educational goals, subject preferences, and time allocation data collected through questionnaires and interviews.

[0007] The early childhood education big data processing module is used to clean, desensitize, standardize, and extract features from the collected multimodal data. It uses a density-based outlier detection algorithm to remove abnormal fluctuations in physiological feature data, desensitizes parent identity information using differential privacy technology, converts behavioral data of different dimensions into a unified scale based on the Z-score standardization method, and extracts temporal features from behavioral interaction data using a convolutional neural network.

[0008] The dynamic user profile building module is used to build a three-dimensional user profile of children based on the processed big data, including growth and development dimensions, learning ability dimensions, and interest and preference dimensions. It uses a hierarchical clustering algorithm to divide the children's growth stages, combines a Bayesian network model to analyze the matching relationship between learning ability and content difficulty, mines interest and preference keywords through a word frequency-inverse document frequency algorithm, and updates the user profile every 7 days based on new data.

[0009] The intelligent recommendation engine module is used to realize personalized content recommendation based on dynamic user profiles. It integrates collaborative filtering algorithm, content feature matching algorithm and deep learning recommendation algorithm to build a hybrid recommendation model. The collaborative filtering algorithm uses the user-based K-nearest neighbor algorithm to filter the preferred content of similar children groups. The content feature matching algorithm calculates the matching degree between content tags and profile tags through cosine similarity. The deep learning recommendation algorithm uses a recurrent neural network model based on attention mechanism to predict content adaptation.

[0010] The interactive feedback optimization module is used to collect real-time feedback data from children on recommended content and parent evaluation data, construct a feedback scoring system, and iteratively optimize the weight parameters of the hybrid recommendation model through a gradient descent algorithm. When the feedback score is lower than a preset threshold, an emergency adjustment mechanism is triggered to regenerate recommended content.

[0011] As a further improvement of the present invention, the multimodal data acquisition module also includes a data synchronization unit. The data synchronization unit uses 5G edge computing technology to realize real-time data synchronization between wearable devices, early education terminals and cloud servers, with a synchronization delay of no more than 500ms, and supports offline data caching function, automatically re-transmitting cached data when the network is restored.

[0012] As a further improvement of the present invention, the early childhood education big data processing module also includes a data security unit. The data security unit uses a symmetric encryption algorithm to encrypt the transmitted data and sets data access permissions based on a role-based access control mechanism. Only administrators are authorized to access the de-identified original data, and parents can only view the child's profile analysis report and recommendation records.

[0013] As a further improvement of the present invention, the dynamic user profile construction module also includes a growth early warning unit. The growth early warning unit compares the child's current developmental data with the standard dataset of the same age group. When a certain indicator is lower than 15% of the standard value, it generates an early warning message and pushes it to the parent's terminal, while providing targeted early education content suggestions.

[0014] As a further improvement of the present invention, the intelligent recommendation engine module also includes a content filtering unit. The content filtering unit conducts compliance review of candidate content based on the Ministry of Education's Early Childhood Education Guidelines, removes content that does not conform to the cognitive level and value orientation of the age group, and marks the difficulty level of the approved content, dividing it into four levels: enlightenment level, basic level, intermediate level and extension level.

[0015] A personalized early childhood education content recommendation method based on big data analysis includes the following steps:

[0016] S1, Multimodal Data Acquisition: Collect physiological characteristic data, behavioral interaction data, environmental scene data, and parent education needs data of children aged 0-6 through wearable devices, early education terminals, and parent-side APPs to establish a multi-source data acquisition channel;

[0017] S2, Big Data Preprocessing: Cleaning, desensitizing, standardizing and feature extraction of the collected multimodal data, removing abnormal data and converting it into a unified format, and extracting temporal and semantic features from the data;

[0018] S3, Dynamic Profile Construction: Based on preprocessed big data, construct a three-dimensional user profile of children, divide the growth stages and analyze learning abilities and interest preferences, and regularly update the profile content according to new data.

[0019] S4, Personalized Recommendation: A hybrid recommendation model is built by integrating collaborative filtering, content feature matching and deep learning algorithms. Based on dynamic user profiles, suitable early education content is selected and pushed to early education terminals.

[0020] S5, Feedback Optimization: Collect children's interaction feedback and parents' evaluation data, build a scoring system and iteratively optimize the recommendation model parameters, and regenerate recommended content when the score is too low.

[0021] As a further improvement of the present invention, the physiological characteristic data in step S1 is collected once per minute, behavioral interaction data is collected in real time, environmental scene data is collected once every 24 hours, and parental education needs data is collected once a month to ensure the timeliness and comprehensiveness of the data.

[0022] As a further improvement of the present invention, the dynamic portrait update in step S3 adopts an incremental learning algorithm, which only extracts features from the newly added data and updates the corresponding dimension parameters of the portrait, without having to rebuild the entire portrait, thus improving the update efficiency.

[0023] As a further improvement of the present invention, the initial values ​​of the weight parameters of the hybrid recommendation model in step S4 are determined by a grid search algorithm. An adaptive learning rate is adopted during the iterative optimization process, and the learning rate is adjusted according to the fluctuation of the feedback data to accelerate the convergence speed of the model.

[0024] As a further improvement of the present invention, the feedback scoring system in step S5 includes a weight of 0.4 for the child's interaction time, a weight of 0.3 for the correct answer rate, and a weight of 0.3 for the parent's satisfaction. The matching score of the recommended content is calculated comprehensively, and a re-recommendation mechanism is triggered when the score is lower than 60.

[0025] The beneficial effects of the present invention are as follows: (1) The present invention has high accuracy in data collection and profile construction, which effectively improves the suitability of recommendations. By integrating multi-dimensional data on physiology, behavior, environment and parental needs through a multi-modal data collection module, and combining it with refined data processing technology to ensure data quality, the three-dimensional dynamic profile is updated in real time to update the child's growth status. This solves the problems of single data dimension and static profile in traditional technology, provides reliable data support for accurate recommendations, and makes early education content more in line with the individual characteristics of children.

[0026] (2) The present invention utilizes a hybrid recommendation and feedback closed-loop mechanism to achieve continuous optimization of recommendation performance. The hybrid recommendation model, constructed by integrating three algorithms, takes into account both group preferences and individual characteristics, significantly improving recommendation accuracy compared to a single algorithm. Combined with a weighted feedback scoring system and iterative parameter optimization, a "recommendation-feedback-optimization" closed loop is formed, which can dynamically adjust the strategy based on children's interaction performance, solving the problems of rigid recommendations and lagging optimization in traditional systems.

[0027] (3) This invention is safe, compliant, and has comprehensive additional functions, taking into account both practicality and security. The data security unit protects privacy through encryption and access control, and the content filtering unit conforms to early childhood education standards; the growth warning unit can promptly alert to developmental abnormalities and provide suggestions, while 5G synchronization and offline caching improve ease of use. It not only meets the needs of personalized early childhood education but also safeguards children's growth and enhances the market applicability of the system. Attached Figure Description

[0028] Figure 1This is a system architecture block diagram of a personalized early childhood education content recommendation system and method based on big data analysis according to the present invention.

[0029] Figure 2 This is a flowchart of the workflow of a personalized early childhood education content recommendation system and method based on big data analysis according to the present invention.

[0030] Figure 3 This is a block diagram of the internal structure of the multimodal data acquisition module of a personalized early childhood education content recommendation system and method based on big data analysis, as described in this invention.

[0031] Figure 4 This is a block diagram of the internal structure of the early childhood education big data processing module of the personalized early childhood education content recommendation system and method based on big data analysis of the present invention.

[0032] Figure 5 This is a block diagram of the internal structure of the dynamic user profile construction module of a personalized early childhood education content recommendation system and method based on big data analysis according to the present invention.

[0033] Figure 6 This is a block diagram of the internal structure of the intelligent recommendation engine module of a personalized early childhood education content recommendation system and method based on big data analysis, according to the present invention.

[0034] Figure 7 This is a flowchart of the interactive feedback optimization module of a personalized early childhood education content recommendation system and method based on big data analysis, according to the present invention. Detailed Implementation

[0035] To make the technical problems, technical solutions, and beneficial effects to be solved by this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the embodiments described herein are merely illustrative and are not intended to limit the scope of this application.

[0036] This invention provides a personalized early childhood education content recommendation system based on big data analysis, comprising: a multimodal data acquisition module, an early childhood education big data processing module, a dynamic user profile construction module, an intelligent recommendation engine module, and an interactive feedback optimization module.

[0037] The multimodal data acquisition module is used to collect physiological characteristic data, behavioral interaction data, environmental scene data, and parental education needs data of children aged 0-6. The physiological characteristic data includes heart rate, attention concentration, and sleep quality data collected through wearable devices; the behavioral interaction data includes touch operation trajectory, content dwell time, and answer accuracy data; the environmental scene data includes family living area, lighting conditions, and noise level data; and the parental education needs data includes educational goals, subject preferences, and time allocation data collected through questionnaires and interviews.

[0038] The early childhood education big data processing module is used to clean, desensitize, standardize, and extract features from the collected multimodal data. It uses a density-based outlier detection algorithm to remove abnormal fluctuations in physiological feature data, desensitizes parent identity information using differential privacy technology, converts behavioral data of different dimensions into a unified scale based on the Z-score standardization method, and extracts temporal features from behavioral interaction data using a convolutional neural network.

[0039] The dynamic user profile building module is used to build a three-dimensional user profile of children based on the processed big data, including growth and development dimensions, learning ability dimensions, and interest and preference dimensions. It uses a hierarchical clustering algorithm to divide the children's growth stages, combines a Bayesian network model to analyze the matching relationship between learning ability and content difficulty, mines interest and preference keywords through a word frequency-inverse document frequency algorithm, and updates the user profile every 7 days based on new data.

[0040] The intelligent recommendation engine module is used to realize personalized content recommendation based on dynamic user profiles. It integrates collaborative filtering algorithm, content feature matching algorithm and deep learning recommendation algorithm to build a hybrid recommendation model. The collaborative filtering algorithm uses the user-based K-nearest neighbor algorithm to filter the preferred content of similar children groups. The content feature matching algorithm calculates the matching degree between content tags and profile tags through cosine similarity. The deep learning recommendation algorithm uses a recurrent neural network model based on attention mechanism to predict content adaptation.

[0041] The interactive feedback optimization module is used to collect real-time feedback data from children on recommended content and parent evaluation data, construct a feedback scoring system, and iteratively optimize the weight parameters of the hybrid recommendation model through a gradient descent algorithm. When the feedback score is lower than a preset threshold, an emergency adjustment mechanism is triggered to regenerate recommended content.

[0042] The multimodal data acquisition module described in this invention also includes a data synchronization unit. The data synchronization unit uses 5G edge computing technology to achieve real-time data synchronization between wearable devices, early education terminals and cloud servers, with a synchronization delay of no more than 500ms. It also supports offline data caching, automatically re-uploading cached data when the network is restored.

[0043] The early childhood education big data processing module described in this invention also includes a data security unit. The data security unit uses a symmetric encryption algorithm to encrypt the transmitted data and sets data access permissions based on a role-based access control mechanism. Only administrators are authorized to access the de-identified original data, and parents can only view the child's profile analysis report and recommendation records.

[0044] The dynamic user profile construction module described in this invention also includes a growth early warning unit. The growth early warning unit compares the child's current developmental data with a standard dataset of the same age group. When a certain indicator is lower than 15% of the standard value, it generates an early warning message and pushes it to the parent's terminal, while providing targeted early education content suggestions.

[0045] The intelligent recommendation engine module described in this invention also includes a content filtering unit. The content filtering unit conducts compliance review of candidate content based on the Ministry of Education's Early Childhood Education Guidelines, removes content that does not conform to the cognitive level and value orientation of the age group, and marks the difficulty level of the approved content, dividing it into four levels: enlightenment level, basic level, intermediate level, and extension level.

[0046] A personalized early childhood education content recommendation method based on big data analysis includes the following steps:

[0047] S1, Multimodal Data Acquisition: Collect physiological characteristic data, behavioral interaction data, environmental scene data, and parent education needs data of children aged 0-6 through wearable devices, early education terminals, and parent-side APPs to establish a multi-source data acquisition channel;

[0048] S2, Big Data Preprocessing: Cleaning, desensitizing, standardizing and feature extraction of the collected multimodal data, removing abnormal data and converting it into a unified format, and extracting temporal and semantic features from the data;

[0049] S3, Dynamic Profile Construction: Based on preprocessed big data, construct a three-dimensional user profile of children, divide the growth stages and analyze learning abilities and interest preferences, and regularly update the profile content according to new data.

[0050] S4, Personalized Recommendation: A hybrid recommendation model is built by integrating collaborative filtering, content feature matching and deep learning algorithms. Based on dynamic user profiles, suitable early education content is selected and pushed to early education terminals.

[0051] S5, Feedback Optimization: Collect children's interaction feedback and parents' evaluation data, build a scoring system and iteratively optimize the recommendation model parameters, and regenerate recommended content when the score is too low.

[0052] In step S1 of this invention, the physiological characteristic data is collected once per minute, behavioral interaction data is collected in real time, environmental scene data is collected once every 24 hours, and parental education needs data is collected once a month to ensure the timeliness and comprehensiveness of the data.

[0053] The dynamic portrait update described in step S3 of this invention adopts an incremental learning algorithm, which only extracts features from newly added data and updates the corresponding dimension parameters of the portrait, without having to rebuild the entire portrait, thus improving update efficiency.

[0054] The initial values ​​of the weight parameters of the hybrid recommendation model described in step S4 of this invention are determined by a grid search algorithm. An adaptive learning rate is used during the iterative optimization process, and the learning rate is adjusted according to the fluctuation of the feedback data to accelerate the convergence speed of the model.

[0055] The feedback scoring system described in step S5 of this invention includes a weight of 0.4 for the child's interaction time, a weight of 0.3 for the accuracy of answering questions, and a weight of 0.3 for the parent's satisfaction. The system comprehensively calculates the suitability score of the recommended content, and triggers a re-recommendation mechanism when the score is below 60.

[0056] Example:

[0057] This embodiment uses the personalized early education service for a 3-year-old boy named Xiaoming (98cm tall, 14kg, in the early childhood education stage) as the application scenario to explain the specific implementation process of the present invention in detail. The equipment, algorithm parameters and data involved are all designed based on the actual early education scenario to ensure that the technical solution is reproducible.

[0058] (I) Preparations before implementation

[0059] (1) Equipment Deployment: A children's smart bracelet (sampling rate 1Hz) supporting heart rate monitoring and attention recognition was provided to Xiaoming as a physiological data collection device; a children's early education tablet (system version Android 12) with a touch sensing module was used as a behavioral interaction terminal; smart home sensors with light and noise detection functions were installed in Xiaoming's family living room and bedroom; parents used a smartphone (iOS 16 system) with the accompanying APP of this invention. All devices are connected to the cloud server through 5G edge nodes, which are deployed at the communication base station in Xiaoming's community to ensure that the data synchronization delay is controlled within 300ms.

[0060] (2) Basic data configuration: The cloud server pre-stores the "Standard Dataset for Development of Children Aged 3-4" (including 20 indicators such as height, weight, and attention span) and the early education content library adapted to the Ministry of Education's "Guidelines for Learning and Development of Children Aged 3-6". The content library covers four major areas: language, science, art, and health. The content has been marked with difficulty levels through content screening units (200 items for beginner level, 150 items for basic level, 100 items for intermediate level, and 50 items for extended level). Each item is accompanied by a tag (such as "Animal cognition - basic level - visual interaction").

[0061] (3) Algorithm parameter initialization: The K-nearest neighbor value of the collaborative filtering algorithm in the hybrid recommendation model is set to 5; the cosine similarity threshold of the content feature matching algorithm is set to 0.6; the hidden layer dimension of the recurrent neural network (LSTM+Attention) based on the attention mechanism is set to 128, and the number of attention heads is 4; the initial weights of the grid search algorithm are determined to be 0.3 for collaborative filtering, 0.3 for content matching, and 0.4 for deep learning; the preset threshold of the feedback scoring system is 60 points, and the weights of each indicator are set as follows: interaction time 0.4, answer accuracy 0.3, and parent satisfaction 0.3.

[0062] (II) System Operation and Method Implementation Process

[0063] Step S1: Multimodal Data Acquisition. The smart bracelet collects Xiaoming's heart rate (normal range 80-120 beats / min) and attention span (simple detection via the bracelet's gyroscope and EEG, with a score of 0-100) every minute, uploading them to the cloud in real time. The early education tablet records Xiaoming's touch trajectory (e.g., frequency of clicking "animal cards"), content dwell time (e.g., 45 seconds watching the "panda recognition" video), and quiz accuracy (e.g., answering 8 / 10 questions correctly in the "animal matching" game). Smart home sensors collect home environment data every 24 hours (living room light 300 lux, noise 45 dB, living area 80 square meters). Parents fill out a monthly questionnaire via the app, specifying educational goals as "improving language expression skills," subject preference as "natural science," and daily early education time as "30 minutes." In offline mode (e.g., when out and about without network access), the device caches 2 hours of Xiaoming's interaction data, which is then re-uploaded within 10 seconds of network recovery.

[0064] Step S2: Big Data Preprocessing. The early childhood education big data processing module first uses a density-based outlier detection algorithm (DBSCAN, neighborhood radius ε = 0.8, minimum sample size MinPts = 5) to remove outliers of 150 beats / min caused by running from Xiaoming's heart rate data; it then uses differential privacy technology (privacy budget ε = 1.0) to desensitize information such as parents' names and phone numbers, replacing them with "Parent A + random code"; it uses Z-score standardization to convert data of different dimensions, such as answer accuracy (80%) and dwell time (45 seconds), into standardized values ​​in the range [-1, 1]; and it uses a 3-layer convolutional neural network (Conv1d + pooling layer) to extract the temporal features of the touch trajectory, outputting a 128-dimensional feature vector. The data security unit uses the AES-256 algorithm to encrypt and transmit data, and sets a two-level permission system of "administrator-parent". The administrator can view the desensitized data, while the parent can only view Xiaoming's profile report through the APP.

[0065] Step S3: Dynamic User Profile Construction. The dynamic user profile construction module builds a three-dimensional profile based on preprocessed data: In the growth and development dimension, Xiaoming is classified as a "mid-3-year-old child" using a hierarchical clustering algorithm (3 clusters). Compared with the standard dataset, his height and weight are within the standard range, and his average attention span is 75 points (higher than the standard value of 70 points). In the learning ability dimension, Bayesian network analysis, combined with answer accuracy and attention data, determines that Xiaoming has the highest suitability for "basic cognitive content." In the interest and preference dimension, the keyword weights "animals" and "interactive games" reach 0.8, identifying animal cognition as the core interest. Seven days later, the module uses an incremental learning algorithm to extract features from only the newly added 500 interaction data points, updating the keyword weight of "plant cognition" in the interest and preference dimension to 0.3. This eliminates the need to reconstruct the entire profile, reducing the update time from 120 seconds for full-scale learning to 15 seconds. Meanwhile, the growth warning unit detected that Xiaoming's participation in "language expression practice" was 18% lower than the standard value, and generated a warning message to be pushed to the parents' APP, along with suggestions for "parent-child picture book reading".

[0066] Step S4: Personalized Recommendation. The intelligent recommendation engine module activates a hybrid recommendation model: a collaborative filtering algorithm filters five children similar to Xiaoming in age and interests, finding that they prefer content related to "animal sound recognition" and "farm scene interaction"; a content feature matching algorithm calculates the cosine similarity between candidate content and Xiaoming's profile tags, achieving a matching score of 0.85 for "basic-animal sound matching"; a deep learning algorithm uses an attention mechanism to focus on "panda"-related content where Xiaoming spends a lot of time, predicting a fit score of 0.92 for the "panda growth process" video. After calculating a comprehensive score based on the initial weights, the content filtering unit verifies that there is no value risk and pushes three content items—"animal sound matching" (basic level), "panda growth process" (basic level), and "parent-child picture book reading guide" (early level)—to the early education tablet, with a total duration of 30 minutes to match the parents' needs.

[0067] Step S5: Feedback Optimization. The interaction feedback optimization module collects real-time data: Xiaoming's interaction time with "Animal Sound Matching" was 12 minutes (weight 0.4, 4.8 points), with a 90% accuracy rate (weight 0.3, 2.7 points). His parents scored 80 points through the app (weight 0.3, 2.4 points), resulting in a comprehensive score of 9.9 points (equivalent to 99 points, exceeding the threshold of 60 points). The model uses an adaptive learning rate (initially 0.01, adjusted to 0.005 due to stable feedback) and iteratively optimizes the weights using a gradient descent algorithm, increasing the deep learning algorithm weights to 0.45. The following day, based on the optimized model, a new "Panda Habits Q&A" (basic level) was added, and Xiaoming's comprehensive score reached 10 points (out of 100), significantly improving recommendation accuracy.

[0068] (III) Verification of Implementation Results

[0069] After one month of operation, data showed that Xiaoming's average interaction time with recommended content increased from 20 minutes to 28 minutes, his answer accuracy increased from 65% to 88%, and his parent satisfaction score increased from 70 to 92. The language practice suggestions pushed by the growth warning unit increased Xiaoming's daily spontaneous expression by an average of 5 times, achieving the dual goals of personalized early education and growth monitoring. Data synchronization latency remained stable at 280ms, offline cache retransmission success rate was 100%, and no data leakage incidents occurred, verifying the system's stability, security, and effectiveness.

[0070] As can be seen from the above embodiments, the system and method of the present invention achieve a closed-loop process for personalized early childhood education through multi-module collaborative design. The multi-modal data acquisition module integrates physiological, behavioral, and other multi-dimensional data, and, together with refined preprocessing technology, provides high-quality data support for dynamic profile construction. This is fully demonstrated in the accurate construction and 7-day incremental update of Xiaoming's 3D profile. The hybrid recommendation model combines the advantages of multiple algorithms and achieves dynamic parameter optimization through a feedback loop, enabling the recommendation accuracy to continuously improve with use. The significant improvement in core indicators such as interaction time and answer accuracy confirms the effectiveness of this mechanism.

[0071] In summary, the personalized early childhood education content recommendation system and method based on big data analysis of the present invention systematically solves the core pain points of traditional early childhood education recommendation technology, such as single data dimension, static and fixed profiles, insufficient recommendation accuracy, and lag in optimization, by constructing a full-process technical system of "multimodal data collection - refined preprocessing - dynamic profile construction - hybrid intelligent recommendation - feedback iterative optimization". Its innovations are concentrated in three aspects: First, it breaks through the limitations of single-dimensional data, integrating multi-source data on physiology, behavior, environment, and parental needs. Combined with targeted processing technologies such as DBSCAN anomaly detection and differential privacy desensitization, it lays a high-quality data foundation for accurate analysis. The 100% offline cache retransmission success rate and zero data leakage cases in the examples demonstrate the reliability of the data processing. Second, it pioneers a three-dimensional dynamic portrait mechanism, using algorithms such as hierarchical clustering and Bayesian networks to accurately depict growth stages, learning abilities, and interests. Combined with incremental learning technology, it reduces the portrait update time from 120 seconds to 15 seconds, solving the problem of traditional static portraits being out of sync with children's growth rhythm. Third, it constructs an intelligent engine of "hybrid recommendation + feedback loop," integrating the advantages of collaborative filtering, content matching, and deep learning algorithms. Combined with a weighted scoring system and adaptive learning rate optimization, the recommendation accuracy continuously improves with usage. As shown in the above examples, within one month, children's interaction time increased by 40% and the correct answer rate increased by 35%, fully verifying the effectiveness of the mechanism.

[0072] Furthermore, this invention enhances the system's practicality and security through additional functions such as low-latency data synchronization, role-based access control, and growth-based early warning, achieved via 5G edge computing. It not only adapts to the personalized needs of children aged 0-6 at different developmental stages but also complies with the Ministry of Education's Early Childhood Education Guidelines, perfectly aligning with the current trend of the early childhood education industry towards "personalization, intelligence, and safety." Its technical solution has been validated in real-world scenarios, demonstrating excellent performance in recommendation adaptability, system stability, and data security. It can be widely applied to scenarios such as home early childhood education terminals, intelligent upgrades of early childhood education institutions, and maternal and infant service platforms, providing a reproducible and scalable technical paradigm for the digital transformation of the early childhood education industry, possessing extremely high academic value and market potential.

[0073] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A personalized early childhood education content recommendation system based on big data analysis, characterized in that, include: The system includes a multimodal data acquisition module, an early childhood education big data processing module, a dynamic user profile building module, an intelligent recommendation engine module, and an interactive feedback optimization module. The multimodal data acquisition module is used to collect physiological characteristic data, behavioral interaction data, environmental scene data, and parental education needs data of children aged 0-6. The physiological characteristic data includes heart rate, attention concentration, and sleep quality data collected through wearable devices; the behavioral interaction data includes touch operation trajectory, content dwell time, and answer accuracy data; the environmental scene data includes family living area, lighting conditions, and noise level data; and the parental education needs data includes educational goals, subject preferences, and time allocation data collected through questionnaires and interviews. The early childhood education big data processing module is used to clean, desensitize, standardize, and extract features from the collected multimodal data. It uses a density-based outlier detection algorithm to remove abnormal fluctuations in physiological feature data, desensitizes parent identity information using differential privacy technology, converts behavioral data of different dimensions into a unified scale based on the Z-score standardization method, and extracts temporal features from behavioral interaction data using a convolutional neural network. The dynamic user profile building module is used to build a three-dimensional user profile of children based on the processed big data, including growth and development dimensions, learning ability dimensions, and interest and preference dimensions. It uses a hierarchical clustering algorithm to divide the children's growth stages, combines a Bayesian network model to analyze the matching relationship between learning ability and content difficulty, mines interest and preference keywords through a word frequency-inverse document frequency algorithm, and updates the user profile every 7 days based on new data. The intelligent recommendation engine module is used to realize personalized content recommendation based on dynamic user profiles. It integrates collaborative filtering algorithm, content feature matching algorithm and deep learning recommendation algorithm to build a hybrid recommendation model. The collaborative filtering algorithm uses the user-based K-nearest neighbor algorithm to filter the preferred content of similar children groups. The content feature matching algorithm calculates the matching degree between content tags and profile tags through cosine similarity. The deep learning recommendation algorithm uses a recurrent neural network model based on attention mechanism to predict content adaptation. The interactive feedback optimization module is used to collect real-time feedback data from children on recommended content and parent evaluation data, construct a feedback scoring system, and iteratively optimize the weight parameters of the hybrid recommendation model through a gradient descent algorithm. When the feedback score is lower than a preset threshold, an emergency adjustment mechanism is triggered to regenerate recommended content.

2. The personalized early childhood education content recommendation system based on big data analysis according to claim 1, characterized in that, The multimodal data acquisition module also includes a data synchronization unit. The data synchronization unit uses 5G edge computing technology to realize real-time data synchronization between wearable devices, early education terminals and cloud servers. The synchronization delay is no more than 500ms, and it supports offline data caching function. When the network is restored, the cached data is automatically re-transmitted.

3. The personalized early childhood education content recommendation system based on big data analysis according to claim 1, characterized in that, The early childhood education big data processing module also includes a data security unit. The data security unit uses a symmetric encryption algorithm to encrypt the transmitted data and sets data access permissions based on a role-based access control mechanism. Only administrators are authorized to access the de-identified original data, and parents can only view the child's profile analysis report and recommendation records.

4. The personalized early childhood education content recommendation system based on big data analysis according to claim 1, characterized in that, The dynamic user profile building module also includes a growth early warning unit. The growth early warning unit compares the child's current developmental data with a standard dataset of the same age group. When a certain indicator is lower than 15% of the standard value, it generates an early warning message and pushes it to the parent's terminal, while providing targeted early education content suggestions.

5. The personalized early childhood education content recommendation system based on big data analysis according to claim 1, characterized in that, The intelligent recommendation engine module also includes a content filtering unit. The content filtering unit conducts compliance review of candidate content based on the Ministry of Education's Early Childhood Education Guidelines, removes content that does not conform to the cognitive level and value orientation of the age group, and marks the difficulty level of the approved content, dividing it into four levels: enlightenment level, basic level, intermediate level, and extension level.

6. A personalized early childhood education content recommendation method based on big data analysis, characterized in that, Includes the following steps: S1, Multimodal Data Acquisition: Collect physiological characteristic data, behavioral interaction data, environmental scene data, and parent education needs data of children aged 0-6 through wearable devices, early education terminals, and parent-side APPs to establish a multi-source data acquisition channel; S2, Big Data Preprocessing: Cleaning, desensitizing, standardizing and feature extraction of the collected multimodal data, removing abnormal data and converting it into a unified format, and extracting temporal and semantic features from the data; S3, Dynamic Profile Construction: Based on preprocessed big data, construct a three-dimensional user profile of children, divide the growth stages and analyze learning abilities and interest preferences, and regularly update the profile content according to new data. S4, Personalized Recommendation: A hybrid recommendation model is built by integrating collaborative filtering, content feature matching and deep learning algorithms. Based on dynamic user profiles, suitable early education content is selected and pushed to early education terminals. S5, Feedback Optimization: Collect children's interaction feedback and parents' evaluation data, build a scoring system and iteratively optimize the recommendation model parameters, and regenerate recommended content when the score is too low.

7. The personalized early childhood education content recommendation method based on big data analysis according to claim 6, characterized in that, The physiological characteristic data in step S1 is collected once per minute, behavioral interaction data is collected in real time, environmental scene data is collected once every 24 hours, and parental education needs data is collected once a month to ensure the timeliness and comprehensiveness of the data.

8. The personalized early childhood education content recommendation method based on big data analysis according to claim 6, characterized in that, The dynamic profile update described in step S3 uses an incremental learning algorithm, which only extracts features from new data and updates the corresponding dimension parameters of the profile, without having to rebuild the entire profile, thus improving update efficiency.

9. The personalized early childhood education content recommendation method based on big data analysis according to claim 6, characterized in that, The initial values ​​of the weight parameters of the hybrid recommendation model described in step S4 are determined by a grid search algorithm. An adaptive learning rate is used during the iterative optimization process, and the learning rate is adjusted according to the fluctuation of the feedback data to accelerate the convergence speed of the model.

10. The personalized early childhood education content recommendation method based on big data analysis according to claim 6, characterized in that, The feedback scoring system described in step S5 includes a weight of 0.4 for the child's interaction time, a weight of 0.3 for the accuracy of the answers, and a weight of 0.3 for the parent's satisfaction. The matching score of the recommended content is calculated comprehensively, and a re-recommendation mechanism is triggered when the score is below 60.

Images