Science popularization education resource recommendation system based on artificial intelligence
By adopting artificial intelligence technology in the popular science education resource recommendation system, comprehensively collect and analyze the science popularization education resources and user behavior data, combine the diverse recommendation algorithm and Bayesian algorithm, solve the problem of low existing system efficiency and not accurate recommendation. It has realized highly personalized and accurate recommendation of science popularization resources.
Patent Information
- Application Number
- CN202411436678.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-15
- Publication Date
- 2025-01-24
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing science popularization education resource recommendation system is inefficient, difficult to cope with massive resources, low accuracy, and lack effective utilization of user behavior data, resulting in insufficient accurate and personalized recommendation results.
Adopt a recommendation system based on artificial intelligence -based science education resources. Through a variety of advanced data collection technologies and algorithms, comprehensively collect resources and user behavior data, use deep learning and other technologies for multi -modal feature extraction, combined with innovative user portrait construction methods and diverse diversification Recommended algorithms to achieve highly personalized recommendations, and optimize the recommendation results through the Bayesian algorithm.
A precise recommendation of popular science education resources is realized, the unique needs of different users, and improving the accuracy and personalization of user experience and recommendation results.
Smart Images

Figure CN119357470A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of popular science education resource recommendation systems, and more specifically, to a popular science education resource recommendation system based on artificial intelligence. Background Art
[0002] With the rapid development of information technology and the growing demand for popularization of scientific knowledge, the dissemination and promotion of popular science education resources have become increasingly important. Traditional methods of recommending popular science education resources are often based on manual classification or simple keyword matching, which has many limitations.
[0003] On the one hand, the manual classification method is inefficient and difficult to cope with the massive amount of popular science education resources. Popular science education resources cover multiple fields, including natural sciences, humanities and social sciences, technology and engineering, and are in various forms, including text, images, audio, video, etc. Manual classification and organization of these resources is not only time-consuming and laborious, but also prone to errors and omissions; on the other hand, the simple keyword matching recommendation method is not very accurate. It can only perform superficial matching based on the keywords entered by the user, and cannot deeply understand the content of the resources and the real needs of the user; in addition, the existing popular science education resource recommendation system lacks effective use of user behavior data. The user's browsing history, learning time, answering status and other behavioral data contain rich information, which can reflect the user's interest preferences, learning progress and knowledge level. However, traditional recommendation systems often do not fully mine these data, resulting in inaccurate and individuated recommendation results. Therefore, we propose a popular science education resource recommendation system based on artificial intelligence to solve the above problems. Summary of the invention
[0004] 1. Technical issues to be solved In view of the problems existing in the prior art, the purpose of the present invention is to provide a science popularization education resource recommendation system based on artificial intelligence. It first comprehensively uses a variety of advanced data collection technologies and algorithms to comprehensively collect science popularization education resource data and user behavior data, providing a rich data basis for accurate recommendation. At the same time, it uses cutting-edge technologies such as deep learning to extract multimodal features of science popularization education resources, which can more accurately grasp the essential characteristics of resources, and through innovative user portrait construction methods and diversified recommendation algorithm combinations, it can achieve highly personalized science popularization education resource recommendations to meet the unique needs of different users. Its interactive interface module not only provides rich functions, but also realizes continuous optimization of recommendation results through the Bayesian algorithm, thereby improving user experience.
[0005] 2. Technical solution
[0006] To solve the above problems, the present invention adopts the following technical solutions.
[0007] The popular science education resource recommendation system based on artificial intelligence includes a data collection module, a resource analysis module, a user portrait module, a recommendation engine module and an interactive interface module, wherein the data collection module, the resource analysis module, the user portrait module, the recommendation engine module and the interactive interface module are operated in sequence; The data collection module is used to collect various types of popular science education resource data and user behavior data, wherein the data collection module includes a web crawler unit, a database interface unit and a user behavior monitoring unit, wherein the database interface unit is connected to an existing popular science education database through data communication, and the user behavior monitoring unit uses an event monitoring mechanism to record the user's browsing history, search history, learning time and answering behavior data, and then uses the breadth-first search algorithm to crawl text, image, audio and video resource data from the popular science education resource website specified in the web crawler unit through these behavior data; The resource analysis module uses an artificial intelligence algorithm to analyze the collected resource data and extract key feature information of the resource data, wherein the key feature information includes key feature information of images in the resource data extracted using an image feature extraction algorithm based on a convolutional neural network (CNN), key feature information of text content in the resource data extracted using a text content feature extraction algorithm based on a recurrent neural network (RNN), and key feature information of audio and video resources in the resource data extracted using an audio and video processing algorithm in deep learning; The user portrait module constructs a user portrait for the user behavior data according to the behavior data monitored by the user behavior monitoring unit, wherein the user portrait includes the user's interest preference, learning progress and knowledge level, classifies the user behavior data through a clustering algorithm, determines the user's interest field, and evaluates the user's learning progress and knowledge level according to the learning time and answer situation; The recommendation engine module generates a personalized list of popular science education resource recommendations for the user through an artificial intelligence recommendation algorithm based on the key feature information and the user portrait. The artificial intelligence recommendation algorithm includes a user-based collaborative filtering algorithm to find user groups with similar interests to the target user and recommend resources based on the behaviors of these users. It also includes a content-based recommendation algorithm to calculate the similarity between resource features and user portraits for recommendation. It also includes a hybrid recommendation algorithm combined with a deep neural network to automatically learn the complex relationship between resources and user features for recommendation. The interactive interface module is used to display recommended popular science education resources to users and receive user feedback information. The interactive interface module is provided with a preview unit, a collection unit, and an evaluation and sharing unit to provide it with resource preview, collection, evaluation and sharing functions. The user's feedback information updates the parameters of the recommendation model through the Bayesian algorithm to optimize the recommendation results.
[0008] Furthermore, the web crawler unit adopts Python's Scrapy framework, sets specific seed URLs and crawling rules, and uses a breadth-first search algorithm to crawl text, images, audio and video data from designated science popularization and education resource websites. During the crawling process, the web page content is parsed to extract information such as titles, texts, image links, audio and video resource addresses, and store them in a local database or distributed storage system.
[0009] Furthermore, the database interface unit establishes a connection with the existing science education database by using the SQLAlchemy library, uses SQL query statements such as "SELECT * FROM resources WHERE category = 'science'" to obtain resource data of a specific category, and sets a scheduled task to regularly update the resource data in the database to ensure the timeliness of the data.
[0010] Furthermore, the user behavior monitoring unit embeds JavaScript code in the front-end page, and adopts an event monitoring mechanism to record the user's browsing history, search history, learning time, answer status and other behavioral data. When the user browses the page, the page browsing event is triggered, and the page URL and browsing time are recorded. When the user searches, the search keywords and search time are recorded. The learning time is recorded by using a timer when playing popular science education videos or audios. The answer status is recorded after the user submits the answer, and the question number, user answer and correct answer are recorded. These behavioral data are sent to the back-end server in real time through Ajax technology for storage and processing.
[0011] Furthermore, the resource analysis module specifically adopts an image feature extraction algorithm based on a convolutional neural network (CNN) for image resources, uses a pre-trained deep learning model such as VGGNet or ResNet, inputs the image into the model, and extracts the feature vector of the image through the convolution layer and the fully connected layer of the model. These feature vectors can represent the color, texture, shape and other features of the image. For video resources, the video is first decomposed into frame images, and then features are extracted for each frame image. Finally, the feature vectors of all frames are fused to obtain the feature vector of the video; The resource analysis module specifically uses a recurrent neural network (RNN) for text resources, and particularly uses a long short-term memory network (LSTM) to perform semantic analysis and feature extraction on the text content, that is, firstly perform word segmentation on the text, and then input the text sequence after word segmentation into the LSTM model, and the hidden state output by the model is used as the feature vector of the text, which can represent the theme, emotion and other features of the text; The resource analysis module uses an audio processing algorithm in deep learning for audio resources, such as Mel Frequency Cepstral Coefficient (MFCC) to extract audio features. The audio signal is first framed, and then the MFCC coefficient is calculated for each frame of audio. Finally, the MFCC coefficients of all frames are combined to obtain a feature vector of the audio.
[0012] Furthermore, the user portrait module specifically includes classifying user behavior data through a clustering algorithm to determine the user's interest areas, using the K-Means clustering algorithm, taking the user's browsing history, search history and other behavior data as input, calculating the similarity between data points, and clustering similar data points into one category. Each cluster represents an interest area. By analyzing the cluster to which the user belongs, the user's interest preference can be determined; It also includes learning time, which is calculated by accumulating the user's learning time on various popular science education resources, and answering status, which is used to evaluate the user's knowledge level by counting the accuracy of the user's answers. Learning time and accuracy of answering questions are used as two dimensions to construct the user's learning progress and knowledge level indicators.
[0013] Furthermore, the recommendation engine module specifically includes a user-based collaborative filtering algorithm to find user groups with similar interests to the target user, recommend resources based on the behaviors of these users, first calculate the similarity between users, use methods such as cosine similarity or Pearson correlation coefficient, and then predict the target user's interest in the resource based on the ratings or browsing records of similar users on the resource, and recommend resources with high interest to the target user; The recommendation engine module specifically includes a content-based recommendation algorithm, which calculates the similarity between resource features and user portraits for recommendation. For text resources, the TF-IDF algorithm is used to calculate the similarity between resource text and user interest keywords. For image, audio and video resources, the Euclidean distance or cosine similarity method is used to calculate the similarity between resource feature vectors and user portrait feature vectors, and the resources are sorted according to the similarity, and the resources with high similarity are recommended to the user. The recommendation engine module specifically also includes a hybrid recommendation algorithm combined with a deep neural network, which automatically learns the complex relationship between resources and user features for recommendation, constructs a deep neural network model, takes resource feature vectors and user portrait feature vectors as input, and outputs a predicted value of the user's interest in the resource. The model is trained using a large amount of user behavior data and resource data, and the model parameters are optimized through a back-propagation algorithm. The trained model can predict the user's interest in resources based on new resources and user information, and make recommendations.
[0014] Furthermore, the preview unit provides a resource preview function, showing a partial content summary for text resources, showing a thumbnail for image resources, and providing a fragment preview for audio and video resources. Users can view the detailed content of the resource by clicking on the preview link; The collection unit provides a collection function, allowing users to collect resources of interest for easy viewing later; The evaluation and sharing unit includes providing an evaluation function, allowing users to rate and comment on the recommended resources. The rating adopts a five-star system, and the comments are in text form. The user's evaluation information can be used to optimize the recommendation results. It also includes providing a sharing function, allowing users to share the recommended resources on social media platforms or send them to other users. The shared link contains a unique identifier of the resource, which facilitates other users to access the resource through the link; User feedback information is used to update the parameters of the recommendation model through the Bayesian algorithm to optimize the recommendation results. When users evaluate or collect recommended resources, the update process of the Bayesian algorithm is triggered. Based on the user's feedback information, the posterior probability of the resource is calculated and the recommendation weight of the resource is updated. As user feedback information continues to accumulate, the recommendation results will increasingly meet user needs.
[0015] The method for recommending popular science education resources based on artificial intelligence includes the following steps: Step 1: Collect science education resource data and user behavior data through the web crawler unit, database interface unit and user behavior monitoring unit of the data acquisition module; Step 2: Analyze science education resources using various artificial intelligence algorithms of the resource analysis module to extract key feature information; Step 3: Build a user profile based on user behavior data through the clustering algorithm of the user profile module; Step 4: Based on key feature information and user portraits, a personalized recommendation list is generated for the user through a variety of recommendation algorithms of the recommendation engine module; Step 5: Display recommended resources to users through the interactive interface module and receive user feedback, and use the Bayesian algorithm to optimize the recommendation results based on the feedback information.
[0016] 3. Beneficial effects
[0017] Compared with the prior art, the advantages of the present invention are: (1) This solution first optimizes the comprehensiveness and innovation of data collection, and adopts a variety of advanced data collection technologies, including using Python's Scrapy framework for web crawling, setting specific seed URLs and crawling rules, and using breadth-first search algorithms to crawl rich and diverse text, image, audio and video data from many designated science popularization education resource websites. This not only enables the acquisition of a wide range of resources, but also ensures the diversity and comprehensiveness of the data; with the help of the SQLAlchemy library, a connection with the existing science popularization education database is established, and precise SQL query statements are used to obtain resource data of specific categories, and regular updates are performed through scheduled tasks to ensure the timeliness and accuracy of the data; in terms of user behavior data collection, JavaScript code is embedded in the front-end page, and the event monitoring mechanism is used to comprehensively record the user's browsing history, search history, learning time, answer status and other multi-dimensional behavior data, and Ajax technology is used to transmit it to the back-end server in real time, providing a rich data source for subsequent user portrait construction and recommendation. This multi-source data collection method is innovative in the field of science popularization education resource recommendation, breaking through the limitations of traditional single data sources, and can more comprehensively understand user needs and resource characteristics; (2) This solution optimizes the depth and accuracy of resource analysis. For image resources, we use an image feature extraction algorithm based on a convolutional neural network (CNN) and a pre-trained deep learning model such as VGGNet or ResNet. After the image is input into the model, its convolutional layer and fully connected layer deeply mine the image's color, texture, shape and other features to extract representative feature vectors. Compared with traditional image analysis methods, this method can more accurately capture the essential features of the image and provide strong support for accurate recommendations. In text resource analysis, we use a recurrent neural network (RNN), especially a long short-term memory network (LSTM), to perform semantic analysis and feature extraction on the text content. First, the text is segmented, and then the text sequence after segmentation is input into the LSTM model. The hidden state output by the model is used as the feature vector of the text, which can effectively capture the deep semantic information of the text, such as the theme and emotion, and overcome the deficiency of traditional text processing methods that can only process surface keywords. For audio resources, the Mel-frequency cepstral coefficient (MFCC) in deep learning is used to extract the features of the audio. The audio signal is framed and the MFCC coefficient is calculated. Finally, the coefficients of all frames are combined to obtain the feature vector of the audio, realizing effective analysis and feature extraction of audio resources. The comprehensive application of this multimodal resource analysis method is uniquely creative in the popular science education resource recommendation system, and can fully and deeply understand the resource content, laying a solid foundation for personalized recommendations. (3) This solution optimizes the scientificity and personalization of user portrait construction. It classifies user behavior data through clustering algorithms (such as K-Means clustering algorithm) to determine the user's interest areas. The user's browsing history, search history and other behavioral data are used as input to calculate the similarity between data points, and similar data points are clustered into one category. Each cluster represents an interest area, thereby accurately determining the user's interest preferences. This data-driven interest classification method avoids the error of subjective judgment and can more scientifically construct a user interest model; at the same time, the user's learning progress and knowledge level are evaluated based on the learning time and answering situation. The learning time is measured by accumulating the user's learning time on various popular science education resources, and the user's knowledge level is evaluated by counting the correct rate of the user's answers. The learning time and the correct rate of the answers are used as two dimensions to construct the user's learning progress and knowledge level indicators. This user portrait construction method that comprehensively considers multiple factors can more comprehensively and accurately reflect the individual characteristics of users and provide an accurate basis for personalized recommendations. It is different from the traditional single-dimensional user portrait construction method and has significant innovation and superiority; (4) This solution optimizes the diversity and integration of recommendation algorithms and adopts a combination of multiple recommendation algorithms, including user-based collaborative filtering algorithms, content-based recommendation algorithms, and hybrid recommendation algorithms combined with deep neural networks. The user-based collaborative filtering algorithm calculates the similarity between users (such as using cosine similarity or Pearson correlation coefficient) to find user groups with similar interests to the target user, recommends resources based on the behavior of these users, and makes full use of the similarity between users for recommendation; the content-based recommendation algorithm makes recommendations by calculating the similarity between resource features and user portraits (such as using the TF-IDF algorithm to calculate the similarity with user interest keywords for text resources, and using Euclidean distance or cosine similarity to calculate the similarity of feature vectors for image, audio and video resources). It can accurately push relevant resources based on user interest preferences; the hybrid recommendation algorithm combined with deep neural networks constructs a deep neural network model, inputs resource feature vectors and user portrait feature vectors, trains with a large amount of user behavior data and resource data, automatically learns the complex relationship between resources and user features, and outputs the predicted value of the user's interest in resources for recommendation. The integrated application of multiple recommendation algorithms can give full play to their respective advantages, make up for the shortcomings of a single algorithm, improve the accuracy and diversity of recommendations, and provide users with science popularization education resource recommendations that better meet their needs. It has outstanding creativity in the application and innovation of recommendation algorithms. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 It is a schematic diagram of the module composition of the present invention; Figure 2 It is a schematic diagram of the module operation of the present invention. DETAILED DESCRIPTION
[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the specification of the present invention; it is obvious that the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments, and all other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making creative work are within the scope of protection of the present invention. Example
[0020] See also Figure 1-Figure 2 ,The science popularization education resource recommendation system based on artificial intelligence ,when working, the web crawler unit of the data acquisition module uses the Scrapy framework of Python, first sets some science popularization education resource websites as seed URLs, and formulates detailed crawling rules, for example, only web pages of specific ,themes (such as physics, chemistry, biology, etc.) are crawled, and the breadth-first search algorithm starts to work, starting from the seed URL, ,gradually traverses the links in the web page, during the crawling process, ,parses the HTML content of the web page, extracts useful information, such as title, text content, picture links, audio and video resource addresses, etc., and then stores the extracted information in a local database (such as MySQL) or a distributed storage system (such as Hadoop HDFS) for subsequent processing, and then the database interface unit establishes a connection with the existing science popularization ,education database through the SQLAlchemy library, for example, connects to a database containing a large number of science popularization ,articles, course videos and other resources, and regularly executes SQL query statements, such as "SELECT * FROM reso urcesWHEREcategory='biology'", obtain resource data of a specific category (such as biology), set a scheduled task, such as updating the resource data in the database on a daily basis to ensure that the latest popular science education resources are obtained, and then the user behavior monitoring unit embeds JavaScript code in the front-end page of the popular science education resource. When the user visits the page, the event monitoring mechanism in the JavaScript code starts to work. For example, when the user clicks on a popular science article to browse, the page browsing event is triggered, and the URL of the page and the browsing start time and end time are recorded to calculate the browsing time. When the user enters a keyword in the search box to search, the search keyword and search time are recorded. When the user is learning popular science education videos or audios, the timer is used to record the playback time as the learning time. When the user completes the question, the question number, user answer and correct answer are recorded to evaluate the user's knowledge level. These behavior data are sent to the back-end server in real time through Ajax technology for storage and processing; Then, the resource analysis module uses an image feature extraction algorithm based on a convolutional neural network (CNN) for the collected image resources. Taking the pre-trained VGGNet or ResNet model as an example, the image is input into the model. The convolution layer of the model automatically extracts low-level features of the image, such as edges and colors, and the fully connected layer combines these low-level features into high-level feature vectors. These feature vectors can well represent the features of the image, such as the shape and texture of the object in the image. For text resources, word segmentation is first performed to divide the text into words or phrases, and then the text sequence after word segmentation is input into a long short-term memory network (LSTM). The LSTM model learns the semantic representation of the text based on the contextual information of the text. The hidden state output by the model is used as the feature vector of the text. This feature vector can reflect the theme, emotion and other features of the text. For audio resources, the Mel frequency cepstral coefficient (MFCC) is used to extract the features of the audio. First, the audio signal is framed and divided into short time periods. Then, the MFCC coefficients are calculated for each frame of audio. These coefficients reflect the energy distribution of the audio at different frequencies. Finally, the MFCC coefficients of all frames are combined to form the feature vector of the audio. Then the user portrait module uses the K-Means clustering algorithm to classify the user's browsing history, search history and other behavioral data. For example, the user's frequently browsed physics experiment videos, physics popular science articles and other behavioral data are used as input, and the similar data points are clustered into one category by calculating the similarity between data points. It is assumed that they are divided into multiple clusters such as physics, chemistry, biology, etc. The user's interest area is determined according to the cluster to which the user's behavior data belongs. For example, if a user's behavior data is mainly concentrated in the physics cluster, then it can be considered that the user is interested in physics. The learning time is calculated by accumulating the user's learning time on various popular science education resources. For example, if the user spends 2 hours watching physics videos and 1 hour reading physics articles, then the user's learning time on physics is 3 hours. The answering situation is evaluated by counting the user's correct answer rate to evaluate the user's knowledge level. For example, if a user answers 80% of the questions correctly in a set of physics test questions, then it can be considered that the user has a certain level of knowledge in physics. The learning time and the correct answer rate are used as two dimensions to construct the user's learning progress and knowledge level indicators. For example, different weights can be set for the learning time and the correct answer rate, and a comprehensive score can be calculated to evaluate the user's learning progress and knowledge level. The recommendation engine module operates in three modes, namely, the collaborative filtering algorithm based on users. First, the similarity between users is calculated, and methods such as cosine similarity or Pearson correlation coefficient can be used. For example, suppose there are users A and B. By comparing their browsing history, collection behavior, etc., the similarity between them is calculated. After finding a user group (such as user B) with similar interests to the target user (such as user A), the target user's interest in the resource is predicted based on the ratings or browsing records of these similar users. For example, if user B gives a high rating to a physics science video and watches it frequently, and user A is similar to user B, then the system will think that user A may also be interested in this video, and recommend it to user A; Content-based recommendation algorithm, for text resources, uses the TF-IDF algorithm to calculate the similarity between the resource text and the user's interest keywords. For example, if the user is interested in "quantum mechanics", the system will calculate the TF-IDF value of the keyword "quantum mechanics" in the popular science article to measure the similarity between the article and the user's interests. For image, audio and video resources, use methods such as Euclidean distance or cosine similarity to calculate the similarity between the resource feature vector and the user portrait feature vector, sort the resources according to the similarity, and recommend resources with high similarity to the user. For example, if the user is interested in videos about space exploration, the system will calculate the similarity between the feature vectors of each space exploration video and the feature vector of the user portrait, and recommend videos with high similarity to the user; Combined with the hybrid recommendation algorithm of deep neural network, a deep neural network model is constructed, with the input as resource feature vector and user portrait feature vector. For example, resource feature vector can include multimodal features such as image, text, audio, etc., and user portrait feature vector can include features such as interest preference, learning progress and knowledge level. The model is trained with a large amount of user behavior data and resource data. During the training process, the model parameters are optimized through the back propagation algorithm so that the model can learn the complex relationship between resources and user features. When there are new resources and user information, the model can predict the user's interest in the resource and make recommendations based on the prediction results. After the above operations, the interactive interface module can display some content summaries on the interactive interface for text resources, so that users can quickly understand the general content of the article. For image resources, thumbnails are displayed, and users can view the complete image by clicking on the thumbnail. For audio and video resources, clip previews are provided, such as the first 10 seconds of the video or the first 30 seconds of the audio, allowing users to decide whether to continue watching or listening. When users see popular science education resources of interest, they can click the favorite button, and the system will store the collected resource information in the user's personal database. Users can view the collected resources at any time in their favorites list for review and re-learning. Users can rate (such as a five-star system) and comment on recommended resources. For example, users can evaluate the content quality, clarity of explanation, etc. of a popular science video. The system will collect user evaluation information and use it for optimization. The recommendation results are optimized. For example, if many users have a low evaluation of a resource, the system will reduce the recommendation weight of the resource. Users can share the recommended resources on social media platforms (such as WeChat, Weibo, etc.) or send them to other users. The shared link contains the unique identifier of the resource. Other users can access the resource by clicking on the link, thereby expanding the dissemination range of popular science education resources. When users evaluate or collect recommended resources, the system will trigger the update process of the Bayesian algorithm. According to the user's feedback information, the posterior probability of the resource is calculated and the recommendation weight of the resource is updated. For example, if a user gives a high score to a resource and collects it, the recommendation weight of the resource will increase; conversely, if the user has a low evaluation of a resource, its recommendation weight will be reduced. With the continuous accumulation of user feedback information, the recommendation results will increasingly meet the needs of users and improve user satisfaction.
[0021] The above is only a preferred specific implementation of the present invention; however, the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solution and its improved conception within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. An artificial intelligence-based popular science education resource recommendation system, characterized in that: It includes a data collection module, a resource analysis module, a user portrait module, a recommendation engine module and an interactive interface module, wherein the data collection module, the resource analysis module, the user portrait module, the recommendation engine module and the interactive interface module are operated in sequence; The data collection module is used to collect various types of popular science education resource data and user behavior data, wherein the data collection module includes a web crawler unit, a database interface unit and a user behavior monitoring unit, wherein the database interface unit is connected to an existing popular science education database through data communication, and the user behavior monitoring unit uses an event monitoring mechanism to record the user's browsing history, search history, learning time and answering behavior data, and then uses the breadth-first search algorithm to crawl text, image, audio and video resource data from the popular science education resource website specified in the web crawler unit through these behavior data; The resource analysis module uses an artificial intelligence algorithm to analyze the collected resource data and extract key feature information of the resource data, wherein the key feature information includes key feature information of images in the resource data extracted using an image feature extraction algorithm based on a convolutional neural network (CNN), key feature information of text content in the resource data extracted using a text content feature extraction algorithm based on a recurrent neural network (RNN), and key feature information of audio and video resources in the resource data extracted using an audio and video processing algorithm in deep learning; The user portrait module constructs a user portrait for the user behavior data according to the behavior data monitored by the user behavior monitoring unit, wherein the user portrait includes the user's interest preference, learning progress and knowledge level, classifies the user behavior data through a clustering algorithm, determines the user's interest field, and evaluates the user's learning progress and knowledge level according to the learning time and answer situation; The recommendation engine module generates a personalized list of popular science education resource recommendations for the user through an artificial intelligence recommendation algorithm based on the key feature information and the user portrait. The artificial intelligence recommendation algorithm includes a user-based collaborative filtering algorithm to find user groups with similar interests to the target user and recommend resources based on the behaviors of these users. It also includes a content-based recommendation algorithm to calculate the similarity between resource features and user portraits for recommendation. It also includes a hybrid recommendation algorithm combined with a deep neural network to automatically learn the complex relationship between resources and user features for recommendation. The interactive interface module is used to display recommended popular science education resources to users and receive user feedback information. The interactive interface module is provided with a preview unit, a collection unit, and an evaluation and sharing unit to provide it with resource preview, collection, evaluation and sharing functions. The user's feedback information updates the parameters of the recommendation model through the Bayesian algorithm to optimize the recommendation results.
2. The artificial intelligence-based popular science education resource recommendation system according to claim 1, characterized in that: The web crawler unit adopts Python's Scrapy framework, sets specific seed URLs and crawling rules, and uses a breadth-first search algorithm to crawl text, images, audio and video data from designated science popularization education resource websites. During the crawling process, the web page content is parsed, and information such as titles, texts, image links, audio and video resource addresses are extracted and stored in a local database or a distributed storage system.
3. The artificial intelligence-based popular science education resource recommendation system according to claim 1, characterized in that: The database interface unit establishes a connection with the existing popular science education database by using the SQLAlchemy library, uses SQL query statements such as "SELECT * FROM resources WHERE category = 'science'" to obtain resource data of a specific category, and sets a scheduled task to regularly update the resource data in the database to ensure the timeliness of the data.
4. The artificial intelligence-based popular science education resource recommendation system according to claim 1, characterized in that: The user behavior monitoring unit embeds JavaScript code in the front-end page and adopts an event monitoring mechanism to record the user's browsing history, search history, learning time, answer status and other behavioral data. When the user browses the page, the page browsing event is triggered and the page URL and browsing time are recorded. When the user searches, the search keywords and search time are recorded. The learning time is recorded by using a timer when playing popular science education videos or audios. The answer status is recorded after the user submits the answer, and the question number, user answer and correct answer are recorded. These behavioral data are sent to the back-end server in real time through Ajax technology for storage and processing.
5. The artificial intelligence-based popular science education resource recommendation system according to claim 1, characterized in that: The resource analysis module specifically adopts an image feature extraction algorithm based on a convolutional neural network (CNN) for image resources. A pre-trained deep learning model such as VGGNet or ResNet is used to input the image into the model, and the feature vectors of the image are extracted through the convolution layer and the fully connected layer of the model. These feature vectors can represent the color, texture, shape and other features of the image. For video resources, the video is first decomposed into frame images, and then features are extracted for each frame image. Finally, the feature vectors of all frames are fused to obtain the feature vector of the video. The resource analysis module specifically uses a recurrent neural network (RNN) for text resources, and particularly uses a long short-term memory network (LSTM) to perform semantic analysis and feature extraction on the text content, that is, firstly perform word segmentation on the text, and then input the text sequence after word segmentation into the LSTM model, and the hidden state output by the model is used as the feature vector of the text, which can represent the theme, emotion and other features of the text; The resource analysis module uses an audio processing algorithm in deep learning for audio resources, such as Mel Frequency Cepstral Coefficient (MFCC) to extract audio features. The audio signal is first framed, and then the MFCC coefficient is calculated for each frame of audio. Finally, the MFCC coefficients of all frames are combined to obtain a feature vector of the audio.
6. The artificial intelligence-based popular science education resource recommendation system according to claim 1, characterized in that: The user portrait module specifically includes classifying user behavior data through a clustering algorithm to determine the user's interest area, using the K-Means clustering algorithm, taking the user's browsing history, search history and other behavior data as input, calculating the similarity between data points, and clustering similar data points into one category. Each cluster represents an interest area. By analyzing the cluster to which the user belongs, the user's interest preference can be determined; It also includes learning time, which is calculated by accumulating the user's learning time on various popular science education resources, and answering status, which is used to evaluate the user's knowledge level by counting the accuracy of the user's answers. Learning time and accuracy of answering questions are used as two dimensions to construct the user's learning progress and knowledge level indicators.
7. The artificial intelligence-based popular science education resource recommendation system according to claim 1, characterized in that: The recommendation engine module specifically includes a user-based collaborative filtering algorithm to find user groups with similar interests to the target user, recommend resources based on the behavior of these users, first calculate the similarity between users, use methods such as cosine similarity or Pearson correlation coefficient, and then predict the target user's interest in the resource based on the ratings or browsing records of similar users on the resource, and recommend resources with high interest to the target user; The recommendation engine module specifically includes a content-based recommendation algorithm, which calculates the similarity between resource features and user portraits for recommendation. For text resources, the TF-IDF algorithm is used to calculate the similarity between resource text and user interest keywords. For image, audio and video resources, the Euclidean distance or cosine similarity method is used to calculate the similarity between resource feature vectors and user portrait feature vectors, and the resources are sorted according to the similarity, and the resources with high similarity are recommended to the user. The recommendation engine module specifically also includes a hybrid recommendation algorithm combined with a deep neural network, which automatically learns the complex relationship between resources and user features for recommendation, constructs a deep neural network model, takes resource feature vectors and user portrait feature vectors as input, and outputs a predicted value of the user's interest in the resource. The model is trained using a large amount of user behavior data and resource data, and the model parameters are optimized through a back-propagation algorithm. The trained model can predict the user's interest in resources based on new resources and user information, and make recommendations.
8. The artificial intelligence-based popular science education resource recommendation system according to claim 1, characterized in that: The preview unit provides a resource preview function, showing a partial content summary for text resources, showing a thumbnail for image resources, and providing a fragment preview for audio and video resources. Users can view the detailed content of the resource by clicking a preview link; The collection unit provides a collection function, allowing users to collect resources of interest for easy viewing later; The evaluation and sharing unit includes providing an evaluation function, allowing users to rate and comment on the recommended resources. The rating adopts a five-star system, and the comments are in text form. The user's evaluation information can be used to optimize the recommendation results. It also includes providing a sharing function, allowing users to share the recommended resources on social media platforms or send them to other users. The shared link contains a unique identifier of the resource, which facilitates other users to access the resource through the link; User feedback information is used to update the parameters of the recommendation model through the Bayesian algorithm to optimize the recommendation results. When users evaluate or collect recommended resources, the update process of the Bayesian algorithm is triggered. Based on the user's feedback information, the posterior probability of the resource is calculated and the recommendation weight of the resource is updated. As user feedback information continues to accumulate, the recommendation results will increasingly meet user needs.
9. A method for recommending popular science education resources based on artificial intelligence, characterized in that: The popular science education resource recommendation system based on artificial intelligence as claimed in claim 1 comprises the following steps: Step 1: Collect science education resource data and user behavior data through the web crawler unit, database interface unit and user behavior monitoring unit of the data acquisition module; Step 2: Analyze science education resources using various artificial intelligence algorithms of the resource analysis module to extract key feature information; Step 3: Build a user profile based on user behavior data through the clustering algorithm of the user profile module; Step 4: Based on key feature information and user portraits, a personalized recommendation list is generated for the user through a variety of recommendation algorithms of the recommendation engine module; Step 5: Display recommended resources to users through the interactive interface module and receive user feedback, and use the Bayesian algorithm to optimize the recommendation results based on the feedback information.
Citation Information
Cited By
Learning resource recommendation method and device based on machine learning, terminal and medium
CN119917746A
Online education platform data storage management method
CN120216770A
AI-enabled data-driven art educational resource recommendation engine
CN120450134A
Education intelligent recommendation method and system
CN120492739A
Data analysis method for optimizing search engine recommendation content
CN121350351A
