An AI-based method and system for recommending educational resources

Abstract

The present invention relates to the field of educational resource recommendation technology, specifically to an educational resource recommendation method and system based on artificial intelligence; first, an educational resource engine is constructed, and then a preliminary user portrait is generated by collecting basic user information, and the user portrait is dynamically updated in combination with the user's personalized needs, and then data is interacted with the educational resource engine based on a data reliability transmission method, educational resource data is called, and collaborative filtering and content-based recommendation methods are used to weightedly fuse the recommended content, provide personalized learning resources for users, and generate learning path planning for users. During the user's learning process, the system monitors the user's learning behavior and optimizes the recommendation model in combination with user feedback. The present invention achieves accurate, efficient, adaptable and reliable educational resource recommendation effects through personalized recommendation, intelligent learning path planning, resource matching optimization, data reliability transmission and adaptive optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of educational resource recommendation, and in particular to an educational resource recommendation method and system based on artificial intelligence. Background Art

[0002] With the rapid development of information technology and artificial intelligence, the demand for digitalization and intelligence of educational resources is growing. Traditional resource recommendation methods can no longer meet the needs of personalized learning. At present, learners are faced with problems such as a wide variety of educational resources, uneven quality, and difficulty in choosing. Especially in the fields of self-study and online education, students often find it difficult to efficiently locate resources that suit their learning needs. On the other hand, artificial intelligence-based recommendation systems have been widely used in e-commerce, social media and other fields, but their application in the field of education is still limited, especially the lack of intelligent recommendation methods that can combine factors such as users' personalized needs, knowledge level, and learning goals.

[0003] A Chinese invention patent with announcement number CN116628339B discloses an educational resource recommendation method and system based on artificial intelligence, the method comprising: determining resource characteristics of educational resources, reading the resource needs of target users and corresponding demand characteristics, and screening the resource characteristics of educational resources according to the demand characteristics to select the target resource characteristics of the educational resources to be recommended, analyzing the historical learning information of the target users to determine the interest characteristics of the target users, respectively determining the matching degree between the interest characteristics and the characteristics of each target resource, and determining the recommendation value corresponding to each educational resource to be recommended based on the matching degree, sorting the recommendation values ​​corresponding to each educational resource to be recommended, outputting the target educational resource corresponding to the optimal recommendation value, and simultaneously monitoring the target user's learning level of the target educational resource in real time, and updating and recommending the learning items in the target educational resource according to the learning level, thereby improving the reliability, timeliness and convenience of educational resource recommendations.

[0004] Although the existing AI-based educational resource recommendation methods and systems have achieved personalized resource recommendations to a certain extent, there are still many problems: first, some systems are relatively rough in constructing user portraits and are unable to deeply analyze users' real needs and learning paths, resulting in insufficient accuracy of recommended content; second, the recommendation algorithm is relatively simple, ignoring the dynamic changes in users' knowledge levels and learning progress, making it difficult to provide resources that match the learning stage; third, the existing systems have weak labeling of educational resources and correlation with knowledge points, the knowledge graph construction is not perfect, and the hierarchical relationship between resources cannot be effectively displayed, and there is a lack of planning for user learning paths. Summary of the invention

[0005] The purpose of the present invention is to propose an educational resource recommendation method and system based on artificial intelligence in response to the problems existing in the background technology.

[0006] The technical solution of the present invention is an educational resource recommendation method based on artificial intelligence, comprising the following specific implementation steps:

[0007] S1. Collect educational resources, label them, and build an educational resource library;

[0008] S2. The user enters basic information through the registration page of the user-side module. After the user logs in, the user-side module guides the user to fill in detailed personalized requirements, and then transmits the basic information and personalized requirements of the user to the user portrait construction module;

[0009] S3, the user portrait construction module generates a preliminary user portrait based on the user's basic information, and classifies the user through the K-means clustering algorithm in combination with the user's personalized needs, constructs a user portrait of the user's learning style, the knowledge points currently mastered, and the weak links, and transmits the constructed user portrait to the educational resource recommendation module and the learning path planning module;

[0010] S4. The educational resource recommendation module analyzes the user's behavior, interests and knowledge mastery through the constructed user portrait, interacts with the educational resource engine based on a data reliability transmission method, calls the educational resources in the educational resource library, and recommends personalized learning resources to the user based on the educational resource recommendation model, generates a list of educational resources, and transmits the list of educational resources to the learning path planning module;

[0011] S5. The learning path planning module analyzes the user's current learning goals, knowledge level and progress through the constructed user portrait, and generates a learning path plan based on the list of educational resources using the learning path planning model, generates a learning plan, and then transmits the learning plan and the list of educational resources to the user-side module;

[0012] S6. Users submit active feedback through the user-side module. After each resource recommendation and learning plan generation, the feedback and update module actively collects user feedback on the recommended resources, and uses the deep Q learning network to update the educational resource recommendation model and learning path planning model based on the user's positive or negative feedback;

[0013] S7. The educational resource recommendation module and the learning path planning module optimize the educational resource recommendation and learning path planning in real time based on user feedback.

[0014] Preferably, the construction process of the education resource library is as follows:

[0015] S21. Collect educational resources;

[0016] S22. After resource collection, clean the data and format the resources;

[0017] S23, structuring the unstructured data and extracting key information;

[0018] S24, call the BERT model to perform semantic analysis on each text resource and automatically generate keywords and topic tags;

[0019] S25. Build a knowledge graph, mine the knowledge points of each subject through the education syllabus and textbook content, then analyze the correlation between the knowledge points, and then associate the labeled resources with the knowledge points in the knowledge graph.

[0020] Preferably, the recommendation process of personalized learning resources is as follows:

[0021] S31. Calculate the cosine similarity between users, find other users with similar interests to the current user, and then calculate the predicted scores of similar users for unlearned resources :

[0022] ;

[0023] ;

[0024] In the formula, r ik and r jk Represents user u i and u j The rating of resource k, R represents the set of resources rated by the user, N(u i ) indicates that the i Similar user set, sim(u i ,u j ) represents user u i and user u j The similarity between

[0025] S32: Based on the resources that the user has learned, use cosine similarity to calculate the similarity between resources, and based on the resources that the user has rated, calculate the predicted rating of similar resources that have not been learned. :

[0026] ;

[0027] ;

[0028] In the formula, r ui and r uj Respectively represent the ratings of user u on resources i and j, U represents the set of all users, N(u) represents the set of resources rated by user u, and sim(j, i) represents the similarity between resources j and i;

[0029] S33, by extracting the data resource tags in the education resource engine, matching the data resource tags with the user portraits, calculating the similarity, and calculating the predicted score for the unlearned resources :

[0030] ;

[0031] ;

[0032] ;

[0033] In the formula, u represents the user's interest vector, d i Represents the feature vector of resource i, ||u||, ||d i || respectively represent the user's interest vector u and the resource's feature vector d i The Euclidean norm of , sim(u,d i ) represents the user's interest vector u and the resource's feature vector d i The similarity function of represents the historical average rating of user u, α represents a balance parameter that controls the weight between similarity and the historical average rating of user u, 0≤α≤1, r ui represents the rating of user u on resource i, R u represents the set of resources that user u has learned, s i Represents the content vector of resource i that user u has learned;

[0034] S34. Use a weighted formula to combine the above prediction scores , and , and score based on the combined predictions , sort the unlearned educational resources, and the weighted formula is as follows:

[0035] ;

[0036] In the formula, w1, w2, and w3 represent the prediction scores respectively. , , The weight of

[0037] S35. Select several resources with the highest scores as personalized learning resources recommended for the user.

[0038] Preferably, the generation process of learning path planning is as follows:

[0039] S41. Extract all relevant knowledge points from the recommended educational resources and define a unique identifier P for each knowledge point. i, and collect its basic information, output the knowledge point list {P1, P2, ..., P n};

[0040] S42, determine the relationship between knowledge points, including pre-knowledge, subsequent knowledge, and similar knowledge, and construct a named entity relationship recognition model to automatically identify these relationships;

[0041] S43. Based on the collected knowledge points and the identified relationships, a knowledge graph structure is established, and a directed graph G(V, E)=({P1, P2, ..., P n}, {R ij |P i , P j ∈V}), that is, we get the triple <knowledge point P i , the relationship R ij , Knowledge Point P j >

[0042] Among them, the node set V represents all knowledge points, and the edge set E represents the relationship between knowledge points, P1, P2, ..., P n Represents several knowledge points, R ij Represents knowledge point P i With knowledge point P j the relationship between;

[0043] S44, for any knowledge point P i Add additional attribute Att i : Difficulty D i , Study time T i 、Recommended resource link i , and select the graph database Neo4j to store the constructed knowledge graph, and store the data of knowledge points and relationships in a structured manner;

[0044] S45. Based on the knowledge graph, analyze all the knowledge points involved in the recommended resources and their interrelationships to form a complete knowledge structure, understand the learning sequence and dependencies, and then use the Dijkstra algorithm to calculate the optimal path from the user's current knowledge level to the target knowledge point. The optimization goal is:

[0045] ;

[0046] In the formula, C i Indicates the learning cost of each knowledge point, and Cost indicates the total cost of the learning path;

[0047] S46. Obtain the generated learning path planning.

[0048] Preferably, the network structure of the named entity relationship recognition model is BERT+BiLSTM+Self attention+CRF.

[0049] Preferably, the process of constructing a user portrait is as follows:

[0050] S61, the user portrait construction module performs preliminary cluster analysis based on the user's basic information, uses the K-means clustering algorithm, and clusters the user's learning goals, interest preferences, and learning level data to assign the user to different learning groups to generate a first-level user portrait;

[0051] S62. Based on the personalized needs of users, the user portrait construction module extracts the feature vector of each user through the feature extraction algorithm, and uses the XGBoost machine learning model to analyze the feature vector to generate a secondary user portrait;

[0052] S63. Through the real-time data processing framework Apache Flink, the user portrait construction module captures and processes the user's behavior data in real time. After each user interacts with the system, the user portrait construction module immediately triggers the portrait update process, updates the user portrait, and obtains the user portrait.

[0053] Preferably, the specific implementation steps of a data reliability transmission method are as follows:

[0054] S71. The educational resource recommendation module performs the following initialization settings:

[0055] S711, randomly select large prime numbers p and q;

[0056] Among them, p≠q, q>3, p>3;

[0057] S712, calculate the primary data encoding parameter D_1_E=p×q×(p-1)×(q-1);

[0058] S713, calculating the primary data analysis parameter D_1_P=p×q;

[0059] S714. Randomly select an integer ε and calculate the encoding factor F_E:

[0060] ;

[0061] Among them, p>2 ε , q>2 ε ;

[0062] S715, select the secondary data encoding parameter D_2_E, and calculate the secondary data parsing parameter D_2_P, so that it satisfies (D_2_E)×(D_2_P)=1 mod (F_E);

[0063] Among them, 1<(D_2_E)<(F_E), gcd(D_2_E, F_E)=1, gcd() represents the greatest common divisor function;

[0064] S716. Publicize the data encoding parameter group {primary data encoding parameter D_1_E, secondary data encoding parameter D_2_E} to the education resource engine;

[0065] S72. The educational resource engine encodes the educational resource data data to be transmitted to obtain the encoded educational resource data Cdata:

[0066] ;

[0067] S73, the educational resource recommendation module calls the encoded educational resource data Cdata in the educational resource engine, and parses the encoded educational resource data Cdata to obtain educational resource data:

[0068] .

[0069] The technical solution of the present invention is: an educational resource recommendation system based on artificial intelligence, which is used to execute the above-mentioned educational resource recommendation method based on artificial intelligence, comprising:

[0070] The user-side module is used to provide user registration, login, information filling, personalized needs and feedback functions, and collect user learning behavior, knowledge level and interest data;

[0071] User portrait building module, used to analyze user input and behavior data and build user portraits;

[0072] The educational resource recommendation module is used to build a recommendation model, integrate content-based recommendation and collaborative filtering recommendation algorithms, and generate personalized educational resource recommendations;

[0073] Educational resource engine, used to collect educational resources and build an educational resource library;

[0074] The learning path planning module is used to build a learning path planning model and plan a learning path for users;

[0075] The feedback and update module is used to collect user learning feedback and update the recommendation model and learning path planning model.

[0076] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0077] The present invention designs an educational resource recommendation method and system based on artificial intelligence. First, an educational resource engine covering a variety of learning resources is constructed, and the resources are labeled. Different knowledge points and resources are associated through knowledge graphs. Then, basic information of users is collected during user registration and login to generate a preliminary user portrait, and the user portrait is dynamically updated in combination with user personalized needs. The educational resource recommendation module interacts with the educational resource engine based on a designed data reliability transmission method and calls educational resource data. After that, the educational resource recommendation module adopts collaborative filtering and content-based recommendation methods, and weightedly integrates the recommended content to provide users with personalized learning resources. The learning path planning module generates a learning path plan for users to ensure that the recommended resources are related to the user's learning path. During the user's learning process, the system monitors the user's learning behavior, optimizes the recommendation model in combination with user feedback, and improves the learning experience. The present invention achieves accurate, efficient, adaptable and reliable educational resource recommendation effects through personalized recommendation, intelligent learning path planning, resource matching optimization, data reliability transmission and adaptive optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 This is an architecture diagram of an artificial intelligence-based educational resource recommendation system proposed by the present invention;

[0079] Figure 2 Network structure diagram for the named entity relationship recognition model. DETAILED DESCRIPTION

[0080] Embodiment 1, as Figure 1 As shown, the present invention proposes an artificial intelligence-based educational resource recommendation system, which includes: a user-end module, a user portrait construction module, a feedback and update module, an educational resource recommendation module, a learning path planning module and an educational resource engine.

[0081] The user-side module provides user registration, login, information filling, personalized needs and feedback functions, and collects user learning behavior, knowledge level and interest data;

[0082] The user portrait building module analyzes user input and behavior data to build user portraits;

[0083] The educational resource recommendation module builds an educational resource recommendation model, integrates content-based recommendation and collaborative filtering recommendation algorithms, generates personalized educational resource recommendations, and ensures the diversity and accuracy of resource recommendations;

[0084] The educational resource engine is used to collect educational resources, including but not limited to texts, videos, online courses, exercises, resource classification including but not limited to subject, grade, difficulty, and build an educational resource library;

[0085] The learning path planning module builds a learning path planning model to plan learning paths for users;

[0086] The feedback and update module collects users' learning feedback, including but not limited to resource usage frequency, completion degree, learning effect evaluation, and updates the recommendation model and learning path planning model.

[0087] In the second embodiment, an educational resource recommendation method based on artificial intelligence proposed in the present invention is applied to an educational resource recommendation system based on artificial intelligence proposed in the first embodiment, and the specific implementation steps are as follows:

[0088] Educational resource library construction phase:

[0089] S101. Collect educational resources through various channels, including but not limited to: web crawlers, open education APIs, school databases, online libraries, and cooperating third-party content providers.

[0090] S102. After resource collection, the data is cleaned, including but not limited to removing duplicate data, deleting invalid links, unifying content in different formats, and formatting resources to ensure that resources from different sources can be processed in a standardized manner in the system.

[0091] S103. Structuring unstructured data (including but not limited to articles and video descriptions) to extract key information, including but not limited to resource title, abstract, author, and release time.

[0092] S104, calling the BERT model to perform semantic analysis on each text resource and automatically generate keywords and topic tags;

[0093] For example, for an article about Newton’s laws, extract “mechanics”, “Newton’s laws”, and “physics” as tags, and store these tags in the database;

[0094] For video resources, we use computer vision technology to analyze the key frames of the video, and combine the title and description of the video to generate tags that match the video content and construct feature vectors.

[0095] Among them, the dimensions covered by label design are as follows:

[0096] Subject classification: including but not limited to mathematics, physics, chemistry, and language;

[0097] Knowledge point association: associated with specific knowledge points in the textbook or syllabus;

[0098] Resource types: including but not limited to texts, videos, exercises, handouts;

[0099] Applicable people: primary school, junior high school, high school, university, graduate students;

[0100] Difficulty level: beginner, intermediate, advanced;

[0101] Resource length: Suitable for short-term study or long-term courses.

[0102] S105. Build a knowledge graph: First, mine the knowledge points of each subject through the education syllabus and textbook content, then analyze the correlation between the knowledge points, including but not limited to the prerequisite knowledge and subsequent extended knowledge, and then associate the labeled resources with the knowledge points in the knowledge graph to obtain the education resource library.

[0103] S106. As the system runs, new educational resources will continue to be added to the resource library. The educational resource engine maintains an automatic update mechanism, regularly captures the latest resources, and automatically connects them according to the tag system and knowledge graph.

[0104] Educational resource recommendation stage:

[0105] S201. The user enters basic information, including but not limited to name, age, learning goals, and learning methods, through the registration page of the user-side module to create a personal account. After the user logs in, the user-side module guides the user to fill in detailed personalized needs, including but not limited to: subjects desired to be studied, specific knowledge points, and time arrangements. The user-side module transmits the above basic information and personalized needs to the user portrait construction module.

[0106] S202, the user portrait construction module generates a preliminary user portrait based on the basic information filled in by the user when registering, and classifies the user through the K-means clustering algorithm in combination with the user's personalized needs, constructs a portrait of the user's learning style, currently mastered knowledge points and weak links, and transmits the constructed user portrait to the education resource recommendation module and the learning path planning module;

[0107] The specific method of constructing user portraits is as follows:

[0108] S2021, the user portrait construction module performs preliminary cluster analysis based on the user's basic information, uses the K-means clustering algorithm to cluster the user's learning goals, interest preferences, and learning level data, assigns the user to different learning groups, and generates a first-level user portrait;

[0109] S2022. Based on the personalized needs of users, the user portrait construction module extracts the feature vector of each user through the feature extraction algorithm, and uses the XGBoost machine learning model to analyze the feature vector to generate a secondary user portrait;

[0110] S2023. Through the real-time data processing framework (Apache Flink), the user portrait construction module captures and processes user behavior data in real time. After each user interaction with the system, the user portrait construction module immediately triggers the portrait update process, updates the user portrait, and obtains the final user portrait.

[0111] S203, the educational resource recommendation module analyzes the user's behavior, interests and knowledge mastery, interacts with the educational resource engine, calls educational resources in the educational resource library, and recommends personalized learning resources to the user. The following are the detailed steps for implementing the recommendation algorithm:

[0112] S2031, User-based collaborative filtering: Calculate the cosine similarity between users, find other users with similar interests to the current user, and then calculate the predicted scores of similar users for unlearned resources ;

[0113] The formula for calculating cosine similarity is as follows:

[0114] ;

[0115] In the formula, r ik and r jk Represents user u i and u j The rating of resource k, R represents the set of resources rated by the user;

[0116] The calculation formula for the prediction score is as follows:

[0117] ;

[0118] In the formula, N(u i ) indicates that the i Similar user set, r jk Represents user u j The score of resource k;

[0119] S2032: Based on the resources that the user has learned, use cosine similarity to calculate the similarity between resources, and based on the resources that the user has rated, calculate the predicted rating for the unlearned resources. ;

[0120] For example, if the algebra course that the user has studied has a high similarity with a geometry course, the educational resource recommendation module will recommend the geometry course;

[0121] The similarity calculation formula is as follows:

[0122] ;

[0123] In the formula, r ui and ruj They represent the ratings of user u on resources i and j respectively, and U represents the set of all users;

[0124] The calculation formula for the prediction score is as follows:

[0125] ;

[0126] Where N(u) represents the set of resources rated by user u;

[0127] S2033. Extract the data resource tags in the education resource engine, match the data resource tags with the user portraits, calculate the similarity, and calculate the predicted score for the unlearned resources. ;

[0128] The formula for predicting the score is as follows:

[0129] ;

[0130] ;

[0131] ;

[0132] In the formula, u represents the user's interest vector, d i Characteristic vector representing the resource, ||u||, ||d i || respectively represent the user's interest vector u and the resource's feature vector d i The Euclidean norm of , sim(u,d i ) represents the user's interest vector u and the resource's feature vector d i The similarity function of represents the historical average rating of user u, α represents a balance parameter that controls the weight between similarity and the historical average rating of user u, 0≤α≤1, r ui represents the rating of user u on resource i, R u represents the set of resources that user u has learned, s i Represents the content vector of resource i that user u has learned;

[0133] S2034, using a weighted formula to combine the above prediction scores, and based on the combined prediction scores , sort the unlearned educational resources, and the weighted formula is as follows:

[0134] ;

[0135] In the formula, w1, w2, and w3 represent the prediction scores respectively. , , The weight of

[0136] It should be noted that the weights w1, w2, and w3 can be adjusted according to model performance and actual needs;

[0137] S2035. Select several resources with the highest scores as recommendation results, generate a list of educational resources, and transmit the list of educational resources to the learning path planning module.

[0138] S204, the learning path planning module dynamically generates a learning path suitable for the user based on the user's current learning goals, knowledge level and progress, and educational resource list. The specific implementation process is as follows:

[0139] S2041. Extract all relevant knowledge points from the recommended educational resources and define a unique identifier P for each knowledge point. i , and collect its basic information, including but not limited to: name, description and content, and output a list of knowledge points {P1, P2, ..., P n};

[0140] S2042. Determine the relationship between knowledge points, including pre-knowledge, subsequent knowledge, and similar knowledge, such as Figure 2 As shown, a BERT+BiLSTM+Self attention+CRF named entity relationship recognition model is constructed to automatically identify these relationships;

[0141] Figure 2 In, BERT represents the pre-trained model;

[0142] T0, T1, T2, ..., T n-2 , T n-1 , T n Represents the word vector output by the BERT pre-training model;

[0143] LSTM (Long Short-Term Memory) stands for long short-term memory network;

[0144] BiLSTM (Bi-directional Long Short-Term Memory) represents a bidirectional long short-term memory network;

[0145] Self attention means self-attention;

[0146] CRF (conditional random field) means conditional random field;

[0147] S2043. Based on the collected knowledge points and the identified relationships, a knowledge graph structure is established, and a directed graph G(V, E)=({P1, P2, ..., P n}, {Rij |P i , P j ∈V}), that is, we get the triple <knowledge point P i , the relationship R ij , Knowledge Point P j >

[0148] Among them, the node set V represents all knowledge points, and the edge set E represents the relationship between knowledge points, P1, P2, ..., P n Represents several knowledge points, R ij Represents knowledge point P i With knowledge point P j the relationship between;

[0149] S2044, for any knowledge point P i Add additional attribute Att i : Difficulty D i , Study time T i 、Recommended resource link i , and select the graph database Neo4j to store the constructed knowledge graph, and store the data of knowledge points and relationships in a structured manner:

[0150] Attribute i ={P i :(Difficulty D i , learning time T i , recommended resource link Link i )};

[0151] S2045. Based on the knowledge graph, analyze all the knowledge points involved in the recommended resources and their interrelationships to form a complete knowledge structure, understand the learning sequence and dependencies, and then use the Dijkstra algorithm to calculate the optimal path from the user's current knowledge level to the target knowledge point:

[0152] ;

[0153] The goal is to minimize the cost while satisfying the dependencies of knowledge points and the learning ability of users;

[0154] In the formula, C i Indicates the learning cost of each knowledge point, and Cost indicates the total cost of the learning path;

[0155] S2046. Generate a learning path, combine the user portrait, generate a specific learning plan, including but not limited to learning duration, learning sequence, goals and tests, to ensure that the user has a clear learning direction and phased goals, and transmit the learning plan and educational resource list to the user-side module.

[0156] S205. The user submits active feedback through the user-side module, including but not limited to marking certain resources as irrelevant and rating educational resources and learning plans.

[0157] S206, after each resource recommendation and learning plan generation, the feedback and update module actively collects user feedback on the recommended resources, including but not limited to the ratings, completion degree, and learning test results of the resources and learning plans, and uses deep Q learning (DQN) technology to update the recommendation model and learning path planning model according to the user's positive or negative feedback;

[0158] For example, resource types with high user ratings will have their recommendation weight increased, and resource types with low user ratings will have their recommendation priority reduced;

[0159] It should be noted that after the user completes each learning task, the feedback and update module collects and analyzes the learning data, calculates the user's mastery of each knowledge point by updating the learning record, and i :

[0160] .

[0161] S207, the educational resource recommendation module and the learning path planning module adjust the recommended educational resources and learning paths in real time based on user feedback. If the user performs well in a certain knowledge point, he or she can skip the subsequent simple content and directly enter a more advanced knowledge point. Otherwise, relevant supplementary materials will be provided to consolidate basic knowledge.

[0162] Embodiment 3, the educational resource recommendation method based on artificial intelligence proposed by the present invention, compared with embodiment 2, further includes a data reliability transmission method, which is used in the interaction process between the educational resource recommendation module and the educational resource engine when generating recommended resources. The specific implementation steps of a data reliability transmission method are as follows:

[0163] S301, the educational resource recommendation module performs the following initialization settings:

[0164] (1) Randomly select large prime numbers p and q;

[0165] Among them, p≠q, q>3, p>3;

[0166] (2) Calculate the first-level data encoding parameter D_1_E = p×q×(p-1)×(q-1);

[0167] (3) Calculate the first-level data analysis parameter D_1_P = p × q;

[0168] (4) Randomly select an integer ε and calculate the encoding factor F_E:

[0169] ;

[0170] Among them, p>2 ε , q>2 ε ;

[0171] (5) Select the secondary data encoding parameter D_2_E and calculate the secondary data parsing parameter D_2_P so that it satisfies (D_2_E)×(D_2_P)=1 mod (F_E);

[0172] Among them, 1<(D_2_E)<(F_E), gcd(D_2_E, F_E)=1, gcd() represents the greatest common divisor function;

[0173] (6) Publicize the data encoding parameter group {first-level data encoding parameter D_1_E, second-level data encoding parameter D_2_E} to the education resource engine.

[0174] S302, the educational resource engine encodes the educational resource data data to be transmitted to obtain the encoded educational resource data Cdata:

[0175] .

[0176] S303, the educational resource recommendation module calls the encoded educational resource data Cdata in the educational resource engine, and parses the encoded educational resource data Cdata to obtain educational resource data:

[0177] .

[0178] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. An educational resource recommendation method based on artificial intelligence, characterized in that: The specific implementation steps include the following: S1. Collect educational resources, label them, and build an educational resource library; S2. The user enters basic information through the registration page of the user-side module. After the user logs in, the user-side module guides the user to fill in detailed personalized requirements, and then transmits the basic information and personalized requirements of the user to the user portrait construction module; S3, the user portrait construction module generates a preliminary user portrait based on the user's basic information, and classifies the user through the K-means clustering algorithm in combination with the user's personalized needs, constructs a user portrait of the user's learning style, the knowledge points currently mastered, and the weak links, and transmits the constructed user portrait to the educational resource recommendation module and the learning path planning module; S4. The educational resource recommendation module analyzes the user's behavior, interests and knowledge mastery through the constructed user portrait, interacts with the educational resource engine based on a data reliability transmission method, calls the educational resources in the educational resource library, and recommends personalized learning resources to the user based on the educational resource recommendation model, generates a list of educational resources, and transmits the list of educational resources to the learning path planning module; S5. The learning path planning module analyzes the user's current learning goals, knowledge level and progress through the constructed user portrait, and generates a learning path plan based on the list of educational resources using the learning path planning model, generates a learning plan, and then transmits the learning plan and the list of educational resources to the user-side module; S6. Users submit active feedback through the user-side module. After each resource recommendation and learning plan generation, the feedback and update module actively collects user feedback on the recommended resources, and uses the deep Q learning network to update the educational resource recommendation model and learning path planning model based on the user's positive or negative feedback; S7, the educational resource recommendation module and the learning path planning module optimize the educational resource recommendation and learning path planning in real time according to the user's feedback; The specific implementation steps of a data reliability transmission method are as follows: S71. The educational resource recommendation module performs the following initialization settings: S711, randomly select large prime numbers p and q; Among them, p≠q, q>3, p>3; S712, calculate the primary data encoding parameter D_1_E=p×q×(p-1)×(q-1); S713, calculating the primary data analysis parameter D_1_P=p×q; S714. Randomly select an integer ε and calculate the encoding factor F_E: ； Among them, p>2 ε , q>2 ε ; S715, select the secondary data encoding parameter D_2_E, and calculate the secondary data parsing parameter D_2_P, so that it satisfies (D_2_E)×(D_2_P)=1 mod (F_E); Among them, 1<(D_2_E)<(F_E), gcd(D_2_E, F_E)=1, gcd() represents the greatest common divisor function; S716. Publicize the data encoding parameter group {primary data encoding parameter D_1_E, secondary data encoding parameter D_2_E} to the education resource engine; S72. The educational resource engine encodes the educational resource data data to be transmitted to obtain the encoded educational resource data Cdata: ； S73, the educational resource recommendation module calls the encoded educational resource data Cdata in the educational resource engine, and parses the encoded educational resource data Cdata to obtain educational resource data: 。 2. The method for recommending educational resources based on artificial intelligence according to claim 1, characterized in that: The construction process of the educational resource library is as follows: S21. Collect educational resources; S22. After resource collection, clean the data and format the resources; S23, structuring the unstructured data and extracting key information; S24, call the BERT model to perform semantic analysis on each text resource and automatically generate keywords and topic tags; S25. Build a knowledge graph, mine the knowledge points of each subject through the education syllabus and textbook content, then analyze the correlation between the knowledge points, and then associate the labeled resources with the knowledge points in the knowledge graph to obtain an education resource library.

3. The method for recommending educational resources based on artificial intelligence according to claim 1, characterized in that: The recommendation process for personalized learning resources is as follows: S31. Calculate the cosine similarity between users, find other users with similar interests to the current user, and then calculate the predicted scores of similar users for unlearned resources : ； ； In the formula, r ik and r jk Represents user u i and u j The rating of resource k, R represents the set of resources rated by the user, N(u i ) indicates that the i Similar user set, sim(u i ,u j ) represents user u i and user u j The similarity between S32: Based on the resources that the user has learned, use cosine similarity to calculate the similarity between resources, and based on the resources that the user has rated, calculate the predicted rating of similar resources that have not been learned. : ； ； In the formula, r ui and r uj Respectively represent the ratings of user u on resources i and j, U represents the set of all users, N(u) represents the set of resources rated by user u, and sim(j, i) represents the similarity between resources j and i; S33, by extracting the data resource tags in the education resource engine, matching the data resource tags with the user portraits, calculating the similarity, and calculating the predicted score for the unlearned resources : ； ； ； In the formula, u represents the user's interest vector, d i Characteristic vector representing the resource, ||u||, ||d i || respectively represent the user's interest vector u and the resource's feature vector d i The Euclidean norm of , sim(u,d i ) represents the user's interest vector u and the resource's feature vector d i The similarity function of represents the historical average rating of user u, α represents a balance parameter that controls the weight between similarity and the historical average rating of user u, 0≤α≤1, r ui represents the rating of user u on resource i, R u represents the set of resources that user u has learned, s i Represents the content vector of resource i that user u has learned; S34. Use a weighted formula to combine the above prediction scores , and , and score based on the combined predictions , sort the unlearned educational resources, and the weighted formula is as follows: ； In the formula, w1, w2, and w3 represent the prediction scores respectively. , , The weight of S35. Select several resources with the highest scores as personalized learning resources recommended for the user.

4. The method for recommending educational resources based on artificial intelligence according to claim 1, characterized in that: The generation process of learning path planning is as follows: S41. Extract all relevant knowledge points from the recommended educational resources and define a unique identifier P for each knowledge point. i , and collect its basic information, output the knowledge point list {P1, P2, ..., P n }; S42, determine the relationship between knowledge points, including pre-knowledge, subsequent knowledge, and similar knowledge, and construct a named entity relationship recognition model to automatically identify these relationships; S43. Based on the collected knowledge points and the identified relationships, a knowledge graph structure is established, and a directed graph G(V, E)=({P1, P2, ..., P n }, {R ij |P i , P j ∈V}), that is, we get the triple <knowledge point P i , the relationship R ij , Knowledge Point P j > Among them, the node set V represents all knowledge points, and the edge set E represents the relationship between knowledge points, P1, P2, ..., P n Represents several knowledge points, R ij Represents knowledge point P i With knowledge point P j the relationship between; S44, for any knowledge point P i Add additional attribute Att i : Difficulty D i , Study time T i 、Recommended resource link i , and select the graph database Neo4j to store the constructed knowledge graph, and store the data of knowledge points and relationships in a structured manner; S45. Based on the knowledge graph, analyze all the knowledge points involved in the recommended resources and their interrelationships to form a complete knowledge structure, understand the learning sequence and dependencies, and then use the Dijkstra algorithm to calculate the optimal path from the user's current knowledge level to the target knowledge point. The optimization goal is: ； In the formula, C i Indicates the learning cost of each knowledge point, and Cost indicates the total cost of the learning path; S46. Obtain the generated learning path planning.

5. The method for recommending educational resources based on artificial intelligence according to claim 4, characterized in that: The network structure of the named entity relationship recognition model is BERT+BiLSTM+Self attention+CRF.

6. The method for recommending educational resources based on artificial intelligence according to claim 1, characterized in that: The process of building a user portrait is as follows: S61, the user portrait construction module performs preliminary cluster analysis based on the user's basic information, uses the K-means clustering algorithm, and clusters the user's learning goals, interest preferences, and learning level data to assign the user to different learning groups to generate a first-level user portrait; S62. Based on the personalized needs of users, the user portrait construction module extracts the feature vector of each user through the feature extraction algorithm, and uses the XGBoost machine learning model to analyze the feature vector to generate a secondary user portrait; S63. Through the real-time data processing framework Apache Flink, the user portrait construction module captures and processes the user's behavior data in real time. After each user interacts with the system, the user portrait construction module immediately triggers the portrait update process, updates the user portrait, and obtains the user portrait.

7. An artificial intelligence-based educational resource recommendation system, which is used to execute an artificial intelligence-based educational resource recommendation method according to any one of claims 1 to 6, characterized in that: include: The user-side module is used to provide user registration, login, information filling, personalized needs and feedback functions, and collect user learning behavior, knowledge level and interest data; User portrait building module, used to analyze user input and behavior data and build user portraits; The educational resource recommendation module is used to build a recommendation model, integrate content-based recommendation and collaborative filtering recommendation algorithms, and generate personalized educational resource recommendations; Educational resource engine, used to collect educational resources and build an educational resource library; The learning path planning module is used to build a learning path planning model and plan a learning path for users; The feedback and update module is used to collect user learning feedback and update the recommendation model and learning path planning model.

Images