Method, device and equipment for personalized recommendation of educational resources and medium

Abstract

The invention relates to the technical field of computers, and provides an educational resource personalized recommendation method and device, equipment and a medium. The method comprises the steps of collecting learning historical data, interest preference data and knowledge mastery degree data of students in real time, and constructing a multi-dimensional student portrait through a machine learning algorithm; based on the multi-dimensional student portraits, matched data are retrieved from various kinds of data of an education resource library through a multi-modal retrieval algorithm, and learning content is generated through fusion and provided for students; and collecting feedback scores of the students on the learning content, taking the feedback scores as rewards, adjusting weight parameters of the multi-modal retrieval algorithm on retrieval of various types of data through a reinforcement learning algorithm, returning to the step of retrieving the matched data until the feedback scores meet a threshold condition, and generating personalized learning content based on the final weight parameters. According to the scheme, the disjunction rate of the recommended learning content and the student portrait is reduced, and the adaptability of the recommended learning content and the student individual is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, device, equipment and medium for personalized recommendation of educational resources. Background Art

[0002] With the rapid development of artificial intelligence (AI), natural language processing (NLP) and multimodal data processing technologies are increasingly being applied in education. In existing educational scenarios, while RAG (Retrieval-Augmented Generation)-based learning content recommendation methods can combine multimodal data to perform knowledge retrieval and text generation, they lack the flexibility to dynamically adjust to individual needs. Furthermore, there's a lack of a closed-loop feedback loop between generated learning content and student feedback, making recommended content easily disconnected from student profiles. This reduces the adaptability of recommended learning content to individual students. Summary of the Invention

[0003] In view of this, the present invention provides a method, apparatus, device and medium for personalized recommendation of educational resources.

[0004] According to a first aspect of the present invention, a method for personalized recommendation of educational resources is provided, comprising: Collect students' learning history data, interest preference data, and knowledge mastery data in real time, and build multi-dimensional student portraits through machine learning algorithms; Based on the multi-dimensional student portrait, matching materials are retrieved from various materials in the educational resource library through a multimodal retrieval algorithm, integrated to generate learning content, and provided to the student; Collect students' feedback scores on the learning content, use them as rewards, and adjust the weight parameters of the multimodal retrieval algorithm for retrieving the various types of materials through a reinforcement learning algorithm, and return to the step of retrieving matching materials until the feedback scores meet the threshold conditions, and generate personalized learning content based on the final weight parameters.

[0005] In some embodiments, the step of using it as a reward and adjusting the weight parameters of the multimodal retrieval algorithm for the retrieval of the various types of information through a reinforcement learning algorithm includes: Using the vector representation of the multi-dimensional student portrait as the state space of the Q-learning algorithm; Using a set of adjustment amounts for weight parameters of the multimodal retrieval algorithm as an action space of the Q-learning algorithm; The feedback score is used as a reward value, a Q-table is updated through the Q-learning algorithm, an optimal action in the action space is selected according to the updated Q-table, and a weight parameter of the multimodal retrieval algorithm is adjusted based on the optimal action.

[0006] In some embodiments, the threshold condition includes: N consecutive feedback scores are greater than a preset score threshold.

[0007] In some embodiments, the step of collecting students' learning history data, interest preference data, and knowledge mastery data in real time and constructing a multi-dimensional student profile through a machine learning algorithm includes: Collect students' study time of various materials in the educational resource library in real time and divide them into several learning mode clusters through clustering algorithms; Calculate the proportion of students' study time for each subject in each learning mode cluster and input it into the decision tree model to obtain the preferred subject label set; The labels of the several learning mode clusters, the preferred subject label sets and the separately obtained student knowledge mastery scores are integrated through machine learning algorithms to generate a multi-dimensional student portrait.

[0008] In some embodiments, the step of retrieving matching materials from various materials in the educational resource library using a multimodal retrieval algorithm based on the multi-dimensional student portrait includes: Extracting semantic features of students' learning needs from the multi-dimensional student portrait; Initializing weight parameters of a multimodal retrieval algorithm for retrieving various types of materials in an educational resource library, and calculating the relevance between the semantic features of the learning needs and the various types of materials based on the weight parameters; Based on the correlation, matching information is screened out.

[0009] In some embodiments, the step of calculating the relevance between the semantic features of the learning requirements and the various types of materials based on the weight parameters includes: Based on the weight parameters, the various types of data are encoded into corresponding dense vectors through the DSE algorithm; The cosine similarities between the dense vector and the semantic features of the learning requirements are calculated respectively, and are used as the correlation between the semantic features of the learning requirements and the various types of information.

[0010] In some embodiments, the step of screening out matching information based on the correlation degree includes: All cosine similarities are jointly modeled using the ColPali algorithm to obtain the probability values ​​corresponding to each type of information; If the probability value of any type of data is greater than or equal to the probability threshold, it will be screened as matching data.

[0011] According to a second aspect of the present invention, there is provided a device for personalized recommendation of educational resources, the device comprising: The first module is used to collect students' learning history data, interest preference data, and knowledge mastery data in real time, and build a multi-dimensional student portrait through machine learning algorithms; The second module is used to retrieve matching materials from various materials in the educational resource library based on the multi-dimensional student portrait using a multimodal retrieval algorithm, fuse them to generate learning content, and provide it to students; The third module is used to collect students' feedback scores on the learning content, use them as rewards, and adjust the weight parameters of the multimodal retrieval algorithm for retrieving the various types of materials through a reinforcement learning algorithm, and return to the step of retrieving matching materials until the feedback scores meet the threshold conditions, and generate personalized learning content based on the final weight parameters.

[0012] According to a third aspect of the present invention, an electronic device is also provided, which includes: at least one processor; and a memory, the memory storing a computer program that can be run on the processor, and the processor executes the aforementioned method for personalized recommendation of educational resources when executing the program.

[0013] According to a fourth aspect of the present invention, a computer-readable storage medium is further provided. The computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the computer program performs the aforementioned method for personalized recommendation of educational resources.

[0014] The above-mentioned method for personalized recommendation of educational resources collects students' learning history data, interest preference data, and knowledge mastery data in real time, and constructs a multi-dimensional portrait through a machine learning algorithm. This solves the problem of dimensional tearing of student portraits caused by traditional RAG, achieves a three-dimensional fusion of learning mode, interest preference, and knowledge level, and improves the accuracy of the portrait's portrayal of student characteristics. Through a reinforcement learning algorithm, the weight parameters of the multimodal retrieval algorithm for retrieving various types of materials in the educational resource library are adjusted. Optimization is performed in a loop until the student's feedback score on the learning content meets the threshold condition. The weight parameters are dynamically updated iteratively based on the feedback, breaking through the limitations of fixed weights, improving the accuracy of multimodal resource matching, and increasing the flexibility of the recommended learning content to adapt to individual students. This establishes a closed-loop optimization mechanism for educational scenarios. The recommended learning content is adaptively updated with the student portrait, reducing the disconnection rate between the recommended learning content and the student portrait and improving the adaptability of the recommended learning content to individual students.

[0015] In addition, the present invention also provides a device for personalized recommendation of educational resources, an electronic device and a computer-readable storage medium, which can also achieve the above-mentioned technical effects and will not be described in detail here. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other embodiments can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 A flowchart of a method for personalized recommendation of educational resources provided by the first embodiment of the present invention; Figure 2 A flowchart of a method for personalized recommendation of educational resources provided by a second embodiment of the present invention; Figure 3 A flowchart of real-time data collection and multi-dimensional student portrait construction in the second embodiment of the present invention; Figure 4 This is a flow chart of generating personalized learning content after multimodal resource retrieval in the second embodiment of the present invention; Figure 5 A flowchart of real-time feedback optimization in a second embodiment of the present invention; Figure 6 A schematic diagram of the structure of a device for personalized recommendation of educational resources provided by a third embodiment of the present invention; Figure 7 FIG4 is a diagram showing the internal structure of an electronic device in a fourth embodiment of the present invention; Figure 8 This is a structural diagram of a computer-readable storage medium in the fifth embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the embodiments of the present invention are further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

[0019] It should be noted that all expressions using "first" and "second" in the embodiments of the present invention are for distinguishing two non-identical entities with the same name or non-identical parameters. It can be seen that "first" and "second" are only for the convenience of expression and should not be understood as limitations on the embodiments of the present invention. Subsequent embodiments will not explain this one by one.

[0020] In one embodiment, please refer to Figure 1 As shown, the present invention provides a method 100 for personalized recommendation of educational resources, specifically comprising: Step 101: Collect students' learning history data, interest preference data, and knowledge mastery data in real time, and construct a multi-dimensional student portrait through a machine learning algorithm; Specifically, learning history data refers to time-series records collected through the learning management system, such as course progress, assignment submission time / scores, and test score distribution. Interest preference data includes the types of learning materials students browse and the topics of interest groups they participate in, and is quantified into a set of preferred subject labels using a decision tree model. Knowledge mastery data is a quantitative score calculated based on the accuracy rate of test questions and the types of incorrect questions. Real-time collection means that when students complete a new course unit, data collection (such as homework duration, unit test scores, etc.) is automatically triggered, and the database is dynamically updated.

[0021] Specifically, machine learning algorithms include clustering algorithms (such as K-Means) and decision tree models. Clustering algorithms use learning time distribution and resource usage preferences as feature vectors to classify learning patterns into clusters. Decision tree models use interest preference data as input and output a set of labels for preferred subjects. Multidimensional student profiles are structured vectors that combine learning pattern cluster labels, preferred subject labels, and knowledge mastery scores.

[0022] In a specific embodiment, students' learning history data, interest preference data, and knowledge mastery data are collected in real time, including: collecting learning history data such as course progress, homework quality, and test score distribution through a learning management system; collecting interest preference data such as the types of learning materials browsed by students and the topics of interest groups they participate in; and obtaining quantitative scores of knowledge mastery based on the accuracy of test questions.

[0023] In another specific embodiment, a multi-dimensional student portrait is constructed through a machine learning algorithm, including: using learning time distribution and resource usage preference as feature vectors, dividing students into several learning mode clusters through the K-Means algorithm; using learning material type and interest group theme as input features of a decision tree model, and generating a preferred subject label set through the decision tree model; when students complete a new course unit, their academic performance, answering time and other data are automatically collected, the feature vector is recalculated and dimensions such as knowledge mastery and learning efficiency are updated.

[0024] In another specific embodiment, the collection method includes direct connection to the database API (Application Programming Interface): regularly pulling course progress, homework submission time / score, and test score distribution through the SQL (Structured Query Language) query interface; log analysis: the server side records learning behavior logs (such as chapter access timestamps) and collects them through Flume / Kafka real-time stream processing.

[0025] Step 102: Based on the multi-dimensional student portrait, matching materials are retrieved from various materials in the educational resource library using a multimodal retrieval algorithm, integrated to generate learning content, and provided to the student; Specifically, multimodal retrieval algorithms include the DSE (Dense Semantic Embedding) algorithm and the ColPali algorithm. The DSE algorithm encodes text / images / audio into dense vectors and calculates cosine similarity with the semantic features of the profile. The ColPali algorithm models the joint probability of cosine similarity and selects resources with a probability value ≥ a threshold. Matching materials refer to multimodal resources that match the subject knowledge points, learning modality preferences, and difficulty level of the profile.

[0026] In a specific embodiment, the fusion generation of learning content includes: inputting the retrieved knowledge fragments into a multimodal large language model (such as GPT-4o, QWen2-VL, InternVL-2) in a format; the model fuses information through the Transformer self-attention mechanism and generates content word by word in combination with the portrait; cross-modal feature fusion, aligning the neural network to unify text (word embedding), image (Convolutional Neural Network, CNN), and audio (Mel spectrum) features.

[0027] In another specific embodiment, based on the multi-dimensional student portrait, the step of retrieving matching materials from various types of materials in the educational resource library through a multimodal retrieval algorithm includes: extracting the semantic features of the student's learning needs from the multi-dimensional student portrait; initializing the weight parameters of the multimodal retrieval algorithm for retrieving various types of materials in the educational resource library, and calculating the correlation between the semantic features of the learning needs and the various types of materials based on the weight parameters; and screening out matching materials based on the correlation.

[0028] In another specific embodiment, the step of calculating the correlation between the semantic features of learning needs and various types of materials based on weight parameters includes: encoding various types of materials into corresponding dense vectors through the DSE algorithm based on the weight parameters; calculating the cosine similarity between the dense vectors and the semantic features of learning needs respectively, and using it as the correlation between the semantic features of learning needs and various types of materials.

[0029] In another specific embodiment, the step of screening out matching data based on the correlation degree includes: performing joint probability modeling on all cosine similarities through the ColPali algorithm to obtain probability values ​​corresponding to each type of data; if the probability value of any type of data is greater than or equal to the probability threshold, it is screened as matching data.

[0030] Step 103: Collect students' feedback scores on the learning content, use them as rewards, and adjust the weight parameters of the multimodal retrieval algorithm for various types of data retrieval through the reinforcement learning algorithm, and return to the step of retrieving matching data until the feedback score meets the threshold condition, and generate personalized learning content based on the final weight parameters.

[0031] Specifically, feedback scores include automatically monitored accuracy, learning time, and student-submitted satisfaction scores. Reinforcement learning algorithms include a state space, an action space, a reward value, and an update mechanism. The vector representation of the student profile is defined as the state space, the set of adjustments to the multimodal retrieval algorithm's weight parameters (such as the DSE vector weights) is defined as the action space, and the feedback score is defined as the reward value. Improvements in accuracy and / or satisfaction correspond to positive rewards, while negative rewards correspond to negative ones. The update mechanism involves updating the Q-table through Q-learning and selecting the optimal action to adjust the weights.

[0032] Specifically, the threshold condition is that the feedback score is greater than or equal to the preset score threshold for N consecutive times, avoiding the impact of misjudgment caused by a single abnormal fluctuation and ensuring the stability of the strategy. The final output content is personalized learning content generated based on the final optimized weight parameters.

[0033] In a specific embodiment, it is used as a reward and the weight parameters of the multimodal retrieval algorithm for retrieving various types of materials are adjusted through a reinforcement learning algorithm, and the step of retrieving matching materials is returned until the feedback score meets the threshold condition. The personalized learning content is generated based on the final weight parameters, including: using the vector representation of the multi-dimensional student portrait as the state space of the Q-learning algorithm; using the set of weight parameter adjustment amounts for the multimodal retrieval algorithm as the action space; updating the Q table with the feedback score as the reward value, and selecting the optimal action to adjust the weight parameters; the updated weight parameters are used for a new round of retrieval to form a closed-loop optimization mechanism; and the cycle is executed until the feedback score is greater than the preset threshold for N consecutive times.

[0034] According to several embodiments of the present invention, the steps of using it as a reward and adjusting the weight parameters of the multimodal retrieval algorithm for various types of data retrieval through a reinforcement learning algorithm include: using the vector representation of the multidimensional student portrait as the state space of the Q-learning algorithm; using the set of adjustment amounts for the weight parameters of the multimodal retrieval algorithm as the action space of the Q-learning algorithm; using the feedback score as the reward value, updating the Q table through the Q-learning algorithm, selecting the optimal action in the action space according to the updated Q table, and adjusting the weight parameters of the multimodal retrieval algorithm based on it.

[0035] The above approach uses the student profile vector as the state space (e.g., a 128-dimensional feature vector) and the weight adjustment as the action space (e.g., an adjustment step of ±0.1), achieving precise parameter-level control. The reward value (feedback score) drives the Q-value update, allowing the weight parameters of the multimodal retrieval algorithm to converge to the optimal solution. The Q-table records the expected benefit of each state-action pair, providing decision transparency for educators. For example, when knowledge mastery is <60, increasing the video weight by 0.1 can result in a +5 reward value, improving the model's interpretability.

[0036] According to several embodiments of the present invention, the steps of collecting students' learning history data, interest preference data, and knowledge mastery data in real time and constructing a multidimensional student profile using a machine learning algorithm include: collecting students' study time for various types of materials in the educational resource library in real time and dividing them into several learning pattern clusters using a clustering algorithm; calculating the proportion of students' study time for each subject within each learning pattern cluster and inputting this into a decision tree model to obtain a set of preferred subject labels; and fusing the labels of several learning pattern clusters, the preferred subject label sets, and the separately obtained student knowledge mastery scores using a machine learning algorithm to generate a multidimensional student profile. The clustering algorithm quantifies study time into discrete pattern clusters, and the decision tree model outputs a confidence score for the preference label (e.g., a probability of a math preference of 0.92), achieving quantification of learning patterns. The fusion of three independent data sources enhances the credibility of the student profile.

[0037] In order to further understand the method of personalized recommendation of educational resources of the present invention, the following is further described in detail in another specific embodiment, the flow chart of which is as follows: Figure 2 shown.

[0038] 1. Real-time data collection and multi-dimensional student portrait construction, the flow chart is as follows Figure 3 As shown in the figure, firstly, data is collected, including learning history data, interest and hobby data, and knowledge mastery data; then clustering algorithms and decision tree algorithms are used to analyze the collected data and construct a multi-dimensional student portrait; then it is determined whether there is new learning progress or feedback. If so, new data is collected, the feature vector is recalculated, and the student portrait is updated, and then the judgment is made again; if not, the process ends. Specifically, real-time data collection includes: (1) Learning history data collection: through the learning management system API, polling and collecting every 5 minutes, course progress percentage (such as 75% for mathematics courses), homework quality score (such as average score 88 / 100), test score distribution (such as algebra correct rate 92%); (2) Interest preference data collection: through front-end tracking, real-time capture, learning resource click type (such as video click ratio 65%), interest group participation theme (such as robotics club); (3) Knowledge mastery data collection: automatically calculated within 30 seconds after the unit test is submitted, knowledge module mastery (such as 85 points in geometry), wrong question type distribution (such as spatial geometry error rate 70%). The construction of multi-dimensional portraits includes: (1) cluster analysis of learning patterns, with the distribution of learning time periods (e.g., 70% between 8 and 10 p.m.) and resource type preferences as features, and the use of the K-Means algorithm to divide learning pattern clusters (e.g., night-time deep learning); (2) decision tree modeling of interest preferences, inputting resource browsing records and interest group data, and outputting a quantitative preference label set (e.g., STEM preference 0.89); (3) dynamic updating of the portrait, when a new unit test is completed, the knowledge mastery vector is recalculated, and the updated portrait is stored in the database (update delay ≤ 1 second).

[0039] 2. Generate personalized learning content after multimodal resource retrieval. The flow chart is as follows: Figure 4 As shown in the figure, a student portrait is obtained, and the DSE algorithm and ColPali algorithm are used for multimodal retrieval to obtain relevant knowledge fragments; then, preliminary content is generated using a pre-trained multimodal large language model; cross-modal feature extraction is then performed on the preliminary content, extracting features of text, image, and audio respectively; the extracted features are then fused to form a unified knowledge representation vector; and finally, personalized learning content is generated.

[0040] Specifically, multimodal resource retrieval includes: (1) semantic feature extraction, extracting the semantics of learning needs from portraits (such as visual explanation of algebraic equations); (2) DSE cross-modal encoding, encoding knowledge base resources into dense vectors, for example, text resources into 768-dimensional semantic vectors and video resources into 2048-dimensional visual vectors; (3) correlation calculation, calculating the cosine similarity between the portrait semantic vector and the resource vector (such as the text resource similarity of 0.91); (4) ColPali probability screening, jointly modeling the multimodal similarity probability formula, P(resource) = 0.6 × text similarity + 0.3 × video similarity + 0.1 × audio similarity for calculation, and the screening condition is a probability value ≥ 0.7 (such as P = 0.795 for a certain animation resource).

[0041] Specifically, personalized content generation includes: (1) multimodal large model input as organization retrieval fragments as structured input: [Knowledge fragment 1] Graphical explanation of algebraic equations (text); [Knowledge fragment 2] Equation solving demonstration (video); [Profile Characteristics] Night study type / High STEM preference / Algebra mastery level 92

[0042] (2) Cross-modal feature fusion, unifying through aligned neural networks: text features (word embedding technology), image features (convolutional neural network), and audio features (Mel spectrum analysis).

[0043] (3) Generate output, output personalized content that matches the profile: Nighttime efficient algebra training: interactive equation solving animation (including AR exercises). The flowchart of the above multimodal retrieval and personalized content generation is as follows: Figure 4 shown.

[0044] 3. Real-time feedback optimization, the flow chart is as follows Figure 5After students start learning, the system simultaneously monitors behavioral data and collects active feedback. Behavioral data includes the accuracy of answering questions and learning time. Active feedback mainly includes satisfaction scores. The collected data is integrated. It is judged whether the adjustment conditions are met. If so, the content generation strategy is adjusted. The learning path is optimized and updated using the Q-learning algorithm. The updated learning path is applied to the subsequent student learning process, forming a closed-loop feedback optimization mechanism. Specifically, (1) Multi-faceted feedback collection, behavioral data monitoring: answering accuracy (such as 94%), learning time (such as 120 minutes for unit learning); active feedback collection: satisfaction score (such as 9 / 10 points); (2) Reinforcement learning optimization, state space: 6-dimensional feature vector of student portrait; action space: multimodal retrieval weight adjustment (such as video weight + 0.1); reward calculation: R = 0.7×{accuracy} + 0.3×{satisfaction}, {example value: 0.928}. Q-learning update: Update the Q table according to the reward value and select the action that maximizes the expected benefit. (3) Closed-loop optimization judgment, continuous optimization until the threshold condition is met: 5 consecutive feedback scores > 8 points (out of 10 points), output the final weight parameters: {text 0.55, video 0.40, audio 0.05}.

[0045] The above-mentioned method of personalized recommendation of educational resources solves the problem of dimensional tearing of student portraits caused by traditional RAG, realizes the three-dimensional fusion of learning mode, interest preference, and knowledge level, improves the accuracy of the portrait's portrayal of student characteristics, and dynamically iterates and updates the weight parameters based on feedback, breaking through the limitations of fixed weights and improving the accuracy of multimodal resource matching. At the same time, a closed-loop optimization mechanism is established in the educational scenario, and the recommended learning content is adaptively updated with the student portrait, reducing the disconnection rate between the recommended learning content and the student portrait.

[0046] In some embodiments, please refer to Figure 6 As shown, the present invention also provides a device 200 for personalized recommendation of educational resources, the device comprising: The first module 201 is used to collect students' learning history data, interest preference data, and knowledge mastery data in real time, and construct a multi-dimensional student portrait through machine learning algorithms; The second module 202 is configured to retrieve matching materials from various materials in the educational resource library based on the multi-dimensional student portrait using a multimodal retrieval algorithm, fuse them to generate learning content, and provide it to the student; The third module 203 is used to collect students' feedback scores on learning content, use them as rewards, and adjust the weight parameters of the multimodal retrieval algorithm for various types of data retrieval through the reinforcement learning algorithm, and return to the step of retrieving matching materials until the feedback scores meet the threshold conditions, and generate personalized learning content based on the final weight parameters.

[0047] It should be noted that each module in the aforementioned apparatus for personalized educational resource recommendation can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the aforementioned modules can be embedded in or independent of a processor in an electronic device in the form of hardware, or can be stored in a memory in the electronic device in the form of software, so that the processor can call and execute the corresponding operations of each of the aforementioned modules.

[0048] According to another aspect of the present invention, an electronic device is provided. The electronic device may be a server. Figure 7 As shown. The electronic device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the electronic device is used to store data. The network interface of the electronic device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the method for personalized recommendation of educational resources described above is implemented.

[0049] According to another aspect of the present invention, a computer readable storage medium is provided. Figure 8 As shown, a computer program is stored thereon, and when the computer program is executed by a processor, the method for personalized recommendation of educational resources described above is implemented.

[0050] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0051] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0052] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for personalized recommendation of educational resources, characterized in that: include: Collect students' learning history data, interest preference data, and knowledge mastery data in real time, and build multi-dimensional student portraits through machine learning algorithms; Based on the multi-dimensional student portrait, matching materials are retrieved from various materials in the educational resource library through a multimodal retrieval algorithm, integrated to generate learning content, and provided to the student; Collect students' feedback scores on the learning content, use them as rewards, and adjust the weight parameters of the multimodal retrieval algorithm for retrieving the various types of materials through a reinforcement learning algorithm, and return to the step of retrieving matching materials until the feedback scores meet the threshold conditions, and generate personalized learning content based on the final weight parameters.

2. The method for personalized recommendation of educational resources according to claim 1, characterized in that: The step of using it as a reward and adjusting the weight parameters of the multimodal retrieval algorithm for the retrieval of the various types of information through a reinforcement learning algorithm includes: Using the vector representation of the multi-dimensional student portrait as the state space of the Q-learning algorithm; Using a set of adjustment amounts for weight parameters of the multimodal retrieval algorithm as an action space of the Q-learning algorithm; The feedback score is used as a reward value, a Q-table is updated through the Q-learning algorithm, an optimal action in the action space is selected according to the updated Q-table, and a weight parameter of the multimodal retrieval algorithm is adjusted based on the optimal action.

3. The method for personalized recommendation of educational resources according to claim 1, characterized in that: The threshold condition includes: N consecutive feedback scores are greater than a preset score threshold.

4. The method for personalized recommendation of educational resources according to claim 1, characterized in that: The step of collecting students' learning history data, interest preference data, and knowledge mastery data in real time and constructing a multi-dimensional student portrait through a machine learning algorithm includes: Collect students' study time of various materials in the educational resource library in real time and divide them into several learning mode clusters through clustering algorithms; Calculate the proportion of students' study time for each subject in each learning mode cluster and input it into the decision tree model to obtain the preferred subject label set; The labels of the several learning mode clusters, the preferred subject label sets and the separately obtained student knowledge mastery scores are integrated through machine learning algorithms to generate a multi-dimensional student portrait.

5. The method for personalized recommendation of educational resources according to claim 1, characterized in that: The step of retrieving matching materials from various materials in the educational resource library using a multimodal retrieval algorithm based on the multi-dimensional student portrait includes: Extracting semantic features of students' learning needs from the multi-dimensional student portrait; Initializing weight parameters of a multimodal retrieval algorithm for retrieving various types of materials in an educational resource library, and calculating the relevance between the semantic features of the learning needs and the various types of materials based on the weight parameters; Based on the correlation, matching information is screened out.

6. The method for personalized recommendation of educational resources according to claim 5, characterized in that: The step of calculating the relevance between the semantic features of the learning requirements and the various types of materials based on the weight parameters includes: Based on the weight parameters, encoding the various types of data into corresponding dense vectors using the DSE algorithm; The cosine similarities between the dense vector and the semantic features of the learning requirements are calculated respectively, and are used as the correlation between the semantic features of the learning requirements and the various types of information.

7. The method for personalized recommendation of educational resources according to claim 6, characterized in that: The step of screening out matching information based on the correlation degree includes: All cosine similarities are jointly modeled using the ColPali algorithm to obtain the probability values ​​corresponding to each type of information; If the probability value of any type of data is greater than or equal to the probability threshold, it will be screened as matching data.

8. A device for personalized recommendation of educational resources, characterized in that: The device comprises: The first module is used to collect students' learning history data, interest preference data, and knowledge mastery data in real time, and build a multi-dimensional student portrait through machine learning algorithms; The second module is used to retrieve matching materials from various materials in the educational resource library based on the multi-dimensional student portrait using a multimodal retrieval algorithm, fuse them to generate learning content, and provide it to the student; The third module is used to collect students' feedback scores on the learning content, use them as rewards, and adjust the weight parameters of the multimodal retrieval algorithm for retrieving the various types of materials through a reinforcement learning algorithm, and return to the step of retrieving matching materials until the feedback scores meet the threshold conditions, and generate personalized learning content based on the final weight parameters.

9. An electronic device, characterized in that: include: at least one processor; as well as A memory storing a computer program executable in the processor, wherein the processor executes the method for personalized recommendation of educational resources according to any one of claims 1 to 7 when executing the program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for personalized recommendation of educational resources according to any one of claims 1 to 7 is performed.

Images