Educational resource diversified recommendation method and system based on long short-term memory neural network

Abstract

The invention discloses an educational resource diversified recommendation method and system based on a long short-term memory neural network, and relates to the technical field of educational resource recommendation, and the method comprises the steps: S1, obtaining the historical learning resource information of learners, and dividing the learners; s2, based on a division result, performing learner representation enhancement and knowledge point mastering modeling; s3, generating a candidate educational resource list based on the knowledge point mastering degree of the learner; s4, based on the candidate educational resource list, calculating educational resource correlation and diversity scores; and S5, fusing the relevance and diversity scores of the educational resources, and recommending a specified number of educational resources for the learner according to the final scores from high to low. A sequence enhancement module is provided, and the learning representation of the inactive learner is enhanced by combining the historical learning behaviors of the inactive learner and the rich historical data of the active learner.

Description

Technical Field

[0001] The present invention relates to the technical field of educational resource recommendation, and in particular to a method and system for diversified educational resource recommendation based on long short-term memory neural network. Background Art

[0002] With the development of information technology and the popularization of online education, personalized learning has gradually become a research focus in the field of educational technology. Personalized learning can significantly improve learning efficiency by dynamically adjusting teaching content to meet learners' abilities, interests and learning goals. In this context, educational resource recommendation systems have emerged, aiming to use artificial intelligence technology to recommend educational resources that match learners' learning needs, so as to optimize learning paths and improve learning outcomes.

[0003] Current educational resource recommendation systems are mainly divided into three categories: knowledge tracking-based, multi-stage-based, and reinforcement learning-based methods. Knowledge tracking-based methods make recommendations by analyzing learners' mastery of knowledge points, but often ignore the diversity of recommended content and are unable to fully meet learners' needs. Multi-stage-based methods attempt to more efficiently predict learners' knowledge status and recommend corresponding educational resources by combining knowledge tracking and neural networks; some methods use knowledge graphs to capture the complex relationship between educational resources, knowledge points, and learners. However, due to the implicitness and variability of learners' learning status, these methods need to rely on a large amount of historical data. This limitation further amplifies the difficulty of recommending inactive learners, because the learning sequences of such learners are usually short and the content is single, and the model is difficult to fully capture their learning needs. Reinforcement learning-based methods dynamically generate recommendations through multi-objective optimization, but they tend to converge to local optimality, resulting in homogeneous recommendations. Summary of the invention

[0004] The purpose of the present invention is to provide a diversified recommendation method for educational resources based on long short-term memory neural networks, so as to solve the problem that the existing educational resource recommendation technology tends to overfit active learners, ignore the majority of inactive learners, resulting in a decrease in the overall performance of learner ability assessment, and does not consider the learners' actual needs for diversity.

[0005] To achieve the above object, the present invention provides a method for recommending diversified educational resources based on a long short-term memory neural network, the steps comprising:

[0006] S1. Obtain learners’ historical learning resource information and divide learners into different groups;

[0007] S2. Based on the results of the division, learner representation enhancement and knowledge point mastery modeling are performed;

[0008] S3. Generate a list of candidate educational resources based on the learner's knowledge mastery;

[0009] S4. Calculate the relevance and diversity scores of educational resources based on the candidate educational resource list;

[0010] S5. Integrate the educational resource relevance and the diversity score, and recommend a specified number of educational resources to the learner according to the final scores from high to low.

[0011] Preferably, the S1 comprises:

[0012] Obtain learners' historical learning resource information through the online learning platform, organize each learner's learning resource information into a sequence in chronological order, and preprocess the sequence to form an initial data set;

[0013] Among them, learners whose number of educational resource interactions is in the top 5% are classified as active learners, and the bottom 95% are classified as inactive learners.

[0014] Preferably, the historical learning resource information includes: learner ID, educational resource serial number learned by the learner, learning time, learning answer results, and knowledge points included in the educational resources; the method of forming the initial data set includes: organizing the questions answered by each user, the knowledge points tested by the questions, the answer results, and the answer time into a learning resource sequence, a knowledge point sequence, a learning resource result sequence, and a learning resource time interval sequence with a length of 200 educational resource interactions in chronological order, and grouping them according to the learner ID; finally, dividing the data, and randomly dividing the data into a training set and a test set in a ratio of 8:2.

[0015] Preferably, S2 includes:

[0016] The learner representation is enhanced by using a sequence enhancement method, which uses the historical learning sequence information of active learners. The steps include: extracting several subsequences from the head user's sequence, encoding the subsequences through a sequence encoder, and training an enhancer to make the representation of the subsequences closer to the representation of the complete user;

[0017] The enhanced learner representation is embedded and spliced ​​with the knowledge points extracted from the historical learning resources, and input into the matrix long short-term memory network mLstm to model the learner's knowledge point mastery status and obtain the learner's mastery of all knowledge points.

[0018] Preferably, a matrix long short-term memory network mLstm is used as the sequence encoder; in the process of training the enhancer, a curriculum learning strategy is adopted to control the weight of each user in the loss according to the difficulty of learning knowledge from active learners; after the enhancer training is completed, the sequence representation of the inactive learner is passed through the enhancer to obtain the representation information of the inactive learner after sequence enhancement.

[0019] Preferably, S3 includes: calculating the difficulty corresponding to each educational resource based on the learner's mastery of knowledge points, and selecting a number of educational resources from the educational resource library as a candidate educational resource list for each learner.

[0020] Preferably, S4 includes:

[0021] Obtaining the embedding of each learner's user features, educational resource features, and knowledge point features from the candidate educational resource list of each learner, inputting them into a bidirectional long short-term memory network, learning context information, and thereby obtaining a relevance score of each educational resource of the overall learner;

[0022] The candidate educational resource list of each learner is classified according to knowledge points, and the embedding of knowledge points, educational resources and learners is obtained. The learning demand distribution of each learner is obtained and the marginal diversity of educational resources is calculated. The diversity score of educational resources is obtained by element-by-element multiplication of the learner's learning demand distribution and the diversity difference of educational resources.

[0023] Preferably, the learning diversity estimator utilizes a long short-term memory network to encode the learner's practice sequence for each knowledge point respectively to obtain the learner's learning pattern representation for each knowledge point; then utilizes an attention mechanism and a multi-layer perceptron to learn the learner's learning demand distribution for different knowledge points; and finally utilizes a probability covering function to calculate the diversity gain of each exercise.

[0024] Preferably, S5 includes: fusing the obtained educational resource relevance and the diversity score through feature fusion to obtain a re-ranking score for each candidate set, and recommending educational resources to learners from high to low according to the final scores of the educational resources.

[0025] The present invention also provides a diversified educational resource recommendation system based on long short-term memory neural network, the system is used to implement the above method, including: an acquisition module, a construction module, a generation module, a calculation module and a fusion module;

[0026] The acquisition module is used to obtain learners' historical learning resource information and divide learners into categories;

[0027] The building module is used to enhance learner representation and model knowledge point mastery based on the segmentation results;

[0028] The generation module is used to generate a candidate education resource list based on the learner's knowledge mastery level;

[0029] The calculation module is used to calculate the relevance and diversity scores of educational resources based on the candidate educational resource list;

[0030] The fusion module is used to fuse the educational resource relevance and the diversity score, and recommend a specified number of educational resources to the learner according to the final scores from high to low.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] (1) Enhanced representation for inactive learners: A major problem with existing educational resource recommendation systems is that they often focus on the data of active learners and ignore the needs of inactive learners. Existing methods rely on a large amount of historical data to train models. Therefore, recommendations for inactive learners are often ineffective because their learning sequences are short and single. In contrast, this application proposes a sequence enhancement module that enhances the learning representation of inactive learners by combining the historical learning behavior of inactive learners with the rich historical data of active learners. This method not only effectively fills the learning behavior characteristics of inactive learners, but also avoids the problem that they cannot obtain accurate recommendations due to data sparsity, thereby improving the adaptability of the recommendation system to inactive learners.

[0033] (2) Multi-stage screening and optimization of candidate educational resources: Existing educational resource recommendation systems usually directly select highly relevant educational resources from the list of candidate educational resources, but do not fully consider the difficulty of educational resources and the distribution of learners' learning needs. In contrast, this application uses multi-stage screening to first screen candidate educational resources based on the learners' mastery of knowledge points, and then further optimizes them based on the learners' learning needs and the diversity of educational resources. This method improves the quality of recommendation results through precise selection of candidate educational resources and diversity optimization, ensuring that the educational resources are not only of moderate difficulty but also have sufficient diversity and challenge.

[0034] (3) Balance between relevance and diversity: Traditional educational resource recommendation methods usually only focus on the relevance of educational resources, which easily leads to the homogenization of recommendation results and lacks attention to the diversity of learners' needs. Especially in methods based on reinforcement learning, the system tends to converge to local optimality, ignoring learners' needs for diverse and extensive educational resources. In contrast, this application uses a relevance estimator and a diversity estimator, and after neural rearrangement, it simultaneously considers the relevance and diversity of educational resources in the recommendation process. This method calculates the diversity score of educational resources and dynamically adjusts the recommendation results based on the distribution of learners' learning needs to ensure that the recommendations not only meet the learners' mastery of knowledge points, but also meet their needs for diversity, thus avoiding the situation of homogenized recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 A schematic diagram of a method flow chart of an embodiment of the present invention;

[0037] Figure 2 A schematic diagram for generating a candidate education resource list according to an embodiment of the present invention;

[0038] Figure 3 A schematic diagram of generating an educational resource recommendation list according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0041] like Figure 1 FIG. 1 is a schematic diagram of the method flow of this embodiment, and the steps include:

[0042] S1. Obtain learners’ historical learning resource information and divide learners into categories.

[0043] The historical learning resource information of learners is obtained from the online learning platform, and the learning resource information of each learner is organized into a sequence in chronological order, and the sequence is preprocessed in a specific way. The preprocessing process includes multiple steps such as data statistics, data cleaning, grouping processing, and serialization processing. Specifically, the original data is first analyzed to count key indicators such as the number of interactions, the number of learners, the number of educational resources, and the number of knowledge points. At the same time, the average number of interactions for each learner, the average number of knowledge points covered by each educational resource, and the number of missing values ​​are calculated. In the data cleaning stage, the missing values ​​and outliers in key fields such as learner id, educational resource id, and knowledge point id are deleted. Next, the data is grouped by learner id to build an independent learning record set for each learner. In the serialization processing stage, the learning resources, knowledge points, answer results, etc. of each learner are arranged in chronological order and unified into a sequence of fixed length 200 for easy model processing. Subsequently, the maximum number of knowledge points in each learner interaction is calculated, and the learner id, educational resource id, knowledge point id, etc. are mapped to integer indexes to simplify data representation and form an initial data set.

[0044] In this embodiment, historical learning resource information includes: learner ID, educational resource serial number learned by the learner, learning time, learning answer results, and knowledge points included in the educational resources; the method of forming an initial data set includes: organizing the questions answered by each user, the knowledge points tested by the questions, the answer results, and the answer time into a learning resource sequence, a knowledge point sequence, a learning resource result sequence, and a learning resource time interval sequence with a length of 200 educational resource interactions in chronological order, and grouping them according to the learner ID; finally, dividing the data, and randomly dividing the data into a training set and a test set in a ratio of 8:2.

[0045] In this embodiment, learners whose educational resource interactions are in the top 5% are classified as active learners, and learners in the bottom 95% are classified as inactive learners.

[0046] S2. Based on the results of the division, learner representation enhancement and knowledge point mastery modeling are carried out.

[0047] By using a sequence enhancement method (this embodiment is implemented by constructing a sequence enhancement module, the structure of which is as follows Figure 2 As shown in the figure, the rich historical learning sequence information of active learners is used to enhance the representation of inactive learners. The sequence enhancement method refers to intercepting a subsequence from the sequence of the head user, encoding the subsequence through a sequence encoder, and training an enhancer to make the representation of the subsequence closer to the representation of the complete user.

[0048] Specifically, from the active learner's historical learning sequence P sThe sequence containing the I most recent interactions is intercepted as a subsequence ,Will As the educational resource interaction sequence of inactive learners, it is intended to simulate the situation where inactive learners have insufficient interaction. This embodiment uses mLstm as the sequence encoder to learn to generate active learners P s Sequence representation of h s , and the subsequence Representation , capturing the learner’s recent learning needs. In order to transfer the knowledge learned by the model from the active learner sequence to the inactive learner, the goal of this embodiment is to generate a complete representation close to the active learner by minimizing the following loss:

[0049] ,

[0050] in, Represents the learner representation enhancer, which is responsible for Outputs a complete sequence representation.

[0051] Since knowledge is transferred from active learners to inactive learners, the performance of the model depends on the quality of the active learner representations, which further depends on the number of relevant interactions among them. Specifically, this embodiment adopts a curriculum learning strategy to gradually train the model from simple to difficult:

[0052] ,

[0053] in represents the loss coefficient of an active learner s; Indicates the current training round; Indicates the maximum round; and Respectively represent the training data The maximum and minimum values ​​of .

[0054] In the early stages of training (i.e. ), the model learns more from active learners with longer sequence lengths than from active learners with shorter sequence lengths. As training progresses, the model starts to learn more from active learners with shorter sequence lengths than from active learners with longer sequence lengths. is a learner representation enhancer based on the input Generate a complete learner representation. By minimizing the loss function ,train Contains enough knowledge to generate complete representations from short practice sequences of learners, and finally acquires inactive learners through sequence enhancers Enhanced representation:

[0055] ,

[0056] in Inactive learners enhanced representation; Represents the entire educational resource sequence based on the learner representations of inactive learners obtained; is a hyperparameter used to control the inactive learner The contribution of the original representation. In addition to the original representation In addition, we also consider using sequence representation enhancers Representations generated from the knowledge gained by active learners can complement and enhance the low-quality representations of inactive learners due to the lack of interaction with educational resources.

[0057] After the enhancer training is completed, the sequence representation of the inactive learner is passed through the enhancer to obtain the representation information of the inactive learner after sequence enhancement. , embedding it with the knowledge points extracted from the learner’s historical learning resources Perform concat operation to get the learner’s feature embedding , input into the matrix long short-term memory network mLstm to model the learner's knowledge point mastery status, in order to obtain the learner's mastery of all knowledge points:

[0058] ,

[0059] Then the optimized mLstm module cost function L K It can be expressed as:

[0060] ,

[0061] Where N is the number of learners; T i represents the time period of educational resources before the i-th learner answers; represents the one-hot encoding of the knowledge point answered at time t+1, and uses r t+1 Indicates the corresponding answer; y t It represents the output of the mLstm module at time t. Finally, the loss function of the knowledge point mastery prediction module training is for:

[0062] ,

[0063] Among them, λ represents a hyperparameter; S H represents the set of active learners.

[0064] S3. Generate a list of candidate educational resources based on the learner’s knowledge mastery.

[0065] like Figure 2 As shown, based on the learner's mastery of each knowledge point, the difficulty of each educational resource is calculated, and N educational resources are selected from the educational resource library as a candidate educational resource list for each learner.

[0066] The specific steps include: after the knowledge point mastery degree prediction module is trained, it is used to predict the learner's mastery degree of different knowledge points, so as to deduce the difficulty of each exercise for different learners according to the following formula:

[0067] ,

[0068] in, Indicates educational resources Difficulty; Indicates the learner's mastery of the i-th knowledge point.

[0069] Finally, a list of candidate educational resources Ω is screened according to the difficulty of educational resources, and 150 candidate educational resources are selected for each learner.

[0070] S4. Based on the list of candidate educational resources, calculate the educational resource relevance and diversity scores.

[0071] like Figure 3 As shown, first, the embedding of each learner's user characteristics, educational resource characteristics, and knowledge point characteristics is obtained from the candidate educational resource list of each learner, and input into a bidirectional long short-term memory network to learn the context information, thereby obtaining the relevance score of each educational resource for the overall learner.

[0072] Specifically, represents the lth (l=1,2,...,L) educational resource in the candidate educational resource list (candidate set) Ω. Through the bidirectional LSTM network, the positive output state of the lth educational resource can be obtained as , the reverse output state is . Final Educational Resources The correlation is expressed as Connections for forward and reverse output states Then the relevance representation vector of each educational resource is stacked Get the correlation representation matrix R C , thereby modeling the relationship between each candidate educational resource and each knowledge point in the candidate subset for learners.

[0073] Afterwards, the candidate educational resource list of each learner is classified according to knowledge points, and the embeddings of knowledge points, educational resources and learners are obtained. The learning demand distribution of each learner is obtained, and the marginal diversity of educational resources is calculated. The difference between the learner's learning demand distribution and the diversity of educational resources is multiplied element by element to obtain the diversity score of educational resources.

[0074] The above-mentioned learning diversity estimator uses the long short-term memory network to encode the learner's practice sequence for each knowledge point respectively, and obtains the learner's learning pattern representation for each knowledge point; then uses the attention mechanism and multi-layer perceptron to learn the learner's learning demand distribution for different knowledge points; finally, the probability covering function is used to calculate the diversity gain of each exercise.

[0075] Specifically, the candidate educational resource list of each learner is classified according to knowledge points, and the embedding of knowledge points, educational resources and learners is obtained. The time dependency of learner records is explicitly encoded in each knowledge point by LSTM to model the interaction within the knowledge points. The final output state of LSTM can be regarded as the encoding of all information contained in the sequence, which has the greatest impact on future predictions. After encoding the learner's learning pattern for each knowledge point separately, the unique signals of the knowledge point are aggregated, and the distribution of learners' learning needs on the knowledge points is obtained through the self-attention mechanism. Then, all knowledge points are Represents the vector superposition to obtain the matrix W. Then use the multi-layer perceptron Generate the distribution of personalized learning requirements for knowledge points ω:

[0076] ,

[0077] Among them, ω represents the possibility that the learner is interested in each knowledge point; represents a self-attention network.

[0078] Given a candidate subset, the diversity of educational resources depends on the difference or novelty between the current educational resource and other educational resources in the candidate subset. Therefore, the probability coverage function is used As a diversity function, the specific formula is as follows:

[0079] ,

[0080] in, represents the knowledge point coverage of educational resources; L represents the number of educational resources; represents the probability that there is at least one educational resource covering knowledge point k in the candidate list Ω.

[0081] In addition, this embodiment defines the marginal diversity of educational resources as , that is, whether there are educational resources in the candidate list The diversity difference of each educational resource in the candidate list Ω can be obtained. ,l=1,2,...,L diversity representation vector .

[0082] ,

[0083] in, represents the element-wise product. Therefore, The kth element of represents the diversity benefit of educational resources in the knowledge point.

[0084] S5. Integrate the relevance and diversity scores of educational resources and recommend a specified number of educational resources to learners based on the final scores from high to low.

[0085] The obtained educational resource relevance and the diversity score are subjected to feature fusion to obtain a re-ranking score for each candidate set, and educational resources are recommended to learners from high to low according to the final scores of the educational resources.

[0086] The scores are predicted by concatenating the correlation matrix and the diversity matrix and then using an MLP to fuse the correlation and diversity:

[0087] ,

[0088] in, Represents the re-ranking score matrix of educational resources.

[0089] Based on the relevance score and diversity score of each educational resource, a re-ranking score of each educational resource for each learner is obtained, and the educational resources are ranked according to the re-ranking score; finally, according to the length K of the recommended list of educational resources, the top K ranked educational resources are recommended to the learner.

[0090] Embodiment 2

[0091] This embodiment also provides a diversified educational resource recommendation system based on long short-term memory neural network, including: an acquisition module, a construction module, a generation module, a calculation module and a fusion module; the acquisition module is used to obtain the learner's historical learning resource information and divide the learners; the construction module is used to enhance the learner representation and model the knowledge point mastery based on the division results; the generation module is used to generate a list of candidate educational resources based on the learner's knowledge point mastery; the calculation module is used to calculate the educational resource relevance and diversity scores based on the candidate educational resource list; the fusion module is used to integrate the educational resource relevance and diversity scores, and recommend a specified number of educational resources to the learners from high to low according to the final scores.

[0092] The following will describe in detail how the present invention solves technical problems in real life in conjunction with this embodiment.

[0093] First, the acquisition module is used to obtain learners' historical learning resource information and divide the learners into categories.

[0094] The historical learning resource information of learners is obtained from the online learning platform, and the learning resource information of each learner is organized into a sequence in chronological order, and the sequence is preprocessed in a specific way. The preprocessing process includes multiple steps such as data statistics, data cleaning, grouping processing, and serialization processing. Specifically, the original data is first analyzed to count key indicators such as the number of interactions, the number of learners, the number of educational resources, and the number of knowledge points. At the same time, the average number of interactions for each learner, the average number of knowledge points covered by each educational resource, and the number of missing values ​​are calculated. In the data cleaning stage, the missing values ​​and outliers in key fields such as learner id, educational resource id, and knowledge point id are deleted. Next, the data is grouped by learner id to build an independent learning record set for each learner. In the serialization processing stage, the learning resources, knowledge points, answer results, etc. of each learner are arranged in chronological order and unified into a sequence of fixed length 200 for easy model processing. Subsequently, the maximum number of knowledge points in each learner interaction is calculated, and the learner id, educational resource id, knowledge point id, etc. are mapped to integer indexes to simplify data representation and form an initial data set.

[0095] In this embodiment, historical learning resource information includes: learner ID, educational resource serial number learned by the learner, learning time, learning answer results, and knowledge points included in the educational resources; the method of forming an initial data set includes: organizing the questions answered by each user, the knowledge points tested by the questions, the answer results, and the answer time into a learning resource sequence, a knowledge point sequence, a learning resource result sequence, and a learning resource time interval sequence with a length of 200 educational resource interactions in chronological order, and grouping them according to the learner ID; finally, dividing the data, and randomly dividing the data into a training set and a test set in a ratio of 8:2.

[0096] In this embodiment, learners whose educational resource interactions are in the top 5% are classified as active learners, and learners in the bottom 95% are classified as inactive learners.

[0097] Then, the building module is used to enhance learner representation and model knowledge point mastery based on the division results.

[0098] By using a sequence enhancement method (this embodiment is implemented by constructing a sequence enhancement module, the structure of which is as follows Figure 2As shown in the figure, the rich historical learning sequence information of active learners is used to enhance the representation of inactive learners. The sequence enhancement method refers to intercepting a subsequence from the sequence of the head user, encoding the subsequence through a sequence encoder, and training an enhancer to make the representation of the subsequence closer to the representation of the complete user.

[0099] Specifically, from the active learner's historical learning sequence P s The sequence containing the I most recent interactions is intercepted as a subsequence ,Will As the educational resource interaction sequence of inactive learners, it is intended to simulate the situation where inactive learners have insufficient interaction. This embodiment uses mLstm as the sequence encoder to learn to generate active learners P s Sequence representation of h s , and the subsequence Representation , capturing the learner’s recent learning needs. In order to transfer the knowledge learned by the model from the active learner sequence to the inactive learner, the goal of this embodiment is to generate a complete representation close to the active learner by minimizing the following loss:

[0100] ,

[0101] in, Represents the learner representation enhancer, which is responsible for Outputs a complete sequence representation.

[0102] Since knowledge is transferred from active learners to inactive learners, the performance of the model depends on the quality of the active learner representations, which further depends on the number of relevant interactions among them. Specifically, this embodiment adopts a curriculum learning strategy to gradually train the model from simple to difficult:

[0103] ,

[0104] in represents the loss coefficient of an active learner s; Indicates the current training round; Indicates the maximum round; and Respectively represent the training data The maximum and minimum values ​​of .

[0105] In the early stages of training (i.e. ), the model learns more from active learners with longer sequence lengths than from active learners with shorter sequence lengths. As training progresses, the model starts to learn more from active learners with shorter sequence lengths than from active learners with longer sequence lengths. is a learner representation enhancer based on the input Generate a complete learner representation. By minimizing the loss function ,train Contains enough knowledge to generate complete representations from short practice sequences of learners, and finally acquires inactive learners through sequence enhancers Enhanced representation:

[0106] ,

[0107] in Inactive learners enhanced representation; Represents the entire educational resource sequence based on the learner representations of inactive learners obtained; is a hyperparameter used to control the inactive learner The contribution of the original representation. In addition to the original representation In addition, we also consider using sequence representation enhancers Representations generated from the knowledge gained by active learners can complement and enhance the low-quality representations of inactive learners due to the lack of interaction with educational resources.

[0108] After the enhancer training is completed, the sequence representation of the inactive learner is passed through the enhancer to obtain the representation information of the inactive learner after sequence enhancement. , embedding it with the knowledge points extracted from the learner’s historical learning resources Perform concat operation to get the learner’s feature embedding , input into the matrix long short-term memory network mLstm to model the learner's knowledge point mastery status, in order to obtain the learner's mastery of all knowledge points:

[0109] ,

[0110] Then the optimized mLstm module cost function L K It can be expressed as:

[0111] ,

[0112] Where N is the number of learners; T i represents the time period of educational resources before the i-th learner answers; represents the one-hot encoding of the knowledge point answered at time t+1, and uses r t+1 Indicates the corresponding answer; y t It represents the output of the mLstm module at time t. Finally, the loss function of the knowledge point mastery prediction module training is for:

[0113] ,

[0114] Among them, λ represents a hyperparameter; S H represents the set of active learners.

[0115] The generation module generates a list of candidate educational resources based on the learner's mastery of knowledge points.

[0116] like Figure 2 As shown, based on the learner's mastery of each knowledge point, the difficulty of each educational resource is calculated, and N educational resources are selected from the educational resource library as a candidate educational resource list for each learner.

[0117] The specific steps include: after the knowledge point mastery degree prediction module is trained, it is used to predict the learner's mastery degree of different knowledge points, so as to deduce the difficulty of each exercise for different learners according to the following formula:

[0118] ,

[0119] in, Indicates educational resources Difficulty; Indicates the learner's mastery of the i-th knowledge point.

[0120] Finally, a list of candidate educational resources Ω is screened according to the difficulty of educational resources, and 150 candidate educational resources are selected for each learner.

[0121] The calculation module calculates the relevance and diversity scores of educational resources based on the list of candidate educational resources.

[0122] like Figure 3 As shown, first, the embedding of each learner's user characteristics, educational resource characteristics, and knowledge point characteristics is obtained from the candidate educational resource list of each learner, and input into a bidirectional long short-term memory network to learn the context information, thereby obtaining the relevance score of each educational resource for the overall learner.

[0123] Specifically, represents the lth (l=1,2,...,L) educational resource in the candidate educational resource list (candidate set) Ω. Through the bidirectional LSTM network, the positive output state of the lth educational resource can be obtained as , the reverse output state is . Final Educational Resources The correlation is expressed as Connections for forward and reverse output states Then the relevance representation vector of each educational resource is stacked Get the correlation representation matrix R C , thereby modeling the relationship between each candidate educational resource and each knowledge point in the candidate subset for learners.

[0124] Afterwards, the candidate educational resource list of each learner is classified according to knowledge points, and the embeddings of knowledge points, educational resources and learners are obtained. The learning demand distribution of each learner is obtained, and the marginal diversity of educational resources is calculated. The difference between the learner's learning demand distribution and the diversity of educational resources is multiplied element by element to obtain the diversity score of educational resources.

[0125] The above-mentioned learning diversity estimator uses the long short-term memory network to encode the learner's practice sequence for each knowledge point respectively, and obtains the learner's learning pattern representation for each knowledge point; then uses the attention mechanism and multi-layer perceptron to learn the learner's learning demand distribution for different knowledge points; finally, the probability covering function is used to calculate the diversity gain of each exercise.

[0126] Specifically, the candidate educational resource list of each learner is classified according to knowledge points, and the embedding of knowledge points, educational resources and learners is obtained. The time dependency of learner records is explicitly encoded in each knowledge point by LSTM to model the interaction within the knowledge points. The final output state of LSTM can be regarded as the encoding of all information contained in the sequence, which has the greatest impact on future predictions. After encoding the learner's learning pattern for each knowledge point separately, the unique signals of the knowledge point are aggregated, and the distribution of learners' learning needs on the knowledge points is obtained through the self-attention mechanism. Then, all knowledge points are Represents the vector superposition to obtain the matrix W. Then use the multi-layer perceptron Generate the distribution of personalized learning requirements for knowledge points ω:

[0127] ,

[0128] Among them, ω represents the possibility that the learner is interested in each knowledge point; represents a self-attention network.

[0129] Given a candidate subset, the diversity of educational resources depends on the difference or novelty between the current educational resource and other educational resources in the candidate subset. Therefore, the probability coverage function is used As a diversity function, the specific formula is as follows:

[0130] ,

[0131] in, represents the knowledge point coverage of educational resources; L represents the number of educational resources; represents the probability that there is at least one educational resource covering knowledge point k in the candidate list Ω.

[0132] In addition, this embodiment defines the marginal diversity of educational resources as , that is, whether there are educational resources in the candidate list The diversity difference of each educational resource in the candidate list Ω can be obtained. ,l=1,2,...,L diversity representation vector .

[0133] ,

[0134] in, represents the element-wise product. Therefore, The kth element of represents the diversity benefit of educational resources in the knowledge point.

[0135] Finally, the fusion module integrates the relevance and diversity scores of educational resources and recommends a specified number of educational resources to learners according to the final scores from high to low.

[0136] The obtained educational resource relevance and the diversity score are subjected to feature fusion to obtain a re-ranking score for each candidate set, and educational resources are recommended to learners from high to low according to the final scores of the educational resources.

[0137] The scores are predicted by concatenating the correlation matrix and the diversity matrix and then using an MLP to fuse the correlation and diversity:

[0138] ,

[0139] in, Represents the re-ranking score matrix of educational resources.

[0140] Based on the relevance score and diversity score of each educational resource, a re-ranking score of each educational resource for each learner is obtained, and the educational resources are ranked according to the re-ranking score; finally, according to the length K of the recommended list of educational resources, the top K ranked educational resources are recommended to the learner.

[0141] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for recommending diversified educational resources based on long short-term memory neural network, characterized in that the steps include: S1. Obtain learners’ historical learning resource information and divide learners into different groups; S2. Based on the results of the division, learner representation enhancement and knowledge point mastery modeling are performed; S3. Generate a list of candidate educational resources based on the learner's knowledge mastery; S4. Calculate the relevance and diversity scores of educational resources based on the candidate educational resource list; S5. Integrate the educational resource relevance and the diversity score, and recommend a specified number of educational resources to the learner according to the final scores from high to low.

2. The method for recommending diversified educational resources based on long short-term memory neural network according to claim 1, characterized in that: The S1 includes: Obtain learners' historical learning resource information through the online learning platform, organize each learner's learning resource information into a sequence in chronological order, and preprocess the sequence to form an initial data set; Among them, learners whose number of educational resource interactions is in the top 5% are classified as active learners, and the bottom 95% are classified as inactive learners.

3. The method for recommending diversified educational resources based on long short-term memory neural network according to claim 2 is characterized in that: The historical learning resource information includes: learner ID, educational resource serial number learned by the learner, learning time, learning answer results and knowledge points included in the educational resources; the method of forming the initial data set includes: organizing the questions answered by each user, the knowledge points tested by the questions, the answer results, and the answer time into a learning resource sequence, a knowledge point sequence, a learning resource result sequence and a learning resource time interval sequence with a length of 200 educational resource interactions in chronological order, and grouping them according to the learner ID; finally, dividing the data, and randomly dividing the data into a training set and a test set in a ratio of 8:

2.

4. The method for recommending diversified educational resources based on long short-term memory neural network according to claim 1, characterized in that: The S2 includes: The learner representation is enhanced by using a sequence enhancement method, which uses the historical learning sequence information of active learners. The steps include: extracting several subsequences from the head user's sequence, encoding the subsequences through a sequence encoder, and training an enhancer to make the representation of the subsequences closer to the representation of the complete user; The enhanced learner representation is embedded and spliced ​​with the knowledge points extracted from the historical learning resources, and input into the matrix long short-term memory network mLstm to model the learner's knowledge point mastery status and obtain the learner's mastery of all knowledge points.

5. The method for recommending diversified educational resources based on long short-term memory neural network according to claim 4 is characterized in that: A matrix long short-term memory network mLstm is used as the sequence encoder; in the process of training the enhancer, a curriculum learning strategy is adopted to control the weight of each user in the loss according to the difficulty of learning knowledge from active learners; after the enhancer training is completed, the sequence representation of the inactive learner is passed through the enhancer to obtain the representation information of the inactive learner after sequence enhancement.

6. The method for recommending diversified educational resources based on long short-term memory neural network according to claim 1, characterized in that: The S3 includes: calculating the difficulty of each educational resource based on the learner's mastery of knowledge points, and selecting a number of educational resources from the educational resource library as a candidate educational resource list for each learner.

7. The method for recommending diversified educational resources based on long short-term memory neural network according to claim 1, characterized in that: The S4 includes: Obtaining the embedding of each learner's user features, educational resource features, and knowledge point features from the candidate educational resource list of each learner, inputting them into a bidirectional long short-term memory network, learning context information, and thereby obtaining a relevance score of each educational resource of the overall learner; The candidate educational resource list of each learner is classified according to knowledge points, and the embedding of knowledge points, educational resources and learners is obtained. The learning demand distribution of each learner is obtained and the marginal diversity of educational resources is calculated. The diversity score of educational resources is obtained by element-by-element multiplication of the learner's learning demand distribution and the diversity difference of educational resources.

8. The method for recommending diversified educational resources based on long short-term memory neural network according to claim 7, characterized in that: The learning diversity estimator uses a long short-term memory network to encode the learner's practice sequence for each knowledge point respectively to obtain the learner's learning pattern representation for each knowledge point; then uses an attention mechanism and a multi-layer perceptron to learn the learner's learning demand distribution for different knowledge points; finally, a probability covering function is used to calculate the diversity gain of each exercise.

9. The method for recommending diversified educational resources based on long short-term memory neural network according to claim 1, characterized in that: The S5 includes: fusing the obtained educational resource relevance and the diversity score through feature fusion to obtain a re-ranking score for each candidate set, and recommending educational resources to learners from high to low according to the final scores of the educational resources.

10. A diversified educational resource recommendation system based on long short-term memory neural network, the system is used to implement the method according to any one of claims 1 to 9, characterized in that: include: Acquisition module, construction module, generation module, calculation module and fusion module; The acquisition module is used to obtain learners' historical learning resource information and divide learners into categories; The building module is used to enhance learner representation and model knowledge point mastery based on the segmentation results; The generation module is used to generate a candidate education resource list based on the learner's knowledge mastery level; The calculation module is used to calculate the relevance and diversity scores of educational resources based on the candidate educational resource list; The fusion module is used to fuse the educational resource relevance and the diversity score, and recommend a specified number of educational resources to the learner according to the final scores from high to low.

Images