An online education platform data storage management method

Abstract

The present application relates to the technical field of data storage, and particularly relates to an online education platform data storage management method, which comprises the following steps: generating education resource interest features by an education resource model based on user access resource data of an online education platform, wherein the education resource model is constructed based on a multi-head self-attention mechanism; generating user interest features by a user interest model based on user time sequence data of the online education platform, wherein the user interest model is constructed based on an improved gated recurrent unit; determining predicted interest projects by an item interest prediction model based on the education resource features and the user interest features, wherein the item interest prediction model is established based on a multi-layer perception machine; and storing data corresponding to the predicted interest projects as hot data in an education resource database. Through the prediction of the education content of interest to the user, the present application realizes the fast access of the user to the education content of interest, and improves the user experience.

Description

Technical Field

[0001] This invention relates to the field of data storage technology, and in particular to a data storage management method for an online education platform. Background Technology

[0002] "Internet + Education" has become the development model for the education industry. The widespread adoption of online education platforms has led to a surge in user numbers, generating massive amounts of data, including user personal information, learning behavior records, teaching videos, and course materials. An efficient data storage mechanism is crucial for ensuring the smooth operation of online education platforms and improving user experience.

[0003] Currently, online education platforms use MySQL data repositories that employ a hot and cold data separation technique. This involves dividing data into hot and cold data based on access frequency and storing them separately. Since hot data is smaller in volume, this method significantly improves data access efficiency and system response speed. However, simply distinguishing data based on access frequency is not accurate enough and cannot meet the personalized access needs of online education platform users. It prevents users from quickly accessing content they might be interested in. Users may have to wait a considerable amount of time while exploring content of interest, as the system accesses and loads cold data from cold tables. This results in a poor user experience and may cause users to cancel their access while waiting for cold data to load, thus reducing their motivation to explore new content.

[0004] Therefore, how to store and manage data that users are interested in on online education platforms is a technical problem that needs to be solved. Summary of the Invention

[0005] To address this, the present invention provides a data storage management method for an online education platform. By employing a multi-head self-attention mechanism, an improved gated loop unit, and a multilayer perceptron, the method accurately identifies content of interest to users on the online education platform and stores and manages it separately. However, accessing content of interest to users requires the platform to access and load large amounts of data from a database, resulting in long waiting times and a poor user experience.

[0006] To achieve the above objectives, this invention proposes a data storage management method for an online education platform. The online education platform is configured with an educational resource database for the separate storage of cold and hot data, comprising:

[0007] The user access resource data of the online education platform is used to generate educational resource interest features through an educational resource model, wherein the educational resource model is constructed based on a multi-head self-attention mechanism.

[0008] The user interest data of the online education platform is used to generate user interest features through a user interest model, wherein the user interest model is constructed based on an improved gated loop unit.

[0009] The educational resource features and the user interest features are used to determine the predicted interest items through a project interest prediction model, wherein the project interest prediction model is based on a multilayer perceptron.

[0010] The data corresponding to the predicted interest items are stored as hot data in the educational resource database.

[0011] Furthermore, the educational resource model includes a multi-head self-attention mechanism, an attention mechanism, and a pooling layer. The process of generating educational resource interest features from user access data of the online education platform through the educational resource model includes:

[0012] The user access resource data is used to generate a resource feature head vector through the multi-head self-attention mechanism;

[0013] The resource feature header vector is used to generate target weights through the attention mechanism;

[0014] The target weights are used to generate the educational resource interest features through the pooling layer.

[0015] Furthermore, the educational resource model also includes an auxiliary loss term for learning the feature representation of the user's access to resource data. The auxiliary loss term is constructed based on the log-likelihood form and an activation function.

[0016] In the above scheme, the multi-head self-attention mechanism is beneficial for capturing a wider range of relevant features, increasing the model's expressive power. Using the multi-head self-attention mechanism to model multiple potential user interests and learn the potential relationships between items at the same time makes interest mining more diverse.

[0017] Furthermore, the user interest model includes a gated recurrent unit module and an attention module. The process of generating user interest features from user time-series data through the user interest model includes:

[0018] The user time-series data is input into the gated loop unit module, the gated loop unit module outputs the user interest features as a hidden state, and the attention module controls the updating of the hidden state by the gated loop unit module.

[0019] Furthermore, the hidden state includes the previous hidden state, the current hidden state, and the transitive hidden state. The process of updating the hidden state by the gated loop unit module through the attention module includes:

[0020] The current hidden state and the target vector are used by the attention module to generate attention weights;

[0021] The transitive hidden state is generated based on the attention weight, the previous hidden state, and the current hidden state, and then the transitive hidden state is passed to the gated loop unit module at the next time step.

[0022] Furthermore, an objective function based on the cross-entropy loss term is constructed, which is used for the optimization prediction of the gated recurrent unit module.

[0023] In the above scheme, the weights of each behavior are adaptively learned through an attention mechanism in the improved gating loop unit, which achieves the latest changes that better express user interests.

[0024] Furthermore, the process of determining predicted interest items by using a project interest prediction model to analyze educational resource characteristics and user interest characteristics includes:

[0025] Construct a loss function based on the negative logarithm function and the probability of accessing interest items, wherein the loss function is used for prediction by the item interest prediction model;

[0026] The educational resource features, user interest features, and target project data are input into the project interest prediction model. The project interest prediction model outputs the predicted interest project access probability and selects the predicted interest project from the target project data based on the predicted interest project access probability.

[0027] Furthermore, the project interest prediction model includes a fully connected layer, at least two layers of Dice activation functions and Sigmoid activation functions. The process of inputting the educational resource features, the user interest features, and the target project data into the project interest prediction model, and the model outputting a predicted probability of accessing the interest project, includes:

[0028] The educational resource features, user interest features, and target project data are sequentially passed through the fully connected layer, at least two layers of Dice activation functions, and Sigmoid activation functions to output the predicted access probability of the interest project.

[0029] In the above scheme, the Dice activation function is used by the multilayer perceptron to enhance the expressive power of low-frequency features and further perceive the evolution of interest, thereby reducing the impact of noise input to the multilayer perceptron.

[0030] Furthermore, the user access resource data includes user information data, user learning behavior data, and target project data;

[0031] The user time-series data includes user historical behavior time-series data, user information data, and target project data.

[0032] Furthermore, the educational resource database includes a cold data table and at least two hot data tables. The process of storing the data corresponding to the predicted interest items as hot data in the educational resource database includes:

[0033] The data corresponding to the predicted interest items are classified according to their access frequency and their correlation with the currently used item set, and stored in multiple hot data tables respectively.

[0034] The cold data and the hot data are periodically converted using a thread pool.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0036] 1. By using a multi-head self-attention mechanism, an improved gating loop unit, and a multilayer perceptron, the content that users are interested in on the online education platform is accurately identified and stored separately. This results in the platform needing to access and load a large amount of data from a database to allow users to access the content they are interested in, leading to long waiting times and a poor user experience.

[0037] 2. Multi-head self-attention mechanism is conducive to capturing a wider range of relevant features, increasing the expressive power of the model. Using multi-head self-attention mechanism to model multiple potential interests of users and the potential relationships between educational resources at the same time makes interest mining more diversified.

[0038] 3. In the improved gating loop unit, the weights of each behavior are adaptively learned through an attention mechanism, which enables a better expression of the latest changes in user interests.

[0039] 4. By using the Dice activation function in a multilayer perceptron, the expressive power of low-frequency features is enhanced, further perceiving the evolution of interest, thereby reducing the impact of noise input to the multilayer perceptron. Attached Figure Description

[0040] Figure 1 This is a flowchart illustrating the data storage management method for an online education platform according to an embodiment of the present invention.

[0041] Figure 2 This is a detailed flowchart illustrating the data storage management method for an online education platform according to an embodiment of the present invention.

[0042] Figure 3 This is a schematic diagram of the model structure of the online education platform data storage management method according to an embodiment of the present invention;

[0043] Figure 4 This is a schematic diagram of an implementation structure of the online education platform data storage management method according to an embodiment of the present invention;

[0044] Figure 5This is a schematic diagram of another implementation of the online education platform data storage management method according to an embodiment of the present invention. Detailed Implementation

[0045] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0046] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0047] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0048] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0049] like Figures 1 to 5 As shown, this invention provides a data storage management method for an online education platform. Through a multi-head self-attention mechanism, an improved gated loop unit, and a multilayer perceptron, the method accurately identifies content of interest to users on the online education platform and stores and manages it separately. However, when users access content of interest, the platform needs to access and load a large amount of data from a database, resulting in long waiting times and a poor user experience.

[0050] like Figures 1 to 5 As shown, this embodiment proposes a data storage management method for an online education platform. The online education platform sets up an educational resource database for the separate storage of cold and hot data, including:

[0051] The user access resource data of the online education platform is used to generate educational resource interest features through an educational resource model, wherein the educational resource model is constructed based on a multi-head self-attention mechanism.

[0052] The user interest data of the online education platform is used to generate user interest features through a user interest model, wherein the user interest model is constructed based on an improved gated loop unit.

[0053] The educational resource features and the user interest features are used to determine the predicted interest items through a project interest prediction model, wherein the project interest prediction model is based on a multilayer perceptron.

[0054] The data corresponding to the predicted interest items are stored as hot data in the educational resource database.

[0055] It is understandable that an educational resource model incorporating a multi-head self-attention mechanism can be used to model users' multi-layered potential interests, and auxiliary loss can be used to aid in the expression and learning of interest features. In the user interest model, user time-series data is added, that is, user access history data is encoded and input in chronological order, thereby enhancing the strength of personalized interest expression.

[0056] Therefore, by accurately identifying educational resources / project data that users are interested in and processing them separately as hot data, online education platforms can make targeted configurations based on the type of project requirements, such as allocating different priorities and the number of threads. This allows platform resources to be more rationally allocated to data processing tasks with different access probabilities, further optimizing and improving the overall performance of the online education platform.

[0057] It is understandable that, such as Figure 4 and 5 As shown, the process of identifying the data corresponding to the predicted interest item can be used to distinguish and construct hot data tables and cold data tables in the initial stage, and can also be used for data conversion between hot data tables and cold data tables that are executed periodically.

[0058] Furthermore, the educational resource model includes a multi-head self-attention mechanism, an attention mechanism, and a pooling layer. The process of generating educational resource interest features from user access data of the online education platform through the educational resource model includes:

[0059] The user access resource data is used to generate a resource feature head vector through the multi-head self-attention mechanism;

[0060] The resource feature header vector is used to generate target weights through the attention mechanism;

[0061] The target weights are used to generate the educational resource interest features through the pooling layer.

[0062] Understandably, multi-head self-attention mechanisms are beneficial for capturing a wider range of relevant features, increasing the model's expressive power, and have already proven effective in various training tasks. Here, we use multi-head self-attention mechanisms to model a variety of potential user interests.

[0063] Specifically, the multi-head self-attention mechanism is as follows:

[0064]

[0065] In the formula, head i W represents the feature head vector of the i-th resource. Q,i W K,i W V,i These are the projection matrices of the i-th head of Q (Query), K (Key), and V (Value), respectively, all of which belong to R, i.e., W. Q,i W K,i W V,i ∈R, R=Multihead(Q,K,V), x1 represents the input user access resource data.

[0066] Furthermore, the educational resource model also includes an auxiliary loss term to learn the feature representation of the user's access to resource data. This auxiliary loss term is constructed based on a log-likelihood form and an activation function, specifically as follows:

[0067]

[0068] In the formula, Loss multihead The auxiliary loss term is defined as follows: N is the number of samples in the training dataset, σ is the sigmoid activation function, and <,> is the inner product used to represent the similarity between the two variables. j ,e′ j+1 Let $\mathbf{j}$ represent the educational resource interest features learned by the model at time $j$, and $\mathbf{j}$ represent the original educational resource vector representation input to the model at time $j$, i.e., the positive sample. Therefore, the auxiliary loss term can help learn feature representations that better reflect the distinguishing characteristics between different resources.

[0069] Understandably, feature interaction in a project interest prediction model refers to the process of combining two or more features to generate new features. Feature interaction primarily involves cross-multiplying features through inner products to generate new features. This method not only reflects users' interest in product combinations but also uncovers potential relationships hidden between features, thereby improving the model's predictive ability.

[0070] In the above scheme, the multi-head self-attention mechanism is beneficial for capturing a wider range of relevant features, increasing the model's expressive power. Using the multi-head self-attention mechanism to model multiple potential user interests and learn the potential relationships between items at the same time makes interest mining more diverse.

[0071] Furthermore, the user interest model includes a gated recurrent unit module and an attention module. The process of generating user interest features from user time-series data through the user interest model includes:

[0072] The user time-series data is input into the gated loop unit module, the gated loop unit module outputs the user interest features as a hidden state, and the attention module controls the updating of the hidden state by the gated loop unit module.

[0073] Furthermore, the hidden state includes the previous hidden state, the current hidden state, and the transitive hidden state. The process of updating the hidden state by the gated loop unit module through the attention module includes:

[0074] The current hidden state and the target vector are used by the attention module to generate attention weights;

[0075] The transitive hidden state is generated based on the attention weight, the previous hidden state, and the current hidden state, and then the transitive hidden state is passed to the gated loop unit module at the next time step.

[0076] Specifically, the gated recurrent unit (GRU) module of the user interest model is:

[0077]

[0078] z t =σ(W z e t +U z h t-1 +b z )

[0079]

[0080] r t =σ(W r e t +U r h t-1 +b r )

[0081] In the formula, h t h t-1 h t'' represents the activation state at time t, the activation state at time t-1, and the updated value of the activation state at time t, respectively. t e t r t Let W represent the update gate, the input representation, and the reset gate at time t, respectively, where σ is the Sigmoid activation function, and W... z U z W h U h W r U r Both are learnable weight vectors, b z b h b r All of these are learnable bias terms.

[0082] Furthermore, an objective function based on the cross-entropy loss term is constructed, which is used for the optimization prediction of the gated recurrent unit module, specifically as follows:

[0083]

[0084] In the formula, Attention t This represents the attention module, softmax(e t ) represents the softmax activation function controlling the input at time t, represented by e. t Whether to update, h t Let represent the activation state at time t, and W represent the weight vector.

[0085] In the above scheme, the weights of each behavior are adaptively learned through an attention mechanism in the improved gating loop unit, which achieves the latest changes that better express user interests.

[0086] Furthermore, the process of determining predicted interest items by using a project interest prediction model to analyze educational resource characteristics and user interest characteristics includes:

[0087] Construct a loss function based on the negative logarithm function and the probability of accessing interest items, wherein the loss function is used for prediction by the item interest prediction model;

[0088] The educational resource features, user interest features, and target project data are input into the project interest prediction model. The project interest prediction model outputs the predicted interest project access probability and selects the predicted interest project from the target project data based on the predicted interest project access probability.

[0089] Specifically, the loss function is:

[0090]

[0091] In the formula, Loss target Let be the loss function, where x represents the input educational resource features and user interest features, y represents the user's access to interest items in the training set, 1-y represents the user's non-access to interest items in the training set, N is the number of samples in the dataset, and O(x) represents the model's output predicted probability of interest item access. Therefore, it increases the accuracy of predicting both the probability of interest item access and the probability of interest item access. This makes the model's learned representation more similar to that of accessed interest items and less similar to that of non-accessed interest items.

[0092] Furthermore, the project interest prediction model includes a fully connected layer, at least two layers of Dice activation functions and Sigmoid activation functions. The process of inputting the educational resource features, the user interest features, and the target project data into the project interest prediction model, and the model outputting a predicted probability of accessing the interest project, includes:

[0093] The educational resource features, user interest features, and target project data are sequentially passed through the fully connected layer, at least two layers of Dice activation functions, and Sigmoid activation functions to output the predicted access probability of the interest project.

[0094] Preferably, the project interest prediction model is configured with two fully connected layers, two Dice activation functions, and one Sigmoid activation function. Therefore, the Dice activation function optimizes the training stability of the hidden layers, the two fully connected layers capture complex features, and the Sigmoid function is used to construct the probabilistic probability of predicting the access probability of the predicted interest items, achieving a balance between gradient control, nonlinear modeling, and prediction accuracy.

[0095] In the above scheme, the Dice activation function is used by the multilayer perceptron to enhance the expressive power of low-frequency features and further perceive the evolution of interest, thereby reducing the impact of noise input to the multilayer perceptron.

[0096] Furthermore, the user access resource data includes user information data, user learning behavior data, and target project data;

[0097] The user time-series data includes user historical behavior time-series data, user information data, and target project data.

[0098] Specifically, the user access resource data is represented as: x1∈(X a ,X b ,X c ), where x1 represents user access to resource data, X a ,X b ,X cThese represent user information data, user learning behavior data, and target project data, respectively.

[0099] Specifically, the user time-series data is represented as: x∈(X d ,X a ,X c ), where x represents user time-series data, X d ,X a ,X c These represent the user's historical behavior time-series data, the user information data, and the target project data, respectively.

[0100] Furthermore, the educational resource database includes a cold data table and at least two hot data tables. The process of storing the data corresponding to the predicted interest items as hot data in the educational resource database includes:

[0101] The data corresponding to the predicted interest items are classified according to their access frequency and their correlation with the currently used item set, and stored in multiple hot data tables respectively.

[0102] The cold data and the hot data are periodically converted using a thread pool.

[0103] Specifically, the access frequency of the data corresponding to the predicted interest items is calculated using a time-decay weighted average to highlight recent popularity:

[0104]

[0105] In the formula, CF(k) represents the access frequency of the data corresponding to the predicted interest item, and c t (k) represents the predicted number of visits to interest item k at time t, where T now ,T first This represents the time when the data corresponding to the predicted interest item is accessed is currently accessed, and the time when the data corresponding to the predicted interest item is accessed for the first time. α represents an adjustment coefficient, preferably 0.3. Therefore, the more times the data is accessed or the closer the access times are, the higher the access frequency value becomes.

[0106] Specifically, the correlation between the data corresponding to the predicted interest item and the currently used item set is calculated using cosine similarity:

[0107]

[0108] In the formula, CR(k) represents the correlation degree, λ represents the weight (preferably 0.5), co(k,U) represents the number of times the currently predicted interest item k and the currently used item set U co-occur, and max k co(k,U) represents the maximum number of times all predicted interest items co-occur with the currently used item set U, cos(vk ,v U ) represents the embedding vector v corresponding to the predicted interest item k. k The average embedding vector v of the currently used itemset U U The cosine similarity.

[0109] The relevance and access frequency are normalized using the maximum and minimum values, respectively, and a comprehensive popularity value is calculated:

[0110]

[0111] In the formula, S(k) represents the comprehensive popularity value, CF(k) represents the access frequency of the data corresponding to the predicted interest item, CR(k) represents the correlation degree, and β represents the exponential weight, preferably 0.4.

[0112] Based on the overall popularity value, the data corresponding to the predicted interest items are categorized and stored in a hot data table:

[0113]

[0114] In the formula, Bucket l This represents the hot data table corresponding to the data storage of the predicted interest item, where l represents the hot data table level number, L represents the total number of hot data table level numbers, and S(k) represents the comprehensive popularity value of the predicted interest item k. max (k) represents the maximum comprehensive popularity value of predicted interest item k. Therefore, it is possible to classify and store the data corresponding to the predicted interest items. Preferably, L is 3, that is, the hot data is divided into a high-popularity hot data table 1, a medium-popularity hot data table 2, and a low-popularity hot data table n according to the above formula.

[0115] The proposed hot and cold data separation scheme aims to improve the system's data access efficiency. By placing a small amount of frequently accessed data in a hot table and storing a large amount of less frequently accessed historical data in a cold table, users can prioritize searching in the hot table when accessing data. Since the hot table contains less data, this method can significantly improve data access efficiency.

[0116] like Figure 4 and 5 As shown, the process of loading hot or cold data on an online education platform is as follows:

[0117] Data lookup: When the lookup function is invoked, the system first searches the hot data table. Because the amount of data in the hot data table remains relatively constant, the lookup process is fast. For example... Figure 4As shown, during the data search process, the database first accesses hot data table 1, then sequentially accesses hot data table 2 until the data is found in hot data table n or a cold data table. Hot data table 1 to hot data table n are sequentially connected. Therefore, more efficient storage and access mechanisms can be used for hot data tables to improve the overall performance of the database. Figure 5 As shown, the database performs multi-threaded parallel access to hot data tables 1 to n, which can further improve access efficiency.

[0118] Data Transfer: A thread pool is used to implement the data transfer functionality in a single thread. During execution, the task is added to the thread pool as a task via an Executor call. The transfer functionality includes: passing in the data whose operation time needs to be changed, copying the data to the hot data table, updating its operation time, and simultaneously deleting the corresponding data from the cold data table. Figure 4 As shown, the cold data table and hot data table 2 are transferred using a thread pool. First, the data is written to hot data table 2, and then the transfer between the hot data tables is performed. Figure 5 As shown, the cold data table and the hot data table group are transferred together through a thread pool, and dynamic allocation is performed when writing to the hot data table group.

[0119] Step-by-step scanning of hot data tables and timed copying: A scheduled task is used to check the amount of data in the data table and all data, and the most recent update time is in the early morning to ensure the freshness of the data.

[0120] like Figure 4 and 5 The thread pool 1 and thread pool 2 shown can dynamically allocate thread resources according to the actual load.

[0121] To simulate a real-world environment, data simulation and stress testing were performed on the database tables. The test plan was based on a multiple-choice question table from an online education management platform. The testing process consisted of the following three steps: selecting the test table and executing the test statements; using the statements to perform the test, gradually increasing the data volume until it reached tens of millions; and gradually increasing the data volume from hundreds of thousands to tens of millions, observing the changes in query performance.

[0122] In this embodiment, the online education platform accurately identifies content of interest to users through a multi-head self-attention mechanism, an improved gated recurrent unit, and a multilayer perceptron. This content is then stored and managed separately. However, accessing this content requires the platform to access and load large amounts of data from a database, resulting in long waiting times and a poor user experience. The multi-head self-attention mechanism helps capture a wider range of relevant features, increasing the model's expressive power. It models various potential user interests and the potential relationships between educational resources at the same time, diversifying interest discovery. The improved gated recurrent unit adaptively learns the weights of each behavior through an attention mechanism, better representing the latest changes in user interests. The multilayer perceptron uses the Dice activation function to enhance the expressive power of low-frequency features, further perceiving the evolution of interests and reducing the impact of noise input to the multilayer perceptron.

[0123] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0124] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A data storage management method for an online education platform, characterized in that, Online education platforms are configured with educational resource databases for separate storage of cold and hot data, including: The user access resource data of the online education platform is used to generate educational resource interest features through an educational resource model, wherein the educational resource model is constructed based on a multi-head self-attention mechanism. The user interest data of the online education platform is used to generate user interest features through a user interest model, wherein the user interest model is constructed based on an improved gated loop unit. The educational resource interest features and the user interest features are used to determine the predicted interest items through a project interest prediction model, wherein the project interest prediction model is based on a multilayer perceptron. The data corresponding to the predicted interest items are stored as hot data in the educational resource database; The educational resource database includes a cold data table and at least two hot data tables. The process of storing the data corresponding to the predicted interest items as hot data in the educational resource database includes: The process of calculating the access frequency of the data corresponding to the predicted interest items using time decay weighting is as follows: ； In the formula, This indicates the access frequency of the data corresponding to the predicted interest item. This indicates the predicted number of visits to interest item k at time t. These represent the time when the data corresponding to the predicted interest item is accessed currently and the time when the data corresponding to the predicted interest item is accessed for the first time, respectively. Indicates the adjustment factor; The process of calculating the correlation between the data corresponding to the predicted interest items and the currently used item set using cosine similarity is as follows: ； In the formula, Indicates the degree of correlation. Indicates weight, This indicates the number of times the currently predicted interest item k co-occurs with the currently used item set U. This represents the embedding vector corresponding to the predicted interest item k. The average embedding vector of the currently used itemset U Cosine similarity; After normalizing the relevance and access frequency using the maximum and minimum values ​​respectively, the overall popularity value is calculated. The process of calculating the overall popularity value is as follows: ； In the formula, This indicates the overall popularity score. This indicates the access frequency of the data corresponding to the predicted interest item. Indicates the degree of correlation. Indicates the index weight; The process of classifying and storing the data corresponding to the predicted interest items in the hot data table based on the comprehensive popularity value is as follows: ； In the formula, This is a hot data table representing the data storage corresponding to the predicted interest item. This indicates the heat data table level designation, where L represents the total number of heat data table level designations. This represents the overall popularity value of the predicted interest item k. This represents the maximum overall popularity value of the predicted interest item k; The cold data and the hot data are periodically converted using a thread pool.

2. The online education platform data storage management method according to claim 1, characterized in that, The educational resource model includes a multi-head self-attention mechanism, an attention mechanism, and a pooling layer. The process of generating educational resource interest features from user access data of online education platforms through the educational resource model includes: The user access resource data is used to generate a resource feature head vector through the multi-head self-attention mechanism; The resource feature header vector is used to generate target weights through the attention mechanism; The target weights are used to generate the educational resource interest features through the pooling layer.

3. The online education platform data storage management method according to claim 2, characterized in that, The educational resource model also includes an auxiliary loss term, which is used to learn the feature representation of the user's access to resource data. The auxiliary loss term is constructed based on the log-likelihood form and the activation function.

4. The online education platform data storage management method according to claim 1, characterized in that, The user interest model includes a gated recurrent unit module and an attention module. The process of generating user interest features from user time-series data through the user interest model includes: The user time-series data is input into the gated loop unit module, the gated loop unit module outputs the user interest features as a hidden state, and the attention module controls the updating of the hidden state by the gated loop unit module.

5. The online education platform data storage management method according to claim 4, characterized in that, The hidden state includes the previous hidden state, the current hidden state, and the transitive hidden state. The process of updating the hidden state by the gated loop unit module through the attention module includes: The current hidden state and the target vector are used by the attention module to generate attention weights; The transitive hidden state is generated based on the attention weight, the previous hidden state, and the current hidden state, and then the transitive hidden state is passed to the gated loop unit module at the next time step.

6. The online education platform data storage management method according to claim 4, characterized in that, A target function based on the cross-entropy loss term is constructed, and the target function is used for the optimization prediction of the gated recurrent unit module.

7. The online education platform data storage management method according to claim 1, characterized in that, The process of determining predicted interest items by using a project interest prediction model to analyze the interest characteristics of educational resources and users includes: Construct a loss function based on the negative logarithm function and the probability of accessing interest items, wherein the loss function is used for prediction by the item interest prediction model; The educational resource interest features, the user interest features, and the target project data are input into the project interest prediction model. The project interest prediction model outputs the predicted interest project access probability and selects the predicted interest project from the target project data based on the predicted interest project access probability.

8. The online education platform data storage management method according to claim 7, characterized in that, The project interest prediction model includes a fully connected layer, at least two layers of Dice activation functions and Sigmoid activation functions. The process of inputting the educational resource interest features, the user interest features, and the target project data into the project interest prediction model, and the project interest prediction model outputting the predicted probability of accessing interest projects, includes: The educational resource interest features, the user interest features, and the target project data are sequentially passed through the fully connected layer, at least two layers of Dice activation functions, and Sigmoid activation functions to output the predicted access probability of the interest project.

9. The online education platform data storage management method according to any one of claims 1 to 8, characterized in that, The user access resource data includes user information data, user learning behavior data, and target project data; The user time-series data includes user historical behavior time-series data, user information data, and target project data.

Images