Page preloading prediction method and device, equipment and storage medium

Abstract

The application discloses a page preloading prediction method and device, equipment and a storage medium, and relates to the technical field of front-end pages. The method comprises the following steps: when a user accesses a current page, sending a page prediction request to a cloud device, so that the cloud device performs page loading prediction on the user based on a first prediction model through group behavior characteristics, and obtains a first page prediction result; determining a source page before the user accesses the current page, and extracting source page characteristics and current page characteristics to perform page loading prediction on the user, and obtaining a second page prediction result; and determining a preloaded page after the current page according to the first page prediction result and the second page prediction result. The first page prediction result reflecting the group behavior trend is combined with the second page prediction result reflecting the individual behavior preference to determine the preloaded page after the current page, the adaptability of individual dynamic changes is improved, and the page preloading effect is improved.

Description

Technical Field

[0001] This application relates to the field of front-end page technology, and in particular to a page preloading prediction method, apparatus, device and storage medium. Background Technology

[0002] Currently, in mobile embedded application scenarios (such as product detail pages on large e-commerce platforms, content landing pages of news and information software, and various lightweight applications with user behavior paths), preloading technology is generally used to optimize user experience and reduce the white screen waiting time when switching pages. Existing page preloading decisions are generally based on prediction using a cloud-based centralized model. For example, the cloud collects massive amounts of user click and redirect sequences, uses deep learning models to train group behavior trends, and thus predicts the page a user might visit next and loads resources in advance.

[0003] However, because cloud-based models rely on large-scale statistical patterns, their outputs represent the average preferences of most people, making it difficult to adapt to individual interests and dynamic changes, thus reducing the effectiveness of preloading. For example, a user might frequently browse tech news during weekday lunch breaks but switch to short video pages frequently in the evening. The cloud model, based on statistics of all users, might categorize this user as a news-oriented user and continuously preload article pages, leading to a lower preloading hit rate in the evening and ultimately significantly reducing the effectiveness of preloading. Summary of the Invention

[0004] The main purpose of this application is to provide a page preloading prediction method, apparatus, device and storage medium, which aims to solve the technical problem that existing cloud prediction models are difficult to adapt to changes in individual dynamic behavior, resulting in a reduction in page preloading effectiveness.

[0005] To achieve the above objectives, this application proposes a page preload prediction method, the method comprising: When a user visits the current page, a page prediction request is sent to a cloud device so that the cloud device can predict the page loading of the user using a first prediction model and obtain a first page prediction result. The first prediction model makes predictions based on group behavior characteristics. Determine the source page before the user accesses the current page, and extract the source page features corresponding to the source page and the current page features corresponding to the current page; Based on the source page features and the current page features, page loading prediction is performed for the user to obtain the second page prediction result; Based on the prediction results of the first page and the prediction results of the second page, the preloaded page following the current page is determined.

[0006] In one embodiment, before the step of sending a page prediction request to a cloud device when a user accesses the current page, the method further includes: Obtain the user's historical page click information, and extract at least two previous page features and the next page features corresponding to each of the previous page features from the historical page click information; The model is trained based on the features of each preceding page and each following page to obtain the second prediction model; The step of predicting page load for the user based on the source page features and the current page features to obtain a second page prediction result includes: Based on the source page features and the current page features, the second prediction model is used to predict page loading for the user, resulting in a second page prediction result.

[0007] In one embodiment, the step of determining the preloaded page following the current page based on the first page prediction result and the second page prediction result further includes: Upon detecting a new page, extract the new page features of the new page; By adding weight tensors corresponding to the new page features to the input and output layers of the second prediction model, an extended second prediction model is obtained. The extended second prediction model is incrementally trained based on the new page features to obtain updated model parameters, and the updated model parameters are stored in IndexedDB. When the user visits the current page again, the IndexedDB is used to predict page load for the user.

[0008] In one embodiment, the step of predicting page load for the user using the IndexedDB when the user visits the current page again includes: When the user visits the current page again, the historical version number of the second prediction model is obtained from the IndexedDB; Compare the historical version number with the version number of the second prediction model itself; If the historical version number is the same as the current version number, then the page loading prediction for the user is performed using the second prediction model; If the historical version number is inconsistent with the current version number, the second prediction model is updated using the IndexedDB, and the updated second prediction model is used to predict page load for the user.

[0009] In one embodiment, the step of predicting page load for the user based on the source page features and the current page features using the second prediction model to obtain a second page prediction result includes: The second prediction model is used to embed the source page features and the current page features respectively to obtain the corresponding source embedding vector and current embedding vector. Obtain the timestamp of the user's access to the current page, and perform attenuation processing on the timestamp to obtain the time weight; The source embedding vector, the current embedding vector, and the temporal weight are concatenated through a fully connected layer to obtain a fused vector; The fusion vector is linearly processed to obtain the page probability after the current page, and the page probability is used as the second page prediction result.

[0010] In one embodiment, the step of determining the preloaded page following the current page based on the first page prediction result and the second page prediction result includes: Obtain the first prediction accuracy of the first prediction model, and weight the first prediction accuracy with the page probability in the first page prediction result to obtain the first page probability sequence; Obtain the second prediction accuracy of the second prediction model, and weight the second prediction accuracy with the page probability in the second page prediction result to obtain the second page probability sequence; The first page probability sequence and the second page probability sequence are merged and rearranged to obtain the page prediction probability distribution; The page with the highest probability from the predicted probability distribution of the pages is selected as the preloaded page after the current page.

[0011] In one embodiment, after determining the preloaded page following the current page based on the first page prediction result and the second page prediction result, the method further includes: The preloaded page is pre-cached locally; Upon detecting the user's page redirection request, extract the first page address corresponding to the page redirection request and the second page address corresponding to the preloaded page; Compare the first page address and the second page address; If the first page address and the second page address are the same, the page redirection request is intercepted, and the pre-cached preloaded page is displayed to the user.

[0012] Furthermore, to achieve the above objectives, this application also proposes a page preload prediction device, the device comprising: The cloud module is used to send a page prediction request to a cloud device when a user visits the current page, so that the cloud device can predict the page loading of the user through a first prediction model and obtain a first page prediction result. The first prediction model makes predictions based on group behavior characteristics. The feature module is used to determine the source page before the user accesses the current page, and to extract the source page features corresponding to the source page and the current page features corresponding to the current page; The prediction module is used to predict page loading for the user based on the source page features and the current page features, and obtain a second page prediction result; The page module is used to determine the preloaded page following the current page based on the first page prediction result and the second page prediction result.

[0013] In addition, to achieve the above objectives, this application also proposes a page preload prediction device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the page preload prediction method as described above.

[0014] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the page preload prediction method described above.

[0015] One or more technical solutions proposed in this application have at least the following technical effects: The page preloading prediction method of this application includes: when a user accesses the current page, sending a page prediction request to a cloud device, so that the cloud device performs page loading prediction for the user through a first prediction model to obtain a first page prediction result, wherein the first prediction model makes predictions based on group behavior features; determining the source page before the user accesses the current page, and extracting the source page features corresponding to the source page and the current page features corresponding to the current page; performing page loading prediction for the user based on the source page features and the current page features to obtain a second page prediction result; and determining the preloaded page after the current page based on the first page prediction result and the second page prediction result.

[0016] This application first triggers a first prediction model on the cloud device when a user accesses the current page to obtain a first page prediction result reflecting the group's behavioral trends. Then, it determines the source page of the current page and extracts the features of the source page and the current page to perform on-device prediction based on the individual-level page features, obtaining a second page prediction result. Compared to existing methods that rely on cloud devices for preloading decisions, this application combines the first page prediction result reflecting the group's behavioral trends with the second page prediction result reflecting individual behavioral preferences to jointly determine the preloaded page after the current page. This improves the adaptability to dynamic changes in individuals, reduces the mismatch between the preloaded page and the actual page the user jumps to, and thus enhances the page preloading effect. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating an embodiment of the page preloading prediction method in this application. Figure 2 The overall flowchart of end-to-cloud collaboration provided in this application; Figure 3 The overall architecture diagram of cloud-based prediction provided in this application; Figure 4 The cloud device model architecture diagram provided in this application; Figure 5 A schematic diagram illustrating the model accuracy of the cloud-based device provided in this application; Figure 6 This is a flowchart illustrating Embodiment 2 of the page preloading prediction method for this application. Figure 7 The overall architecture diagram for client-side prediction provided in this application; Figure 8 A schematic diagram illustrating the model accuracy of the client-side device provided in this application; Figure 9 A schematic diagram of the model weight expansion provided in this application; Figure 10 This is a flowchart illustrating the third embodiment of the page preloading prediction method in this application. Figure 11A diagram illustrating the combined accuracy of the data provided in this application; Figure 12 This is a block diagram of the module structure of the page preloading prediction device in an embodiment of this application; Figure 13 This is a schematic diagram of the hardware operating environment involved in the page preloading prediction device in the embodiments of this application.

[0020] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0022] The main solution proposed in this application is as follows: Currently, in mobile embedded application scenarios (such as product detail pages of large e-commerce platforms, content landing pages of news and information software, and various lightweight applications with user behavior paths), preloading technology is generally used to optimize user experience and reduce the white screen waiting time during page transitions. Existing methods for making page preloading decisions are generally based on prediction using a centralized cloud model. For example, the cloud collects massive amounts of user click and redirect sequences, uses deep learning models to train group behavior trends, and thus predicts the page a user might visit next and loads resources in advance.

[0023] However, because cloud-based models rely on large-scale statistical patterns, their outputs represent the average preferences of most people, making it difficult to adapt to individual interests and dynamic changes, thus reducing the effectiveness of preloading. For example, a user might frequently browse tech news during weekday lunch breaks but switch to short video pages frequently in the evening. The cloud model, based on statistics of all users, might categorize this user as a news-oriented user and continuously preload article pages, leading to a lower preloading hit rate in the evening and ultimately significantly reducing the effectiveness of preloading.

[0024] To address the aforementioned issues, this application provides a page preloading prediction method. First, when a user accesses the current page, a first prediction model on a cloud device is triggered to perform a prediction, obtaining a first page prediction result reflecting group behavior trends. Then, the source page of the current page is determined, and features of both the source page and the current page are extracted. Based on these individual-level page features, a second page prediction result is obtained. Compared to existing methods that rely on cloud devices for preloading decisions, this application combines the first page prediction result reflecting group behavior trends with the second page prediction result reflecting individual behavioral preferences to jointly determine the preloaded page following the current page. This improves adaptability to dynamic individual changes and reduces the mismatch between the preloaded page and the user's actual page, thereby enhancing the page preloading effect.

[0025] It should be noted that the executing entity of this application embodiment can be an electronic device with data processing, network communication, and program execution functions, such as a smartphone, tablet computer, smart wearable device, or other page preloading prediction device capable of executing the page preloading prediction method of this application. This embodiment does not limit this. The following uses a page preloading prediction device (hereinafter referred to as the prediction device) as an example to describe this embodiment and the following embodiments.

[0026] Based on this, this application proposes a page preloading prediction method according to a first embodiment, referring to... Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the page preloading prediction method of this application. In this embodiment, the page preloading prediction method may include steps S10 to S40: Step S10: When a user accesses the current page, a page prediction request is sent to a cloud device so that the cloud device can predict the page loading of the user using a first prediction model and obtain a first page prediction result. The first prediction model makes predictions based on group behavior characteristics.

[0027] It should be noted that the page prediction request can be a data packet sent by the prediction device to the cloud device to trigger the page loading prediction function on the cloud side. This data packet may contain the page identifier and user identifier of the currently accessed page, which are used by the cloud device to identify the current prediction context. The cloud device can be a server device or server cluster deployed on the network side. Compared to the prediction device mentioned above, the cloud device has the capability to process large-scale group behavior data.

[0028] Understandably, the first prediction model could be a deep learning model deployed on a cloud device, built on the Transformer architecture, used to learn from the historical page access sequences of a group of users and capture group behavior trends, thereby making probability predictions for the next hop page.

[0029] For details, please refer to Figure 2 , Figure 2 The overall flowchart for end-to-cloud collaboration provided in this application illustrates that this embodiment employs an end-to-cloud collaborative architecture for page preloading prediction, including cloud-based prediction and client-side prediction. For cloud-based prediction, please refer to... Figure 3 , Figure 3 The overall architecture diagram of cloud prediction provided in this application consists of two parts: behavior sequence processing and cloud sequence prediction, which realizes the function of capturing group behavior trends.

[0030] In practical use, when the prediction device detects a user's interaction with the current page, it triggers a page prediction request. This request is then sent to the cloud device via the network to instruct it to initiate the prediction process. Upon receiving the page prediction request, the cloud device continuously consumes real-time data from the event tracking message queue through its behavior sequence processing module. After validating the consumed event tracking data, it constructs a behavior sequence and, using an ID mapping dictionary, converts the page identifier into a model-recognizable vector index. Finally, the processed group behavior features are input into the first prediction model. The first prediction model performs inference calculations based on these group behavior features and outputs the first page prediction result for the current user's current page access context. Finally, the first page prediction result is sent back to the prediction device.

[0031] For details, please refer to Figure 4 , Figure 4 The diagram shows the model architecture of the cloud device provided in this application. The first prediction model, also known as the cloud sequence prediction model, is a self-designed model based on the Transformer architecture. Figure 4As shown, the first prediction model first extracts features from each page access record in the group behavior features to obtain user behavior input features. Then, it uses the model's page matrix and time matrix to multiply the behavior features and page features in the user behavior input features respectively. Simultaneously, the model generates corresponding learnable position features for each position in the sequence (the position features introduced here are learnable and can compensate for the Transformer's dependence on absolute position), and concatenates the position features with the user behavior input features through multi-dimensional expansion to form the concatenated features. Subsequently, a fully connected layer (i.e., FC1) is used to perform linear transformation calculation on the concatenated features to obtain the calculation result. To facilitate the processing of long sequence information, the sequence features can then be fed into an encoding module consisting of 4 Transformer encoders. Each Transformer encoder uses an 8-head attention mechanism to perform attention calculation on the sequence features and performs non-linear activation processing through the ELU activation function to extract the dependencies and access patterns between positions within the sequence to enhance the model's expression, thereby obtaining the sequence features. Finally, the sequence features are normalized (i.e., normalized) to convert the output value into a probability distribution, and then linearly calculated through a fully connected layer (i.e., FC2) to obtain the prediction result for the first page.

[0032] It's important to note that cloud-based sequence prediction models require rapid response times for prediction results in practical applications. Besides demanding a streamlined model parameter set, the main challenge lies in acquiring user behavior sequences in real-time to serve as the context for prediction. Therefore, it's necessary to consume event tracking data in real-time, validate the data for validity, and then construct behavior sequences before storing them in a queue. With hundreds of millions of event tracking data points generated daily, it's crucial to retain only data from potential target user groups to ensure efficient use of storage resources. Therefore, while consuming event tracking data, a scheduled task dynamically maintains the active user table in the queue, assessing, cleaning, and updating group data. The potential target user table is also periodically updated to a memory cache to reduce the impact on the queue cluster. Furthermore, data is padded to the end of the sequence before the model processes it, further reducing information loss and ultimately achieving low-latency, high-concurrency user behavior sequence collection.

[0033] Because event tracking data can lose information about user behavior such as entering a page, returning, and then entering another page, a significant portion of the initial training data for cloud-based models contains erroneous data. This embodiment designs a method where, each time a prediction is triggered, the user's next page is recorded in a data table. Once sufficient data has accumulated, all actual next-page tracking data is retrieved to filter a new batch of sequence training data, cleaning up the stale data and improving model performance. Figure 5 As shown, Figure 5This is a schematic diagram of the model accuracy of the cloud device provided in this application; after the training data has undergone data cleaning, supplementation, sampling and other operations and multiple rounds of training, the preferred accuracy of the cloud sequence prediction model can reach 65%.

[0034] Step S20: Determine the source page before the user accesses the current page, and extract the source page features corresponding to the source page and the current page features corresponding to the current page.

[0035] It should be noted that the current page can be the webpage or application interface that the device is currently displaying to the user and that the user is browsing. The source page can be the previous page that the user viewed before accessing the current page, i.e., the original page the user was on when performing the navigation operation.

[0036] Understandably, source page characteristics can be information data used to uniquely identify and describe the attributes of the source page, such as the source page's page identifier, page type, and the functional module to which the page belongs. Current page characteristics can also be information data used to uniquely identify and describe the aforementioned attributes of the current page, such as the current page's page identifier, page type, and the functional module to which the page belongs.

[0037] In practical use, after successfully sending a page prediction request to the cloud device, the prediction device will simultaneously initiate the page information extraction process on the local client side. Specifically, by listening for user-triggered page redirection events or page load completion events, it obtains the user's access path record before accessing the current page from the context information carried by the events, parses and determines the source page. Then, it extracts the source page features from the source page's page metadata and the current page features from the current page's page metadata.

[0038] Step S30: Based on the source page features and the current page features, perform page loading prediction for the user to obtain the second page prediction result.

[0039] It should be understood that the second page prediction result can be the prediction output obtained by the prediction device after performing page loading prediction on the local end. This prediction output is different from the first page prediction result returned by the cloud device and is used to reflect the prediction judgment of the next hop page based on the individual user's access behavior locally.

[0040] In practical use, after extracting the aforementioned source page features and current page features, the prediction device can use these features as input data for local prediction to perform inference calculations, thereby assessing the probability of a user jumping from the current page to various candidate pages. After completing the inference calculations, a prediction output containing the probability distribution information of each candidate page is generated, and this prediction output is determined as the second page prediction result.

[0041] Step S40: Based on the first page prediction result and the second page prediction result, determine the preloaded page following the current page.

[0042] It's important to note that a preloaded page is a page whose data or resources are pre-loaded based on predictions during the user's current page visit, before the user actually navigates to that page. A preloaded page is different from the current page the user is currently viewing, and also different from the source page the user has already visited; it refers to the next page the user is most likely to visit after the current page.

[0043] In practical use, such as Figure 2 As shown, after receiving the first page prediction result from the cloud device and the second page prediction result generated locally, the prediction device can make a weighted decision based on the first and second page prediction results to determine which page should be preloaded after the current page. Finally, the redirect is cached, achieving progressive page preloading. This breaks the limitations of traditional single cloud computing, realizes the dynamic integration of group characteristics and personalized characteristics, and achieves proactive page preloading by predicting user behavior.

[0044] Furthermore, such as Figure 2 As shown, when a user actually navigates to another device, the system can also report the prediction results and perform fine-grained caching and feature collection for continuous training of the cloud model.

[0045] Furthermore, considering the scenario of displaying a preloaded page to the user, in this embodiment, after the step of determining the preloaded page following the current page based on the first page prediction result and the second page prediction result, the method further includes: Step S51: Perform local pre-caching of the preloaded page.

[0046] Step S52: If the user's page redirection request is detected, extract the first page address corresponding to the page redirection request and the second page address corresponding to the preloaded page.

[0047] It should be noted that a page redirection request can be a network request generated by the routing module of the predictive device or the browser engine after the user performs an interactive operation such as clicking a link, clicking a button, or swiping back on the current page, and is used to request the loading of the target page.

[0048] Understandably, the first page address can be a Uniform Resource Locator (URL) carried in the page redirection request, representing the target page the user actually intends to access. The second page address can be a URL corresponding to a preloaded page.

[0049] Step S54: If the first page address and the second page address are the same, then the page redirection request is intercepted, and the pre-cached preloaded page is displayed to the user.

[0050] Specifically, after determining the preloaded page, the prediction device initiates an asynchronous resource retrieval request to the server address of the preloaded page, downloading and storing the required HTML document and related resources in the local cache area. It then continuously monitors user interactions to detect if the user has triggered a page redirection request. If a page redirection request is detected, the device extracts the first page address corresponding to the target page the user actually requests from the request parameters or routing information of the page redirection request. Simultaneously, it reads the second page address corresponding to the previously determined and cached preloaded page from local storage. Next, the first and second page addresses are compared to determine if they are identical. If they match, it means the page the user actually wants to access is the pre-cached preloaded page, triggering an interception mechanism to block the page redirection request and prevent it from re-initiating resource loading on the network side. Finally, the device directly reads the pre-cached preloaded page resource data from the local cache and immediately renders and displays the pre-cached preloaded page to the user, allowing the browser to display the page information immediately.

[0051] In this embodiment, by using the aforementioned progressive preloading, the pre-cached preloaded page is directly returned to the browser for rendering when the user actually navigates to it. This makes the acquisition of page content completely independent of real-time network requests, so that the page can be displayed the instant the user clicks to navigate. This completely eliminates the white screen delay caused by waiting for network response and resource loading in traditional solutions, realizes the instant presentation of page switching, and greatly improves the user experience.

[0052] This application provides a page preloading prediction method. First, when a user accesses the current page, a first prediction model on a cloud device is triggered to predict the page, obtaining a first page prediction result reflecting group behavior trends. Then, the source page of the current page is determined, and features of both the source page and the current page are extracted. Based on these individual-level page features, a second page prediction result is obtained. Compared to existing methods that rely on cloud devices for preloading decisions, this embodiment combines the first page prediction result reflecting group behavior trends with the second page prediction result reflecting individual behavioral preferences to jointly determine the preloaded page following the current page. This improves adaptability to dynamic individual changes and reduces the mismatch between the preloaded page and the actual page the user jumps to, thereby enhancing the page preloading effect.

[0053] Based on the first embodiment of this application, in the second embodiment of this application, the same or similar content as the above embodiment can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 6 , Figure 6 This is a flowchart illustrating Embodiment 2 of the page preloading prediction method for this application. To predict page loading for users on the client side, such as... Figure 6 As shown, in this embodiment, before the step of sending the page prediction request to the cloud device when the user accesses the current page, the method further includes: Step S01: Obtain the user's historical page click information, and extract at least two previous page features and the next page features corresponding to each previous page feature from the historical page click information.

[0054] It should be noted that historical page click information can be log data of all page jump behaviors triggered by the user during past visits, which are recorded and stored locally by the predictive device.

[0055] Understandably, the previous page feature can be the page identifier or page attribute data that serves as the starting page for the user's navigation in the user's historical page click information; that is, the feature information of the original page where the user was when they initiated the navigation operation from a certain page. The next page feature can be the page identifier of the target page that the user actually arrives at after navigating from the aforementioned previous page, corresponding to the previous page feature.

[0056] Step S02: Train the model based on the features of each preceding page and each following page to obtain the second prediction model.

[0057] It should be understood that model training can be a process of iteratively updating model parameters using labeled training sample data and the backpropagation algorithm.

[0058] The step of predicting page load for the user based on the source page features and the current page features to obtain a second page prediction result includes: Step S31: Based on the source page features and the current page features, perform page loading prediction for the user using the second prediction model to obtain the second page prediction result.

[0059] Specifically, such as Figure 7 As shown, Figure 7 The diagram below shows the overall architecture of the client-side prediction provided in this application. To enable the second prediction model on the client side to function correctly on low-end devices, a simple multilayer neural network model (i.e., a multilayer perceptron) can be built in the user's browser, allowing for training and prediction in a local environment.

[0060] In practical use, various user click operations are typically reported through a data tracking mechanism. Therefore, before deploying local edge prediction functionality, the prediction device can read the user's historical page click information stored in a local persistent database. Then, it extracts at least two previous page features (i.e., features of the previous page) sequentially from the historical page click information, either chronologically or by record entry, and simultaneously extracts the corresponding next page features for each previous page feature, forming at least two sets of previous-next page paired samples. Next, the extracted previous and next page features are used as new training datasets to construct input features and labels for training a local lightweight model. Then, the previous page features are used as model input, and the corresponding next page features are used as the target output. Model training is performed iteratively by calculating the loss function and updating the model weights, ultimately obtaining the trained second prediction model. After training the second prediction model, when performing local prediction, the aforementioned source page features and current page features are used as input data to the second prediction model for inference calculation. The second prediction model outputs the probability distribution of the next page, obtaining the second page prediction result.

[0061] It's important to note that after each training session, training data can be stored in IndexedDB, including the model version, model weights, training data, and a mapping dictionary. This ensures that the client-side model can independently update and continuously optimize itself locally, enabling personalized predictions without relying on a remote server.

[0062] After model structure adjustments, data cleaning, and parameter tuning, the client-side model's training data metrics over the past month are as follows: Figure 8 As shown, Figure 8 This diagram illustrates the model accuracy of the client-side device provided in this application. Over the course of one month, after 12 iterations, both training and validation errors decreased continuously, and the training and validation accuracy tended to stabilize.

[0063] It is worth noting that the first prediction model mentioned above is deployed on cloud devices, and its training data comes from the collective behavioral sequences of a massive number of users. It is a public model shared by all users, reflecting the common access patterns of the group of users. Unlike the first prediction model, the second prediction model is deployed and runs on the prediction device itself, i.e., locally on the user's terminal. Its training data comes from the user's own page browsing history. The training, prediction, and updates of the model are all completed on the user's terminal and are not shared with other users. In other words, each user's terminal maintains an independent and user-specific second prediction model. The second prediction models of different users are isolated from each other in terms of initial parameter values, training samples, and iterative update paths, and do not affect each other. This ensures accurate characterization of individual behavioral patterns while avoiding the data security risks associated with uploading user privacy data to the cloud.

[0064] Furthermore, initially, due to insufficient training data from users, the client-side cannot make predictions and can rely solely on the model on the cloud device. When users collect sufficient training data locally or the collection time reaches a threshold, the client-side model training can be automatically triggered, and the model parameters can be saved to the user's local database for use in the next prediction.

[0065] Furthermore, when a new page appears, in this embodiment, the step of determining the preloaded page after the current page based on the first page prediction result and the second page prediction result further includes: extracting new page features of the new page when a new page is detected; adding weight tensors corresponding to the new page features to the input and output layers of the second prediction model to obtain an extended second prediction model; incrementally training the extended second prediction model based on the new page features to obtain updated model parameters, and storing the updated model parameters in IndexedDB; when the user visits the current page again, using the IndexedDB to predict page loading for the user.

[0066] It should be noted that a new page can be a page that the prediction device detects for the first time during operation and has not yet been included in the page dictionary of the second prediction model, such as a newly launched feature page or a page that has not been visited by users before. New page features can be information data used to uniquely identify the new page.

[0067] Understandably, a weight tensor can be a multidimensional array structure used in a neural network model to store connection weights. In the input layer, it corresponds to the mapping matrix from page identifiers to embedding vectors, and in the output layer, it corresponds to the mapping matrix from hidden layers to the output probabilities of each page category.

[0068] It should be understood that IndexedDB can be a locally stored persistent database used to store parameter files and version information for the second predictive model.

[0069] Specifically, when new pages emerge and the second prediction model cannot recognize the new page IDs, it needs to expand its dimensions to accommodate the ever-growing page dictionary. For example... Figure 9 As shown, Figure 9 The diagram illustrating the model weight expansion provided in this application illustrates how, to preserve knowledge from historical training, a new normal distribution tensor can be randomly generated in both the input and output layers. The original embedding vector dimension remains unchanged, but the number of rows is increased to match the new page dictionary dimension. Then, based on the size of the old weight tensor, data exceeding the old weight dimension is concatenated with the old weights to obtain a new weight tensor. In this way, the probability distribution output by the model can adapt to the new page ID in real time.

[0070] Furthermore, as more pages are visited, the augmented dimensional weights are filled with normally distributed values, and the probability of each page being the next page is known, given the previous and current pages. The final model is stored directly in the user's browser's local IndexedDB, allowing for local predictions without accessing a remote server. The local model is continuously retrained and its data is overwritten and saved back to the original IndexedDB as the training data grows.

[0071] In this embodiment, by adding weight tensors corresponding to the new page to the input and output layers to expand the model dimensions, while retaining the original embedding dimensions and the old weights that have already been trained, the model will not forget the features of the historical pages it has learned while learning the knowledge of the new page. This avoids the waste of computing resources and the problem of knowledge forgetting caused by retraining the entire model due to the emergence of new pages.

[0072] Furthermore, considering model iteration, in this embodiment, the step of predicting page load for the user using the IndexedDB when the user revisits the current page includes: obtaining the historical version number of the second prediction model from the IndexedDB when the user revisits the current page; comparing the historical version number with the current version number of the second prediction model; if the historical version number matches the current version number, then predicting page load for the user using the second prediction model; if the historical version number does not match the current version number, then updating the second prediction model using the IndexedDB, and predicting page load for the user using the updated second prediction model.

[0073] It should be explained that the historical version number can be a version identifier information stored in IndexedDB that is associated with the last saved model parameter file, used to characterize the version status of the second prediction model parameters stored in IndexedDB.

[0074] Understandably, the version number can be the version identifier information carried by the second prediction model program file currently loaded or deployed in the memory of the prediction device, used to characterize the version status of the currently executable model.

[0075] In this embodiment, the prediction device first obtains the version number of the edge model stored in the local IndexedDB and compares it with the current version number of the model. If they match, it indicates that the local model is the latest version, and the local page mapping map and the complete model weight file are loaded directly from the IndexedDB. Then, it directly enters the ready state for prediction. If they do not match, it indicates that the local model has expired or the current model file has been updated, and the update process is triggered. At this time, the local model version number is first updated to the latest version number specified in the file, and then the old page mapping map, historical training data, and old model weight file stored locally are cleared in sequence. Finally, a new model structure matching the current version is recreated in the edge memory, waiting for subsequent incremental training to fill in the data before it can be used for prediction. The whole process ensures that the edge model avoids compatibility issues between the old and new versions by full cleanup and reconstruction when the version changes, ensuring that only one valid model is retained on the same user device, preventing confusion or loading anomalies caused by data residue.

[0076] Based on the first and / or second embodiments of this application, in the third embodiment of this application, the content that is the same as or similar to that in embodiments one and two above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 10 , Figure 10 This is a flowchart illustrating Embodiment 3 of the page preloading prediction method for this application. To obtain the aforementioned second page prediction result, as... Figure 10 As shown, in this embodiment, the step of predicting page loading for the user based on the source page features and the current page features using the second prediction model to obtain a second page prediction result includes: Step S311: The source page features and the current page features are embedded using the second prediction model to obtain the corresponding source embedding vector and current embedding vector.

[0077] It should be noted that embedding can be an operation that maps discrete categorical features to continuous dense vector representations. This is achieved by converting high-dimensional, sparse, unique codes into low-dimensional real-valued vectors to capture semantic similarity relationships between pages. The source embedding vector can be the dense vector representation obtained after embedding the aforementioned source page features. The current embedding vector can be the dense vector representation obtained after embedding the aforementioned current page features.

[0078] Step S312: Obtain the timestamp of the user accessing the current page, and perform attenuation processing on the timestamp to obtain the time weight.

[0079] Understandably, a timestamp can be the moment information recorded by the prediction device when a user visits the current page. The decay process can be the process of converting the timestamp into a weighted value related to the time interval. This weighted value decreases as the time interval increases according to a power law of decay coefficient, reflecting the decreasing influence of historical access behavior on the current prediction over time.

[0080] It should be understood that time weight can be a numerical scalar obtained by decaying a timestamp, used to characterize the importance of the current access time relative to the time reference of the model training data.

[0081] Step S313: The source embedding vector, the current embedding vector, and the time weight are concatenated through a fully connected layer to obtain a fused vector.

[0082] It should be noted that a fused vector can be a vector obtained by concatenating multiple vectors or scalars from different sources along the vector dimension.

[0083] Step S314: Perform linear processing on the fusion vector to obtain the page probability after the current page, and use the page probability as the second page prediction result.

[0084] Understandably, linear processing can be a calculation process involving matrix multiplication of the input vector and the weight matrix, plus a bias term, used to map the fused vector to a value in the output space. The page probability can be the probability distribution value of each candidate page output by the second prediction model mentioned above as the user's next-hop destination.

[0085] Specifically, the second prediction model can define three input layers: source page features, current page features, and timestamps. First, the source page features and current page features are embedded into 16-dimensional embedding vectors (i.e., the source embedding vector and the current embedding vector). The timestamp input layer does not directly use the timestamp value, but instead obtains a time weight (i.e., w = t^diff) by raising the power of the decay coefficient, where the power diff is the difference between the timestamp of the training data and the number of days since the training time. Then, the time weight passes through a fully connected layer and is concatenated with the two flattened 16-dimensional embedding vectors to obtain a 33-dimensional fusion vector. To prevent overfitting, the fusion vector first passes through a Dropout layer (a type of anti-overfitting network layer), and then through a 64-dimensional fully connected layer for linear calculation, outputting a probability distribution as the second page prediction result.

[0086] In this embodiment, the discrete features of the source page and the current page are mapped into dense vectors through the embedding layer, preserving the semantic relationship between pages. At the same time, a time weight with attenuation is introduced, so that the behavior closer to the current moment has a greater impact on the prediction result. This enables dynamic perception of the evolution trend of user interests over time, achieving accurate capture and efficient modeling of individual behavioral preferences.

[0087] Furthermore, in order to obtain the aforementioned preloaded page, in this embodiment, the step of determining the preloaded page following the current page based on the first page prediction result and the second page prediction result includes: Step S41: Obtain the first prediction accuracy of the first prediction model, and weight the first prediction accuracy with the page probability in the first page prediction result to obtain the first page probability sequence.

[0088] It should be noted that the first prediction accuracy can be a measure of the accuracy of the first prediction model in predicting pages for a group of users, obtained from the statistical analysis of historical predictions by cloud devices. It reflects the prediction confidence of the first prediction model in the current time period. The first page probability sequence can be an ordered set of adjusted probability values ​​corresponding to each candidate page in the first page prediction results, obtained after weighted processing.

[0089] Step S42: Obtain the second prediction accuracy of the second prediction model, and weight the second prediction accuracy with the page probability in the second page prediction result to obtain the second page probability sequence.

[0090] It should be understood that the second prediction accuracy can be a measure of the historical accuracy of the second prediction model when predicting pages for the current user, obtained from statistics on the local side of the prediction device. It is used to reflect the prediction confidence of the second prediction model on local data. The second page probability sequence can be an ordered set of adjusted probability values ​​corresponding to each candidate page in the second page prediction result, obtained after weighted processing.

[0091] Step S43: Merge and rearrange the first page probability sequence and the second page probability sequence to obtain the page prediction probability distribution.

[0092] Understandably, the predicted probability distribution of a page can be a complete probability distribution that includes each candidate page and its final combined probability value, obtained after merging and rearranging.

[0093] Step S44: Select the page with the highest probability from the predicted probability distribution of the pages as the preloaded page after the current page.

[0094] Specifically, such as Figure 11 As shown, Figure 11 This diagram illustrates the accuracy merging process provided in this application. Due to variations in user behavior data, the performance of the client-side model differs significantly. Therefore, a dynamic weight calculation scheme is needed to balance the weights between the client and cloud for users with varying data conditions. After each training iteration, the client-side model stores a copy of its accuracy data locally. Combined with the cloud-based accuracy data, these two dynamic accuracies (i.e., the first and second prediction accuracies) are used as weights. The page probabilities predicted by the two models are multiplied by these weights to obtain the first and second page probability sequences. The probability values ​​belonging to the same page in the two sequences are then merged, and all merged candidate pages are rearranged in descending order of their final probability values ​​to obtain a complete page prediction probability distribution. Finally, the candidate page with the highest probability value is selected from the page prediction probability distribution and designated as the preloaded page following the current page.

[0095] Because this embodiment uses the accuracy of each of the two prediction models as dynamic weights, it weights the cloud-based group prediction results and the edge-based individual prediction results separately before merging and rearranging them. This results in the prediction results of either model having a higher current accuracy having a greater weight in the final decision, while the influence of the model is weakened when the accuracy is lower. This achieves adaptive adjustment of the prediction credibility of both ends, avoids the problem of overgeneralization in fixed weight or simple averaging schemes, and improves the accuracy of preloaded page selection.

[0096] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the page preloading prediction method of this application. Any simple modifications based on this technical concept are within the scope of protection of this application. Furthermore, all actions involving the acquisition of signals, information, or data in this application are performed in accordance with the relevant data protection regulations of the country where the application is located and with authorization from the owner of the relevant device.

[0097] This application also provides a page preload prediction device. Please refer to... Figure 12 , Figure 12 This is a block diagram of the module structure of the page preloading prediction device according to an embodiment of this application; in this embodiment, the page preloading prediction device includes: The cloud module 101 is used to send a page prediction request to a cloud device when a user accesses the current page, so that the cloud device can predict the page loading of the user through a first prediction model and obtain a first page prediction result. The first prediction model makes predictions based on group behavior characteristics. The feature module 102 is used to determine the source page before the user accesses the current page, and extract the source page features corresponding to the source page and the current page features corresponding to the current page; Prediction module 103 is used to predict page loading for the user based on the source page features and the current page features, and obtain a second page prediction result; Page module 104 is used to determine the preloaded page following the current page based on the first page prediction result and the second page prediction result.

[0098] This embodiment first triggers a first prediction model on the cloud device when a user accesses the current page to make a prediction, obtaining a first page prediction result reflecting the trend of group behavior. Then, it determines the source page of the current page and extracts the features of the source page and the current page to perform on-device prediction based on the page features at the individual level, obtaining a second page prediction result. Compared with existing methods that rely on cloud devices for preloading decisions, this embodiment can combine the first page prediction result reflecting the trend of group behavior with the second page prediction result reflecting individual behavioral preferences to jointly determine the preloaded page after the current page. This improves the adaptability to dynamic changes of individuals, reduces the mismatch between the preloaded page and the actual page the user jumps to, and thus improves the page preloading effect.

[0099] In one implementation, the prediction module 103 is further configured to acquire the user's historical page click information, and extract at least two previous page features and corresponding next page features from the historical page click information; and train a model based on the previous page features and the next page features to obtain a second prediction model. The prediction module 103 is further configured to predict page loading for the user based on the source page features and the current page features using the second prediction model, and obtain a second page prediction result.

[0100] In one implementation, the prediction module 103 is further configured to: extract new page features of the new page when a new page is detected; add weight tensors corresponding to the new page features to the input and output layers of the second prediction model to obtain an expanded second prediction model; perform incremental training on the expanded second prediction model based on the new page features to obtain updated model parameters, and store the updated model parameters in IndexedDB; when the user visits the current page again, use the IndexedDB to predict page loading for the user.

[0101] In one implementation, the prediction module 103 is further configured to, when the user visits the current page again, obtain the historical version number of the second prediction model from the IndexedDB; compare the historical version number with the current version number of the second prediction model; if the historical version number matches the current version number, then predict page loading for the user using the second prediction model; if the historical version number does not match the current version number, then update the second prediction model using the IndexedDB, and predict page loading for the user using the updated second prediction model.

[0102] In one implementation, the prediction module 103 is further configured to embed the source page features and the current page features respectively using the second prediction model to obtain corresponding source embedding vectors and current embedding vectors; obtain the timestamp of the user accessing the current page and perform decay processing on the timestamp to obtain a time weight; concatenate the source embedding vector, the current embedding vector, and the time weight through a fully connected layer to obtain a fusion vector; perform linear processing on the fusion vector to obtain the page probability after the current page, and use the page probability as the second page prediction result.

[0103] In one implementation, the page module 104 is further configured to: obtain a first prediction accuracy of the first prediction model; weight the first prediction accuracy with the page probability in the first page prediction result to obtain a first page probability sequence; obtain a second prediction accuracy of the second prediction model; weight the second prediction accuracy with the page probability in the second page prediction result to obtain a second page probability sequence; merge and rearrange the first page probability sequence and the second page probability sequence to obtain a page prediction probability distribution; and select the page with the highest probability from the page prediction probability distribution as the preloaded page after the current page.

[0104] In one implementation, the page module 104 is further configured to perform local pre-caching of the preloaded page; upon detecting a user's page jump request, extracting the first page address corresponding to the page jump request and the second page address corresponding to the preloaded page; comparing the first page address and the second page address; if the first page address and the second page address are consistent, intercepting the page jump request and displaying the pre-cached preloaded page to the user.

[0105] Other embodiments or specific implementations of the page preloading prediction device in this application can be found in the above-described method embodiments, and will not be repeated here.

[0106] The page preloading prediction device provided in this application, employing the page preloading prediction method in the above embodiments, can solve the technical problem that existing cloud prediction models are difficult to adapt to changes in individual dynamic behavior, leading to a reduction in page preloading effectiveness. Compared with the prior art, the beneficial effects of the page preloading prediction device provided in this application are the same as those of the page preloading prediction method provided in the above embodiments, and other technical features in the page preloading prediction device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0107] This application provides a page preload prediction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the page preload prediction method in the above embodiments.

[0108] The following is for reference. Figure 13 , Figure 13This is a schematic diagram of the hardware operating environment involved in the page preload prediction device in the embodiments of this application, illustrating a structural schematic diagram suitable for implementing the page preload prediction device in the embodiments of this application. The page preload prediction device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, personal digital assistants (PDAs), portable media players (PMPs), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 13 The page preload prediction device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0109] like Figure 13 As shown, the page preload prediction device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in the read-only memory 1002 or a program loaded from the storage device 1003 into the random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the page preload prediction device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the page preload prediction device to communicate wirelessly or wiredly with other devices to exchange data. While the figures show page preload prediction devices with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0110] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0111] The page preloading prediction device provided in this application, employing the page preloading prediction method in the above embodiments, can solve the technical problem that existing cloud prediction models are difficult to adapt to changes in individual dynamic behavior, leading to a reduction in page preloading effectiveness. Compared with the prior art, the beneficial effects of the page preloading prediction device provided in this application are the same as those of the page preloading prediction method provided in the above embodiments, and other technical features in this page preloading prediction device are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0112] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0113] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0114] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the page preload prediction method in the above embodiments.

[0115] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, radio frequency (RF), etc., or any suitable combination thereof.

[0116] The aforementioned computer-readable storage medium may be included in the page preload prediction device; or it may exist independently and not assembled into the page preload prediction device.

[0117] The aforementioned computer-readable storage medium carries one or more programs that, when executed by a page preloading prediction device, cause the page preloading prediction device to: send a page prediction request to a cloud device when a user accesses the current page, so that the cloud device performs page loading prediction for the user using a first prediction model to obtain a first page prediction result, wherein the first prediction model makes predictions based on group behavior features; determine the source page before the user accesses the current page, and extract the source page features corresponding to the source page and the current page features corresponding to the current page; perform page loading prediction for the user based on the source page features and the current page features to obtain a second page prediction result; and determine the preloaded page after the current page based on the first page prediction result and the second page prediction result.

[0118] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0119] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation that may be implemented in systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0120] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0121] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described page preloading prediction method. This solves the technical problem that existing cloud-based prediction models struggle to adapt to changes in individual dynamic behavior, leading to reduced page preloading effectiveness. Compared to the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the page preloading prediction method provided in the above embodiments, and will not be elaborated upon here.

[0122] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the page preload prediction method described above.

[0123] The computer program product provided in this application can solve the technical problem that existing cloud prediction models are unable to adapt to changes in individual dynamic behavior, resulting in reduced page preloading effectiveness. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the page preloading prediction method provided in the above embodiments, and will not be repeated here.

[0124] The above description is only a part of the embodiments of this application and does not limit the scope of protection of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included in the scope of protection of this application.

Claims

1. A page preload prediction method, characterized in that, The method includes: When a user visits the current page, a page prediction request is sent to a cloud device so that the cloud device can predict the page loading of the user using a first prediction model and obtain a first page prediction result. The first prediction model makes predictions based on group behavior characteristics. Determine the source page before the user accesses the current page, and extract the source page features corresponding to the source page and the current page features corresponding to the current page; Based on the source page features and the current page features, page loading prediction is performed for the user to obtain the second page prediction result; Based on the prediction results of the first page and the prediction results of the second page, the preloaded page following the current page is determined.

2. The method as described in claim 1, characterized in that, Before the step of sending the page prediction request to the cloud device when the user visits the current page, the method further includes: Obtain the user's historical page click information, and extract at least two previous page features and the next page features corresponding to each of the previous page features from the historical page click information; The model is trained based on the features of each preceding page and each following page to obtain the second prediction model; The step of predicting page load for the user based on the source page features and the current page features to obtain a second page prediction result includes: Based on the source page features and the current page features, the second prediction model is used to predict page loading for the user, resulting in a second page prediction result.

3. The method as described in claim 2, characterized in that, The step of determining the preloaded page following the current page based on the first page prediction result and the second page prediction result further includes: Upon detecting a new page, extract the new page features of the new page; By adding weight tensors corresponding to the new page features to the input and output layers of the second prediction model, an extended second prediction model is obtained. The extended second prediction model is incrementally trained based on the new page features to obtain updated model parameters, and the updated model parameters are stored in IndexedDB. When the user visits the current page again, the IndexedDB is used to predict page load for the user.

4. The method as described in claim 3, characterized in that, The step of predicting page load for the user when the user visits the current page again using the IndexedDB includes: When the user visits the current page again, the historical version number of the second prediction model is obtained from the IndexedDB; Compare the historical version number with the version number of the second prediction model itself; If the historical version number is consistent with the current version number, then the page loading prediction for the user is performed using the second prediction model; If the historical version number is inconsistent with the current version number, the second prediction model is updated using the IndexedDB, and the updated second prediction model is used to predict page load for the user.

5. The method as described in claim 2, characterized in that, The step of predicting page loading for the user based on the source page features and the current page features using the second prediction model to obtain a second page prediction result includes: The second prediction model is used to embed the source page features and the current page features respectively to obtain the corresponding source embedding vector and current embedding vector. Obtain the timestamp of the user's access to the current page, and perform attenuation processing on the timestamp to obtain the time weight; The source embedding vector, the current embedding vector, and the temporal weight are concatenated through a fully connected layer to obtain a fused vector; The fusion vector is linearly processed to obtain the page probability after the current page, and the page probability is used as the second page prediction result.

6. The method as described in claim 2, characterized in that, The step of determining the preloaded page following the current page based on the first page prediction result and the second page prediction result includes: Obtain the first prediction accuracy of the first prediction model, and weight the first prediction accuracy with the page probability in the first page prediction result to obtain the first page probability sequence; Obtain the second prediction accuracy of the second prediction model, and weight the second prediction accuracy with the page probability in the second page prediction result to obtain the second page probability sequence; The first page probability sequence and the second page probability sequence are merged and rearranged to obtain the page prediction probability distribution; The page with the highest probability from the predicted probability distribution of the pages is selected as the preloaded page after the current page.

7. The method according to any one of claims 1 to 6, characterized in that, After the step of determining the preloaded page following the current page based on the first page prediction result and the second page prediction result, the method further includes: The preloaded page is pre-cached locally; Upon detecting the user's page redirection request, extract the first page address corresponding to the page redirection request and the second page address corresponding to the preloaded page; Compare the first page address and the second page address; If the first page address and the second page address are the same, the page redirection request is intercepted, and the pre-cached preloaded page is displayed to the user.

8. A page preload prediction device, characterized in that, The device includes: The cloud module is used to send a page prediction request to a cloud device when a user visits the current page, so that the cloud device can predict the page loading of the user through a first prediction model and obtain a first page prediction result. The first prediction model makes predictions based on group behavior characteristics. The feature module is used to determine the source page before the user accesses the current page, and to extract the source page features corresponding to the source page and the current page features corresponding to the current page; The prediction module is used to predict page loading for the user based on the source page features and the current page features, and obtain a second page prediction result; The page module is used to determine the preloaded page following the current page based on the first page prediction result and the second page prediction result.

9. A page preload prediction device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the page preload prediction method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the page preloading prediction method as described in any one of claims 1 to 7.

Images