Application preheating terminal system based on artificial intelligence

Abstract

The invention discloses an application preheating terminal system based on artificial intelligence, and belongs to the technical field of application preheating. In the system, a full-amount data analysis module collects application use data of all authorized users as full-amount data, conjoint analysis based on CNN and LSTM is carried out on the full-amount data, universal use habits and universal use trends oriented to all authorized users are obtained, and a universal preheating strategy is generated; the personalized data training module is used for training a personalized preheating model according to the use data of the user on the application, and the personalized preheating model is used for performing data optimization on the use data of the user and predicting the use habit of the user on the application; and the full data and personalized data fusion module fuses the universal preheating strategy and the predicted use habit of the user on the application to preheat the application. According to the method, large-scale global data and the personalized end-side model are combined, so that the scheduling of the preheating resources is more accurate.

Description

Technical Field

[0001] The present invention belongs to the technical field of application preheating, and particularly relates to an application preheating terminal system based on artificial intelligence. Background Art

[0002] In modern mobile terminals, in order to speed up the application startup speed, preheating technology is usually adopted, that is, some resources of the application are pre-loaded in advance, such as dependent dynamic libraries and other common components. Traditional preheating methods are often based on general rules and cannot be optimized according to each user's usage habits. However, the traditional preheating method based on general rules cannot be optimized according to each user's usage habits. This may lead to waste of some preheating resources, or the application is not fully preheated when starting, affecting the system performance and user experience. Summary of the Invention

[0003] In view of the above technical problems, the present invention proposes an application preheating solution based on artificial intelligence.

[0004] In a first aspect of the present invention, an application preheating terminal system based on artificial intelligence is proposed. The system includes: a full-scale data analysis module located on the cloud side, and a personalized data training module and a full-scale data and personalized data fusion module located on the terminal side; wherein: On the cloud side, the full-scale data analysis module is configured to: collect the application usage data of all authorized users as full-scale data, perform joint analysis of the full-scale data based on CNN and LSTM, obtain the general usage habits and general usage trends for all authorized users, and generate a general preheating strategy; On the terminal side, the personalized data training module is configured to: train a personalized preheating model according to the user's usage data of the application, and the personalized preheating model optimizes the user's usage data and predicts the user's usage habits of the application; On the terminal side, in response to the preheating process of the application when the terminal is started, the full-scale data and personalized data fusion module is configured to: receive the general preheating strategy, call the trained personalized preheating model to predict the user's usage habits of the application, and fuse the general preheating strategy and the predicted user's usage habits of the application to preheat the application.

[0005] According to the system of the first aspect of the present invention, on the cloud side, the full-scale data analysis module is configured to: collect the application usage data of all authorized users as full-scale data; specifically including: Record the startup time when the user starts the application, record the stop time when the application enters the background, collect the user's application usage data within a preset time interval and perform desensitization processing on it, and upload the desensitized application usage data to the cloud as full-scale data during idle time.

[0006] For the system according to the first aspect of the present invention, on the cloud side, the full - scale data analysis module is configured to: perform a joint analysis of the full - scale data based on CNN and LSTM to obtain the general usage habits and general usage trends for all authorized users, and generate a general pre - heating strategy; specifically including: Comprehensively analyze the full - scale data using CNN and LSTM, extract the startup patterns and resource requirements of the applications, further generate an application startup heat map, obtain the general usage habits and general usage trends for all authorized users from the heat map, and generate a general pre - heating strategy, and send the general pre - heating strategy to the terminal at a predetermined time.

[0007] For the system according to the first aspect of the present invention, on the cloud side: Construct a hybrid model using CNN and LSTM; the convolutional neural network CNN extracts spatial features from the sequence data and inputs the spatial features into the long short - term memory network LSTM to capture the long - term dependencies in the time dimension; When training the hybrid model, the training data set is characterized as , represents the i - th group of input sequence data, represents the pre - heating hit rate achieved by the pre - heating strategy generated based on the i - th group of sequence data, and N represents the number of sequence data.

[0008] For the system according to the first aspect of the present invention, on the terminal side, the personalized data training module is configured to: train a personalized pre - heating model according to the usage data of the user for the application; where: The training process adopts the incremental learning method, and the model parameters of the personalized pre - heating model are , the collected usage data of the user for the application is , represents the i - th group of usage data of the user for the application, represents the pre - heating hit rate achieved by the predicted usage habits of the user for the application based on the i - th group of usage data of the user for the application, and M represents the number of usage data of the user for the application; Update the model parameters of the personalized pre - heating model by the stochastic gradient descent method: , where, represents the updated model parameters, η represents the learning rate, represents the loss function.

[0009] For the system according to the first aspect of the present invention, on the terminal side, the usage habits of the user for the application predicted by the personalized pre - heating model include: multi - application switching frequency, common usage time periods, geographical locations.

[0010] For the system according to the first aspect of the present invention, on the terminal side, in response to the terminal starting the preheating process of the application, the full - volume data and personalized data fusion module is configured to: Receive a general preheating strategy and obtain features extracted from the cloud ; Call the trained personalized preheating model to predict the user's usage habits of the application and obtain features extracted from the terminal ; Fuse the general preheating strategy and the predicted user's usage habits of the application, and splice and into a comprehensive feature vector ; Wherein: , Further introduce a weight adjustment mechanism:

[0011] Wherein, represents a weight adjustment coefficient.

[0012] The second aspect of the present invention proposes an application preheating terminal method based on artificial intelligence, and the method includes: On the cloud side: Collect the application usage data of all authorized users as full - volume data, conduct a joint analysis of the full - volume data based on CNN and LSTM to obtain the general usage habits and general usage trends for all authorized users, and generate a general preheating strategy; On the terminal side: Train a personalized preheating model according to the user's usage data of the application. The personalized preheating model optimizes the user's usage data and predicts the user's usage habits of the application; Receive the general preheating strategy, call the trained personalized preheating model to predict the user's usage habits of the application, and fuse the general preheating strategy and the predicted user's usage habits of the application to preheat the application.

[0013] The third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor. When the processor executes the computer program stored in the memory, it implements an application preheating terminal system according to the second aspect of the present disclosure.

[0014] The fourth aspect of the present invention discloses a computer - readable storage medium. When the computer program stored on the computer - readable storage medium is executed by the processor, it implements an application preheating terminal system according to the second aspect of the present disclosure.

[0015] In summary, the solution provided by the present invention uses a deep learning algorithm, combines the application usage data of a large number of authorized users and the personalized usage behaviors of individual users, and performs intelligent and customized preheating on the applications on the mobile terminal. This method can not only improve the preheating hit rate, but also effectively reduce energy consumption and improve the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0017] Figure 1 It is a schematic diagram of the application preheating process according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, rather than all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] The first aspect of the present invention proposes an application preheating terminal system based on artificial intelligence. The system includes: a full-scale data analysis module on the cloud side, and a personalized data training module and a full-scale data and personalized data fusion module on the terminal side. Among them (as Figure 1 shown): On the cloud side, the full-scale data analysis module is configured to: collect the application usage data of all authorized users as full-scale data, perform joint analysis of the full-scale data based on CNN and LSTM, obtain the general usage habits and general usage trends for all authorized users, and generate a general preheating strategy; On the terminal side, the personalized data training module is configured to: train a personalized preheating model according to the user's usage data of the application, and the personalized preheating model optimizes the user's usage data and predicts the user's usage habits of the application; On the terminal side, in response to the terminal starting the preheating process of the application, the full-scale data and personalized data fusion module is configured to: receive the general preheating strategy, call the trained personalized preheating model to predict the user's usage habits of the application, and fuse the general preheating strategy and the predicted user's usage habits of the application to preheat the application.

[0020] In some embodiments, on the cloud side, the full - volume data analysis module is configured to: collect the application usage data of all authorized users as full - volume data; specifically including: Record the start time when the user launches the application, record the stop time when the application enters the background, collect the application usage data of the user within a preset time interval and perform desensitization processing on it, and upload the desensitized application usage data to the cloud as full - volume data during off - peak hours.

[0021] In some embodiments, on the cloud side, the full - volume data analysis module is configured to: perform joint analysis on the full - volume data based on CNN and LSTM to obtain the general usage habits and general usage trends for all authorized users, and generate a general pre - heating strategy; specifically including: Use CNN and LSTM to comprehensively analyze the full - volume data, extract the startup mode and resource requirements of the application, further generate an application startup heat map, obtain the general usage habits and general usage trends for all authorized users from the heat map, and generate a general pre - heating strategy, and send the general pre - heating strategy to the terminal at a predetermined time.

[0022] In some embodiments, on the cloud side: Build a hybrid model using CNN and LSTM; the convolutional neural network CNN extracts spatial features from sequence data and inputs the spatial features into the long short - term memory network LSTM to capture long - term dependencies in the time dimension; When training the hybrid model, the training dataset is characterized as , represents the i - th group of input sequence data, represents the pre - heating hit rate achieved by the pre - heating strategy generated based on the i - th group of sequence data, and N represents the number of sequence data.

[0023] Specifically, for full - volume data analysis, the system avoids the problem of no data in cold start by collecting the application usage data of all authorized users. The cloud dataset comprehensively analyzes a large number of applications, extracts the startup mode and resource requirements of various applications, thereby ensuring the basic accuracy of pre - heating at the big - data level.

[0024] Specifically, for high-performance data collection of authorized user applications, a strategy of recording during peak usage hours and reporting during off-peak hours is adopted to achieve the goal of high-performance data collection. Specifically, when an authorized user starts an application, the start time is logged, and when the application enters the background, the stop time is logged. By this method, the application usage records of the current authorized user within 24 hours are recorded, including multi-dimensional data such as the start time, single-use duration, total use duration, and number of starts, and during off-peak hours (nighttime, when the mobile phone is connected to charging and the Wi-Fi is on), the data is desensitized, and after ignoring sensitive user identity information, it is uploaded to the cloud.

[0025] Specifically, the cloud analyzes the overall application usage habits. The convolutional neural network (CNN) and long short-term memory network (LSTM) are used to analyze the behavior habits of all authorized users. We adopt a scoring method, that is, define key metrics (such as the number of application starts and usage duration) and assign weights, quantify user behavior into scores, and then generate a heat map of application starts to analyze the overall usage habits and trends of applications. The trained cloud strategy will be sent to the device side during specific time periods (such as system updates) for regular updates. When the device side performs a cold start, it will preheat the application based on the full amount of data.

[0026] Specifically, the entire system architecture draws on the idea of federated learning. Each participating party independently trains a model using local data and uploads the desensitized data to the central server. The server aggregates the information from all parties to optimize the global model and distributes the updated model to each participating party, and continuously iterates and optimizes in this way to form an efficient and secure closed-loop learning system.

[0027] The cloud collects application usage data from a large number of authorized terminals, covering multi-dimensional information such as the start time, stop time, and usage duration of the application.

[0028] Among them, the warm-up hit rate refers to the ratio of the number of times a user uses a warmed-up application to the total number of times the warmed-up application is used within a certain period of time. Its calculation formula is:

[0029] For the training dataset , represents the i-th group of input sequence data, represents the warm-up hit rate achieved by the warm-up strategy generated based on the i-th group of sequence data, and N represents the number of sequence data.

[0030] After the data collection is completed, a data cleaning algorithm is used to accurately judge the validity of the data in combination with context information such as time zones and user behavior patterns, so as to reduce the situation of mistakenly deleting valid data. Subsequently, the data is normalized based on the distribution to ensure the comparability and consistency of the data.

[0031] In actual training, it is found that the warm-up hit rate not only has local features and spatial correlations with the frequency and habits of users using applications, but also shows trends and dependencies in time. For example, after using Application A, users often immediately use Application B. To comprehensively consider discrete feature points and time dependencies, the cloud uses CNN and LSTM algorithms to build a hybrid model. Among them, the convolutional neural network CNN can efficiently process sequence data and extract spatial features from it. After extracting spatial features through convolutional operations, the results are input into the long short-term memory network LSTM to capture long-term dependencies in the time dimension, and then more comprehensively and accurately mine users' application usage patterns.

[0032] In some embodiments, on the terminal side, the personalized data training module is configured to: train a personalized warm-up model according to the usage data of the application by the user; where: The training process adopts an incremental learning method, and the model parameters of the personalized warm-up model are , and the collected usage data of the application by the user is , represents the i-th group of usage data of the application by the user, represents the warm-up hit rate achieved by the usage habits of the application predicted based on the i-th group of usage data of the application by the user, and M represents the number of usage data of the application by the user; Update the model parameters of the personalized warm-up model by the stochastic gradient descent method:

[0033] Where, represents the updated model parameters, η represents the learning rate, represents the loss function.

[0034] In some embodiments, on the terminal side, the usage habits of the application predicted by the personalized warm-up model include: multi-application switching frequency, common time periods, and geographical locations.

[0035] Specifically, for multi-dimensional personalized model training: the end-side system trains a personalized warm-up model according to multi-dimensional parameters such as the user's device usage habits, such as multi-application switching frequency, common time periods, geographical locations, and historical warm-up hit rates. This model can be optimized according to the usage habits of different users to ensure that the warm-up resources can effectively cover the actual needs.

[0036] Specifically, the edge side predicts the usage habits of target users' applications. Through the lightweight training model built into the system, combined with the user's personal data on the user side and the user's usage habits, personalized portrait analysis of user behavior is carried out to predict and analyze the usage habits of target users. The personal data involved in this part is only used on the edge side and will not be uploaded to the cloud to ensure the security of personal information.

[0037] Specifically, edge-side training first performs data collection and preprocessing. The edge side mainly collects personalized application usage data of local users. When preprocessing the data, a more refined data simplification strategy is adopted, such as sampling in combination with data importance evaluation. According to the importance of the application usage data for the warm-up strategy, the data is hierarchically sampled to retain key data and reduce the storage and computing pressure of unimportant data.

[0038] Specifically, the model is trained according to the usage habits and data of local users. Incremental learning is adopted, and the existing model parameters on the local side , collect new local data , represents the i-th group of usage data of the user for the application, represents the warm-up hit rate of the user's usage habits for the application predicted based on the i-th group of usage data of the user for the application, and M represents the number of usage data of the user for the application.

[0039] Update the model parameters through stochastic gradient descent:

[0040] Among them, represents the updated model parameters, η represents the learning rate, represents the loss function.

[0041] In some embodiments, on the terminal side, in response to the terminal starting the warm-up process for the application, the full-data and personalized-data fusion module is configured to: Receive the general warm-up strategy and obtain the features extracted from the cloud ; Call the trained personalized warm-up model to predict the usage habits of the user for the application and obtain the features extracted from the terminal ; Fuse the general warm-up strategy and the predicted usage habits of the user for the application, and and are concatenated into a comprehensive feature vector ; Among them:

[0042] Further introduce a weight adjustment mechanism:

[0043] Among them, represents the weight adjustment coefficient.

[0044] Specifically, for the combination of full-scale data and personalized data: The system combines the full-scale user data with the personalized usage behavior of a single user to form a more accurate preheating strategy. This combination can significantly improve the preheating hit rate, reduce unnecessary resource loading, and further enhance the response speed of the device and the user experience.

[0045] Specifically, the combination of cloud data and personalized data predicts application usage habits. Through the issued strategy, it enters the edge-side prediction training set in the form of input. The edge side will combine the cloud data and the edge-side prediction analysis to conduct a more comprehensive and customized prediction and analysis of the user's application usage habits. This method can make full use of the global data in the cloud and the personalized data on the edge side to improve the accuracy of prediction and the efficiency of application preheating.

[0046] Specifically, for the feature fusion between the cloud and the edge side: Among them, for feature-level fusion: After training the full-scale feature model, the cloud first compresses the model through techniques such as quantization, pruning, and knowledge distillation, and then sends it to the device side. At the same time, the cloud uses feature selection algorithms (such as information gain, chi-square test) to screen out the features crucial for the preheating strategy and send them down.

[0047] The feature vector extracted from the cloud ,, and the local feature vector extracted on the edge side , will and be concatenated into a comprehensive feature vector : .

[0048] Taking the concatenated feature vector as the input for the subsequent preheating strategy prediction model. By introducing a weight adjustment mechanism, the fusion weight is adjusted according to the current data features. The weighted fused feature vector can be expressed as:

[0049] Among them, ∈[0,1] is the weight adjustment coefficient.

[0050] Specifically, taking the daily application usage duration and startup frequency of the user as an example, the feature vector extracted from the cloud =[120, 8] (representing the daily usage duration of a certain application for 120 seconds and 8 startups respectively). The local feature vector extracted on the device side =[5, 1] (such as the installation location code 5 of the application on the device and the user-defined label 1). The comprehensive feature vector is obtained . Finally, input the concatenated feature vectors into the warm-up strategy prediction model to obtain a decision on whether to apply warm-up.

[0051] In some embodiments, regarding decision-level fusion: In the manner of decision-level fusion, two models make predictions separately, and then the prediction results are integrated through a voting mechanism. If the voting result exceeds a predetermined threshold, the operating system will warm up the specified application. Specifically, the edge device makes independent predictions using the global model features obtained from the cloud and the personalized model features trained locally. For example, in a classification problem, if the global model at the edge predicts the category as , and the local model at the edge predicts the category as , the final predicted category is determined through a voting mechanism. For example: The model predicts that the prediction result of the compressed cloud model is "warm-up". The prediction of the edge model that fuses the key gain features is "warm-up". In a simple majority vote, both models predict "warm-up", and the voting result exceeds the predetermined threshold. Therefore, the operating system will warm up this application.

[0052] The solution proposed by the present invention has the following technical advantages: (1) Accuracy: By combining large-scale global data and personalized edge models, the scheduling of warm-up resources is made more accurate, effectively reducing the problem of applications not being warmed up at startup. (2) Efficiency: Through an intelligent warm-up strategy, unnecessary resource preloading is reduced, significantly lowering system energy consumption while increasing the startup speed of applications. (3) Adaptability: Based on the continuous training of deep learning models, the system can continuously adapt to users' usage habits and environmental changes, ensuring the long-term effectiveness of the warm-up strategy.

[0053] The present invention not only improves the startup speed of mobile terminal applications but also optimizes the utilization rate of system resources through intelligent resource scheduling, reduces power consumption, and greatly improves the overall performance of the user experience and device performance.

[0054] The second aspect of the present invention proposes an application warm-up terminal method based on artificial intelligence, and the method includes: On the cloud side: Collect the application usage data of all authorized users as full-scale data, conduct a joint analysis of the full-scale data based on CNN and LSTM to obtain the general usage habits and general usage trends for all authorized users, and generate a general warm-up strategy; On the terminal side: Train a personalized warm-up model according to the user's usage data of the application. The personalized warm-up model optimizes the user's usage data and predicts the user's usage habits of the application; Receive a general preheating strategy, call the trained personalized preheating model to predict the user's usage habits of the application, and fuse the general preheating strategy and the predicted user's usage habits of the application to preheat the application.

[0055] A third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it implements an application preheating terminal system based on artificial intelligence in the second aspect of the present disclosure.

[0056] A fourth aspect of the present invention discloses a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, it implements an application preheating terminal system based on artificial intelligence in the second aspect of the present disclosure.

[0057] In summary, the solution provided by the present invention uses a deep learning algorithm, combines the application usage data of a large number of authorized users and the personalized usage behaviors of individual users, and preheats the applications on the mobile terminal in an intelligent and customized manner. This method can not only improve the preheating hit rate, but also effectively reduce energy consumption and improve the user experience.

[0058] Please note that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification. The above embodiments only represent several implementation manners of the present application. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be pointed out that for those of ordinary skill in the art, without departing from the concept of the present application, several deformations and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. An application preheating terminal system based on artificial intelligence, characterized in that, The system includes: a full - scale data analysis module on the cloud side, and a personalized data training module and a full - scale data and personalized data fusion module on the terminal side; wherein: On the cloud side, the full - scale data analysis module is configured to: collect the application usage data of all authorized users as full - scale data, perform a joint analysis of the full - scale data based on CNN and LSTM, obtain the general usage habits and general usage trends for all authorized users, and generate a general pre - heating strategy; On the terminal side, the personalized data training module is configured to: train a personalized pre - heating model according to the user's application usage data, and the personalized pre - heating model optimizes the user's usage data and predicts the user's application usage habits; On the terminal side, in response to the terminal starting the pre - heating process of the application, the full - scale data and personalized data fusion module is configured to: receive the general pre - heating strategy, call the trained personalized pre - heating model to predict the user's application usage habits, and fuse the general pre - heating strategy and the predicted user's application usage habits to pre - heat the application.

2. The application preheating terminal system based on artificial intelligence according to claim 1, characterized in that, On the cloud side, the full - scale data analysis module is configured to: collect the application usage data of all authorized users as full - scale data; specifically including: Record the start time when the user starts the application, record the stop time when the application enters the background, collect the user's application usage data within a preset time interval and perform desensitization processing on it, and upload the desensitized application usage data to the cloud as full - scale data during off - peak hours.

3. An application preheating terminal system based on artificial intelligence according to claim 2, characterized in that, On the cloud side, the full - scale data analysis module is configured to: perform a joint analysis of the full - scale data based on CNN and LSTM, obtain the general usage habits and general usage trends for all authorized users, and generate a general pre - heating strategy; specifically including: Comprehensively analyze the full - scale data using CNN and LSTM, extract the start mode and resource requirements of the application, further generate an application start heat map, obtain the general usage habits and general usage trends for all authorized users from the heat map, and generate a general pre - heating strategy, and send the general pre - heating strategy to the terminal at a predetermined time.

4. The application preheating terminal system based on artificial intelligence according to claim 3, characterized in that, On the cloud side: Build a hybrid model using CNN and LSTM; the convolutional neural network CNN extracts spatial features from sequence data and inputs the spatial features into the long short - term memory network LSTM to capture long - term dependencies in the time dimension; When training a hybrid model, the training data set is characterized as , represents the i-th set of sequence data of the input, represents the warm-up hit rate achieved by the warm-up strategy generated based on the i-th set of sequence data, and N represents the number of sequence data.

5. The application preheating terminal system based on artificial intelligence according to claim 4, characterized in that, On the terminal side, the personalized data training module is configured to: train a personalized pre - heating model according to the user's application usage data; wherein: The training process adopts the incremental learning method, and the model parameters of the personalized warm-up model are , and the collected usage data of the application by users is , represents the i-th group of usage data of the application by the user, represents the warm-up hit rate of the usage habits of the application predicted based on the i-th group of usage data of the application by the user, and M represents the quantity of the usage data of the application by the user; Update the model parameters of the personalized pre - heating model by the stochastic gradient descent method: ， Among them, represents the updated model parameters, η represents the learning rate, represents the loss function.

6. The application preheating terminal system based on artificial intelligence according to claim 5, wherein, On the terminal side, the application usage habits predicted by the personalized pre - heating model include: multi - application switching frequency, common time periods, and geographical locations.

7. An application preheating terminal system based on artificial intelligence according to claim 6, characterized in that, On the terminal side, in response to the terminal starting the pre - heating process of the application, the full - scale data and personalized data fusion module is configured to: Receive a general warm-up strategy and obtain features extracted from the cloud ; Call the trained personalized warm-up model to predict the user's usage habits of the application and obtain the features extracted from the terminal ; Integrate the general preheating strategy and the usage habits of users obtained by prediction for the application, and and are concatenated into a comprehensive feature vector ; where: ， Further introduce a weight adjustment mechanism: ， Among them, represents the weight adjustment coefficient.

8. An application preheating terminal method based on artificial intelligence, characterized in that, The method includes: On the cloud side: Collect the application usage data of all authorized users as the full-scale data, conduct a joint analysis based on CNN and LSTM on the full-scale data, obtain the general usage habits and general usage trends for all authorized users, and generate a general preheating strategy; On the terminal side: Train a personalized preheating model according to the user's usage data of the application. The personalized preheating model optimizes the data of the user's usage data and predicts the user's usage habits of the application; Receive the general preheating strategy, call the trained personalized preheating model to predict the user's usage habits of the application, and preheat the application by integrating the general preheating strategy and the predicted user's usage habits of the application.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it implements a method for preheating an application terminal based on artificial intelligence according to claim 8.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium. When the computer program is executed by the processor, it implements a method for preheating an application terminal based on artificial intelligence according to claim 8.

Images