Cloud mobile phone recycling method and device, storage medium and server

Abstract

The invention discloses a cloud mobile phone recycling method and device, a storage medium and a server, and relates to the technical field of cloud computing, and the method comprises the steps: responding to a condition that the occupancy rate of cloud mobile phone resources is greater than a preset occupancy rate, and taking user tags meeting a preset search condition in a user tag library as candidate user tags; and then, inputting the candidate user tag into the target recovery prediction model to obtain a predicted recovery probability of the candidate user tag corresponding to the first cloud mobile phone instance. And then, carrying out recovery operation on the first cloud mobile phone instance of which the predicted recovery probability is greater than the target recovery probability. According to the embodiment of the invention, efficient recovery of the cloud mobile phone can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cloud computing, and particularly relates to a cloud phone recycling method and device, a storage medium and a server. BACKGROUND

[0002] With the continuous development of cloud computing technology, cloud phones, as cloud virtual devices based on virtualization technology, have been widely used in multiple scenarios. However, since the number of cloud phones in a server is limited, if a cloud phone is occupied for a long time without being used after the cloud phone establishes a connection with a cloud client, other cloud clients may fail to establish a connection with the cloud phone, thereby affecting the use experience of other users.

[0003] Therefore, how to recycle a cloud phone that is occupied for a long time without being used to ensure that other cloud clients can establish a connection with the cloud phone, thereby improving the utilization rate of the cloud phone, is a problem to be solved. SUMMARY

[0004] Therefore, the present application provides a cloud phone recycling method and device, a storage medium and a server, which can realize efficient recycling of cloud phones and improve the utilization rate of cloud phones.

[0005] In a first aspect, the present application provides a cloud phone recycling method applied to a server, the server being configured to manage a plurality of cloud phone instances, and the method comprising: in response to the occupancy rate of cloud resources being greater than a preset occupancy rate, taking a user tag in a user tag library that meets a preset search condition as a candidate user tag; wherein the user tag library comprises user tags corresponding to each first cloud phone instance, and the first cloud phone instance is a cloud phone instance in the plurality of cloud phone instances that has established a connection with a cloud client; inputting the candidate user tag into a target recycling prediction model to obtain a predicted recycling probability of the first cloud phone instance corresponding to the candidate user tag; recycling the first cloud phone instance with a predicted recycling probability greater than a target recycling probability.

[0006] In a second aspect, the present application provides a cloud phone recycling device, comprising: a tag determination unit configured to, in response to the occupancy rate of cloud resources being greater than a preset occupancy rate, take a user tag in a user tag library that meets a preset search condition as a candidate user tag; wherein the user tag library comprises user tags corresponding to each first cloud phone instance, and the first cloud phone instance is a cloud phone instance in the plurality of cloud phone instances that has established a connection with a cloud client; a probability determination unit configured to input the candidate user tag into a target recycling prediction model to obtain a predicted recycling probability of the first cloud phone instance corresponding to the candidate user tag. The cloud machine recycling unit is configured to recycle the first cloud mobile instance whose predicted recycling probability is greater than the target recycling probability.

[0007] In a third aspect, the present application provides a computer readable storage medium, having stored thereon a computer program, which, when executed by a processor, implements the method of the first aspect or the method of the second aspect.

[0008] In a fourth aspect, the present application provides a server, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor implements the method of the first aspect or the method of the second aspect when executing the computer program.

[0009] In a fifth aspect, the present application provides a computer program product, having stored thereon a computer program, which, when executed by a processor, implements the method of the first aspect or the method of the second aspect.

[0010] The present application provides a cloud mobile recycling method, device, storage medium and server. When the occupancy rate of cloud machine resources is greater than the preset occupancy rate, it indicates that the usage of cloud mobile in the server is high. Therefore, in order to reduce the situation that the cloud mobile is occupied for a long time without being used, the server can determine the cloud mobile instance that can be recycled in the server by predicting the recyclable probability (i.e. predicted recycling probability) of the cloud mobile instance corresponding to the user label of the cloud mobile instance. In this way, efficient recycling of the cloud mobile can be realized, the waste of cloud machine resources caused by the idle time of the cloud mobile for too long can be reduced, the connection between other cloud machine clients and the cloud mobile can be ensured, the utilization rate of the cloud mobile is improved, and the user's use experience is improved.

[0011] The above description is only a summary of the technical solutions of the present application. In order to enable the technical means of the present application to be more clearly understood, and to be implemented according to the content of the description, and in order to enable the above and other purposes, characteristics and advantages of the present application to be more apparent and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0012] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the present application and, together with the specification, serve to explain the principles of the present application.

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, for those skilled in the art, other drawings can also be obtained from these drawings without any creative labor.

[0014] Figure 1 A flowchart illustrating a cloud phone recycling method provided by related technologies is shown. Figure 2 This paper shows a schematic diagram of the structure of a cloud mobile phone recycling system provided in an embodiment of this application; Figure 3 A flowchart illustrating a cloud phone recycling method provided in an embodiment of this application is shown; Figure 4 A flowchart illustrating another cloud phone recycling method provided in an embodiment of this application is shown; Figure 5 This illustration shows a flowchart of a method for determining a target exit period according to an embodiment of this application; Figure 6 A flowchart illustrating a training method for a recycling prediction model provided in an embodiment of this application is shown. Figure 7 This illustration shows a process diagram of a cloud phone recycling procedure provided in an embodiment of this application; Figure 8 This paper shows a schematic diagram of the structure of a cloud phone recycling device provided in an embodiment of this application; Figure 9 A schematic diagram of another cloud phone recycling device provided in an embodiment of this application is shown. Detailed Implementation

[0015] To facilitate the explanation of the embodiments of this application, some technical terms and technical means related to the embodiments of this application, as well as the application scenarios of the embodiments of this application, will be introduced first below.

[0016] Cloud phones are virtual devices implemented in the cloud based on virtualization technology. A server manages multiple virtualized containers, each running a customized operating system. Each virtualized container is essentially a cloud phone instance. Combined with hardware emulation capabilities, the functionality of a cloud phone is essentially the same as a real device. Users access the cloud phone through a cloud phone client installed on their physical device. The server then establishes a connection between any cloud phone and the cloud phone client, ensuring that users can experience the cloud phone as if it were a real device.

[0017] In some cases, since the number of cloud phones managed by the server is limited, if a cloud phone is occupied and not used for a long time after establishing a connection with a cloud client, other cloud clients may be unable to establish a connection because the cloud phone is occupied, thus affecting the user experience of other users.

[0018] Therefore, to ensure that cloud clients can establish a normal connection with cloud phones, the server can reclaim cloud phones that have been occupied and unused for a long time, so that cloud clients can establish a connection with the reclaimed cloud phones. In other words, the server can determine which cloud phones to reclaim based on the time limit of each cloud phone.

[0019] In some embodiments, such as Figure 1 As shown, the process of server-side cloud phone recycling can specifically include: At preset time intervals, the server identifies cloud phones with idle time exceeding a preset idle time as cloud phones to be recycled. Next, the server backs up the cloud machine data of the cloud phones to be recycled. Then, the server initializes the cloud phones to be recycled after the data backup is complete. Next, the server changes the cloud machine status of the initialized cloud phones to a pending connection state. Finally, in response to any connection request from a cloud machine client, the server connects the cloud machine client to the cloud phone in the pending connection state.

[0020] The preset time period and preset idle time can both be preset according to actual needs, and the specific values ​​are not limited.

[0021] It should be noted that the cloud phones to be recycled mentioned above are determined based on the idle time of each cloud phone in the server. While this allows for the recycling of cloud phones, the requirements for setting the preset idle time are relatively high. It's possible that a preset idle time set too high (e.g., 3 days) could result in cloud phones remaining idle for a long time without being recycled, or a preset idle time set too low (e.g., 2 hours) could lead to frequent recycling of cloud phones, causing excessive server load. This results in a low recycling and reuse rate of cloud resources, leading to wasted cloud resources and ultimately impacting the user experience.

[0022] Therefore, to improve the recycling and reuse rate of cloud server resources and thus enhance the user experience, this application provides a cloud phone recycling method. In this method, in response to a cloud server resource occupancy rate exceeding a preset occupancy rate, user tags matching preset search conditions in a user tag library are selected as candidate user tags. The user tag library includes user tags corresponding to each first cloud phone instance, where a first cloud phone instance is a cloud phone instance that has already established a connection with a cloud server client among multiple cloud phone instances. The server can then input the candidate user tags into a target recycling prediction model to obtain the predicted recycling probability of the first cloud phone instance corresponding to the candidate user tag. Finally, the server can perform a recycling operation on first cloud phone instances whose predicted recycling probability is greater than the target recycling probability.

[0023] In this embodiment, when the occupancy rate of cloud server resources exceeds a preset occupancy rate, it indicates a high usage of cloud phone instances on the server. Therefore, to reduce the situation where cloud phones are occupied but not used for extended periods, the server can predict the recyclability probability of cloud phone instances by using the user tags corresponding to those instances (i.e., predict the recyclability probability). This allows for efficient cloud phone recycling, reducing the waste of cloud server resources caused by prolonged idle time, ensuring that other cloud server clients can establish connections with the cloud phones, improving cloud phone utilization, and ultimately enhancing the user experience.

[0024] Furthermore, the server only inputs user tags corresponding to a portion of the first cloud phone instances into the target recycling prediction model to predict recycling probabilities, rather than inputting user tags corresponding to every first cloud phone instance into the target recycling prediction model. This not only reduces server resource waste and improves server resource utilization, but also, because user tags are determined based on user habits, allows the generated predicted recycling probabilities to be combined with user habits, improving the relevance of the predicted recycling probabilities to users and thus ensuring a better user experience.

[0025] Furthermore, since the triggering condition for cloud phone recycling is based on the occupancy rate of cloud server resources, the server will only execute the cloud phone recycling operation when the occupancy rate of cloud server resources is greater than a preset occupancy rate. This maximizes the reuse of cloud server resources, reduces the occurrence of ineffective recycling due to low usage of cloud phones on the server, and improves the recycling and reuse rate of cloud server resources.

[0026] In some examples, the server in this application embodiment can be a single server, a server cluster, a distributed server, a centralized server, a cloud server, or a computer, etc., and there is no specific limitation.

[0027] The embodiments of this application will now be described in more detail with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0028] Figure 2 This is a schematic diagram of the cloud phone recycling system according to an embodiment of this application. Figure 2 As shown, the cloud phone recycling system may include a data acquisition module, a data analysis module, a model prediction module, a cloud phone recycling module, and an information feedback module.

[0029] The data acquisition module is used to collect the first data of the first cloud phone instance. The first cloud phone instance is the cloud phone instance among multiple cloud phone instances that has already established a connection with the cloud machine client. The first data includes user behavior data and / or cloud machine usage data. User behavior data may include the user's unique identification number (ID), user cloud machine ID, cloud machine login time, application name used, application usage duration, cloud machine login region, cloud machine specifications, and user level, etc. Cloud machine usage data may include the time of entering the cloud machine, the time of exiting the cloud machine, etc. The cloud machine login region refers to the region to which the cloud phone instance's Internet Protocol (IP) address belongs. Cloud machine specifications may include the cloud phone instance's resolution, memory capacity, central processing unit (CPU) model, etc. User levels may include regular users, monthly card users, quarterly card users, annual card users, and star-level users, etc.

[0030] The data analysis module is used to analyze the aforementioned first data to obtain the user tags corresponding to the first cloud phone instance. These user tags may include user ID, target logout time period (or logout time tag), user's city of residence (or login area tag), user's occupation (or occupation tag), user preference items (or preference tags), and the preferred applications corresponding to those preferences (or secondary preference tags).

[0031] In some embodiments, the target exit time period is determined based on the time of entering and exiting the cloud server. The user's city of residence is determined based on the cloud server login region. User preferences and corresponding preferred applications are determined based on the application name and usage duration.

[0032] The model prediction module is used to train a target recycling prediction model and, through this model, predict the probability (or predicted recycling probability) that the first cloud phone instance is recyclable. In some embodiments, the model prediction module can acquire a sample dataset. This sample dataset includes multiple sample data points and their corresponding actual recycling probabilities, and includes historical user tags. Then, for each sample data point, the sample data is input into a pre-built recycling prediction model to obtain the predicted recycling probability. Finally, based on the predicted and actual recycling probabilities of the sample data, the parameters of the pre-built recycling prediction model are adjusted to obtain the target recycling prediction model.

[0033] Subsequently, upon receiving the user tag corresponding to the first cloud phone instance, the model prediction module can directly input the user tag corresponding to the first cloud phone instance into the target recycling prediction model to obtain the predicted recycling probability of the first cloud phone instance.

[0034] The cloud server recycling module is used to recycle the first cloud phone instance whose predicted recycling probability is greater than the target recycling probability. Specifically, upon receiving the predicted recycling probability of the first cloud phone instance, it can determine whether the predicted recycling probability is greater than the target recycling probability. If the predicted recycling probability is greater than the target recycling probability, it indicates that the first cloud phone instance is unlikely to be used in the short term, meaning it will not be accessed by users again in the near future. Therefore, the cloud server recycling module can recycle the first cloud phone instance.

[0035] The information feedback module monitors whether the first cloud phone instance that has been reclaimed is accessed again within a preset time interval. Specifically, after reclaiming a first cloud phone instance whose predicted reclamation probability is greater than the target reclamation probability, the information feedback module adjusts the probability fine-tuning value of the first cloud phone instance based on whether it has been accessed again within the preset time interval, thereby optimizing the target reclamation prediction model. The probability fine-tuning value determines the number of decision trees in the target reclamation prediction model. The number of decision trees is positively correlated with the predicted reclamation probability.

[0036] In one implementation, such as Figure 2 As shown, the data acquisition module can collect the first data of each first cloud phone instance. Then, the data acquisition module can execute step a, sending the first data of each first cloud phone instance to the data analysis module. Afterwards, the data analysis module can analyze the first data of each first cloud phone instance to obtain the user tags corresponding to each first cloud phone instance, and store these user tags in the user tag library.

[0037] Subsequently, in response to the cloud server resource occupancy rate exceeding the preset occupancy rate, the cloud server recycling module can execute step b, sending a tag selection instruction to the data analysis module. The data analysis module can then select user tags from the user tag library that meet preset search criteria, and designate these user tags as candidate user tags. Next, the data analysis module can execute step c, sending the candidate user tags to the model prediction module. The model prediction module can then input the candidate user tags into the target recycling prediction model to obtain the predicted recycling probability of the first cloud phone instance corresponding to the candidate user tags. Next, the model prediction module can execute step d, sending the predicted recycling probability of the first cloud phone instance corresponding to the candidate user tags to the cloud server recycling module. Finally, the cloud server recycling module can recycle the first cloud phone instance whose predicted recycling probability is greater than the target recycling probability.

[0038] Next, the cloud server recycling module can execute step e, sending a cloud server monitoring command to the information feedback module. Then, the information feedback module can adjust the probability fine-tuning value of the first recycled cloud phone instance based on whether it has been accessed again within a preset time interval. Afterwards, the information feedback module can execute step f, sending the adjusted probability fine-tuning value to the model prediction module, so that the information feedback module can optimize the target recycling prediction model based on this adjusted probability fine-tuning value.

[0039] This embodiment provides a method for recycling cloud phones, such as... Figure 3 As shown, the method includes: S301, in response to the cloud machine resource occupancy rate being greater than the preset occupancy rate, uses user tags in the user tag library that meet the preset search conditions as candidate user tags.

[0040] In some cases, a server can calculate the utilization rate of cloud server resources. This utilization rate is the ratio between the number of primary cloud phone instances and the total number of cloud phone instances managed by the server. The server manages multiple cloud phone instances. The primary cloud phone instance is the one among the multiple cloud phone instances that has already established a connection with a cloud phone client. In other words, the utilization rate of cloud server resources is used to characterize the usage level of multiple cloud phone instances. For example, if the total number of cloud phone instances managed by the server is 100, and the number of primary cloud phone instances is 80, then the utilization rate of cloud server resources is 80 ÷ 100 = 80%.

[0041] Next, the server can determine whether the occupancy rate of the cloud machine resources is greater than the preset occupancy rate. If the occupancy rate is greater than the preset rate, it indicates that the usage of cloud phone instances on the server is high. Therefore, to achieve accurate recycling of cloud phone instances, the server can obtain a user tag library to select user tags that match preset search criteria. If the occupancy rate of the cloud machine resources is less than or equal to the preset rate, it indicates that the usage of cloud phone instances on the server is low. Therefore, to reduce unnecessary resource waste during recycling, the server will not recycle the first cloud phone instance.

[0042] The user tag library includes user tags corresponding to each First Cloud Phone instance. User tags may include user ID, target logout time period, user's city of residence, user's occupation, user's preferred items, and the preferred applications corresponding to the user's preferred items. The target logout time period is the time period during which the user of the First Cloud Phone instance frequently logs out of the cloud server, that is, the First Cloud Phone instance is logged out more often during the target logout time period than during other time periods.

[0043] In some cases, using the user tag library as shown in Table 1 as an example, user 10000's first cloud phone instance corresponds to user's permanent residence city A, user's occupation is student, user's preferred activity is games, and the user's preferred applications are game application 1, game application 2, and game application 3. User 20002's first cloud phone instance corresponds to user's permanent residence city C, user's occupation is doctor, user's preferred activity is videos, and the user's preferred applications are video application 1, video application 2, and video application 3.

[0044] Table 1

[0045] Specifically, after obtaining the aforementioned user tag library, the server can determine whether each user tag in the library meets preset search criteria. These preset search criteria are used to initially filter the first cloud phone instances corresponding to the user tags. Afterward, the server can treat all user tags that meet the preset search criteria as candidate user tags. There can be one or more candidate user tags.

[0046] In one implementation, the aforementioned preset search criteria may include at least one of the following: (1) Search for user tags whose target exit time is the target time period from the user tag library. The target time period is the time period to which the cloud machine resource occupancy rate belongs when it is greater than the preset occupancy rate.

[0047] (2) Search the user tag library for user tags whose target exit time period is any candidate time period. The candidate time period includes a preset number of time periods before and / or after the target time period. The preset number can be set according to the actual situation. For example, taking the target time period as 5 o'clock to 6 o'clock and the preset number as 2, the candidate time periods can be 3 o'clock to 4 o'clock and 4 o'clock to 5 o'clock, or the candidate time periods can be 4 o'clock to 5 o'clock and 6 o'clock to 7 o'clock, or the candidate time periods can be 6 o'clock to 7 o'clock and 7 o'clock to 8 o'clock.

[0048] (3) Search the user tag library for the target exit time period with a preset character. The preset character is used to indicate that the first cloud phone instance corresponding to the user tag does not have a period of frequent cloud machine exit. For example, the preset character can be "full".

[0049] It should be noted that the above-mentioned preset search conditions may include one or more of the above conditions in combination. If the above-mentioned preset search conditions include multiple conditions, the server may perform the judgment simultaneously or in a preset execution order, without any limitation. For example, the server may first search for user tags whose target exit time is the target time period in the user tag library, then search for user tags whose target exit time is any candidate time period in the user tag library, and finally search for user tags whose target exit time is a preset character in the user tag library.

[0050] S302, input the candidate user tags into the target recycling prediction model to obtain the predicted recycling probability of the first cloud phone instance corresponding to the candidate user tags.

[0051] Specifically, after obtaining the aforementioned candidate user tags, the server can input these tags into the target recycling prediction model to obtain the predicted recycling probability of the first cloud phone instance corresponding to the candidate user tag. The target recycling prediction model is a pre-trained recycling prediction model. This model is used to predict the recyclability probability of the first cloud phone instance. The training process of this recycling prediction model will be described in detail below.

[0052] S303, perform a recycling operation on the first cloud phone instance whose predicted recycling probability is greater than the target recycling probability.

[0053] Specifically, after obtaining the predicted recovery probability, the server can determine whether this predicted recovery probability is greater than the target recovery probability. The target recovery probability is positively correlated with the user level of the user to which the first cloud phone instance belongs. That is, the higher the user level, the higher the target recovery probability. For example, the target recovery probability is 70% for ordinary users, 80% for monthly card users, 85% for quarterly card users, 90% for annual card users, and 95% for star-level users.

[0054] In some embodiments, if the predicted recycling probability is greater than the target recycling probability, it indicates that the first cloud phone instance is unlikely to be accessed again, meaning it will not be accessed by users again in the short term. Therefore, to ensure that other cloud machine clients can establish normal connections with the cloud phone instance, the server can recycle the first cloud phone instance whose predicted recycling probability is greater than the target recycling probability, so that cloud machine clients can establish connections with the recycled cloud phone. This achieves efficient recycling of cloud phones, reduces the waste of cloud machine resources caused by prolonged idle time, improves the utilization rate of cloud phones, and ultimately enhances the user experience.

[0055] In other embodiments, if the predicted recovery probability is less than or equal to the target recovery probability, it indicates that the first cloud phone instance is still likely to be accessed again, meaning it may not be accessed by the user again for a short period. Therefore, to ensure that the first cloud phone instance can be accessed normally by the user, the server may not recover the first cloud phone instance whose predicted recovery probability is less than or equal to the target recovery probability. This reduces the server load caused by frequent cloud phone recovery, improves the utilization rate of recovery resources, and thus improves the user experience.

[0056] In one implementation, the recycling process of the first cloud phone instance can specifically include: the server backing up the cloud machine data of the first cloud phone instance whose predicted recycling probability is greater than the target recycling probability. Then, the server initializes the first cloud phone instance after the data backup is complete. Next, the server changes the cloud machine status of the initialized first cloud phone instance to a pending connection state. Then, in response to a connection request from any cloud machine client, the server can connect the cloud machine client to the first cloud phone instance in the pending connection state.

[0057] In some cases, the number of cloud phone instances with a predicted recovery probability greater than the target recovery probability may be excessive. Therefore, to reduce the number of cloud phone instances to be recovered, the server can pre-set a preset recovery quantity. This way, after determining that the predicted recovery probability is greater than the target recovery probability, the server can designate the first cloud phone instance with the predicted recovery probability greater than the target recovery probability as the second cloud phone instance. Then, if the number of second cloud phone instances exceeds the preset recovery quantity, the server can determine the preset recovery quantity of second cloud phone instances according to their predicted recovery probabilities, from highest to lowest. The server can then perform the recovery operation on the preset number of second cloud phone instances. This reduces the number of cloud phones recovered but not effectively used, minimizes the number of recovered cloud phone instances, reduces unnecessary waste of recovery resources, and thus improves the utilization rate of recovery resources.

[0058] In one implementation, after the first cloud phone instance is reclaimed, the server can continue to monitor whether it receives access requests from cloud machine clients connected to the reclaimed first cloud phone instance within a preset time interval (e.g., 1 hour). In other words, it determines whether the reclaimed first cloud phone instance has been accessed again within the preset time interval. Subsequently, the server can adjust the probability fine-tuning value of the first cloud phone instance based on whether it has been accessed again within the preset time interval. Specifically, for example... Figure 4 As shown, cloud phone recycling methods may also include: S304, for each first cloud phone instance whose predicted recovery probability is greater than the target recovery probability, determine whether the first cloud phone instance is accessed again within a preset time interval.

[0059] Specifically, after reclaiming the first cloud phone instance whose predicted reclamation probability is greater than the target reclamation probability, the server can determine whether each first cloud phone instance has been accessed again within a preset time interval. If the first cloud phone instance is accessed again within the preset time interval, it means that the prediction result was not accurate, i.e., the predicted reclamation probability was not determined accurately. Therefore, the server can execute step S305. If the first cloud phone instance is not accessed again within the preset time interval, it means that the prediction result was accurate, i.e., the predicted reclamation probability was determined relatively accurately. Therefore, the server can continue to execute step S306 to further determine the adjustment strategy for the probability fine-tuning value.

[0060] S305 adds a preset value to the probability fine-tuning value of the first cloud phone instance.

[0061] The probability fine-tuning value determines the number of decision trees in the target recovery prediction model, and the number of decision trees is positively correlated with the predicted recovery probability. In other words, the larger the probability fine-tuning value of the first cloud phone instance, the fewer the number of decision trees, and the lower the predicted recovery probability.

[0062] In some cases, the process by which the server adjusts the number of decision trees in the target recovery prediction model using probability fine-tuning values ​​can be obtained through the following expression: Expression 1; Where, N adj N represents the number of decision trees after adjustment. opt This represents the original number of decision trees, and α is the adjusted probability fine-tuning value.

[0063] In some embodiments, after determining that the first cloud phone instance has been accessed again within a preset time interval, the server can add a preset value to the probability fine-tuning value of the first cloud phone instance. The preset value can be pre-set according to actual conditions, as long as it is less than 1; the specific value is not limited. For example, the preset value can be 0.1, 0.15, etc. In this way, by increasing the probability fine-tuning value, the number of decision trees can be reduced, thereby lowering the predicted recovery probability and providing a basis for accurately determining the predicted recovery probability subsequently.

[0064] For example, taking the initial number of decisions as 20 and the preset value as 0.1, after determining that the first cloud phone instance is accessed again within the preset time interval, the server can add 0.1 to the probability fine-tuning value (e.g., 0) of the first cloud phone instance, that is, the adjusted probability fine-tuning value is 0.1. Then, the server can input the adjusted probability fine-tuning value 0.1 and the original number 20 into the above expression one to obtain the adjusted number of decision trees as 18.

[0065] S306, determine whether the probability fine-tuning value of the first cloud phone instance is greater than zero.

[0066] Specifically, after determining that the first cloud phone instance has not been accessed again within the preset time interval, the server can further determine whether the probability fine-tuning value of the first cloud phone instance is greater than zero. If the probability fine-tuning value of the first cloud phone instance is greater than zero, it means that the number of decision trees in the target recovery prediction model can be further increased to improve the prediction accuracy of the recovery probability. Therefore, the server can execute step S307 to reduce the probability fine-tuning value of the first cloud phone instance. If the probability fine-tuning value of the first cloud phone instance is equal to zero, it means that the number of decision trees in the target recovery prediction model has reached its maximum value, that is, the number of decision trees cannot be increased further. Therefore, the server can execute step S308, that is, not adjust the probability fine-tuning value of the first cloud phone instance.

[0067] S307 subtracts a preset value from the probability adjustment value of the first cloud phone instance.

[0068] Specifically, after determining that the probability fine-tuning value of the first cloud phone instance is greater than zero, the server can subtract a preset value from the probability fine-tuning value of the first cloud phone instance. In this way, by lowering the probability fine-tuning value, the number of decision trees can be increased, thereby improving the predicted recovery probability and providing a foundation for the subsequent accurate determination of the predicted recovery probability.

[0069] For example, taking the initial number of decisions as 18 and the preset value as 0.1, after determining that the first cloud phone instance has not been accessed again within the preset time interval and that the probability fine-tuning value of the first cloud phone instance (e.g., 0.1) is greater than 0, the server can subtract 0.1 from the probability fine-tuning value of the first cloud phone instance, resulting in an adjusted probability fine-tuning value of 0. Then, the server can input the adjusted probability fine-tuning value of 0 and the original number of 18 into the above expression one to obtain an adjusted decision tree number of 20.

[0070] S308 does not adjust the probability fine-tuning value of the first cloud phone instance.

[0071] Specifically, after determining that the probability fine-tuning value of the first cloud phone instance is equal to zero, the server does not need to adjust the probability fine-tuning value of the first cloud phone instance. This reduces the possibility of over-adjustment of the probability fine-tuning value, providing a foundation for accurately determining the predicted recovery probability in the future.

[0072] In one implementation, the process of generating the aforementioned user tag library may specifically include: obtaining first data of the first cloud phone instance. This first data includes user behavior data and / or cloud machine usage data. User behavior data may include user ID, user cloud machine ID, cloud machine login time, application name used, application usage duration, cloud machine login region, cloud machine specifications, and user level, etc. Cloud machine usage data may include cloud machine entry time, cloud machine exit time, etc. Subsequently, the first data is analyzed to obtain the user tags corresponding to the first cloud phone instance.

[0073] In some cases, user tags will be used as an example, including the target exit time period, and combined with... Figure 5 The process for determining the target exit time period is shown in the diagram, detailing how the server determines the target exit time period.

[0074] S501, obtain the first data of the first cloud phone instance. The first data includes the time of entering the cloud server and the time of exiting the cloud server.

[0075] S502, combining the entry time of the cloud server, counts the time periods to which the exit time of the cloud server belongs, and obtains the number of times the first cloud phone instance exits the cloud server in each time period.

[0076] In this embodiment, considering that the cloud server exit time is generally accurate to the second or millisecond level, i.e., the cloud server exit time can be 16:47:55, the server can calculate the number of times the first cloud phone instance exits the cloud server in each time period according to a preset time unit. The preset time unit can be pre-set according to actual needs. In this embodiment, the preset time unit can be an hour, meaning the server can calculate the number of times the cloud server exits in each of the 24 time periods. In other embodiments, the preset time unit can also be a minute, a day, etc., and is not specifically limited.

[0077] In some cases, if there are no corresponding exit times within a time period (e.g., 3 PM to 4 PM), it means that the user did not exit the cloud server during that time period. Therefore, the server can record the number of exit times for that time period as 0. If there is a corresponding exit time within a time period but no corresponding entry time, it means that the user did not re-enter the cloud server during that time period. Therefore, the server can record the number of exit times for that time period as 1. If there are both corresponding exit times and corresponding entry times within a time period, it means that the user may have entered or exited the cloud server instance multiple times during that time period. Therefore, the server can use the number of exit times for that time period as the total number of exit times for that time period.

[0078] For example, if the cloud server entry time includes 7:05:51 and 16:50:25, and the cloud server exit time includes 16:47:55, 16:55:32, and 6:47:55, the cloud server can determine the time period to which the cloud server exit time 16:47:55 belongs as 16:00 to 17:00, the cloud server can determine the time period to which the cloud server exit time 16:55:32 belongs as 16:00 to 17:00, and the cloud server can determine the time period to which the cloud server exit time 6:47:55 belongs as 6:00 to 7:00.

[0079] It can be seen that within the 4 PM to 5 PM timeframe, there are two exit times for the cloud server (16:47:55 and 16:55:32) and one entry time for the cloud server (16:50:25). Therefore, the server can determine the number of cloud server exits during the 4 PM to 5 PM timeframe as 2. Furthermore, within the 6 AM to 7 AM timeframe, there is only one exit time (6:47:55), and no entry time. Therefore, the server can determine the number of cloud server exits during the 6 AM to 7 AM timeframe as 1. Simultaneously, the server can determine the number of cloud server exits for all timeframes outside of the 4 PM to 5 PM and 6 AM to 7 AM timeframes as 0.

[0080] In some cases, taking users with ID 10000 and ID 20002 as examples, as shown in Table 2, the number of times user 10000's first cloud phone instance exited the cloud server in various time periods were 1, 0, 1, 9, 7...1, 0, 0. That is, the number of times the cloud server exited was 1 from 0:00 to 1:00, 0 from 1:00 to 2:00, 1 from 2:00 to 3:00, 9 from 3:00 to 4:00, 7 from 4:00 to 5:00, 1 from 21:00 to 22:00, 0 from 22:00 to 23:00, and 0 from 23:00 to 0:00. The number of times user 20002's first cloud phone instance exited the cloud server in different time periods were 0, 0, 1, 0, 1...10, 8, 0. That is, the number of times the cloud server exited was 0 from 0:00 to 1:00, 0 from 1:00 to 2:00, 1 from 2:00 to 3:00, 0 from 3:00 to 4:00, 1 from 4:00 to 5:00, 10 from 21:00 to 22:00, 8 from 22:00 to 23:00, and 0 from 23:00 to 0:00.

[0081] Table 2

[0082] S503: Based on the preset calculation strategy and the number of times the first cloud phone instance exits the cloud machine in each time period, the target exit time period of the first cloud phone instance is obtained.

[0083] Specifically, after obtaining the number of times the first cloud phone instance exited the cloud server in each time period, the server can determine the target exit period for the first cloud phone instance based on a preset calculation strategy and the number of times the first cloud phone instance exited the cloud server in each time period. The preset calculation strategy may include a calculation strategy for the average, a calculation strategy for the weighted average, a calculation strategy for the moving average, a calculation strategy for Gaussian smoothing, etc.

[0084] In one implementation, when the preset calculation strategy is an average calculation strategy, the server can sum the number of times the first cloud phone instance exits the cloud server in each time period to obtain the total number of exits. Then, the server can determine the average number of exits based on the total number of exits and the number of time periods where the number of exits is greater than zero. Finally, the server can select the time periods where the number of exits exceeds the average number of exits as the target exit periods.

[0085] In some cases, the server uses an average calculation strategy to determine the target exit period, which can be obtained through the following expression two: Expression 2; Where i represents the time period, f(i) represents the target exit time period, and x i denoted by i, where i represents the number of times the user exits the cloud server, n represents the number of time periods in which the number of times the user exits the cloud server is greater than zero, a is the amplification factor, and b is the fine-tuning factor.

[0086] For example, taking the average number of exits as 6 as an example, as shown in Table 2, the target exit time periods for the first cloud phone instance of user 10000 are 3-4 and 4-5, and the target exit time periods for the first cloud phone instance of user 20002 are 21-22 and 22-23.

[0087] In another implementation, where the preset calculation strategy is a weighted average calculation strategy, the server can calculate the weighted average number of exits of the cloud machine in each time period according to the prediction weight of each time period, and obtain the weighted average number of exits.

[0088] In some cases, the server uses a weighted average calculation strategy. The process of determining the weighted average number of exits can be obtained through the following expression three: Expression 3; in, w represents the weighted average number of exits. i This represents the prediction weight for time period i.

[0089] Afterwards, the server can use the time period in which the number of times the cloud machine exits is greater than the weighted average number of exits as the target exit period.

[0090] In some cases, the process by which the server determines the target exit period based on a weighted average number of exits can be obtained through the following expression four: Expression 4; Where f(i) represents the target exit period.

[0091] In another implementation, where the preset calculation strategy is a moving average, for each time period, the first number of cloud machine exits within a first preset time period before that time period and the second number of cloud machine exits within a first preset time period after that time period are determined. Then, the server can calculate the average number of exits for the interval to which that time period belongs by averaging the first number of cloud machine exits, the second number of cloud machine exits, and the total number of cloud machine exits within that time period. Finally, the server can select the time period corresponding to the interval where the average number of exits is greater than the first preset average number of exits as the target exit time period.

[0092] In some cases, the server uses a moving average calculation strategy. The process of determining the average number of exits over a period can be obtained through the following expression five: Expression 5; in, This represents the average number of exits within the interval to which time period i belongs; w is equivalent to a window used to represent the first preset time period.

[0093] In another implementation, with the above-mentioned preset calculation strategy being Gaussian smoothing, the target weights for each time period are determined based on the Gaussian function.

[0094] In some cases, the server employs a Gaussian smoothing calculation strategy, and the process of determining the target weights for each time period can be obtained through the following expression six: Expression 6; Where x represents the time period, G x σ represents the target weight for time period x, and σ is the standard deviation, which controls the width of the Gaussian function G().

[0095] Next, for each time period, the server can sum the number of cloud machine exits within that time period and the number of cloud machine exits within a second preset time period following that time period to obtain the target number of cloud machine exits. Then, the server can determine the target weighted average number of exits for that time period based on this target number of exits and the target weight for that time period. Finally, the server can define the time periods where the target average number of exits is greater than the second preset average number of exits as the target exit time periods.

[0096] In some cases, the process by which the server determines the average number of exits after Gaussian smoothing can be obtained through the following expression seven: Expression 7; in, G represents the number of times the target weighted average is calculated for time period i. i Let x represent the target weight for time period i, n represent the second preset time period, and x represent the target weight for time period i. (i+n) This indicates the number of times the target has exited the cloud server.

[0097] It should be noted that the process for determining the other tags included in the user tags, and the process for determining the target exit time period mentioned above, will not be elaborated here.

[0098] This embodiment also provides a method for training a recycling prediction model, such as... Figure 6 As shown, the method includes: S601, Obtain the sample dataset.

[0099] The sample dataset includes multiple sample data sets and their corresponding actual recall probabilities. The sample data includes historical user tags. Historical user tags may include historical user ID, historical logout time period, historical user's city of residence, historical user's occupation, historical user's preferred items, and the corresponding preferred applications for those items.

[0100] For example, the above sample dataset can be represented by the following set: [(3-4 PM, 4-5 PM, City A, Students, Games, Sports, ..., Game App 1, Game App 2, Game App 3, ..., 0), ..., (21:00-22:00, 22:00-23:00, City C, Doctor, Video, Cycling, ..., Video App 1, Video App 2, Video App 3, ..., 0)).

[0101] S602, for each sample data, input the sample data into the pre-built recovery prediction model to obtain the predicted recovery probability corresponding to the sample data.

[0102] The recovery prediction model includes multiple decision trees, and the number of decision trees is positively correlated with the predicted recovery probability. In other words, the fewer the decision trees, the lower the predicted recovery probability.

[0103] Specifically, feature extraction is performed on the sample data to obtain the corresponding feature vector. The feature vector can be obtained through X. i [x (i,1) x (i,2) , ..., x (i,n) The data is represented by a vector, i, where i is the sample data. Then, the feature vectors corresponding to the sample data are input into each decision tree to obtain the prediction results of each decision tree.

[0104] In some cases, the process of determining the prediction results of the above decision tree can be obtained through the following expression: Expression 8; in, Let T represent the prediction result of decision tree i, j represent the number of decision trees, and T represent the prediction result of decision tree i. j Let X represent the i-th decision tree. i This represents the feature vector corresponding to the sample data.

[0105] Then, the prediction results of each decision tree are processed to obtain the predicted recovery probability corresponding to the sample data.

[0106] In some cases, the process of determining the predicted recovery probability can be obtained through the following expression nine: Expression Nine; in, This represents the predicted recovery probability corresponding to the sample data.

[0107] S603, based on the predicted recovery probability and the actual recovery probability corresponding to the sample data, adjust the parameters of the pre-built recovery prediction model to obtain the target recovery prediction model.

[0108] The objective function of the aforementioned recovery prediction model includes a loss function and a regularization function. The loss function characterizes the difference between the predicted recovery probability and the actual recovery probability, aiming to make the predicted recovery probability as close as possible to the actual recovery probability. The regularization function constrains the complexity of the recovery prediction model to reduce the possibility of overfitting.

[0109] In some cases, the objective function of the above recycling prediction model can be obtained by the following expression: Expression 10; Where Obj represents the objective function of the recovery prediction model. Let Ω represent the loss function and Ω represent the regularization function.

[0110] In some embodiments, considering that the above-described recycling prediction model is a binary classification prediction model, meaning that when performing binary classification prediction, the model typically outputs a probability value representing the likelihood of belonging to the target category (recyclable category), a logarithmic loss function is used as the loss function for the recycling prediction model during training to adapt it to binary classification prediction models. Specifically, this loss function can be obtained through the following expression eleven: Expression 11; in, Let y represent the loss function. i y represents the actual recovery probability. i ^ This indicates the predicted recovery probability.

[0111] In other embodiments, the above regular function can be obtained by expression twelve as shown below: Expression twelve; Among them, Ω(T) j ) represents the regularization function; T represents the number of leaf nodes in the decision tree; γ is a constant representing the fixed penalty for each leaf node; λ is the penalty coefficient for the weights, which depends on the output value (i.e., the weight) of the leaf node; w j This represents the weight of decision tree j.

[0112] In one implementation, in response to a cloud server resource occupancy rate exceeding a preset occupancy rate, user tags matching preset search criteria from a user tag library are selected as candidate user tags. This user tag library includes user tags corresponding to each first cloud phone instance, where a first cloud phone instance is one of multiple cloud phone instances that has already established a connection with a cloud server client. The server then inputs the candidate user tags into a target recycling prediction model to obtain the predicted recycling probability of the first cloud phone instance corresponding to each candidate user tag. Finally, the server can perform a recycling operation on first cloud phone instances whose predicted recycling probability is greater than the target recycling probability. The following will combine... Figure 2 The structure shown and Figure 7 The cloud phone recycling process shown details how the server recycles the first cloud phone instance.

[0113] S7001, In response to the cloud machine resource occupancy rate being greater than the preset occupancy rate, the cloud machine recycling module sends a tag search instruction to the data analysis module.

[0114] The tag lookup instruction is used to enable the data analysis module to determine candidate user tags from the user tag library.

[0115] In some embodiments, the cloud server recycling module can calculate the occupancy rate of cloud server resources. The occupancy rate is the ratio between the number of first cloud phone instances and the total number of cloud phone instances managed by the server. The cloud server recycling module then determines whether this occupancy rate is greater than a preset occupancy rate. If the occupancy rate is greater than the preset occupancy rate, it indicates that the usage of cloud phone instances on the server is high. Therefore, to achieve accurate recycling of cloud phone instances, the cloud server recycling module can send a tag lookup command to the data analysis module. If the occupancy rate is less than or equal to the preset occupancy rate, it indicates that the usage of cloud phone instances on the server is low. Therefore, to reduce unnecessary waste of recycling resources, the cloud server recycling module does not recycle the first cloud phone instance.

[0116] S7002. When the data analysis module receives a tag search instruction from the cloud machine recycling module, it selects user tags in the user tag library that meet the preset search conditions as candidate user tags.

[0117] Specifically, after receiving the tag search instruction from the cloud server recycling module, the data analysis module can obtain the user tag library and determine whether each user tag in the user tag library meets the preset search conditions. The preset search conditions are used to initially filter the first cloud phone instance corresponding to the user tag. Afterwards, the data analysis module can treat all user tags that meet the preset search conditions as candidate user tags. There can be one or more candidate user tags.

[0118] S7003, the data analysis module sends candidate user tags to the model prediction module.

[0119] S7004. Upon receiving the candidate user tags sent by the data analysis module, the model prediction module inputs the candidate user tags into the target recycling prediction model to obtain the predicted recycling probability of the first cloud phone instance corresponding to the candidate user tags.

[0120] The target recycling prediction model is a pre-trained recycling prediction model. This model is used to predict the recyclability probability of the first cloud phone instance.

[0121] S7005, The model prediction module sends the predicted recycling probability of the first cloud phone instance to the cloud phone recycling module.

[0122] S7006. Upon receiving the predicted recovery probability of the first cloud mobile phone instance sent by the model prediction module, the cloud mobile phone recovery module performs a recovery operation on the first cloud mobile phone instance whose predicted recovery probability is greater than the target recovery probability.

[0123] Specifically, after receiving the predicted recovery probability of the first cloud phone instance from the model prediction module, the cloud device recovery module can determine whether the predicted recovery probability is greater than the target recovery probability. The target recovery probability is positively correlated with the user level of the user to whom the first cloud phone instance belongs. In other words, the higher the user level, the higher the target recovery probability.

[0124] In some embodiments, if the predicted recycling probability is greater than the target recycling probability, it indicates that the first cloud phone instance is unlikely to be accessed again, meaning it will not be accessed by users again in the short term. Therefore, to ensure that other cloud machine clients can establish normal connections with the cloud phone instance, the cloud machine recycling module can recycle the first cloud phone instance whose predicted recycling probability is greater than the target recycling probability, so that cloud machine clients can establish connections with the recycled cloud phone. This achieves efficient recycling of cloud phones, reduces the waste of cloud machine resources caused by prolonged idle time, improves the utilization rate of cloud phones, and ultimately enhances the user experience.

[0125] In other embodiments, if the predicted recovery probability is less than or equal to the target recovery probability, it indicates that the first cloud phone instance is still likely to be accessed again, meaning it may not be accessed by the user again for a short period. Therefore, to ensure that the first cloud phone instance can be accessed normally by the user, the cloud server recovery module may not recover the first cloud phone instance whose predicted recovery probability is less than or equal to the target recovery probability. This reduces the server load caused by frequent cloud phone recovery, improves the utilization rate of recovery resources, and thus improves the user experience.

[0126] S7007, the cloud machine recycling module sends cloud machine monitoring commands to the information feedback module.

[0127] Specifically, after reclaiming the first cloud phone instance whose predicted reclamation probability is greater than the target reclamation probability, the cloud machine reclamation module can send a cloud machine monitoring command to the information feedback module. This cloud machine monitoring command is used by the information feedback module to determine whether the reclaimed first cloud phone instance has been accessed again within a preset time interval.

[0128] S7008: Upon receiving a cloud server monitoring instruction from the cloud server recycling module, the information feedback module adjusts the probability fine-tuning value of the first cloud mobile phone instance based on whether the first cloud mobile phone instance is accessed again within a preset time interval.

[0129] Specifically, after receiving the cloud server monitoring command from the cloud server recycling module, the information feedback module can adjust the probability fine-tuning value of the first cloud phone instance based on whether it has been accessed again within a preset time interval. The probability fine-tuning value determines the number of decision trees in the target recycling prediction model, and the number of decision trees is positively correlated with the predicted recycling probability. In other words, the larger the probability fine-tuning value of the first cloud phone instance, the fewer the number of decision trees, and the lower the predicted recycling probability.

[0130] In some embodiments, the information feedback module can determine whether each first cloud phone instance has been accessed again within a preset time interval. If the first cloud phone instance is accessed again within the preset time interval, it indicates that the prediction result was not accurate, meaning the predicted recovery probability was not determined correctly. Therefore, the information feedback module can add a preset value to the probability fine-tuning value of the first cloud phone instance. In this way, by increasing the probability fine-tuning value, the number of decision trees can be reduced, thereby lowering the predicted recovery probability and providing a basis for accurately determining the predicted recovery probability in the future.

[0131] In other embodiments, if the first cloud phone instance is not accessed again within a preset time interval, it indicates that the prediction result has been hit, meaning the predicted recovery probability is relatively accurate. Therefore, the information feedback module can further determine whether the probability fine-tuning value of the first cloud phone instance is greater than zero. If the probability fine-tuning value of the first cloud phone instance is greater than zero, it means that the number of decision trees in the target recovery prediction model can be further increased to improve the prediction accuracy of the recovery probability. Therefore, the information feedback module can subtract a preset value from the probability fine-tuning value of the first cloud phone instance. In this way, by lowering the probability fine-tuning value, the number of decision trees can be increased, thereby improving the predicted recovery probability and providing a foundation for subsequent accurate determination of the predicted recovery probability.

[0132] In some other embodiments, if the probability fine-tuning value of the first cloud phone instance is zero, it indicates that the number of decision trees in the target recovery prediction model has reached its maximum, meaning the number of decision trees cannot be increased further. Therefore, the information feedback module does not adjust the probability fine-tuning value of the first cloud phone instance. This reduces the possibility of over-adjustment of the probability fine-tuning value, providing a foundation for accurately determining the predicted recovery probability in the future.

[0133] S7009, The information feedback module sends the adjusted probability fine-tuning value to the model prediction module.

[0134] In some cases, the information feedback module can also send the adjusted probability fine-tuning value to the data analysis module so that the data analysis module can record the adjusted probability fine-tuning value in the user tag corresponding to the first cloud phone instance for easy subsequent use.

[0135] S7010: Upon receiving the adjusted probability fine-tuning value from the information feedback module, the model prediction module adjusts the number of decision trees in the target recovery prediction model based on the adjusted probability fine-tuning value.

[0136] Specifically, after receiving the adjusted probability fine-tuning value from the information feedback module, the model prediction module can adjust the number of decision trees in the target recovery prediction model based on the adjusted probability fine-tuning value. The probability fine-tuning value is negatively correlated with the number of decision trees in the target recovery prediction model; that is, the larger the probability fine-tuning value, the fewer decision trees are in the target recovery prediction model.

[0137] Furthermore, this embodiment provides a cloud phone recycling device, such as... Figure 8 As shown, the device includes: a tag determination unit 81, a probability determination unit 82, and a cloud machine recycling unit 83.

[0138] The tag determination unit 81 is used to respond to the fact that the occupancy rate of cloud machine resources is greater than the preset occupancy rate, and to use user tags in the user tag library that meet the preset search conditions as candidate user tags; wherein, the user tag library includes user tags corresponding to each first cloud phone instance, and the first cloud phone instance is a cloud phone instance among multiple cloud phone instances that has established a connection with the cloud machine client. The probability determination unit 82 is used to input the candidate user tags into the target recycling prediction model to obtain the predicted recycling probability of the first cloud phone instance corresponding to the candidate user tags; The cloud phone recycling unit 83 is used to recycle the first cloud phone instance whose predicted recycling probability is greater than the target recycling probability.

[0139] Furthermore, in one possible implementation of this embodiment, such as Figure 8As shown, the tag determination unit 81 is further configured to use user tags in the user tag library whose target exit time is the target time period as candidate user tags; wherein, the target time period is the time period to which the cloud machine resource occupancy rate belongs when it is greater than a preset occupancy rate; and / or, Determine a predetermined number of candidate time periods before and / or after the target time period, and use user tags from the user tag library whose target exit time period is any of the candidate time periods as candidate user tags; and / or, User tags in the user tag library whose target exit time period is a preset character are selected as candidate user tags.

[0140] Furthermore, in one possible implementation of this embodiment, such as Figure 8 As shown, the tag determination unit 81 is also used to obtain the first data of the first cloud phone instance; wherein, the first data includes the time of entering the cloud phone and the time of exiting the cloud phone; By combining the entry time into the cloud server, the time periods to which the exit time belongs are statistically analyzed to obtain the number of times the first cloud phone instance exits the cloud server in each time period; Based on the preset calculation strategy and the number of times the first cloud phone instance exits the cloud machine in each time period, the target exit time period is obtained.

[0141] Furthermore, in one possible implementation of this embodiment, such as Figure 8 As shown, the tag determination unit 81 is also used to add up the number of times the first cloud phone instance exits the cloud machine in each time period to obtain the total number of times it exits the cloud machine; The average number of exits is determined based on the total number of cloud server exits and the number of time periods in which the number of cloud server exits is greater than zero. The period during which the number of times the cloud server is exited exceeds the average number of exits will be designated as the target exit period.

[0142] Furthermore, in one possible implementation of this embodiment, such as Figure 8 As shown, the cloud phone recycling unit 83 is also used to use the first cloud phone instance with a predicted recycling probability greater than the target recycling probability as the second cloud phone instance. If the number of cloud machines in the second cloud phone instance is greater than the preset number to be recycled, the preset number of second cloud phone instances to be recycled is determined from high to low according to the predicted recycling probability of each second cloud phone instance. Perform the recycling operation on the second cloud phone instance with the preset recycling quantity.

[0143] Furthermore, in one possible implementation of this embodiment, such as Figure 9 As shown, the original mobile phone recycling device also includes a numerical adjustment unit 84.

[0144] The numerical adjustment unit 84 is used to determine whether the first cloud phone instance is accessed again within a preset time interval for each first cloud phone instance whose predicted recovery probability is greater than the target recovery probability. If the first cloud phone instance is accessed again within a preset time interval, the probability fine-tuning value of the first cloud phone instance is increased by a preset value; wherein, the probability fine-tuning value is used to determine the number of decision trees in the target recovery prediction model, and the number of decision trees is positively correlated with the predicted recovery probability.

[0145] Furthermore, in one possible implementation of this embodiment, such as Figure 9 As shown, the numerical adjustment unit 84 is also used to, if the probability fine-tuning value of the first cloud phone instance is greater than zero and the first cloud phone instance has not been accessed again within a preset time interval, subtract a preset value from the probability fine-tuning value of the first cloud phone instance; or, If the first cloud phone instance is not accessed again within a preset time interval, and the probability fine-tuning value of the first cloud phone instance is equal to zero, the probability fine-tuning value of the first cloud phone instance will not be adjusted.

[0146] Furthermore, in one possible implementation of this embodiment, such as Figure 9 As shown, the original mobile phone recycling device also includes a model training unit 80.

[0147] Model training unit 80 is used to acquire sample datasets; the sample datasets include multiple sample data and the actual recovery probabilities corresponding to the sample data, and the sample data include historical user tags; For each sample data, the sample data is input into a pre-built recycling prediction model to obtain the predicted recycling probability corresponding to the sample data; Based on the predicted recovery probability and the actual recovery probability corresponding to the sample data, the parameters of the pre-constructed recovery prediction model are adjusted to obtain the target recovery prediction model.

[0148] Furthermore, in one possible implementation of this embodiment, such as Figure 9 As shown, the model training unit 80 is also used to extract features from the sample data to obtain the feature vector corresponding to the sample data. The feature vectors corresponding to the sample data are input into each decision tree to obtain the prediction results of each decision tree; The prediction results of each decision tree are processed to obtain the predicted recovery probability corresponding to the sample data.

[0149] It should be noted that other corresponding descriptions of the various functional units involved in the cloud phone recycling device provided in this embodiment can be found in the description of the cloud phone recycling method in the above embodiments, and will not be repeated here.

[0150] Based on the cloud phone recycling method shown in the above embodiments, this embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method shown in the above embodiments.

[0151] Based on the methods shown in the above embodiments, this embodiment also provides a computer program product on which a computer program is stored, and when the computer program product is executed by a processor, it implements the methods shown in the above embodiments.

[0152] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0153] Based on the method shown in the above embodiments, and Figure 8 In order to achieve the above objectives, the virtual device embodiment shown in this application also provides a server, such as a cloud server, which includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the method shown in the above embodiment.

[0154] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0155] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0156] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0157] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented using software plus necessary general-purpose hardware platforms, or it can be implemented through hardware. Compared with current related technologies, when the occupancy rate of cloud machine resources is greater than the preset occupancy rate, it indicates that the usage of cloud phones in the server is relatively high. Therefore, in order to reduce the situation where cloud phones are occupied for a long time without being used, the server can predict the recyclability probability of cloud phone instances by using the user tags corresponding to the cloud phone instances (i.e., predict the recycling probability), and determine the cloud phone instances in the server that can be recycled. In this way, efficient recycling of cloud phones can be achieved, reducing the waste of cloud machine resources caused by excessive idle time of cloud phones, ensuring that other cloud machine clients can establish connections with cloud phones, improving the utilization rate of cloud phones, and thus improving the user experience.

[0158] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0159] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A cloud phone recycling method, characterized in that, Applied to a server configured to manage a plurality of cloud phone instances, the method comprises: in response to the occupancy rate of the cloud machine resource being greater than a preset occupancy rate, taking a user tag in a user tag library that meets a preset search condition as a candidate user tag; wherein the user tag library comprises user tags corresponding to each first cloud phone instance, and the first cloud phone instance is a cloud phone instance in the plurality of cloud phone instances that has established a connection with a cloud machine client; inputting the candidate user tag into a target recycling prediction model to obtain a predicted recycling probability of the first cloud phone instance corresponding to the candidate user tag; performing a recycling operation on the first cloud phone instance whose predicted recycling probability is greater than a target recycling probability.

2. The method of claim 1, wherein, The user tag comprises a target exit time period, and the taking of the user tag in the user tag library that meets the preset search condition as the candidate user tag comprises: taking a user tag in the user tag library whose target exit time period is a target time period as the candidate user tag; wherein the target time period is a time period to which a time corresponding to a time when the occupancy rate of the cloud machine resource is greater than the preset occupancy rate belongs; and / or, determining a preset number of candidate time periods before and / or after the target time period, and taking a user tag in the user tag library whose target exit time period is any of the candidate time periods as the candidate user tag; and / or, taking a user tag in the user tag library whose target exit time period is a preset character as the candidate user tag.

3. The method of claim 2, wherein, The method further comprises: obtaining first data of the first cloud phone instance; wherein the first data comprises a cloud machine entry time and a cloud machine exit time; combining the cloud machine entry time, counting a time period to which the cloud machine exit time belongs to obtain the number of times of exiting the cloud machine of the first cloud phone instance in each time period; obtaining the target exit time period according to a preset calculation strategy and the number of times of exiting the cloud machine of the first cloud phone instance in each time period.

4. The method of claim 3, wherein, In the case where the preset calculation strategy is an average number calculation strategy, the obtaining of the target exit time period according to the preset calculation strategy and the number of times of exiting the cloud machine of the first cloud phone instance in each time period comprises: adding the number of times of exiting the cloud machine of the first cloud phone instance in each time period to obtain a total number of times of exiting the cloud machine; determining an average number of times of exiting the cloud machine according to the total number of times of exiting the cloud machine and the number of time periods in which the number of times of exiting the cloud machine is greater than zero; taking a time period in which the number of times of exiting the cloud machine is greater than the average number of times of exiting the cloud machine as the target exit time period.

5. The method according to any one of claims 1 to 4, characterized in that, The performing of the recycling operation on the first cloud phone instance whose predicted recycling probability is greater than the target recycling probability comprises: taking the first cloud phone instance whose predicted recycling probability is greater than the target recycling probability as a second cloud phone instance; in the case where the number of cloud machines of the second cloud phone instance is greater than a preset recycling number, determining the second cloud phone instances of the preset recycling number from high to low according to the predicted recycling probabilities of the second cloud phone instances; performing a recycling operation on the second cloud phone instances of the preset recycling number.

6. The method according to any one of claims 1 to 4, characterized in that, The target recycling probability is positively correlated with a user level of a user to which the first cloud phone instance belongs.

7. The method according to any one of claims 1 to 4, characterized in that, The method further comprises: For each first cloud mobile phone instance with a predicted recovery probability greater than a target recovery probability, determining whether the first cloud mobile phone instance is accessed again within a preset time interval; In a case where the first cloud mobile phone instance is accessed again within a preset time interval, adding a preset value to a probability fine-tuning value of the first cloud mobile phone instance; wherein the probability fine-tuning value is used to determine a number of decision trees in the target recovery prediction model, and the number of decision trees is positively correlated with the predicted recovery probability.

8. The method of claim 7, wherein, The method further comprises: In a case where the first cloud mobile phone instance is not accessed again within a preset time interval, if the probability fine-tuning value of the first cloud mobile phone instance is greater than zero, subtracting a preset value from the probability fine-tuning value of the first cloud mobile phone instance; or, In a case where the first cloud mobile phone instance is not accessed again within a preset time interval, if the probability fine-tuning value of the first cloud mobile phone instance is equal to zero, not adjusting the probability fine-tuning value of the first cloud mobile phone instance.

9. The method of claim 7, wherein, The method further comprises: Obtaining a sample data set; wherein the sample data set comprises a plurality of sample data and real recovery probabilities corresponding to the sample data, and the sample data comprises historical user labels; For each sample data, inputting the sample data into a pre-constructed recovery prediction model to obtain a predicted recovery probability corresponding to the sample data; According to the predicted recovery probability corresponding to the sample data and the real recovery probability corresponding to the sample data, adjusting parameters of the pre-constructed recovery prediction model to obtain the target recovery prediction model.

10. The method of claim 9, wherein, The recovery prediction model comprises a plurality of decision trees, and the number of decision trees is positively correlated with the predicted recovery probability; the inputting of the sample data into the pre-constructed recovery prediction model to obtain the predicted recovery probability corresponding to the sample data comprises: performing feature extraction on the sample data to obtain a feature vector corresponding to the sample data; inputting the feature vector corresponding to the sample data into each decision tree to obtain a prediction result of each decision tree; processing the prediction results of each decision tree to obtain the predicted recovery probability corresponding to the sample data.

11. A cloud phone recycling device, comprising: Comprise: A label determination unit is configured to, in response to an occupancy rate of cloud machine resources being greater than a preset occupancy rate, take a user label in a user label library that meets a preset search condition as a candidate user label; wherein the user label library comprises user labels corresponding to each first cloud mobile phone instance, and the first cloud mobile phone instance is a cloud mobile phone instance in the plurality of cloud mobile phone instances that has established a connection with a cloud machine client; A probability determination unit is configured to input the candidate user label into a target recovery prediction model to obtain a predicted recovery probability of a first cloud mobile phone instance corresponding to the candidate user label; A cloud machine recovery unit is configured to perform a recovery operation on the first cloud mobile phone instance with a predicted recovery probability greater than a target recovery probability.

12. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program, which when executed by the processor, implements the method of any one of claims 1 to 10.

13. A server comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, The processor, when executing the computer program, implements the method of any one of claims 1 to 10.

14. A computer program product having stored thereon a computer program, characterized in that, The computer program product, which when executed by the processor, implements the method of any one of claims 1 to 10.

Images