Methods and systems for preloading applications and generating prediction models

1. An application preloading method comprising:

obtaining application usage state information of a terminal and contextual information of the terminal;

inputting the application usage state information and the contextual information into a pre-generated prediction model, the prediction model being configured to predict application startup and to calculate at least one prediction value for the application startup, wherein the prediction model is pre-generated according to usage association information of applications within a predetermined time period and contextual information of the terminal corresponding to the usage association information; and

determining an application to be started according to the at least one prediction value, and preloading the application to be started;

wherein generating the prediction model for predicting the application startup comprises:

obtaining a user behavior sample within the predetermined time period, wherein the user behavior sample comprises usage association information of at least two applications, wherein the usage association information comprises application usage state information at each sampling point in time within the predetermined time period and a time sequence of using applications within the predetermined time period;

extracting the contextual information of the terminal corresponding to the usage association information of the at least two applications at each sampling point in time within the predetermined time period; and

inputting the usage association information and the contextual information as training data into a predetermined algorithm model, and obtaining the prediction model for predicting the application startup through training:

wherein the algorithm model comprises an input layer, a hidden layer, and an output layer; and

wherein:

a cardinality of elements in the input layer is determined according to a vector dimension of the usage association information and the contextual information of the terminal,

the hidden layer is fully connected with the inout layer, and

a cardinality of elements in the output layer is determined according to a number of applications,

an error function used by the algorithm model is a cross entrpy loss function that is indicated as,

J

=

∑

k

=

1

C

⁢

y

k

⁢

⁢

log

⁡

(

y

^

k

)

,

wherein y k indicates a standard value of an application usage state,

ŷ k indicates a prediction value of the application usage state, and

J indicates a cross entropy of the algorithm model; and

C=M+1, wherein M indicates the number of applications.

2. The application preloading method according to claim 1 , wherein the application usage state information comprises information of an application currently in use, or information indicating that no application is currently in use.

3. The application preloading method according to claim 1 , wherein obtaining the user behavior sample within the predetermined time period comprises:

sorting applications according to usage frequencies of the applications within the predetermined time period;

determining at least two target applications according to a result of the sorting; and

determining the usage association information based on usage state information of the target applications.

4. The application preloading method according to claim 3 , wherein determining the usage association information based on usage state information of the target applications comprises:

sampling usage logs of the target applications according to a predetermined sampling period to determine usage state information of the target applications at each sampling point in time; and

associating the usage state information of the target applications according to the sampling points in time to determine the usage association information.

5. The application preloading method according to claim 1 , wherein the contextual information comprises one or more of:

scene information indicating an environmental state in which the terminal is located, or

state information of the terminal.

6. The application preloading method according to claim 5 , wherein:

the scene information comprises at least one of time information or location information; and

the state information comprises one or more of:

information indicating a display screen status of on or off,

power quantity information,

network connection information, or

information indicating whether terminal is in a charging status.

7. A prediction model generation method comprising:

obtaining a user behavior sample within a predetermined time period, wherein the user behavior sample comprises usage association information of at least two applications, wherein the usage association information comprises application usage state information at each sampling point in time within the predetermined time period and a time sequence of using applications within the predetermined time period;

extracting contextual information of a terminal corresponding to the usage association information of the at least two applications at each sampling point in time within the predetermined time period;

inputting the usage association information and the extracted contextual information as training data into a predetermined algorithm model; and

obtaining the prediction model for predicting application startup through training:

wherein the algorithm model comprises an input layer, a hidden layer, and an output layer; and

wherein:

a cardinality of elements in the input layer is determined according to a vector dimension of the usage association information and the contextual information of the terminal,

the hidden layer is fully connected with the inout layer, and

a cardinality of elements in the output layer is determined according to a number of applications,

an error function used by the algorithm model is a cross entrpy loss function that is indicated as,

J

=

∑

k

=

1

C

⁢

y

k

⁢

log

⁡

(

y

^

k

)

,

wherein y k indicates a standard value of an application usage state,

ŷ k indicates a prediction value of the application usage state, and

J indicates a cross entropy of the algorithm model; and

C=M+1, wherein M indicates the number of applications.

8. The prediction model generation method according to claim 7 , wherein obtaining the user behavior sample within the predetermined time period comprises:

sorting applications according to usage frequencies of the applications within the predetermined time period;

determining at least two target applications according to a result of the sorting; and

determining the usage association information based on usage state information of the target applications.

9. The prediction model generation method according to claim 8 , wherein the determining the usage association information based on usage state information of the target applications comprises:

sampling usage logs of the target applications according to a predetermined sampling period to determine usage state information of the target applications at each sampling point in time; and

associating the usage state information of the target applications according to the sampling points in time to determine the usage association information.

10. The prediction model generation method according to claim 9 , wherein inputting the usage association information and the contextual information as training data into the predetermined algorithm model, and obtaining the prediction model for predicting application startup comprises:

training the algorithm model according to the usage state information corresponding to each sampling point in time in the usage association information and the contextual information of the terminal, to obtain the prediction model for predicting the application startup.

11. The prediction model generation method according to claim 7 ,

wherein the at least two applications are included in an application usage record; and

the method further comprises:

deleting an invalid application usage record entry in the application usage record within the predetermined time period.

12. The prediction model generation method according to claim 7 , wherein the contextual information comprises one or more of:

scene information indicating an environmental state in which the terminal is located, or

state information of the terminal.

13. The prediction model generation method according to claim 12 , wherein:

the scene information comprises one or more of time information or location information; and

the state information comprises one or more of:

information indicating a display screen status of on or off,

power quantity information,

network connection information, or

information indicating whether the terminal is in a charging status.

14. A terminal comprising: a processor, and a memory storing a computer program that is executable by the processor to perform steps of:

obtaining application usage state information of a terminal and contextual information of the terminal;

inputting the application usage state information and the contextual information into a pre-generated prediction model, the prediction model being configured to predict application startup and to calculate at least one prediction value for the application startup, wherein the prediction model is pre-generated according to usage association information of applications within a predetermined time period and contextual information of the terminal corresponding to the usage association information; and

determining an application to be started according to the at least one prediction value, and preloading the application to be started;

wherein generating the prediction model for predicting the application startup comprises:

obtaining a user behavior sample within the predetermined time period, wherein the user behavior sample comprises usage association information of at least two applications, wherein the usage association information comprises application usage state information at each sampling point in time within the predetermined time period and a time sequence of using applications within the predetermined time period;

extracting the contextual information of the terminal corresponding to the usage association information of the at least two applications at each sampling point in time within the predetermined time period; and

inputting the usage association information and the contextual information as training data into a predetermined algorithm model, and obtaining the prediction model for predicting the application startup through training:

wherein the algorithm model comprises an input layer, a hidden layer, and an output layer; and

wherein:

a cardinality of elements in the input layer is determined according to a vector dimension of the usage association information and the contextual information of the terminal,

the hidden layer is fully connected with the inout layer, and

a cardinality of elements in the output layer is determined according to a number of applications,

an error function used by the algorithm model is a cross entrpy loss function that is indicated as,

J

=

∑

k

=

1

C

⁢

y

k

⁢

log

⁡

(

y

^

k

)

,

wherein y k indicates a standard value of an application usage state,

ŷ k indicates a prediction value of the application usage state, and

J indicates a cross entropy of the algorithm model; and

C=M+1, wherein M indicates the number of applications.

15. The terminal according to claim 14 , wherein the application usage state information comprises information of an application currently in use, or information indicating that no application is currently in use.

16. The terminal according to claim 14 , wherein obtaining the user behavior sample within the predetermined time period comprises:

sorting applications according to usage frequencies of the applications within the predetermined time period;

determining at least two target applications according to a result of the sorting; and

determining the usage association information based on usage state information of the target applications.

17. The terminal according to claim 16 , wherein determining the usage association information based on usage state information of the target applications comprises:

sampling usage logs of the target applications according to a predetermined sampling period to determine usage state information of the target applications at each sampling point in time; and

associating the usage state information of the target applications according to the sampling points in time to determine the usage association information.