Application recommendation method, probability prediction model generation method and terminal equipment
By acquiring multiple types of data from terminal devices and inputting them into a pre-trained probability prediction model, the target category is determined, thus solving the problem of low application recommendation accuracy and achieving more accurate application recommendations and improved user experience.
Patent Information
- Application Number
- CN202310601181.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-25
- Publication Date
- 2024-11-26
AI Technical Summary
In existing technologies, the accuracy of app recommendations is low and the user experience is poor because users open apps at different times on different days.
By acquiring multi-class data from the terminal device at the current moment, inputting it into a pre-trained probabilistic prediction model, determining the target category in the candidate application set, and displaying recommendation information on the interface, the accuracy of recommendations is improved by using probabilistic prediction model generation methods to train and filter sample data.
It improves the accuracy of app recommendations, ensuring that terminal devices can accurately display candidate applications that users want to use, thus enhancing the user experience.
Smart Images

Figure CN119025764A_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of terminal technology, and in particular relates to an application recommendation method, a probability prediction model generation method and a terminal device. Background Art
[0002] With the increasing popularity of terminal devices such as smart phones, more and more applications (Application, APP) are installed on terminal devices, which greatly meets the convenience of users' lives.
[0003] Currently, recommended applications for the current period are usually predicted based on applications opened by users during the same period on historical dates corresponding to the current period. However, due to the large differences in applications opened by users during the same period on different dates, the accuracy of currently recommended applications is low, resulting in a poor user experience. Summary of the invention
[0004] The present application provides an application recommendation method, a probability prediction model generation method and a terminal device, which can ensure the accuracy of the application recommended by each terminal device.
[0005] In a first aspect, the present application provides an application recommendation method, the method comprising:
[0006] Acquire first data of the terminal device at the current moment, the first data including N types of data, the N types of data being data associated with a candidate application program in the set of candidate applications to be started, where N is an integer greater than or equal to 1;
[0007] Input the first data into a pre-trained probability prediction model to obtain second data of the terminal device, the pre-trained probability prediction model is trained based on multiple groups of first sample data, each group of first sample data in the multiple groups of first sample data includes first information and label information of each candidate application in the candidate application set, the first information includes information of the first application, first time difference information, first association relationship, second association relationship, identification of the first candidate application and at least one item of time information for opening the first candidate application; wherein the first application is an application opened before opening the first candidate application, the first time difference information is the time difference information between opening the first candidate application and opening the first application, the first association relationship is the association relationship between opening the first candidate application and opening the first application, the second association relationship is the association relationship between the location information and the number of times the first candidate application is opened, and the second data is used to represent the probability value of opening each candidate application in the candidate application set at the current moment; according to the second data, obtain a target category in the candidate application set, the candidate application set includes multiple categories of candidate applications, and the target category is the category with the highest opening probability value among the multiple categories of candidate applications;
[0008] Recommendation information is displayed on the first interface according to the target category, and the recommendation information includes information of at least one candidate application corresponding to the target category.
[0009] In the above method, by inputting N types of data associated with the current moment and the candidate applications in the candidate application set to be opened into a pre-trained probability prediction model, the probability value of each candidate application in the candidate application set being opened at the current moment can be accurately obtained, so that the target category in the candidate application set can be obtained according to the probability value of opening each candidate application. In this way, the obtained target category is more real and accurate. Based on this, the terminal device can display the information of at least one candidate application corresponding to the target category on the first interface according to the accurate target category, so that the user can use the candidate application corresponding to the recommended information or the sub-service in the candidate application as soon as possible, thereby improving the accuracy of application recommendation and ensuring the user's experience.
[0010] Based on the above description, each terminal device predicts its own target category in the above manner, which can ensure the accuracy of the target category corresponding to each terminal device and ensure that each terminal device can accurately display the recommended information.
[0011] In combination with the first aspect, in some implementations of the first aspect, the second data includes multiple probability values, the multiple probability values correspond one-to-one to the candidate applications in the candidate application set, and the target category is obtained according to the second data, including:
[0012] Determine M probability values greater than a first preset threshold value among the multiple probability values, where M is an integer greater than or equal to 1;
[0013] According to the categories of the candidate applications corresponding to the M probability values, a target category is obtained.
[0014] In the above method, the target category can be determined according to the category of the candidate application corresponding to the maximum probability value, thereby ensuring the accuracy of the determined target category.
[0015] In combination with the first aspect, in some implementations of the first aspect, the second data includes multiple probability values, the multiple probability values correspond one-to-one to the candidate applications in the candidate application set, and the target category is obtained according to the second data, including:
[0016] Adding probability values of candidate applications of the same category among the multiple probability values to obtain multiple first probability values, wherein the multiple first probability values correspond one-to-one to the candidate applications of the multiple categories;
[0017] Determine K first probability values greater than a second preset threshold value among the multiple first probability values, where K is an integer greater than or equal to 1;
[0018] According to the categories of the candidate applications corresponding to the K probability values, the target category is obtained.
[0019] In the above method, the target category can be determined based on the category of the candidate application corresponding to the maximum cumulative sum among the cumulative sums of the probability values of the candidate application in the same category, thereby ensuring the accuracy of the determined target category.
[0020] In combination with the first aspect, in some implementations of the first aspect, the second data includes multiple probability values, the multiple probability values correspond one-to-one to the candidate applications in the candidate application set, and the target category is obtained according to the second data, including:
[0021] Averaging the probability values of the candidate applications of the same category among the multiple probability values to obtain multiple second probability values, wherein the multiple second probability values correspond one-to-one to the candidate applications of the multiple categories;
[0022] Determine P second probability values greater than a third preset threshold value among the multiple second probability values, where P is an integer greater than or equal to 1;
[0023] A target category is obtained according to the categories of the candidate application programs corresponding to the P second probability values.
[0024] In the above method, the target category can be determined based on the category of the candidate application corresponding to the maximum average value among the average values of the probability values of the candidate applications of the same category, thereby ensuring the accuracy of the determined target category.
[0025] In conjunction with the first aspect, in certain implementations of the first aspect, displaying recommendation information on the first interface according to the target category includes:
[0026] Sending an application push request to the server, the application push request including the terminal identification and target category of the terminal device;
[0027] Receiving an application recommendation result sent by the server, the application recommendation result including an identifier of a target application program;
[0028] According to the application recommendation result, the recommendation information is displayed in the application recommendation area in the first interface.
[0029] In the above method, based on the identifier of the target application included in the application recommendation results sent by the server, the recommendation information corresponding to the identifier of the target application can be displayed in the application recommendation area in the first interface, thereby ensuring that the terminal device can accurately display the recommendation information of the candidate applications that the user wants to use, facilitating the user to use the candidate applications they want to use as soon as possible, improving the accuracy of the recommended applications, and ensuring the user experience.
[0030] In a second aspect, the present application provides a method for generating a probability prediction model, the method comprising:
[0031] Acquire multiple groups of original sample data of the terminal device, each group of original sample data in the multiple groups of original sample data includes original information of a first candidate application in a set of candidate applications opened within a preset number of historical days, and the original information includes an identifier of the first candidate application, time information of opening the first candidate application, and location information of opening the first candidate application;
[0032] According to the multiple groups of original sample data, multiple groups of first sample data of the terminal device are obtained, each group of the multiple groups of first sample data includes first information and label information of each candidate application in the set of candidate applications, and the first information includes at least one of information of the first application, first time difference information, first association relationship, second association relationship and original information; wherein the first application is an application opened before the first candidate application is opened, the first time difference information is time difference information between opening the first candidate application and opening the first application, the first association relationship is an association relationship between opening the first candidate application and opening the first application, and the second association relationship is an association relationship between location information and number information of opening the first candidate application;
[0033] The probability prediction model to be trained is trained according to the multiple groups of first sample data to obtain a probability prediction model.
[0034] In the above method, by obtaining the first information and label information of each candidate application in the set of candidate applications to be opened, the probability prediction model to be trained is trained, and a probability prediction model with higher accuracy can be obtained, thereby preparing for predicting the probability value of each candidate application in the set of candidate applications to be opened at the current moment through the probability prediction model.
[0035] In combination with the second aspect, in some implementations of the second aspect, the multiple groups of first sample data include multiple groups of label information, the first information in the multiple groups of first sample data includes multiple types of data, and the probability prediction model to be trained is trained according to the multiple groups of first sample data to obtain the probability prediction model, including:
[0036] Screening multiple types of data in multiple groups of first sample data to obtain multiple groups of second sample data of terminal devices;
[0037] Inputting multiple sets of second sample data into the probability prediction model to be trained to obtain multiple sets of prediction probability values, each set of prediction probability values being used to represent the probability value of starting each candidate application in the candidate application set when starting the first candidate application;
[0038] The model parameters of the probability prediction model to be trained are trained according to the differences between the multiple groups of prediction probability values and the multiple groups of label information to obtain the probability prediction model.
[0039] In the above method, after obtaining multiple groups of first sample data, multiple types of data in the multiple groups of first sample data may be screened to facilitate screening out data associated with candidate applications in the set of candidate applications to be started.
[0040] In conjunction with the second aspect, in some implementations of the second aspect, screening multiple types of data in multiple groups of first sample data to obtain multiple groups of second sample data of the terminal device includes:
[0041] Multiple categories of data in multiple groups of first sample data are screened according to correlation values to obtain multiple groups of second sample data of the terminal device; the correlation values are used to characterize the association relationship between each category of data in the multiple categories of data and the multiple groups of label information.
[0042] In the above method, data associated with the candidate application in the set of candidate application programs to be started can be screened out by using the correlation value used to characterize the association relationship between each type of data in the multiple types of data and the multiple groups of tag information.
[0043] In conjunction with the second aspect, in some implementations of the second aspect, screening multiple types of data in multiple groups of first sample data to obtain multiple groups of second sample data of the terminal device includes:
[0044] Screening multiple types of data in multiple groups of first sample data according to the correlation value to obtain H types of data, where H is an integer greater than or equal to 1;
[0045] Inputting the H types of data into a logistic regression model to obtain H weight values corresponding to the H types of data, wherein one weight value among the H weight values is used to characterize the degree of association between a type of data in the H types of data and a candidate application in the startup candidate set;
[0046] Based on H weight values, H types of data are screened to obtain Q types of data, where Q is an integer greater than or equal to 1;
[0047] According to the Q-type data, multiple groups of second sample data of the terminal device are obtained.
[0048] In the above method, a first screening is performed on multiple categories of data based on relevant values to obtain H category data associated with the candidate applications in the candidate set for activation, and then a second screening is performed on the H category data to obtain Q category data with the highest degree of association with the candidate applications in the candidate set for activation. Thus, the probability prediction model to be trained is trained based on the Q category data with the highest degree of association with the candidate applications in the candidate set for activation, so that a probability prediction model with higher accuracy can be obtained.
[0049] In combination with the second aspect, in certain implementations of the second aspect, the correlation value includes a first correlation value and / or a second correlation value, the first correlation value is used to characterize the linear relationship between each type of data in the multiple types of data and the multiple groups of label information; the second correlation value is used to characterize the monotonic relationship between each type of data in the multiple types of data and the multiple groups of label information.
[0050] In combination with the second aspect, in some implementations of the second aspect, the original information also includes network status information.
[0051] In conjunction with the second aspect, in some implementations of the second aspect, the first information further includes external device information, memory status information, information of the second application, first duration information, information of the third application, basic information of the opened candidate application, first opening times information, second time difference information, first opening times information, second duration information, third association relationship and first ratio information;
[0052] Among them, the second application is an application opened in the same period of the historical date corresponding to the period of time when the first candidate application is opened, the first duration information is the duration information of opening the first application, the third application is the application with the highest number of openings within the first preset duration before the first candidate application is opened, the first opening number information is the number of times the first candidate application is opened within the second preset time before the first candidate application is opened, the second time difference information is the time difference information between opening the candidate application and opening the first candidate application, the first opening number information is the number of times the first candidate application is opened within the third preset time before the first candidate application is opened, the second duration information is the opening time information within the third time before the first candidate application is opened, the third association relationship is the association relationship between the external device and the opening of the first candidate application, and the first ratio information is the ratio of the number of days the first candidate application is opened within the same period of the first historical days to the first historical days.
[0053] Among them, obtaining multiple types of data that may be associated with the candidate applications in the set of candidate applications for opening can facilitate better training of the probability prediction model to be trained based on the multiple types of data, so as to obtain a probability prediction model with higher accuracy.
[0054] In a third aspect, the present application provides a terminal device, which is used to execute the application recommendation method in the first aspect and any possible implementation manner of the first aspect.
[0055] In a possible implementation, the terminal device includes: a first acquisition module, a first obtaining module and a display module.
[0056] A first acquisition module, used to acquire first data of the terminal device at the current moment, the first data including N types of data, the N types of data being data associated with a candidate application in the set of enabled candidate applications, where N is an integer greater than or equal to 1;
[0057] A first obtaining module is used to input the first data into a pre-trained probability prediction model to obtain second data of the terminal device, wherein the pre-trained probability prediction model is obtained by training based on multiple groups of first sample data, wherein each group of first sample data in the multiple groups of first sample data includes first information and label information of each candidate application in the set of candidate applications for opening, and the first information includes at least one of information of the first application, first time difference information, first association relationship, second association relationship, an identifier of the first candidate application, and time information for opening the first candidate application; wherein the first application is an application opened before opening the first candidate application, the first time difference information is the time difference information between opening the candidate application and opening the first application, the first association relationship is the association relationship between opening the candidate application and opening the first application, and the second association relationship is the association relationship between location information and number information for opening the candidate application, and the second data is used to represent the probability value of opening each candidate application in the set of candidate applications at the current moment;
[0058] The first obtaining module is further used to obtain a target category in the candidate application set according to the second data, the candidate application set includes multiple categories of candidate applications, and the target category is a category with the highest activation probability value among the multiple categories of candidate applications;
[0059] The first display module is used to display recommendation information on the first interface according to the target category, where the recommendation information includes information of at least one candidate application corresponding to the target category.
[0060] In conjunction with the third aspect, in some implementations of the third aspect, the second data includes multiple probability values, the multiple probability values correspond one-to-one to the candidate applications in the candidate application set, and the first obtaining module is specifically used to:
[0061] Determine M probability values greater than a first preset threshold value among the multiple probability values, where M is an integer greater than or equal to 1;
[0062] According to the categories of the candidate applications corresponding to the M probability values, a target category is obtained.
[0063] In conjunction with the third aspect, in some implementations of the third aspect, the second data includes multiple probability values, the multiple probability values correspond one-to-one to the candidate applications in the candidate application set, and the first obtaining module is specifically used to:
[0064] Adding probability values of candidate applications of the same category among the multiple probability values to obtain multiple first probability values, wherein the multiple first probability values correspond one-to-one to the candidate applications of the multiple categories;
[0065] Determine K first probability values greater than a second preset threshold value among the multiple first probability values, where K is an integer greater than or equal to 1;
[0066] According to the categories of the candidate applications corresponding to the K probability values, the target category is obtained.
[0067] In conjunction with the third aspect, in some implementations of the third aspect, the second data includes multiple probability values, the multiple probability values correspond one-to-one to the candidate applications in the candidate application set, and the first obtaining module is specifically used to:
[0068] Averaging the probability values of the candidate applications of the same category among the multiple probability values to obtain multiple second probability values, wherein the multiple second probability values correspond one-to-one to the candidate applications of the multiple categories;
[0069] Determine P second probability values greater than a third preset threshold value among the multiple second probability values, where P is an integer greater than or equal to 1;
[0070] A target category is obtained according to the categories of the candidate application programs corresponding to the P second probability values.
[0071] In conjunction with the third aspect, in some implementations of the third aspect, the display module is specifically configured to:
[0072] Sending an application push request to the server, the application push request including the terminal identification and target category of the terminal device;
[0073] Receiving an application recommendation result sent by the server, the application recommendation result including an identifier of a target application program;
[0074] According to the application recommendation result, the recommendation information is displayed in the application recommendation area in the first interface.
[0075] In a fourth aspect, the present application provides a terminal device, which is used to execute the probability prediction model generation method in the second aspect and any possible implementation of the second aspect.
[0076] In a possible implementation manner, the terminal device includes: a second acquiring module and a second obtaining module.
[0077] A second acquisition module is used to acquire multiple groups of original sample data of the terminal device, each group of original sample data in the multiple groups of original sample data includes original information of a first candidate application in a set of candidate applications opened within a preset number of historical days, and the original information includes an identifier of the first candidate application, time information of opening the first candidate application, and location information of opening the first candidate application;
[0078] A second obtaining module is used to obtain multiple groups of first sample data of the terminal device according to the multiple groups of original sample data, each group of first sample data in the multiple groups of first sample data includes first information and label information of each candidate application in the set of candidate application openings, and the first information includes at least one of information of the first application, first time difference information, a first association relationship, a second association relationship and original information; wherein the first application is an application opened before the first candidate application is opened, the first time difference information is the time difference information between opening the candidate application and opening the first application, the first association relationship is the association relationship between opening the candidate application and opening the first application, and the second association relationship is the association relationship between location information and number information of opening the candidate application;
[0079] The second obtaining module is used to train the probability prediction model to be trained according to multiple groups of first sample data to obtain the probability prediction model.
[0080] In conjunction with the fourth aspect, in some implementations of the fourth aspect, the multiple groups of first sample data include multiple groups of label information, the first information in the multiple groups of first sample data includes multiple types of data, and the second obtaining module is specifically used to:
[0081] Screening multiple types of data in multiple groups of first sample data to obtain multiple groups of second sample data of terminal devices;
[0082] Inputting multiple sets of second sample data into the probability prediction model to be trained to obtain multiple sets of prediction probability values, each set of prediction probability values being used to represent the probability value of starting each candidate application in the candidate application set when starting the first candidate application;
[0083] The model parameters of the probability prediction model to be trained are trained according to the differences between the multiple groups of prediction probability values and the multiple groups of label information to obtain the probability prediction model.
[0084] In conjunction with the fourth aspect, in some implementations of the fourth aspect, the second obtaining module is specifically configured to:
[0085] The multiple categories of data in the multiple groups of first sample data are screened according to the correlation value to obtain multiple groups of second sample data of the terminal device; the correlation value is used to characterize the association relationship between each category of data in the multiple categories of data and the multiple groups of label information.
[0086] In conjunction with the fourth aspect, in some implementations of the fourth aspect, the second obtaining module is specifically configured to:
[0087] Screening multiple types of data in multiple groups of first sample data according to the correlation value to obtain H types of data, where H is an integer greater than or equal to 1;
[0088] Inputting H types of data into the logistic regression model, obtaining H weight values corresponding to the H types of data, wherein one weight value among the H weight values is used to represent the degree of association between a type of data in the H types of data and a candidate application in the candidate set for starting;
[0089] Based on H weight values, H types of data are screened to obtain Q types of data, where Q is an integer greater than or equal to 1;
[0090] According to the Q-type data, multiple groups of second sample data of the terminal device are obtained.
[0091] In combination with the fourth aspect, in certain implementations of the fourth aspect, the correlation value includes a first correlation value and / or a second correlation value, the first correlation value is used to characterize the linear relationship between each type of data in the multiple types of data and the multiple groups of label information; the second correlation value is used to characterize the monotonic relationship between each type of data in the multiple types of data and the multiple groups of label information.
[0092] In combination with the fourth aspect, in certain implementations of the fourth aspect, the original information also includes network status information.
[0093] In conjunction with the fourth aspect, in some implementations of the fourth aspect, the first information further includes external device information, memory status information, information of the second application, first duration information, information of the third application, basic information of the opened candidate application, first opening times information, second time difference information, first opening times information, second duration information, third association relationship and first ratio information;
[0094] Among them, the second application is an application opened in the same period of the historical date corresponding to the period of time when the first candidate application is opened, the first duration information is the duration information of opening the first application, the third application is the application with the highest number of openings within the first preset duration before the first candidate application is opened, the first opening number information is the number of times the first candidate application is opened within the second preset time before the first candidate application is opened, the second time difference information is the time difference information between opening the first candidate application and opening the first candidate application, the first opening number information is the number of times the first candidate application is opened within the third preset time before the first candidate application is opened, the second duration information is the opening time information within the third time before the first candidate application is opened, the third association relationship is the association relationship between the external device and the opening of the first candidate application, and the first ratio information is the ratio of the number of days the first candidate application is opened within the same period of the first historical days to the first historical days.
[0095] In a fifth aspect, the present application provides a terminal device comprising one or more processors, a memory and one or more computer programs, wherein the one or more computer programs are stored in the memory, and when the computer programs are executed by one or more processors, the terminal device executes the application recommendation method in the first aspect and any possible implementation of the first aspect, and / or executes the probability prediction model generation method in the second aspect and any possible implementation of the second aspect.
[0096] In a sixth aspect, the present application provides a chip system comprising a processor for calling and running a computer program from a memory, so that a terminal device equipped with the chip system executes the application recommendation method in the first aspect and any possible implementation of the first aspect, and / or executes the probability prediction model generation method in the second aspect and any possible implementation of the second aspect.
[0097] In the seventh aspect, the present application provides a computer-readable storage medium, including a computer program. When the computer program runs on a terminal device, the terminal device executes the application recommendation method in the first aspect and any possible implementation of the first aspect, and / or executes the probability prediction model generation method in the second aspect and any possible implementation of the second aspect.
[0098] In an eighth aspect, the present application provides a computer program product. When the computer program product runs on a computer, it enables the computer to execute the application recommendation method in the first aspect and any possible implementation of the first aspect, and / or execute the probability prediction model generation method in the second aspect and any possible implementation of the second aspect.
[0099] It can be understood that the beneficial effects of the third to eighth aspects mentioned above can be found in the relevant descriptions in the first and / or second aspects mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS
[0100] Figure 1 A schematic diagram of a human-computer interaction interface provided by the prior art;
[0101] Figure 2 A schematic diagram of the structure of a terminal device provided in one embodiment of the present application;
[0102] Figure 3 A schematic diagram of a software architecture of a terminal device provided in one embodiment of the present application;
[0103] Figure 4 A flowchart of a probability prediction model generation method provided in one embodiment of the present application;
[0104] Figure 5 A flowchart of a probability prediction model generation method provided in one embodiment of the present application;
[0105] Figure 6 A flowchart of an application recommendation method provided in one embodiment of the present application;
[0106] Figure 7 A flowchart of an application recommendation method provided in one embodiment of the present application;
[0107] Figure 8 A flowchart of an application recommendation method provided in one embodiment of the present application;
[0108] Fig. 9 A schematic diagram of the structure of a terminal device provided in one embodiment of the present application;
[0109] Fig.10 A schematic diagram of the structure of a terminal device provided in one embodiment of the present application. DETAILED DESCRIPTION
[0110] In this application, "at least one" means one or more, and "plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c alone can represent: a, b, c, a and b, a and c, or a, b, and c, where a, b, and c can be single or multiple. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.
[0111] Application recommendation, also known as service recommendation or service suggestion, recommends applications or sub-services in applications to users on the user interface (UI) to enable users to use them more conveniently.
[0112] The applications in the above application recommendations are usually third-party applications (Application, APP), and each application may include multiple sub-services. For example, a short video application includes sub-services such as recommended popular short videos, local popular short videos, live broadcasts, and shopping. The terminal device can recommend the entire application or one or more sub-services in the application. When the terminal device recommends a sub-service in the application, the terminal device can display the icon corresponding to the sub-service in the application. When the user clicks the icon, the terminal device can directly display the interface corresponding to the sub-service.
[0113] Among them, in the initial state, the terminal device can display an application recommendation area on an interface, or can display or close the application recommendation area based on the user's instructions, so that applications can be recommended to users through the application recommendation area. The application recommendation area also includes an operation service area, and paid applications can be recommended in the operation service area.
[0114] In addition, the application recommendation area can be located in the main interface of the terminal device, the display interface of the drop-down status bar, the interface corresponding to the negative one screen, or the interface where the global search function is located, etc. This application does not limit the interface where the application recommendation area is located. The application recommendation area can be located in any area on the aforementioned interface, and this application does not limit the area where the application recommendation area is located on the interface of the terminal device.
[0115] For example, Figure 1In the present invention, the application recommendation area is displayed on the main interface of the terminal device, and the user can use the application recommended in the application recommendation area directly on the main interface through the application icon in the application recommendation area.
[0116] Currently, recommended applications for the current period are usually predicted based on applications opened by users during the same period on historical dates corresponding to the current period. However, due to the large differences in applications opened by users during the same period on different dates, the accuracy of currently recommended applications is low, resulting in a poor user experience.
[0117] In view of the above problems, the present application may provide an application recommendation method, a probability prediction model generation method, a terminal device, a chip system, a computer-readable storage medium, and a computer program product, which may pre-acquire various data that may be associated with the opening of candidate applications, train the probability prediction model to be trained, and obtain the probability prediction model; and, when the terminal device needs to display the recommended information, obtain the data associated with the candidate applications in the candidate application set, input the pre-trained probability prediction model, and obtain the probability value of each candidate application in the candidate application set being opened at the current moment, thereby, determining the target category in the candidate application set according to the probability value of each candidate application, and displaying the recommended information; thereby, the determined target category can be made as reasonable and accurate as possible, and each terminal device can be guaranteed to accurately display the recommended information, thereby improving the user experience.
[0118] Among them, the above-mentioned application recommendation method and probability prediction model generation method are both applied to terminal devices, and the terminal device can be a terminal device with display screen hardware and corresponding software support.
[0119] For example, the terminal device may be a mobile phone, a tablet computer, a vehicle-mounted device, a laptop computer, a wearable device, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), a smart car, a smart TV, a robot, or the like.
[0120] It should be pointed out that in some possible implementations, the terminal device may also be called an electronic device, user equipment (UE), etc., which is not limited in the embodiments of the present application.
[0121] For ease of explanation, Figure 2 In the figure, the terminal device 100 is taken as a mobile phone as an example for illustration.
[0122] like Figure 2As shown, the terminal device 100 may include a processor 101, a communication module 102, a display screen 103, a camera 104, a sensor 105, an internal memory 106, a USB interface 107, an external memory interface 108, a charging management module 109, a power management module 110, and a battery 111, etc.
[0123] The processor 101 may include one or more processing units, for example, the processor 101 may include an application processor (AP), a modem processor, a graphics processor, an image signal processor (ISP), a controller, a memory, a video stream codec, a digital signal processor (DSP), a baseband processor, and / or a neural-network processing unit (NPU), etc. Different processing units may be independent devices or integrated in one or more processors 101.
[0124] The controller may be the nerve center and command center of the terminal device 100. The controller may generate an operation control signal according to the instruction operation code and the timing signal to complete the control of fetching and executing instructions.
[0125] The processor 101 may also be provided with a memory for storing instructions and data.
[0126] In some embodiments, the memory in the processor 101 is a cache memory. The memory can store instructions or data that the processor 101 has just used or cyclically used. If the processor 101 needs to use the instruction or data again, it can be directly called from the memory. This avoids repeated access, reduces the waiting time of the processor 101, and thus improves the efficiency of the system.
[0127] In some embodiments, the processor 101 may include one or more interfaces. The interface may include an inter-integrated circuit (I2C) interface, an inter-integrated circuits sound (I2S) interface, a pulse code modulation (PCM) interface, a universal asynchronous receiver / transmitter (UART) interface, a mobile industry processor interface (MIPI), a general-purpose input / output (GPIO) interface, a subscriber identity module (SIM) interface, and / or a universal serial bus (USB) interface, etc.
[0128] The modem processor may include a modulator and a demodulator. Among them, the modulator is used to modulate the low-frequency baseband signal to be sent into a medium-high frequency signal. The demodulator is used to demodulate the received electromagnetic wave signal into a low-frequency baseband signal. The demodulator then transmits the demodulated low-frequency baseband signal to the baseband processor for processing. After the low-frequency baseband signal is processed by the baseband processor, it is passed to the application processor. The application processor outputs a sound signal through an audio device (not limited to a speaker 170A, a receiver 170B, etc.), or displays an image or video through a display screen 103. In some embodiments, the modem processor may be an independent device. In other embodiments, the modem processor may be independent of the processor 101 and be set in the same device as the mobile communication module 150 or other functional modules.
[0129] The communication module 102 may include antenna 1 and antenna 2, a mobile communication module, and / or a wireless communication module.
[0130] Each antenna in the terminal device 100 can be used to cover a single or multiple communication frequency bands. Different antennas can also be reused to improve the utilization of the antennas. For example, antenna 1 can be reused as a diversity antenna for a wireless local area network. In some other embodiments, the antenna can be used in combination with a tuning switch.
[0131] The mobile communication module can provide solutions for wireless communications including 2G / 3G / 4G / 5G, etc., applied to the terminal device 100. The mobile communication module may include at least one filter, a switch, a power amplifier, a low noise amplifier (LNA), etc. The mobile communication module may receive electromagnetic waves from the antenna 1, and perform filtering, amplification, and other processing on the received electromagnetic waves, and transmit them to the modulation and demodulation processor for demodulation. The mobile communication module may also amplify the signal modulated by the modulation and demodulation processor, and convert it into electromagnetic waves for radiation through the antenna 1. In some embodiments, at least some of the functional modules of the mobile communication module may be arranged in the processor 101. In some embodiments, at least some of the functional modules of the mobile communication module may be arranged in the same device as at least some of the modules of the processor 101.
[0132] The wireless communication module can provide wireless communication solutions including wireless local area networks (WLAN) (such as wireless fidelity (Wi-Fi) networks), bluetooth (BT), global navigation satellite system (GNSS), frequency modulation (FM), near field communication (NFC), infrared (IR), etc., which are applied to the terminal device 100. The wireless communication module can be one or more devices integrating at least one communication processing module. The wireless communication module receives electromagnetic waves via antenna 2, modulates and filters the electromagnetic wave signals, and sends the processed signals to processor 101. The wireless communication module can also receive the signal to be sent from processor 101, modulate the frequency, amplify it, and convert it into electromagnetic waves for radiation through antenna 2.
[0133] The display screen 103 is used to display images or videos in the human-computer interaction interface. The display screen 103 includes a display panel. The display panel can be a liquid crystal display (LCD), an organic light-emitting diode (OLED), an active-matrix organic light-emitting diode or an active-matrix organic light-emitting diode (AMOLED), a flexible light-emitting diode (FLED), MiniLED, MicroLED, Micro-OLED, quantum dot light emitting diodes (QLED), etc. In some embodiments, the terminal device 100 may include 1 or N display screens 103, where N is a positive integer greater than 1.
[0134] The camera 104 is used to capture still images or videos. The object generates an optical image through the lens and projects it onto the photosensitive element. The photosensitive element can be a charge coupled device (CCD) or a complementary metal-oxide-semiconductor (CMOS) phototransistor. The photosensitive element converts the optical signal into an electrical signal, and then passes the electrical signal to the ISP to be converted into a digital image signal. The ISP outputs the digital image signal to the DSP for processing. The DSP converts the digital image signal into an image signal in a standard RGB, YUV or other format. In some embodiments, the terminal device 100 may include 1 or N cameras 104, where N is a positive integer greater than 1.
[0135] The charging management module 109 is used to receive charging input from a charger, where the charger can be a wireless charger or a wired charger.
[0136] In some wired charging embodiments, the charging management module 109 may receive charging input from a wired charger through the USB interface 107 .
[0137] In some wireless charging embodiments, the charging management module 109 may receive wireless charging input via a wireless charging coil of the terminal device 100. While the charging management module 109 charges the battery 111, it may also power the terminal device 100 via the power management module 110.
[0138] The power management module 110 is used to connect the battery 111, the charging management module 109 and the processor 101. The power management module 110 receives input from the battery 111 and / or the charging management module 109, and supplies power to the processor 101, the internal memory 106, the external memory, and the communication module 102. The power management module 110 can also be used to monitor parameters such as battery capacity, battery cycle number, and battery health status (leakage, impedance).
[0139] In some other embodiments, the power management module 110 may also be disposed in the processor 101 .
[0140] In other embodiments, the power management module 110 and the charging management module 109 may also be disposed in the same device.
[0141] The external memory interface 108 can be used to connect an external memory card, such as a Micro SD card, to expand the storage capacity of the terminal device 100. The external memory card communicates with the processor 101 through the external memory interface 108 to implement a data storage function. For example, files such as music and video streams are stored in the external memory card.
[0142] The internal memory 106 can be used to store computer executable program codes, which include instructions. The processor 101 executes various functional applications and data processing of the terminal device 100 by running the instructions stored in the internal memory 106. The internal memory 106 can include a program storage area and a data storage area.
[0143] The program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc. The data storage area may store data created during the use of the terminal device 100 (such as audio data, a phone book, etc.), etc. In addition, the internal memory 106 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, a universal flash storage (UFS), etc.
[0144] The sensor 105 in the terminal device 100 may include image sensors, touch sensors, pressure sensors, gyroscope sensors, air pressure sensors, magnetic sensors, acceleration sensors, distance sensors, proximity light sensors, ambient light sensors, fingerprint sensors, temperature sensors, bone conduction sensors and other components to realize sensing and / or acquisition functions for different signals.
[0145] Optionally, the terminal device 100 may also include peripheral devices, such as a mouse, buttons, indicator lights, a keyboard, a speaker, a microphone, etc.
[0146] The buttons include a power button, a volume button, etc. The buttons may be mechanical buttons or touch buttons. The terminal device 100 may receive button inputs and generate key signal inputs related to user settings and function controls of the terminal device 100.
[0147] The indicator may be a light indicator, which may be used to indicate the charging status and the change in the power level, and may also be used to indicate messages, missed calls, and notifications, etc.
[0148] It is understandable that the structure illustrated in this embodiment does not constitute a specific limitation on the terminal device 100 .
[0149] In other embodiments, the terminal device 100 may include more or fewer components than shown, or combine some components, or split some components, or arrange the components differently. The components shown may be implemented in hardware, software, or a combination of software and hardware.
[0150] Please refer to Figure 3 , is a schematic diagram of a software architecture of a terminal device provided in an embodiment of the present application. The application recommendation method provided in an embodiment of the present application is applied to Figure 2 When the terminal device 100 is shown, the software in the terminal device 100 can be divided into the following Figure 3 The application layer 201, the hardware abstraction layer (HAL) 202 and the driver layer 203 are shown.
[0151] A plurality of applications may be installed in the application layer 201 , which may include a camera application for implementing a shooting function.
[0152] The camera application refers to a computer program that runs on an operating system and can perform a photo-taking task. When the format of the camera application is an Android application package (APK) format, the camera application can run on an Android operating system.
[0153] In the embodiment of the present application, the camera application may be an application with a shooting function. For example, an application with the function of sharing short videos can be used as the camera application in the embodiment of the present application if it has a shooting function.
[0154] The hardware abstraction layer 202 is a software located between the operating system kernel and the hardware circuit, and is generally used to abstract the hardware to enable the interaction between the operating system and the hardware circuit at the logical layer. In the embodiment of the present application, the hardware abstraction layer 202 may include a camera hardware abstraction layer (camera HAL) that enables the interaction between the camera application and the image sensor at the logical layer.
[0155] The driver layer 203 may be installed with a plurality of drivers for driving the hardware.
[0156] It should be noted that the application layer 201, the hardware abstraction layer 202 and the driver layer 203 may also include other contents, which are not specifically limited here.
[0157] Based on the above scenario description, the present application takes the terminal device as an example, combined with the accompanying drawings and application scenarios, to elaborate on the probability prediction model generation method provided by the embodiment of the present application.
[0158] See also Figure 4 , Figure 4 A flow chart of a probability prediction model generation method provided in an embodiment of the present application is shown.
[0159] like Figure 4 As shown, the probability prediction model generation method provided by the present application may include:
[0160] S301. Acquire multiple groups of original sample data from a terminal device.
[0161] Among them, each group of original sample data in the multiple groups includes the original information of the first candidate application in the candidate application set opened within a preset historical number of days, and the original information includes the identification of the first candidate application, the time information of opening the first candidate application, the location information of opening the first candidate application, and the network status information of opening the first candidate application.
[0162] In some embodiments, the preset number of historical days is pre-set. For example, the preset number of historical days is 90 days.
[0163] It should be understood that the above description uses 90 days as an example; the duration corresponding to the preset number of historical days in this application is not limited.
[0164] The candidate applications in the candidate application set in each terminal device are the same. The candidate applications in the candidate application set are third-party applications (Application, APP). The candidate applications in the candidate application set are applications in all applications in the terminal device that can be used to recommend to the user.
[0165] It should be noted that the candidate applications in the candidate application set here are the candidate applications installed in the terminal device among all the candidate applications in the candidate application set.
[0166] Assume that all candidate applications in the candidate application set in each terminal include short video APP1, short video APP2, short video APP3, long video APP1, long video APP2, long video APP3, shopping APP1, shopping APP2, shopping APP3, learning APP1, learning APP2 and learning APP3, the candidate applications installed in the terminal device include short video APP1, short video APP2, long video APP2, long video APP3 and shopping APP1.
[0167] Then, each set of original sample data in the multiple sets of original sample data includes the original information of opening short video APP1, short video APP2, long video APP2, long video APP3 or shopping APP1 within a preset historical number of days.
[0168] In addition, the first candidate application is a candidate application that is opened every time within a preset number of historical days.
[0169] Exemplarily, the time information for opening the first candidate application may be the time information in subsequent Tables 2 and 3; the location information for opening the first candidate application may be the geographic location information in subsequent Tables 2 and 3; and the network status information for opening the first candidate application may be the network status information in subsequent Tables 2 and 3.
[0170] Exemplarily, the time information for launching the first candidate application includes the date and time (happentime), the day of the week corresponding to the date (weekday), whether the date is a holiday (holiday), and the hour of the date (hour); the location information for launching the first candidate application includes the base station identifier (cellid), the service set identifier (wifi-ssid), the basic service set identifier (wifi-bassid), and the city identifier.
[0171] In a possible implementation, the original information further includes network status information. For example, the network status information may refer to the information of “network-status” shown in the subsequent Table 1.
[0172] S302. Obtain multiple groups of first sample data of the terminal device according to the multiple groups of original sample data.
[0173] Among them, each group of first sample data in multiple groups of first sample data includes first information and label information of each candidate application in the set of candidate applications, and the first information includes at least one of the information of the first application, the first time difference information, the first association relationship, the second association relationship and the original information; wherein, the first application is an application opened before the first candidate application is opened, the first time difference information is the time difference information between opening the first candidate application and opening the first application, the first association relationship is the association relationship between opening the first candidate application and opening the first application, and the second association relationship is the association relationship between the location information and the number of times the first candidate application is opened.
[0174] Exemplarily, the first information of each candidate application may be the context information and candidate information in the subsequent Table 2; the label information may be the label information in the subsequent table.
[0175] Exemplarily, the information of the first application may be the App opened for the last x times in the subsequent Tables 2 and 3; the first time difference information may be the time difference between the App opened for the last x times in the subsequent Tables 2 and 3; the first association relationship may be the App association in the subsequent Tables 2 and 3; the second association relationship may be the click rate / percentage of the external device in the subsequent Tables 2 and 3.
[0176] Optionally, in one implementation, the first information further includes external device information, memory status information, information of the second application, first duration information, information of the third application, basic information of the first candidate application opened, first opening times information, second time difference information, first opening times information, second duration information, third association relationship and first ratio information;
[0177] Among them, the second application is an application opened in the same period of the historical date corresponding to the period of time when the first candidate application is opened, the first duration information is the duration information of opening the first application, the third application is the application with the highest number of openings within the first preset duration before the first candidate application is opened, the first opening number information is the number of times the first candidate application is opened within the second preset time before the first candidate application is opened, the second time difference information is the time difference information between opening the first candidate application and opening the first candidate application, the first opening number information is the number of times the candidate application is opened within the third preset time before the first candidate application is opened, the second duration information is the opening time information within the third time before the first candidate application is opened, the third association relationship is the association relationship between the external device and the opening of the first candidate application, and the first ratio information is the ratio of the number of days the first candidate application is opened within the same period of the first historical days to the first historical days.
[0178] Exemplarily, the external device information may be the external device information as in the subsequent Tables 2 and 3; the memory status information may be the memory status information as in the subsequent Tables 2 and 3; the information of the second application may be the top 1-3 applications opened during the same period of historical dates as in the subsequent Tables 2 and 3; the first duration information may be the time difference of the App opened for the previous x times as in the subsequent Tables 2 and 3; the basic information of the opened candidate application may be the basic information of the candidate App; the first opening times information may be the total number of times / duration of the candidate App as in the subsequent Tables 2 and 3; the second duration information may be the opening time and opening time ratio of the candidate App at X time as in the subsequent Tables 2 and 3; the third association relationship may be the external device click rate / ratio as in the subsequent Tables 2 and 3; the first ratio information may be the ratio of opening days in the same historical period as in the subsequent Tables 2 and 3.
[0179] S303: Train the probability prediction model to be trained according to the multiple groups of first sample data to obtain a probability prediction model.
[0180] Based on S302, the terminal device may obtain multiple groups of first sample data including first information and label information, thereby training the probability prediction model to be trained based on the first sample data to obtain the probability prediction model.
[0181] In some embodiments, the probabilistic prediction model is a light GBM model.
[0182] It should be understood that the above description uses the probability prediction model as a lightGBM model as an example; this application does not limit the type of probability prediction model.
[0183] Optionally, in one implementation, multiple groups of first sample data include multiple groups of label information, the first information in the multiple groups of first sample data includes multiple categories of data, and the probability prediction model to be trained is trained based on the multiple groups of first sample data to obtain the probability prediction model, including: inputting multiple groups of second sample data into the probability prediction model to be trained to obtain multiple groups of prediction probability values, each group of prediction probability values is used to represent the probability value of opening each candidate application in the candidate application set when the first candidate application is opened; training model parameters of the probability prediction model to be trained based on the differences between the multiple groups of prediction probability values and the multiple groups of label information to obtain the probability prediction model.
[0184] In one example, multiple types of data in multiple groups of first sample data are screened to obtain multiple groups of second sample data of the terminal device, including:
[0185] The multiple categories of data in the multiple groups of first sample data are screened according to the correlation value to obtain multiple groups of second sample data of the terminal device; the correlation value is used to characterize the association relationship between each category of data in the multiple categories of data and the multiple groups of label information.
[0186] In an embodiment of the present application, multiple categories of data in multiple groups of first sample data are filtered through relevant values, and the terminal device can obtain data associated with candidate applications in the set of candidate applications for opening. Therefore, the terminal device trains the probability prediction model to be trained based on the filtered data associated with the candidate applications in the set of candidate applications for opening, and can obtain a more accurate probability prediction model.
[0187] In another example, multiple types of data in multiple groups of first sample data are screened to obtain multiple groups of second sample data of the terminal device, including:
[0188] According to the correlation values, multiple categories of data in multiple groups of first sample data are screened to obtain H category data, where H is an integer greater than or equal to 1; the H category data are input into a logistic regression model to obtain H weight values corresponding to the H category data, and one weight value among the H weight values is used to characterize the degree of association between a category of data in the H category data and a candidate application in the candidate set for starting; based on the H weight values, the H category data are screened to obtain Q category data, where Q is an integer greater than or equal to 1; based on the Q category data, multiple groups of second sample data of the terminal device are obtained.
[0189] Optionally, the above implementation can refer to the following Figure 5 The relevant description of S905 is not repeated here.
[0190] In an embodiment of the present application, a first screening is performed on multiple categories of data in multiple groups of first sample data through correlation values, and the terminal device can obtain H category data associated with candidate applications in the set of candidate applications to be opened. After the H category data is input into the logistic regression model, the H category data and H weight values are obtained, and the H category data is screened for a second time based on the H weight values. The terminal device can obtain Q category data with a high correlation with the candidate applications in the set of candidate applications to be opened. Therefore, the terminal device can obtain accurate multiple groups of second sample data of the terminal device based on the Q category data. Therefore, the terminal device trains the probability prediction model to be trained based on the Q category data obtained after two screenings, which can reduce the memory consumption of the terminal device when training and inferring the model (calling the model during prediction), and reduce the power consumption of the terminal device when training and inferring the model.
[0191] Optionally, in one implementation, the correlation value includes a first correlation value and / or a second correlation value, the first correlation value is used to characterize the linear relationship between each type of data in the multiple types of data and the multiple groups of label information; the second correlation value is used to characterize the monotonic relationship between each type of data in the multiple types of data and the multiple groups of label information.
[0192] In the method for generating a probability prediction model of the present application, multiple groups of original sample data of the terminal device are obtained to obtain multiple groups of first sample data of the terminal device. Since each group of first sample data in the multiple groups of first sample data includes first information and label information for opening each candidate application in the candidate application set, the first information includes multiple categories of data related to opening the candidate applications in the candidate application set, and the label information is used to indicate the usage status of all candidate applications in the candidate application set when opening a candidate application. In this way, by opening the first information of each candidate application in the candidate application set, the probability prediction model to be trained is trained, and a probability prediction model with higher accuracy can be obtained, thereby preparing for the terminal device to predict the probability value of opening each candidate application in the candidate application set at the current moment through the probability prediction model.
[0193] Combine the following Figure 5 The probability prediction model generation method provided in the embodiment of the present application is described by way of example.
[0194] S401. Obtain original data.
[0195] The original data can be Figure 4 The original sample data shown in .
[0196] Assume that in multiple sets of original data, the time information for opening a candidate application includes the date and time, the day of the week corresponding to the date, whether the date is a holiday, and the hours in the date; the location information for opening the first candidate application includes the base station identifier, the service set identifier, the basic service set identifier, and the city identifier.
[0197] Then, taking two sets of original data on March 24, 2022 as an example, the two sets of original data of the terminal device on March 24, 2022 are shown in Table 1 below. The first set of original data includes 10 types of features of opening an APP at 9:04:10 on March 24, 2022, namely: the date and time of opening the APP, the day of the week corresponding to the date of opening the APP, whether the date of opening the APP is a holiday, the hour of the date of opening the APP, the name of the opened APP (Appname), the base station data of the opened APP, the service set identifier of the opened APP, and the basic service set identifier of the opened APP. , the city identification of the APP opening, and the network status information (netword-status) of the APP opening; the second group of raw data includes 10 types of features of the APP opening at 9:24:13 on March 24, 2022, namely: the date and time when the APP was opened, the day of the week corresponding to the date when the APP was opened, whether the date when the APP was opened is a holiday, the hour in the date when the APP was opened, the name of the APP opening, the base station identification of the APP opening, the service set identification of the APP opening, the basic service set identification of the APP opening, the city where the APP was opened, and the network status information of the APP opening.
[0198] Table 1 Two groups of original data
[0199]
[0200] Among them, taking the first set of original data as an example, Table 1 is explained as follows:
[0201] The first column is the date and time (happentime) when the candidate application is opened. For example, the first column of the first row corresponding to happentime is 2022 / 3 / 249:04:10, which means that the date of opening the candidate application is March 24, 2022, and the specific time is 9:04:10.
[0202] The second column is the weekday corresponding to March 24, 2022; for example, 4 in the second column of the first row corresponding to weekday indicates that the date for launching the candidate application is Thursday.
[0203] The third column indicates whether March 24, 2022 is a holiday (holiday); for example, the 0 in the third column of the first row corresponding to holiday indicates that the date of opening the candidate application is not a holiday, where the terminal device can use the identifier "0" to indicate a non-holiday and the identifier "1" to indicate a holiday.
[0204] The fourth column is the hour at 9:04:10 on March 24, 2022; for example, the 9 in the fourth column of the first row corresponding to hour indicates that the hour on the date when the candidate application is opened this time is the 9th hour.
[0205] The fifth column is the name of the candidate application (Appname) opened at 9:04:10 on March 24, 2022; for example, APP1 in the fifth column of the first row corresponding to Appname indicates that the candidate application opened this time is APP1.
[0206] The sixth column is the base station identifier (cellid) of the candidate application that was launched at 9:04:10 on March 24, 2022; for example, 191974580 in the sixth column of the first row corresponding to the cellid indicates that the base station number connected to the terminal device when the candidate application is launched this time is 191974580, where the base station identifier can be the base station number.
[0207] The seventh column is the service set identifier (wifi-ssid) of the candidate application that was enabled at 9:04:10 on March 24, 2022; for example, employee-mobile in the seventh column of the first row corresponding to wifi-ssid indicates that the router number connected to the terminal device when the candidate application was enabled this time is employee-mobile, where the service set identifier can be the router number.
[0208] The eighth column is the basic service set identifier (wifi-bassid) of the candidate application that was enabled at 9:04:10 on March 24, 2022; for example, ec:c0:1b:f2:61:13 in the eighth column of the first row corresponding to wifi-bassid indicates that the router number connected to the terminal device when the candidate application was enabled this time is ec:c0:1b:f2:61:13, where the service set identifier can be the router number.
[0209] The ninth column is the city identifier of the candidate application that was opened at 9:04:10 on March 24, 2022; for example, the city corresponding to the first row and tenth column is City A, which means that the city where the terminal device is located when the candidate application is opened this time is City A.
[0210] The tenth column is the network status information (netword-status) of the candidate application opened at 9:04:10 on March 24, 2022; for example, the wifi in the tenth column of the first row corresponding to netword-status indicates that the network connected to the terminal device when the candidate application is opened this time is a wireless network.
[0211] Based on the above description, the terminal device obtains multiple groups of original data, each group of original data includes original information of the first candidate application in the candidate application set opened within a preset historical number of days, and can prepare data for the terminal device to train the probability prediction model to be trained.
[0212] S402: Create features and construct samples based on the original data to obtain multiple groups of algorithm sample data.
[0213] Each of the multiple sets of algorithm sample data includes multiple types of features and label data for each candidate application in the candidate application set. Each type of feature corresponds to the data corresponding to the feature. The multiple sets of algorithm sample data can be Figure 4 The first sample data shown in . The multi-class feature data can be Figure 4 The tag data can be Figure 4 The label information shown in .
[0214] Based on the description in Table 1, the multiple groups of original data obtained by the terminal device include 10 types of features. Based on this, the terminal device can create features again based on the 10 types of features to obtain more types of features.
[0215] For example, based on the 10 types of features of the original data, feature creation is performed to obtain features of more categories. See the multiple columns of data corresponding to the features of the context dimension and the features of the candidate dimension in Table 2.
[0216] In addition, after creating features, the terminal device can construct samples for each set of raw data, and expand a set of raw data from 1 candidate application data to L, where L is the number of candidate applications in the candidate application set. Through this expansion method, the features of the context dimension can be quickly introduced, and the features of the candidate dimension can be quickly expanded.
[0217] For example, when the value of L is 5, after obtaining multiple groups of original data, the terminal device can expand each group of original data from 1 candidate application data to 5.
[0218] It should be understood that the above description uses the value of L as 5 for example; the value of L in this application is not limited.
[0219] For example, the features of the context dimension of APP1 obtained after sample construction using the first set of original data include:
[0220] 1. Time information of opening APP1, including: the date and time of opening APP1 (happentime), the day of the week corresponding to the date of opening APP1 (weekday), and whether the date of opening APP1 is a holiday (holiday).
[0221] 2. Information about the external device when APP1 is turned on, including: whether the terminal device is connected to an external device (e.g., Bluetooth) when APP1 is turned on, and the type of the connected external device.
[0222] 3. The location information of APP1 is turned on, including: the base station identifier (cellid) of APP1, the service set identifier (wifi-ssid) of APP1 and the basic service set identifier (wifi-bassid) of APP1. The location identifier can be obtained through a clustering algorithm (for example, the distance is 200 meters).
[0223] 4. Open the network status information (network-status) of APP1, including: when starting APP1, the terminal device is in a state of connecting to a wireless network (wifi), in a state of connecting to traffic, or in a state of not being connected to the Internet.
[0224] 5. Open the memory status information of APP1, including the size of the memory used when APP1 is opened, the total memory size, and the ratio of the size of the used memory to the total memory size.
[0225] 6. The city information of opening APP1 includes the city where the terminal device is located when starting APP1. The city information of opening APP1 can also be divided into the location information of opening APP1.
[0226] 7. Information of the second application, including: applications opened during the same period of the historical date corresponding to the period of time when APP1 was opened.
[0227] For example, the previous application with the highest number of uses is from 12:00 to 13:00 on the day before APP1 is opened; the top two applications with the highest number of uses are also from 20:00 to 21:00 three days before APP1 is opened.
[0228] 8. Information of the first application, including: an application opened before opening APP1. For example, a candidate application opened once before opening APP1.
[0229] 9. First time difference information, including: time difference information between launching the candidate application and launching the first application.
[0230] For example, the time difference between the last application opened before opening APP1 and the time difference between the last application opened before opening APP1.
[0231] 10. First duration information, including: the first duration information is duration information for opening the first application.
[0232] For example, the opening time of the application that was opened last time before APP1 was opened.
[0233] 11. Information of the third application, including: the application with the highest number of opening times within the first preset time period before opening APP1.
[0234] For example, the application that was opened most times in the hour before APP1 was opened.
[0235] For example, the features of the context dimension of APP1 obtained after sample construction using the first set of original data include:
[0236] 1. Basic information of the opened APP1, including: the name of APP1, the type of candidate application, whether the candidate application is a high-frequency application of the user (top 10% by frequency), and the time difference between the installation time of the candidate application and the current time (min).
[0237] 2. First opening count information, including: information on the number of times the candidate application is opened within a second preset time period before APP1 is opened.
[0238] For example, the total usage time (min) of APP1, the total number of times the candidate applications are used, the average usage time of the candidate applications, and the average number of times the candidate applications are used.
[0239] 3. The second time difference information includes: the time difference information between opening APP1 and the last time APP1 was opened.
[0240] For example, the time difference between APP1 and the last n times it was opened, and the average usage time difference (min) of APP1 in the last n days.
[0241] 4. The second opening frequency information includes: the number of times APP1 is opened within the third preset time period before APP1 is opened.
[0242] For example:
[0243] a) The number of times APP1 was opened in the previous hour;
[0244] b) Number of times APP1 is opened in the first hour / total number of times it is opened = click rate;
[0245] c) The number of times APP1 was opened in the previous 3 days;
[0246] d) Number of times APP1 was opened in the previous 3 days / total number of times it was opened = click rate (click_rate_recent_3day).
[0247] It is understandable that the time the user has opened a candidate application before opening the candidate application may have an impact on the currently opened APP 1. For example, if the user opened short video APP 1 three times in the previous hour, it means that short video APP 1 is a high-frequency application, then the user may also open short video APP 1 at the current moment.
[0248] 5. The second duration information includes: the activation duration information within the third duration before APP1 is activated.
[0249] a) The duration of APP1 being opened in the previous day;
[0250] b) The proportion of APP1's opening time in the previous day = the opening time of APP1 / the total opening time of all candidate applications in the candidate application set.
[0251] It is understandable that the time the user opened the candidate application the day before may have an impact on the candidate application currently opened. For example, if the user opened shopping app 1 for a long time the day before or three days before, then the user may also open shopping app 1 on the current date.
[0252] 6. The first association relationship includes: an association relationship between opening APP1 and opening the first application program.
[0253] The ratio of the number of times APP1 is opened after the previous K-th application is opened, for example:
[0254] a) The ratio of the number of times APP1 is opened after the first application is opened; for example, after opening chat APP1, the number of times short video APP1 is opened next time is 50 times, and after opening chat APP1, the number of times all other applications are opened next time is 100 times, then the ratio is 50 / 100=0.5;
[0255] b) The ratio of the number of times APP1 is opened after the second application is opened; for example: after opening chat APP1, the next two times short video APP1 is opened are 50 times, and after opening chat APP1, the next two times all other applications are opened are 100 times, then the ratio is 50 / 100=0.5.
[0256] It is understandable that the application opened by the user last time may have an impact on the candidate application opened currently. For example, if the user opened shopping app 1 last time, then the current user is more likely to open shopping app such as shopping app 2, because from the opening of shopping app 1 in the previous order, the user may have the intention to shop before (the shopping app 1 was opened before and the opening time was long), and may have the intention to shop again later (the time difference from the current time is not long, and the user may continue to shop).
[0257] 7. The second association relationship includes: an association relationship between the external device and the opened APP1.
[0258] a) The click rate of the application during the Bluetooth headset connection. For example, if the application is used 100 times during the Bluetooth connection, and WeChat is used 50 times, then the value is 50 / 100=0.5;
[0259] b) The percentage of times the Bluetooth headset is used during connection. For example, if the Bluetooth is connected 10 times in total and WeChat is used 6 times out of these 10 connections, then the value is 6 / 10=0.6.
[0260] It is understandable that the candidate applications opened by the user when the external device is connected or not connected may be different. For example, when the user is connected to an external device, the user uses short video APP1 and long video APP1 more frequently, and when the user is not connected to an external device, the user uses learning APP1 and shopping APP1 more frequently.
[0261] Among them, according to experimental proof, the association between the external device and the opening of the candidate application can well reflect the regularity of the opening of the candidate application, and is very effective data for improving the accuracy of prediction. 8. The third association includes: the association between the location information and the number of times of opening APP1.
[0262] For example:
[0263] a) Click-through rate of the location identifier (SSID is clustered to cellID, for example, cellIDs within 200 meters are clustered into one category) of the candidate App opened in the previous day. For example, if 100 applications were used in the current geographic location identifier in the previous day, and 50 of them were short video APP1, then the value is 50 / 100=0.5;
[0264] b) The percentage of usage of the location identifier (ssid is clustered with cellid) for opening the candidate app in the previous three days. For example, the current geographic location identifier was reached 10 times in the previous three days, and short video APP1 was used 7 of the 10 times at the location identifier, then the value is 7 / 10=0.7.
[0265] It is understandable that the candidate applications opened by users in different locations may be different. For example, users open long video APP1 more often at home, and users open short video APP1 more often at work.
[0266] Among them, according to experimental proof, the correlation between the location information and the number of times the candidate application is opened can well reflect the regularity of the opening of the candidate application, and is very effective data for improving the accuracy of prediction.
[0267] 9. First ratio information, including: the ratio of the number of days during which APP1 is opened in the same period of the first historical days to the first historical days.
[0268] For example:
[0269] a) The percentage of days with the same hour of use in the last three days. For example, in the last three days, from 12:00 to 1:00, this period of time appeared 3 times. Among these 3 times, chat APP1 was used 2 times, so the value is 2 / 3=0.667.
[0270] Among them, according to experimental proof, the ratio of the number of days the candidate application is opened in the same period of the first historical days to the first historical days can well reflect the regularity of the launch of the candidate application, and is very effective data for improving the accuracy of prediction.
[0271] Next, we will continue to combine Table 1 and take the two sets of original data on March 24, 2022 as an example to illustrate the algorithm sample data. The two sets of algorithm sample data of the terminal device on March 24, 2022 are shown in Table 2 below. The first set of algorithm sample data includes the time at 9:04:10 on March 24, 2022, when APP1 was turned on, multiple categories of feature data (information in the context dimension and information in the candidate dimension) and label data (l ab le) of turning on APP1 and turning on other APPs in the candidate application set except APP1; the second set of algorithm sample data includes the time at 9:24:13 on March 24, 2022, when APP2 was turned on, multiple categories of feature data (features in the context dimension and features in the candidate dimension) and label data (l ab le) of turning on APP2 and turning on other APPs in the candidate application set except APP2.
[0272] Table 2 Sample data of two sets of algorithms
[0273]
[0274]
[0275] Among them, taking the first group of algorithm sample data as an example, Table 2 is explained as follows:
[0276] For example, the data in the weekday column can be regarded as one type of data among multiple types of data. It can be seen that this type of data includes the weekday data in multiple groups of algorithm sample data.
[0277] In the candidate dimension information, the 0.5 in the eighth column of the first row corresponding to click_rate_recent_1day indicates that the click rate of APP1 in the previous day is 0.5, where the number of times APP1 is opened in the previous day / the total number of times it is opened = the click rate.
[0278] In the candidate dimension information, the 0.572 in the ninth column of the first row corresponding to click_rate_recent_3day indicates that the click rate of APP1 in the first three days is 0.572, where the number of times APP1 is opened in the first three days / the total number of times it is opened = the click rate.
[0279] In the candidate dimension information, the 1 in the first row and the tenth column corresponding to label indicates that APP1 is turned on.
[0280] Similarly, the 1 in the eighth column of the second row corresponding to click_rate_recent_1day in the candidate information indicates that the click rate of APP1 in the previous day was 1.
[0281] In the candidate dimension information, the 1 in the second row and ninth column corresponding to click_rate_recent_3day indicates that the click rate of APP1 in the previous three days is 1.
[0282] The 0 in the second row and tenth column corresponding to label in the candidate dimension information indicates that APP2 is not enabled.
[0283] Similarly, the 0.25 in the eighth column of the seventh row corresponding to click_rate_recent_1day in the candidate information indicates that the click rate of APP1 in the previous day was 0.25.
[0284] In the candidate dimension information, the 0.286 in the seventh row and ninth column corresponding to click_rate_recent_3day indicates that the click rate of APP1 in the first three days is 0.286.
[0285] In the candidate dimension information, the 1 in the first row and the tenth column corresponding to label indicates that APP1 is turned on.
[0286] Similarly, the 0 in the eighth column of the seventh row corresponding to click_rate_recent_1day in the candidate dimension indicates that the click rate of APP1 in the previous day was 0.
[0287] In the candidate dimension information, the 0.143 in the seventh row and ninth column corresponding to click_rate_recent_3day indicates that the click rate of APP1 in the first three days was 0.143.
[0288] The 1 in the seventh row and tenth column corresponding to label in the candidate dimension information indicates that APP2 is turned on.
[0289] In addition, for the label data: in each group of algorithm sample data, if it is necessary to predict the recommended information displayed each time the first interface is displayed within the target time period, the candidate application that is opened each time can be used as a positive sample in the label data here (the label data is 1), and the remaining other candidate applications can be used as negative samples (the label data is 0); if it is necessary to predict the recommended information displayed within the target time period, the candidate application that is opened within the next x minutes (for example, 0.5 hours) can be used as a positive sample (the label data is 1), and the remaining other candidate applications can be used as negative samples (the label data is 0).
[0290] Based on the above description, the features of the context dimension and the features of the candidate dimension are described in detail in combination with the following Table 3, wherein Table 3 includes feature categories, multi-category features, and feature processing of multi-category features.
[0291] Table 3 Schematic diagram of the features of the context dimension and the candidate dimension
[0292]
[0293]
[0294]
[0295]
[0296]
[0297] Among them, Table 3 is explained as follows:
[0298] In Table 3, multiple types of data can be divided into continuous and categorical types. Continuous data can directly participate in the training of the probability prediction model to be trained, and the terminal device can convert the categorical data into numbers and other forms to participate in the training of the probability prediction model to be trained. For example, the data of whether to connect to wifi is converted into 0 or 1, where 0 means not connected to wifi and 1 means connected to wifi.
[0299] S403: According to the multi-category features, features for inputting into the probability prediction model to be trained are obtained.
[0300] Among them, the data corresponding to the multi-class features can be Figure 4 The first information in the first sample data shown in includes multiple types of data.
[0301] S404: performing feature selection on multiple categories of features used to input into the probability prediction model.
[0302] Exemplarily, the multiple columns of data in the context dimension and the multiple columns of data in the candidate dimension in Table 2 are multi-category features obtained through S403; wherein, one column of the multiple columns of data in the context dimension and the multiple columns of data in the candidate dimension in Table 2 can be regarded as a category of features; feature selection can be performed on the features in Table 2.
[0303] It is understandable that, for each terminal device, among the multiple types of data corresponding to the first information, one part is valid data associated with the candidate application in the candidate set, and the other part is invalid data unrelated to the candidate application in the candidate set.
[0304] For example, for a terminal device that opens a candidate application program at a relatively regular time, the data corresponding to the time dimension is valid data, and the remaining data may be invalid data.
[0305] Specifically, user 1 to whom terminal device 1 belongs has a relatively regular schedule, and the time for starting the candidate application in the terminal device is also relatively regular. Therefore, terminal device 1 is a terminal device that starts the candidate application regularly.
[0306] For another example, for terminal devices that have candidate applications opened at relatively regular spatial locations, the data corresponding to the spatial dimension is valid data, and the remaining data may be invalid data.
[0307] Specifically, the data traffic in the subscriber identity module (SIM) card in the terminal device 2 is relatively small. Therefore, the user 2 to whom the terminal device 2 belongs will only open the candidate application in the terminal device at a location with a wireless network (wifi) such as home and company. Therefore, the terminal device 2 is a terminal device that opens the candidate application at a relatively regular spatial location.
[0308] Therefore, after obtaining multiple categories of data corresponding to the first information, the terminal device needs to filter the multiple categories of data to filter out valid data associated with the candidate applications in the candidate set to be opened, so as to reduce the amount of data when training the probability prediction model to be trained, thereby reducing the memory consumption of the terminal device during training and reducing the power consumption of the terminal device during training.
[0309] In some embodiments, the terminal device may perform feature selection by following the steps below:
[0310] The multi-class features are screened according to the Pearson correlation coefficient and the Spelman correlation coefficient to obtain H-class features;
[0311] Input H features into the logistic regression model to obtain H weight values corresponding to the H features.
[0312] Based on H weight values, H features are screened to obtain Q features.
[0313] Among them, the Pearson correlation coefficient can be Figure 4 The first correlation value shown in ; the Spelman correlation coefficient can be Figure 4 The first correlation value shown in ; the H-type feature can be Figure 4 The H-type data shown in ; the Q-type characteristics can be Figure 4 The Q class data shown in .
[0314] In some embodiments, the terminal device screens multiple types of features to obtain H types of features in the following steps:
[0315] Determine multiple Pearson correlation coefficients of multiple groups of algorithm sample data, where the multiple Pearson correlation coefficients correspond one-to-one to multiple categories of features;
[0316] Determine a first number of first absolute values of a plurality of absolute values corresponding to a plurality of Pearson correlation coefficients that are greater than a fourth preset threshold;
[0317] Determine multiple Spelman correlation coefficients of multiple groups of algorithm sample data, where the multiple Spelman correlation coefficients correspond to multiple types of features one by one;
[0318] Determine a second number of second absolute values of the plurality of absolute values corresponding to the plurality of Spelman correlation coefficients that are greater than a fifth preset threshold;
[0319] According to the features of the first quantity category corresponding to the first absolute value of the first quantity and the features of the same category among the features of the second quantity category corresponding to the second absolute value of the second quantity, H features are obtained, where H is an integer greater than or equal to 1.
[0320] Among them, the H-type feature is the H-type feature associated with the candidate application program in the candidate application set for opening among the multiple types of features.
[0321] For example, the number of first absolute values greater than the fourth threshold among the multiple absolute values corresponding to the multiple Pearson correlation coefficients is 10, the number of second absolute values greater than the fifth preset threshold among the multiple absolute values corresponding to the multiple Spelman correlation coefficients is 8, and among the 10 categories of features corresponding to the 10 first absolute values and the 8 categories of features corresponding to the 8 second absolute values, 5 categories of data belong to the same category. Therefore, the features obtained by screening the multiple categories of features according to the correlation values are 5 categories of features.
[0322] In some embodiments, when the Pearson correlation coefficient is a Pearson correlation coefficient, the formula for the terminal device to determine the Pearson correlation coefficient of one type of feature among the multiple types of features is:
[0323]
[0324]
[0325]
[0326]
[0327] Where n represents the number of multi-class features, μx represents the first average value of one class of features in the multi-class features, μy represents the first average value of multiple label data corresponding to multiple sets of label data, σy represents the first standard deviation of one class of features in the multi-class features, σy represents the second standard deviation of multiple label data corresponding to multiple sets of label data, cov(x, y) represents the covariance of one class of features and multiple label data, ρ x,y Represents the Pearson correlation coefficient between a class of features and multiple label data.
[0328] In combination with Table 2, the calculation of the Pearson correlation coefficient is explained by taking the type of features corresponding to the weekday column as an example: the terminal device can calculate the first average value of the type of features corresponding to the weekday column, and calculate the first standard deviation of the type of features corresponding to the weekday column based on the first average value; calculate the second average value of the multiple label data corresponding to the label column, and calculate the first standard deviation of the multiple label data corresponding to the label column; calculate the covariance of the type of features corresponding to the weekday column and the multiple label data corresponding to the label column; the terminal device then calculates the ratio of the covariance to the product of the first standard deviation and the second standard deviation to obtain the Pearson correlation coefficient of the type of features corresponding to the weekday column.
[0329] Among them, when one type of feature is one variable and multiple label data are another variable, the value of the Pearson correlation coefficient is between [-1, 1]. The more correlated the two variables are, the closer the Pearson correlation coefficient will be to -1 or 1; correspondingly, the correlation strength corresponding to the Pearson correlation coefficient is as follows:
[0330] 0.8-1.0 The two variables are strongly correlated;
[0331] 0.6-0.8 The two variables are strongly correlated;
[0332] 0.4-0.6 The two variables are moderately correlated;
[0333] 0.2-0.4 The two variables are weakly correlated;
[0334] 0.0-0.2 The two variables are very weakly correlated or not correlated.
[0335] That is to say, when the Pearson correlation coefficient is the Pearson correlation coefficient, the closer the value of the Pearson correlation coefficient corresponding to a type of feature calculated based on the above Pearson correlation coefficient formula is to -1 or 1, the more relevant the type of feature is to the candidate application in the set of candidate applications to be started.
[0336] For example, when the Pearson correlation coefficient is the Pearson correlation coefficient, the number of multi-category features is 20, and the fourth preset threshold is 0.2, after the terminal device calculates the 20 Pearson correlation coefficients corresponding to the multi-category features, it can filter out the first absolute values of the first quantity greater than 0.2 from the 20 absolute values corresponding to the 20 Pearson correlation coefficients. Correspondingly, the terminal device can determine the first quantity category of data corresponding to the first absolute value of the first quantity; when the second quantity is 10, the data filtered out by the Pearson correlation coefficient is 10 categories.
[0337] In some embodiments, when the Spelman correlation coefficient is the Spelman correlation coefficient, the formula for the terminal device to determine the Spelman correlation coefficient is:
[0338]
[0339] in, Represents the first average value of one type of feature among multiple types of features. represents the first average value of multiple label data corresponding to multiple groups of label data, and ρ represents the Spelman correlation coefficient between a class of features and multiple label data.
[0340] In combination with Table 2, the calculation of the Spelman correlation coefficient is explained by taking a type of feature corresponding to the weekday column as an example: the terminal device can calculate the first average value of the type of feature corresponding to the weekday column; calculate the second average value of multiple label data corresponding to the label column; and calculate the Spelman correlation coefficient of the type of feature corresponding to the weekday column based on the first average value and the second average value.
[0341] Among them, when one type of feature is one variable and multiple label data are another variable, the value of the Spelman correlation coefficient is between [-1, 1]. The more correlated the two variables are, the closer the Spelman correlation coefficient will be to -1 or 1. The correlation strength corresponding to the Spelman correlation coefficient is as follows:
[0342] 0.8-1.0 The two variables are strongly correlated;
[0343] 0.6-0.8 The two variables are strongly correlated;
[0344] 0.4-0.6 The two variables are moderately correlated;
[0345] 0.2-0.4 The two variables are weakly correlated;
[0346] 0.0-0.2 The two variables are very weakly correlated or not correlated.
[0347] That is, the closer the value of the Spelman correlation coefficient corresponding to a type of feature calculated based on the above Spelman correlation coefficient formula is to -1 or 1, the more relevant the type of feature is to the candidate application in the set of candidate application programs to be started.
[0348] For example, when the number of multi-category features is 20 and the fifth preset threshold is 0.2, after calculating the 20 Spelman correlation coefficients corresponding to the multi-category features, the terminal device can filter out the second absolute values of the second quantity that are greater than 0.2 from the 20 absolute values corresponding to the 20 Pearson correlation coefficients. Correspondingly, the terminal device can determine the second quantity category of data corresponding to the second absolute values of the second quantity; when the second quantity is 8, the data filtered out by the Spelman correlation coefficient is 8 categories.
[0349] In some embodiments, the terminal device may directly filter multiple features in multiple groups of algorithm sample data according to correlation values to obtain multiple groups of second sample data of the terminal device. In other words, the terminal device may not execute S402 and S403, and filter the H-type features obtained according to the correlation values to obtain multiple groups of second sample data of the terminal device.
[0350] Based on the description of S401, the terminal device can perform a first screening of multiple types of features to screen out H-type features associated with candidate applications in the candidate application set. Based on this, the terminal device can also perform a second screening to facilitate screening out the H-type features that have the highest degree of association with the candidate applications in the candidate application set.
[0351] In some embodiments, the terminal device may input H types of features into a logistic regression model to obtain H weight values corresponding to the H types of features, and one of the H weight values is used to characterize the degree of association between a type of feature in the H types of features and a candidate application in the candidate set for starting.
[0352] Among them, the terminal device can pre-train the logistic regression model, and store the pre-trained logistic regression model in the terminal device and / or in a storage device that communicates with the terminal device, so that the terminal device can call the logistic regression model to predict the correlation degree of data.
[0353] In some embodiments, screening H types of features based on H weight values to obtain Q types of features includes the following steps:
[0354] Determine H absolute values corresponding to the H weight values;
[0355] Accumulate the H absolute values to obtain a first accumulated value;
[0356] Determine a ratio of each of the H absolute values to the first accumulated value;
[0357] Select Q ratios from a plurality of ratios in descending order, where Q is an integer greater than or equal to 1;
[0358] If the accumulated value of the P ratios is greater than a sixth preset threshold, Q type features corresponding to the Q ratios are determined.
[0359] Among them, the Q-category features are the first Q-category features in the H-category features that have the highest correlation with the candidate applications in the startup candidate set.
[0360] Based on the above description, the terminal device uses the sixth preset threshold to screen out Q-type features that account for a certain proportion of H-type features, ensuring that features with a high degree of correlation with candidate applications in the candidate set are screened out, which can improve the accuracy of the training of the probability prediction model to be trained.
[0361] Assume that after the Pearson correlation coefficient and the Spelman correlation coefficient are used to screen multiple categories of features, 5 categories of features remain. Then, by inputting the 5 categories of features into the logistic regression model, 5 weight values are obtained to indicate the degree of association between each category of features and the candidate applications in the candidate set for starting.
[0362] Therefore, taking the 5 weight values corresponding to the 5 types of features as an example, the screening of Q-type features from H-type features is explained. The terminal device screens Q-type features from H-type features as shown in Table 4, including the names of the 5 types of features (feature), the 5 weight values corresponding to the 5 types of features (fea_important), the absolute values of the 5 weight values corresponding to the 5 types of features (fea_important-abs), and the ratios of the 5 absolute values corresponding to the 5 types of features to the accumulated values of the 5 absolute values (fea_importance_normal).
[0363] Table 4 Screening process of five types of features
[0364] feature fea_important fea_important-abs fea_importance_normal fea1 1.7 1.7 23.94% fea2 0.5 0.5 7.04% fea3 -1.5 1.5 21.13% fea4 3 3 42.25% fea5 -0.4 0.4 5.63%
[0365] Among them, the following is described about Table 4:
[0366] The first column is the names of the five types of features obtained by screening the Pearson correlation coefficient and the Spelman correlation coefficient; for example, fea1 represents the sixth candidate information: the first association relationship.
[0367] The second column contains the five weight values corresponding to the five types of features. For example, 1.7 in the second column of the first row indicates that the weight value of the feature fea1 is 1.7.
[0368] The third column is the absolute value of the five weight values corresponding to the five types of features. For example, 1.7 in the third column of the first row means that the absolute value corresponding to the weight value of 1.7 for the feature fea1 is 1.7.
[0369] The fourth column is the ratio of the 5 absolute values corresponding to the 5 types of features to the cumulative value of the 5 absolute values; for example, 23.94% in the fourth column of the first row means that the ratio of the absolute value of 1.7 corresponding to the weight value of 1.7 of the feature fea1 to the cumulative value of the 5 absolute values is 1.7 / (1.7+0.5+1.5+3+0.4)=23.94%.
[0370] Continuing with Table 4, in descending order, the first ratio of 42.25% is screened in the feature_importance_abs column. If the ratio is less than 90%, the first two ratios are screened; if the cumulative sum of the first two ratios (42.25% + 23.94%) is less than 90%, the first three ratios are screened; if the cumulative sum of the first three ratios (42.25% + 23.94% + 21.13%) is less than 90%, the first four ratios are screened; if the cumulative sum of the first four ratios (42.25% + 23.94% + 21.13% + 7.04% = 94.37%) is greater than 90%, the screening is discontinued, and it is determined that the four types of features corresponding to the first four ratios are fea4, fea1, fea3 and fea2. Then, these four types of features are the Q-type features finally screened out by the terminal device.
[0371] Based on the above description, the terminal device obtains Q-type features with a high degree of correlation with the candidate applications in the candidate set for activation through a second screening. Therefore, the terminal device can accurately train the probability prediction model to be trained based on the Q-type features with a high degree of correlation with the candidate applications in the candidate set for activation according to the Q-type data.
[0372] S405. Train the probability prediction model to be trained according to the features obtained by feature selection to obtain a probability prediction model.
[0373] The specific implementation of S405 can be found in Figure 4 The relevant description of S303 in will not be repeated here.
[0374] Based on the above scenario description, the present application takes the terminal device as an example, and combines the accompanying drawings and application scenarios to elaborate on the application recommendation method provided in the embodiment of the present application.
[0375] See also Figure 6 , Figure 6 A flow chart of an application recommendation method provided in an embodiment of the present application is shown.
[0376] like Figure 6 As shown, the application recommendation method provided by this application may include:
[0377] S501. Obtain the first data of the terminal device at the current moment.
[0378] The current time is the start time of the target time period, or the time when the first interface is displayed in the target time period. The target time period is a leisure time period, which is used to indicate that the user is in a leisure state where a candidate application in the candidate application set installed on the terminal device can be used.
[0379] Among them, the first interface is an interface for recommending applications.
[0380] Exemplarily, the first interface is the main interface of the terminal device. When the first interface is the main interface of the terminal device, the specific implementation of the first interface can be found in Figure 1 The description of the interface in is not repeated here.
[0381] The first data includes N types of data, where the N types of data are data associated with the candidate application programs in the candidate application set to be started, and N is an integer greater than or equal to 1.
[0382] In some embodiments, in the method of generating a probability prediction model, the terminal device screens multiple types of data in multiple groups of first sample data according to relevant values. Then, the N types of data are the H types of data screened according to relevant values in the method of generating a probability prediction model.
[0383] For example, in the method of generating a probability prediction model, after screening the data once, the screened H-type data is divided into 3 categories, namely, the weekday corresponding to the date in the time information, the association between opening a candidate application and the application opened before opening the candidate application, and network status information. The N-type data is divided into 3 categories, namely, the weekday corresponding to the date in the time information, the association between opening a candidate application and the application opened before opening the candidate application, and network status information.
[0384] In other embodiments, in the method for generating a probability prediction model, the terminal device performs a first screening of multiple categories of data in multiple groups of first sample data according to relevant values, and after obtaining H category data, the H category data is input into a logistic regression model to obtain H weight values corresponding to the H category data, and the H category data is screened a second time based on the H weight values to obtain Q category data. Then, the N category data is the Q category data screened out after screening the data twice in the method for generating a probability prediction model.
[0385] For example, in the method of generating a probability prediction model, after filtering the data twice, the Q-category data filtered out is 2 categories, namely the weekday corresponding to the date in the time information, and the network status information; the N-category data is 2 categories, namely the weekday corresponding to the date in the time information, and the network status information.
[0386] S502: Input the first data into a pre-trained probability prediction model to obtain second data of the terminal device.
[0387] Among them, the pre-trained probability prediction model is obtained by training based on multiple groups of first sample data, each group of first sample data in the multiple groups of first sample data includes first information and label information of each candidate application in the set of opened candidate applications, and the first information includes external device information, memory status information, information of the first application, information of the second application, first time difference information, first duration information, information of the third application, basic information of the opened candidate application, first opening number information, second time difference information, second opening number information, second duration information, first association relationship, second association relationship, third association relationship, first ratio information and at least one of the original information.
[0388] In some embodiments, the specific implementation method of the original information can refer to the relevant description of the original information in S301 above; the specific implementation method of the first information can refer to the relevant description of the first information in S302 above, which will not be repeated here.
[0389] The second data is used to represent the probability value of starting each candidate application in the candidate application set at the current moment.
[0390] In addition, the terminal device can Figure 4 The method and Figure 5 The method pre-trains the probability prediction model to be trained, and stores the pre-trained probability prediction model in the terminal device and / or in a storage device that communicates with the terminal device, so that the terminal device can call the pre-trained probability prediction model to predict the probability value of each candidate application in the candidate application set being opened at the current moment.
[0391] Based on the above description, the terminal device inputs the N types of data included in the first data and associated with the candidate applications in the candidate application set into a pre-trained probability prediction model to obtain the probability value of opening each candidate application in the candidate application set at the current moment, so as to prepare data for determining the target category in the candidate application set.
[0392] S503: Obtain a target category in the candidate application set according to the second data.
[0393] The candidate application set includes candidate applications of multiple categories.
[0394] For example, the candidate application set includes three categories of candidate applications, namely short video category, long video category and shopping category. The short video category includes short video APP1, short video APP2 and short video APP3; the long video category includes long video APP1, long video APP2 and long video APP3; the shopping category includes shopping APP1, shopping APP2 and shopping APP3.
[0395] The target category is the category with the highest activation probability value among multiple categories of candidate applications. The terminal device obtains the target category in the candidate application set based on the second data, and multiple implementation methods may be used.
[0396] It should be understood that the second data includes a plurality of probability values, and the plurality of probability values correspond one-to-one to the candidate applications in the candidate application set.
[0397] In some embodiments, the terminal device determines M probability values greater than a first preset threshold value among multiple probability values, where M is an integer greater than or equal to 1; and obtains a target category based on the categories of the candidate applications corresponding to the M probability values.
[0398] For example, when M is equal to 1, the candidate application corresponding to one probability value among multiple probability values that is greater than the first preset threshold is short video APP1, and the category corresponding to short video APP1 is the short video category, then the short video category is the target category.
[0399] In other embodiments, the terminal device adds the probability values of candidate applications of the same category in multiple probability values to obtain multiple first probability values, and the multiple first probability values correspond one-to-one to the candidate applications of the multiple categories; determines K first probability values greater than a second preset threshold among the multiple first probability values, where K is an integer greater than or equal to 1; and obtains the target category based on the categories of the candidate applications corresponding to the K probability values.
[0400] For example, when K is equal to 1, the candidate application set includes 3 categories of candidate applications, the first probability values are 3, and the category corresponding to the first probability value among the 3 first probability values that is greater than the second preset threshold is the short video category, then the short video category is the target category.
[0401] In other embodiments, the terminal device averages the probability values of candidate applications of the same category among multiple probability values to obtain multiple second probability values, and the multiple second probability values correspond one-to-one to the candidate applications of the multiple categories; determines P second probability values among the multiple second probability values that are greater than a third preset threshold, where P is an integer greater than or equal to 1; and obtains the target category based on the categories of the candidate applications corresponding to the P second probability values.
[0402] For example, when P is equal to 1, the candidate application set includes 3 categories of candidate applications, the second probability values are 3, and the category corresponding to one of the 3 second probability values that is greater than the second preset threshold is the short video category, then the short video category is the target category.
[0403] Based on any one of the above three implementation methods, the terminal device can obtain accurate target categories in the candidate application set, so that the terminal device can display accurate recommendation information based on the target category.
[0404] S504: Display recommendation information on the first interface according to the target category.
[0405] The recommendation information includes information of at least one candidate application corresponding to the target category.
[0406] After the terminal device determines the target category, it can accurately display information of at least one candidate application corresponding to the target category on the first interface according to the target category, so that the terminal device can accurately recommend candidate applications in the candidate application set or sub-services in the candidate applications to the user.
[0407] In the application recommendation method of the present application, the terminal device can prepare data for the terminal device to predict the target category in the candidate application set that the terminal device needs to display at the current moment by obtaining the first data associated with the candidate application in the candidate application set at the current moment of the terminal device; thus, the terminal device can input the first data into a pre-trained probability prediction model to obtain the second data of the terminal device used to represent the probability value of each candidate application in the candidate application set at the current moment, then the terminal device can obtain the target category in the candidate application set according to the probability value of each candidate application, so that the target category obtained by the terminal device is more real and accurate. Based on this, the terminal device can display the information of at least one candidate application corresponding to the target category on the first interface according to the target category, so that the user can use the candidate application or the sub-service in the candidate application corresponding to the recommended information as soon as possible, thereby improving the accuracy of application recommendation and ensuring the user's experience.
[0408] Based on the above description, each terminal device predicts its own target category in the above manner, which can ensure the accuracy of the target category corresponding to each terminal device and ensure that each terminal device can accurately display the recommended information.
[0409] Based on the above Figure 6 In the description of the illustrated embodiment, when the electronic device displays recommendation information on the first interface according to the target category, the server may determine which candidate applications in the target category can be recommended as recommendation information.
[0410] Next, combine Figure 7 , which introduces in detail the specific implementation process of displaying recommended information on the first interface according to the target category of this application.
[0411] See also Figure 7 , Figure 7 A schematic diagram of a process for displaying recommended information on a first interface according to a target category provided by an embodiment of the present application is shown.
[0412] like Figure 7 As shown, the implementation process of displaying recommended information on the first interface according to the target category provided by the present application may include:
[0413] S601: Send an application push request to a server.
[0414] The application push request includes the terminal identification and target category of the terminal device.
[0415] The terminal identification of the terminal device is used to uniquely identify the terminal device. The terminal identification of the terminal device may include but is not limited to the terminal number; for example, when the terminal device is a mobile phone, the terminal number is the serial number of the mobile phone.
[0416] It should be understood that the above description uses the terminal identifier of the terminal device as the terminal number as an example; the present application does not limit the type of the terminal identifier of the terminal device.
[0417] In some embodiments, after the device manufacturer of the terminal device determines which candidate applications are paid applications, the paid applications can be marked with identifiers and stored in the server, so that after the server receives the push request sent by the terminal device, it can send the identifiers of the paid applications corresponding to the target category to the terminal device.
[0418] In this way, after determining the target category, the terminal device can send a push request to the server with the terminal identification and the target category of the terminal device, so that the server can send the identification of the paid application to the terminal device according to the target category.
[0419] S602: Receive application recommendation results sent by the server.
[0420] The application recommendation result includes the identifier of the target application program.
[0421] In some embodiments, the identification of the target application may be an identification of a paid application among all the applications in the target category.
[0422] Based on the above description, after receiving the push request sent by the terminal device, the server can determine all candidate applications in the target category and send the application recommendation result including the identification of the target application to the terminal device according to the terminal identification of the terminal device.
[0423] Thus, the terminal device can receive the application recommendation result sent by the server, so that the terminal device can display the recommendation information according to the application recommendation result.
[0424] Assume that the target category is the short video category, the short video category includes short video APP1, short video APP2 and short video APP3, the identification of the application stored in the server is the identification of the paid application, and the identification of the paid application includes the identification of short video APP1, the identification of short video APP2, the identification of long video APP2, the identification of long video APP3 and the identification of shopping APP1.
[0425] Then, after receiving the target category included in the application recommendation results sent by the server, the server can determine that short video APP1 and short video APP2 belong to the target category. Thus, the server can use the identifier of video APP1 and the identifier of short video APP2 as the identifier of the target application and send the application recommendation results including the identifier of video APP1 and the identifier of short video APP2 to the terminal device.
[0426] In addition, the identifier of the target application may be one or more; this application does not limit the number of identifiers of the target application.
[0427] In addition, the identifier of the target application may include the package name of the target application, and may also include the name of a sub-service in the target application; this application does not limit the type of the identifier of the target application.
[0428] S603: Display recommendation information in the application recommendation area of the first interface according to the application recommendation result.
[0429] Among them, the terminal device can display the application recommendation area in a variety of ways.
[0430] In some embodiments, the terminal device may display an application recommendation area on the first interface in an initial state.
[0431] In some other embodiments, the terminal device may display the application recommendation area on the first interface based on the user's instruction. Figure 1 The description of the application recommendation area of the interface is not repeated here.
[0432] The recommendation information may be in various forms.
[0433] In some embodiments, the recommendation information is an icon of a target application in a target category, and when the target application is identified as one, the application recommendation area displays an icon of the target application; when the target application is identified as multiple, the application recommendation area displays icons of multiple target applications sorted according to preset rules.
[0434] For example, when the target application is short video APP1 and the identifier of the target application includes the package name of short video APP1, the recommended information displayed in the application recommendation area is the icon of short video APP1.
[0435] It should be understood that the above description is made by taking the target application as short video APP1 and the recommended information as the icon of short video APP1 as an example; this application does not limit the type of target application and the type of recommended information.
[0436] In other embodiments, the recommendation information is an icon corresponding to a sub-service of a target application in a target category, and when the target application is identified as one, the application recommendation area displays an icon corresponding to the sub-service in the target application; when the target application is identified as multiple, the application recommendation area displays icons corresponding to the sub-services in the target application sorted according to preset rules.
[0437] For example, when the target application is short video APP1, short video APP1 includes 4 sub-services, namely recommended popular short videos, local popular short videos, live broadcast and shopping, and the identifier of the target application includes the package name of short video APP1 and the name of the recommended popular short video service in short video APP1, the recommendation information displayed in the application recommendation area is the icon corresponding to the recommended popular short video service in short video APP1.
[0438] It should be understood that the above description is based on an example in which the target application is short video APP1 and the recommended information is the icon corresponding to the recommended popular short video service in short video APP1; this application does not limit the type of target application and the type of recommended information.
[0439] In addition, when displaying recommendation information in the application recommendation area in the first interface based on the application recommendation results, the terminal device can determine the target application installed in the terminal device, so that the terminal device can display recommendation information based on the target application installed in the terminal device.
[0440] For example, when the target applications include short video APP1 and short video APP2, and the target application installed on the terminal device is short video APP1, the terminal device can display the recommendation information corresponding to short video APP1.
[0441] In the present application, after the terminal device determines the target category, it can send an application push request to the server and receive an application recommendation result including the identifier of the target application sent by the server. As a result, the terminal device can display the recommendation information corresponding to the identifier of the target application in the application recommendation area in the first interface according to the identifier of the target application included in the application recommendation result, thereby ensuring that the terminal device can accurately display the recommendation information of the candidate applications that the user wants to use, facilitating the user to use the candidate applications they want to use as soon as possible, improving the accuracy of the recommended applications, and ensuring the user experience.
[0442] Based on the above description, in a specific embodiment, it is assumed that:
[0443] 1. The terminal device is a mobile phone;
[0444] 2. The terminal device screens the multiple categories of data included in the first information in the multiple groups of first sample data twice, the first screening using the Pearson correlation coefficient and the Spelman correlation coefficient, and the second screening using the weight value used to indicate the correlation relationship between each category of data in the multiple categories of data and the multiple groups of label information.
[0445] Based on the above assumptions, the mobile phone can execute the probability prediction model generation method provided in this application.
[0446] The generation of the mobile phone execution probability prediction model may include the following steps:
[0447] Step 11: Obtain multiple groups of original sample data from the terminal device.
[0448] Among them, each group of original sample data in the multiple groups includes the original information of the first candidate application in the candidate application set opened within a preset historical number of days, and the original information includes the identification of the first candidate application, the time information of opening the first candidate application, and the location information of opening the first candidate application.
[0449] Step 12: Obtain multiple groups of first sample data of the terminal device according to the multiple groups of original sample data.
[0450] Each of the plurality of sets of first sample data includes first information and label information for opening each candidate application in the candidate application set.
[0451] Step 13: Screen multiple types of data included in the first information in multiple groups of first sample data according to the Pearson correlation coefficient and the Spelman correlation coefficient to obtain H types of data.
[0452] Step 14: Input H types of data into the logistic regression model to obtain H weight values corresponding to the H types of data.
[0453] Step 15: Filter H types of data based on H weight values to obtain Q types of data.
[0454] Step 16: Obtain multiple groups of second sample data of the terminal device according to the Q-type data.
[0455] Step 17: Input multiple groups of second sample data into the probability prediction model to be trained to obtain multiple groups of prediction probability values.
[0456] Each set of predicted probability values is used to represent the probability value of starting each candidate application in the candidate application set when the first candidate application is started.
[0457] Step 18: training the model parameters of the probability prediction model to be trained according to the differences between the multiple groups of predicted probability values and the multiple groups of label information to obtain the probability prediction model.
[0458] Based on the above description, in a specific embodiment, it is assumed that:
[0459] 1. The terminal device is a mobile phone;
[0460] 2. The N-type data is 2 types of data, namely the weekday corresponding to the date in the time information, and the network status information;
[0461] 3. The candidate application set includes candidate applications in three categories, namely, a short video category, a long video category and a shopping category. The short video category includes short video APP1, short video APP2 and short video APP3, the long video category includes long video APP1 and long video APP2, and the shopping category includes shopping APP1;
[0462] 4. The identification of the application stored in the service pool of the server is the identification of the paid application, and the identification of the paid application includes the identification of the short video APP1, the identification of the short video APP2, the identification of the long video APP2, the identification of the long video APP3 and the identification of the shopping APP1;
[0463] 5. The target application is short video APP1.
[0464] like Figure 8 As shown, based on the above assumptions, the mobile phone can execute the application recommendation method provided by this application.
[0465] The method for executing application recommendation on a mobile phone may include the following steps:
[0466] Step 21, predetermine the target time period (leisure time period).
[0467] Among them, the mobile phone can predict the target time period of the next day based on the time information of the daily user opening the candidate application in the candidate application set in the historical days, and after the mobile phone obtains the target time period, it can register a time fence according to the target time period.
[0468] Step 22. At the start time of the target time period, or at the time when the first interface is displayed within the target time period (triggering decision), obtain the first data of the terminal device at the current time. The first data includes 2 types of data. The 2 types of data are the day of the week corresponding to the date and network status information.
[0469] Step 23: Input the first data into a pre-trained probability prediction model to obtain five probability values corresponding to the five candidate applications.
[0470] Step 24. According to the category corresponding to the probability value with the highest probability value among the five probability values, the target category in the candidate application set is obtained. Among them, the probability value corresponding to the short video APP1 is 0.8, and the probability value corresponding to the long video APP1 is 0.6. Therefore, the target category is the short video category.
[0471] Among them, when the mobile phone determines the short video category, it can determine the short video APP (application matching) corresponding to the short video category.
[0472] Step 25: Send an application push request to the server, where the application push request includes the terminal identifier and target category of the terminal device.
[0473] Step 26: Receive the application recommendation result sent by the server, where the application recommendation result includes the package name of the short video APP1 in the short video category and the name of the recommended popular short video service in the short video APP1.
[0474] Step 27. During the target time period, based on the package name of the video APP1 and the name of the recommended popular short video service in the short video APP1, the icon corresponding to the recommended popular short video service in the short video APP1 is displayed in the application recommendation area in the first interface (reach decision).
[0475] Exemplarily, the present application also provides a terminal device.
[0476] Next, combine Fig. 9 , a terminal device provided in an embodiment of the present application is described in detail.
[0477] See also Fig. 9 , Fig. 9 A schematic block diagram of a terminal device provided in an embodiment of the present application is shown.
[0478] like Fig. 9As shown, the terminal device can exist independently or be integrated into other devices to achieve mutual communication with the above-mentioned terminal devices, and is used to implement the operations corresponding to the terminal device in the embodiments of the above-mentioned method. The terminal device of the present application may include: a first acquisition module 701, a first obtaining module 702 and a display module 703.
[0479] A first acquisition module 701 is used to acquire first data of the terminal device at the current moment, where the first data includes N types of data, where the N types of data are data associated with a candidate application in the set of candidate applications to be started, and N is an integer greater than or equal to 1;
[0480] The first obtaining module 702 is used to input the first data into a pre-trained probability prediction model to obtain the second data of the terminal device, the pre-trained probability prediction model is obtained by training based on multiple groups of first sample data, each group of first sample data in the multiple groups of first sample data includes first information and label information of each candidate application in the candidate application set, the first information includes at least one of the information of the first application, the first time difference information, the first association relationship, the second association relationship, the identifier of the first candidate application and the time information of starting the first candidate application; wherein the first application is an application opened before the first candidate application is opened, the first time difference information is the time difference information between opening the first candidate application and opening the first application, the first association relationship is the association relationship between opening the first candidate application and opening the first application, the second association relationship is the association relationship between opening the first candidate application and opening the first application, and the second association relationship is the association relationship between the location information and the number of times the first candidate application is opened, and the second data is used to represent the probability value of opening each candidate application in the candidate application set at the current moment;
[0481] The first obtaining module 702 is further used to obtain a target category in the candidate application set according to the second data, the candidate application set includes multiple categories of candidate applications, and the target category is the category with the highest activation probability value among the multiple categories of candidate applications;
[0482] The display module 703 is used to display recommendation information on the first interface according to the target category, where the recommendation information includes information of at least one candidate application corresponding to the target category.
[0483] In some embodiments, the second data includes a plurality of probability values, and the plurality of probability values correspond one-to-one to the candidate applications in the candidate application set. The first obtaining module 702 is specifically configured to:
[0484] Determine M probability values greater than a first preset threshold value among the multiple probability values, where M is an integer greater than or equal to 1;
[0485] According to the categories of the candidate applications corresponding to the M probability values, a target category is obtained.
[0486] In some embodiments, the second data includes a plurality of probability values, and the plurality of probability values correspond one-to-one to the candidate applications in the candidate application set. The first obtaining module 702 is specifically configured to:
[0487] Adding probability values of candidate applications of the same category among the multiple probability values to obtain multiple first probability values, wherein the multiple first probability values correspond one-to-one to the candidate applications of the multiple categories;
[0488] Determine K first probability values greater than a second preset threshold value among the multiple first probability values, where K is an integer greater than or equal to 1;
[0489] According to the categories of the candidate applications corresponding to the K probability values, the target category is obtained.
[0490] In some embodiments, the second data includes a plurality of probability values, and the plurality of probability values correspond one-to-one to the candidate applications in the candidate application set. The first obtaining module 702 is specifically configured to:
[0491] Averaging the probability values of the candidate applications of the same category among the multiple probability values to obtain multiple second probability values, wherein the multiple second probability values correspond one-to-one to the candidate applications of the multiple categories;
[0492] Determine P second probability values greater than a third preset threshold value among the multiple second probability values, where P is an integer greater than or equal to 1;
[0493] A target category is obtained according to the categories of the candidate application programs corresponding to the P second probability values.
[0494] In some embodiments, the display module 703 is specifically used to:
[0495] Sending an application push request to the server, the application push request including the terminal identification and target category of the terminal device;
[0496] Receiving an application recommendation result sent by the server, the application recommendation result including an identifier of a target application program;
[0497] According to the application recommendation result, the recommendation information is displayed in the application recommendation area in the first interface.
[0498] Exemplarily, the present application also provides a terminal device.
[0499] Next, combine Fig.10 , a terminal device provided in an embodiment of the present application is described in detail.
[0500] See also Fig.10 , Fig.10 A schematic block diagram of a terminal device provided in an embodiment of the present application is shown.
[0501] like Fig.10 As shown, the terminal device can exist independently or be integrated into other devices to achieve mutual communication with the above-mentioned terminal devices, and is used to implement the operations corresponding to the terminal device in the embodiments of the above-mentioned method. The terminal device of the present application may also include: a second acquisition module 801 and a second obtaining module 802.
[0502] The second acquisition module 801 is used to acquire multiple groups of original sample data of the terminal device, each group of original sample data in the multiple groups of original sample data includes original information of the first candidate application in the candidate application set opened within a preset number of historical days, and the original information includes the identifier of the first candidate application, the time information of opening the first candidate application, and the location information of opening the first candidate application;
[0503] The second obtaining module 802 is used to obtain multiple groups of first sample data of the terminal device according to the multiple groups of original sample data, each group of first sample data in the multiple groups of first sample data includes first information and label information of each candidate application in the set of candidate application openings, and the first information includes at least one of information of the first application, first time difference information, first association relationship, second association relationship and original information; wherein the first application is an application opened before opening the first candidate application, the first time difference information is time difference information between opening the first candidate application and opening the first application, the first association relationship is an association relationship between opening the first candidate application and opening the first application, and the second association relationship is an association relationship between location information and number information of opening the first candidate application;
[0504] The second obtaining module 802 is used to train the probability prediction model to be trained according to multiple groups of first sample data to obtain the probability prediction model.
[0505] In some embodiments, the multiple sets of first sample data include multiple sets of label information, the first information in the multiple sets of first sample data includes multiple types of data, and the second obtaining module 802 is specifically used to:
[0506] Screening multiple types of data in multiple groups of first sample data to obtain multiple groups of second sample data of terminal devices;
[0507] Inputting multiple sets of second sample data into the probability prediction model to be trained to obtain multiple sets of prediction probability values, each set of prediction probability values being used to represent the probability value of starting each candidate application in the candidate application set when starting the first candidate application;
[0508] The model parameters of the probability prediction model to be trained are trained according to the differences between the multiple groups of prediction probability values and the multiple groups of label information to obtain the probability prediction model.
[0509] In some embodiments, the second obtaining module 802 is specifically configured to:
[0510] The multiple categories of data in the multiple groups of first sample data are screened according to the correlation value to obtain multiple groups of second sample data of the terminal device; the correlation value is used to characterize the association relationship between each category of data in the multiple categories of data and the multiple groups of label information.
[0511] In some embodiments, the second obtaining module 802 is specifically configured to:
[0512] Screening multiple types of data in multiple groups of first sample data according to the correlation value to obtain H types of data, where H is an integer greater than or equal to 1;
[0513] Inputting H types of data into the logistic regression model, obtaining H weight values corresponding to the H types of data, wherein one weight value among the H weight values is used to represent the degree of association between a type of data in the H types of data and a candidate application in the candidate set for starting;
[0514] Based on H weight values, H types of data are screened to obtain Q types of data, where Q is an integer greater than or equal to 1;
[0515] According to the Q-type data, multiple groups of second sample data of the terminal device are obtained.
[0516] In some embodiments, the correlation value includes a first correlation value and / or a second correlation value, the first correlation value is used to characterize the linear relationship between each type of data in the multiple types of data and the multiple groups of label information; the second correlation value is used to characterize the monotonic relationship between each type of data in the multiple types of data and the multiple groups of label information.
[0517] In some embodiments, the original information also includes network status information.
[0518] In some embodiments, the first information further includes external device information, memory status information, information of the second application, first duration information, information of the third application, basic information of the opened candidate application, first opening times information, second time difference information, first opening times information, second duration information, third association relationship and first ratio information;
[0519] Among them, the second application is an application opened in the same period of the historical date corresponding to the period of time when the first candidate application is opened, the first duration information is the duration information of opening the first application, the third application is the application with the highest number of openings within the first preset duration before the first candidate application is opened, the first opening number information is the number of times the first candidate application is opened within the second preset time before the first candidate application is opened, the second time difference information is the time difference information between opening the first candidate application and opening the first candidate application, the first opening number information is the number of times the first candidate application is opened within the third preset time before the first candidate application is opened, the second duration information is the opening time information within the third time before the first candidate application is opened, the third association relationship is the association relationship between the external device and the opening of the first candidate application, and the first ratio information is the ratio of the number of days the first candidate application is opened within the same period of the first historical days to the first historical days.
[0520] Exemplarily, the present application provides a terminal device, comprising one or more processors, a memory and one or more computer programs, wherein the one or more computer programs are stored in the memory, and when the computer programs are executed by one or more processors, the terminal device executes the application recommendation method and / or the probability prediction model generation method in the foregoing embodiments.
[0521] Exemplarily, the present application provides a chip system, which includes a processor for calling and running a computer program from a memory, so that a terminal device equipped with the chip system executes the application recommendation method and / or the probability prediction model generation method in the foregoing embodiments.
[0522] Exemplarily, the present application provides a computer-readable storage medium, including a computer program, which, when executed on a terminal device, enables the terminal device to execute the application recommendation method and / or the probability prediction model generation method in the foregoing embodiments.
[0523] Exemplarily, the present application provides a computer program product, which, when executed on a computer, enables the computer to execute the application recommendation method and / or the probability prediction model generation method in the foregoing embodiments.
[0524] In the above embodiments, all or part of the functions can be implemented by software, hardware, or a combination of software and hardware. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer codes or instructions. When the computer program code or instructions are loaded and executed on a computer, the process or function according to the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer code or instructions can be stored in a computer-readable storage medium. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integration. Available media can be magnetic media, (e.g., floppy disk, hard disk, tape), optical media (e.g., DVD), or semiconductor media (e.g., solid state disk (SSD)), etc.
[0525] A person skilled in the art can understand that to implement all or part of the processes in the above-mentioned embodiments, the processes can be completed by a computer program to instruct the relevant hardware, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-mentioned method embodiments. The aforementioned storage medium includes: a read-only memory (ROM) or a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.
Claims
1. An application recommendation method, characterized in that: The method comprises: Acquire first data of the terminal device at the current moment, where the first data includes N types of data, where the N types of data are data associated with a candidate application program in the set of candidate applications to be started, and N is an integer greater than or equal to 1; Input the first data into a pre-trained probability prediction model to obtain second data of the terminal device, wherein the pre-trained probability prediction model is trained based on multiple groups of first sample data, wherein each group of first sample data in the multiple groups of first sample data includes first information and label information of each candidate application in the set of candidate applications for opening, and the first information includes at least one of information of the first application, first time difference information, first association relationship, second association relationship, an identifier of the first candidate application, and time information for opening the first candidate application; wherein the first application is an application opened before opening the first candidate application, the first time difference information is the time difference information between opening the first candidate application and opening the first application, the first association relationship is the association relationship between opening the first candidate application and opening the first application, the second association relationship is the association relationship between opening the first candidate application and opening the first application, and the second data is used to represent the probability value of opening each candidate application in the set of candidate applications at the current moment; According to the second data, a target category in the candidate application set is obtained, the candidate application set includes multiple categories of candidate applications, and the target category is a category with the highest activation probability value among the multiple categories of candidate applications; Recommendation information is displayed on the first interface according to the target category, where the recommendation information includes information of at least one candidate application corresponding to the target category.
2. The method according to claim 1, characterized in that The second data includes a plurality of probability values, and the plurality of probability values correspond one-to-one to the candidate applications in the candidate application set. The obtaining of the target category according to the second data includes: Determine M probability values greater than a first preset threshold value among the multiple probability values, where M is an integer greater than or equal to 1; The target category is obtained according to the categories of the candidate application programs corresponding to the M probability values.
3. The method according to claim 1, characterized in that The second data includes a plurality of probability values, and the plurality of probability values correspond one-to-one to the candidate applications in the candidate application set. The obtaining of the target category according to the second data includes: Adding probability values of candidate applications of the same category among the multiple probability values to obtain multiple first probability values, wherein the multiple first probability values correspond one-to-one to the candidate applications of the multiple categories; Determine K first probability values greater than a second preset threshold value among the multiple first probability values, where K is an integer greater than or equal to 1; The target category is obtained according to the categories of the candidate application programs corresponding to the K probability values.
4. The method according to claim 1, characterized in that: The second data includes a plurality of probability values, and the plurality of probability values correspond one-to-one to the candidate applications in the candidate application set. The obtaining of the target category according to the second data includes: Averaging the probability values of the candidate applications of the same category among the multiple probability values to obtain multiple second probability values, wherein the multiple second probability values correspond one-to-one to the candidate applications of the multiple categories; Determine P second probability values greater than a third preset threshold value among the multiple second probability values, where P is an integer greater than or equal to 1; The target category is obtained according to the categories of the candidate application programs corresponding to the P second probability values.
5. The method according to any one of claims 1 to 4, characterized in that The displaying of recommendation information on the first interface according to the target category includes: Sending an application push request to a server, wherein the application push request includes a terminal identifier of the terminal device and the target category; Receiving an application recommendation result sent by the server, wherein the application recommendation result includes an identifier of a target application program; According to the application recommendation result, recommendation information is displayed in the application recommendation area in the first interface.
6. A method for generating a probability prediction model, characterized in that: The method comprises: Acquire multiple groups of original sample data of the terminal device, each group of original sample data in the multiple groups of original sample data includes original information of a first candidate application in a set of candidate applications opened within a preset number of historical days, the original information including an identifier of the first candidate application, time information of opening the first candidate application, and location information of opening the first candidate application; According to the multiple groups of original sample data, multiple groups of first sample data of the terminal device are obtained, each group of first sample data in the multiple groups of first sample data includes first information and label information of each candidate application in the set of candidate applications for opening, the first information includes information of the first application, first time difference information, a first association relationship, a second association relationship and at least one of the original information; wherein the first application is an application opened before opening the first candidate application, the first time difference information is the time difference information between opening the first candidate application and opening the first application, the first association relationship is the association relationship between opening the first candidate application and opening the first application, and the second association relationship is the association relationship between the location information and the number of times the first candidate application is opened; The probability prediction model to be trained is trained according to the multiple groups of first sample data to obtain a probability prediction model.
7. The method according to claim 6, characterized in that The multiple groups of first sample data include multiple groups of label information, the first information in the multiple groups of first sample data includes multiple types of data, and the probability prediction model to be trained is trained according to the multiple groups of first sample data to obtain the probability prediction model, including: Screening the multiple types of data in the multiple groups of first sample data to obtain multiple groups of second sample data of the terminal device; Inputting the plurality of groups of second sample data into the probability prediction model to be trained to obtain a plurality of groups of prediction probability values, each group of prediction probability values being used to represent the probability value of starting each candidate application in the candidate application set when starting the first candidate application; The model parameters of the probability prediction model to be trained are trained according to the differences between the multiple groups of predicted probability values and the multiple groups of label information to obtain the probability prediction model.
8. The method according to claim 7, characterized in that The screening of the multiple types of data in the multiple groups of first sample data to obtain multiple groups of second sample data of the terminal device includes: The multiple categories of data in the multiple groups of first sample data are screened according to the correlation value to obtain multiple groups of second sample data of the terminal device; the correlation value is used to characterize the association relationship between each category of data in the multiple categories of data and the multiple groups of label information.
9. The method according to claim 7, characterized in that: The screening of the multiple types of data in the multiple groups of first sample data to obtain multiple groups of second sample data of the terminal device includes: Screening the multiple types of data in the multiple groups of first sample data according to the correlation values to obtain H types of data, where H is an integer greater than or equal to 1; Inputting the H types of data into a logistic regression model to obtain H weight values corresponding to the H types of data, wherein one weight value among the H weight values is used to characterize the degree of association between a type of data in the H types of data and the opening of a candidate application in the candidate set; The H-type data are screened based on the H weight values to obtain Q-type data, where Q is an integer greater than or equal to 1; A plurality of groups of second sample data of the terminal device are obtained according to the Q-type data.
10. The method according to claim 8 or 9, characterized in that: The correlation value includes a first correlation value and / or a second correlation value, wherein the first correlation value is used to characterize the linear relationship between each type of data in the multiple types of data and the multiple groups of label information; the second correlation value is used to characterize the monotonic relationship between each type of data in the multiple types of data and the multiple groups of label information.
11. The method according to any one of claims 7 to 10, characterized in that The original information also includes network status information.
12. The method according to any one of claims 7 to 11, characterized in that The first information also includes external device information, memory status information, second application information, first duration information, third application information, basic information of the opened candidate application, first opening times information, second time difference information, first opening times information, second duration information, third association relationship and first ratio information; Among them, the second application is an application opened in the same period of the historical date corresponding to the period of time when the first candidate application is opened, the first duration information is the duration information of opening the first application, the third application is the application with the highest number of openings within the first preset duration before opening the first candidate application, the first opening number information is the number of times the first candidate application is opened within the second preset duration before opening the first candidate application, the second time difference information is the time difference information between opening the first candidate application and opening the first candidate application, the first opening number information is the number of times the candidate application is opened within the third preset duration before opening the first candidate application, the second duration information is the opening duration information within the third duration before opening the first candidate application, the third association relationship is the association relationship between the external device and the opening of the first candidate application, and the first ratio information is the ratio of the number of days the first candidate application is opened within the same period of the first historical days to the first historical days.
13. A terminal device, characterized in that: The terminal device includes a module for executing the application recommendation method as described in any one of claims 1 to 5, and / or a module for executing the probability prediction model generation method as described in any one of claims 6 to 12.
14. A terminal device, characterized in that: include: One or more processors, a memory, and one or more computer programs; wherein the one or more computer programs are stored in the memory, and when the computer programs are executed by the one or more processors, the terminal device executes the application recommendation method as described in any one of claims 1 to 5, and / or the probability prediction model generation method as described in any one of claims 6 to 12.
15. A chip system, characterized in that: The chip system includes a processor for calling and running a computer program from a memory, so that a terminal device equipped with the chip system executes the application recommendation method as described in any one of claims 1 to 5, and / or the probability prediction model generation method as described in any one of claims 6 to 12.
16. A computer-readable storage medium comprising a computer program, characterized in that: When the computer program runs on a terminal device, the terminal device executes the application recommendation method as described in any one of claims 1 to 5, and / or the probability prediction model generation method as described in any one of claims 6 to 12.
Citation Information
Patent Citations
Application program prediction model establishment and preloading method and device, medium and terminal
CN109766138A
Application recommendation method and device, storage medium and electronic equipment
CN111277706A
Application prediction method, electronic equipment and storage medium
CN115016854A
Application program recommendation method and electronic equipment
CN115981520A
