Data processing method and device

Abstract

The present application discloses a data processing method and device, belonging to the field of computer technology. The method comprises: obtaining a plurality of feature data; automatically performing two different nonlinear combination processing on the plurality of feature data to obtain two groups of processed data, wherein the two groups of processed data include a group of high-order data and a group of low-order data, the high-order data is related to the nonlinear combination of m feature data in the plurality of feature data, and the low-order data is related to the nonlinear combination of n feature data in the plurality of feature data, m≥3, m>n≥2; determining predicted data according to a plurality of target data, the plurality of target data includes two groups of processed data. The present application solves the problem of low accuracy of predicted data determined by the related art, improves the accuracy of predicted data, and the present application is used for data processing.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a data processing method and device. Background Art

[0002] With the development of computer technology, when people analyze things that can change dynamically (such as the number of visits to a website, the number of downloads of a file, the click-through rate of an application icon in an application store, or the click-through rate of a video on a video website), they can usually obtain multiple feature data of the thing (feature data is used to characterize the attributes of the thing), and process the multiple feature data to obtain predictive data used to characterize the development direction of the thing.

[0003] In the related art, people can input multiple feature data of an object into a data processing device, and the data processing device can perform high-order nonlinear combination processing on the multiple feature data of the object to obtain multiple high-order data, and determine prediction data based on the multiple high-order data, wherein the high-order data is a nonlinear combination of at least three feature data in the multiple feature data. For example, the data processing device can perform high-order nonlinear combination processing on multiple feature data of an application in an application store to obtain prediction data for the click rate of the icon used to characterize the application, and the recommendation priority of the application can be determined based on the prediction data.

[0004] Since the prediction data in the related technology is obtained by the data processing device based on multiple high-order data, and accurate prediction needs to be based on a large amount of data, the amount of data of the above multiple high-order data still cannot meet the accurate prediction requirements. Therefore, the accuracy of the prediction data determined by the related technology is relatively low. Summary of the invention

[0005] In order to solve the problem of low accuracy of prediction data determined by related technologies, the present application provides a data processing method and device. The technical solution is as follows:

[0006] In a first aspect, a data processing method is provided, the method comprising: acquiring a plurality of feature data; automatically performing two different nonlinear combination processing on the plurality of feature data to obtain two groups of processed data, wherein the two groups of processed data include a group of high-order data and a group of low-order data, the high-order data is related to a nonlinear combination of m feature data in the plurality of feature data, and the low-order data is related to a nonlinear combination of n feature data in the plurality of feature data, m≥3, m>n≥2; determining predicted data according to a plurality of target data, the plurality of target data including the two groups of processed data.

[0007] In the related art, a group of processed data (including at least one high-order data) can be obtained by performing a nonlinear combination process (high-order nonlinear combination process) on multiple feature data; while the present application can automatically perform two different nonlinear combination processes (high-order nonlinear combination process and low-order nonlinear combination process) on multiple feature data and obtain two groups of processed data. That is, the amount of data of the two groups of processed data obtained in the present application is greater than the amount of data of the one group of processed data obtained in the related art, and accurate prediction needs to be based on a large amount of data to be realized, therefore, the accuracy of the predicted data obtained based on the two groups of processed data in the present application is higher than the accuracy of the predicted data obtained based on the one group of processed data in the related art.

[0008] Optionally, the automatically performing two different nonlinear combination processing on the multiple feature data to obtain two groups of processed data includes: automatically performing high-order nonlinear combination processing on the multiple feature data to obtain the group of high-order data; and automatically performing low-order nonlinear combination processing on the multiple feature data to obtain the group of low-order data.

[0009] It should be noted that after obtaining multiple feature data, if the number of high-order data obtained by performing high-order nonlinear combination processing on a part of the multiple feature data is the first number, and the number of low-order data obtained by performing low-order nonlinear combination processing on another part of the multiple feature data is the second number, then the number of target data based on which the prediction data is finally determined is equal to the sum of the first number and the second number. If the multiple data are processed by high-order nonlinear combination processing, the number of high-order data obtained is greater than the first number, and the multiple data are also processed by low-order nonlinear combination processing, the number of low-order data obtained is greater than the second number, and the number of target data based on which the prediction data is finally determined is also greater than the sum of the first number and the second number. That is, when the multiple feature data are processed by high-order nonlinear combination processing and low-order nonlinear combination processing respectively, the number of target data based on which the prediction data is finally determined will be more, and the accuracy of the prediction data will be higher.

[0010] Optionally, after acquiring the multiple feature data, the method further includes: automatically performing linear combination processing on the multiple feature data to obtain a set of first auxiliary data, wherein the first auxiliary data is related to the linear combination of at least two feature data among the multiple feature data, and the multiple target data also include: the at least one first auxiliary data.

[0011] That is, the present application can not only perform high-order nonlinear combination processing and low-order nonlinear combination processing on multiple feature data, but also perform linear combination processing on multiple feature data to obtain a set of first auxiliary data, thereby further increasing the number of target data on which the final prediction data is based, and further improving the accuracy of the prediction data.

[0012] Optionally, the automatically performing low-order nonlinear combination processing on the multiple feature data to obtain the set of low-order data includes: automatically performing low-order nonlinear combination processing on the multiple feature data through a processing module based on a factor decomposition machine FM to obtain the set of low-order data; the automatically performing linear combination processing on the multiple feature data to obtain a set of first auxiliary data includes: automatically performing linear combination processing on the multiple feature data through a processing module based on FM to obtain the set of first auxiliary data; or, the automatically performing low-order nonlinear combination processing on the multiple feature data to obtain the set of low-order data includes: automatically performing low-order nonlinear combination processing on the multiple feature data through a processing module based on a domain-aware decomposition machine FFM to obtain the set of low-order data; the automatically performing linear combination processing on the multiple feature data to obtain a set of first auxiliary data includes: automatically performing linear combination processing on the multiple feature data through a processing module based on FFM to obtain the set of first auxiliary data.

[0013] Optionally, before determining the prediction data based on multiple target data, the method also includes: combining expert experience and the multiple feature data to determine x empirical feature data, x≥1, the x empirical feature data including: data obtained from the multiple feature data in combination with expert experience, and / or, data obtained by performing nonlinear combination processing on at least two of the multiple feature data in combination with expert experience; automatically processing the x empirical feature data to obtain a set of second auxiliary data, the multiple target data also including: the set of second auxiliary data, wherein, when x=1, the second auxiliary data is related to one empirical feature data, when x>1, the second auxiliary data is related to a linear combination of at least two of the x empirical feature data, and / or the second auxiliary data is related to a nonlinear combination of at least two of the x empirical feature data.

[0014] That is, the present application can determine x empirical feature data that have a greater impact on the predicted data based on the guidance of expert experience, and process the x empirical feature data to obtain a set of second auxiliary data. And the target data based on which the predicted data is finally determined also includes the second auxiliary data, thereby further increasing the number of target data based on which the predicted data is finally determined, and further improving the accuracy of the predicted data.

[0015] Optionally, the multiple feature data are multiple sparse feature data. After acquiring the multiple feature data, the method includes: converting the multiple sparse feature data into multiple dense feature data; automatically performing two different nonlinear combination processing on the multiple feature data to obtain two groups of processed data, including: automatically performing two different nonlinear combination processing on the multiple dense feature data to obtain the two groups of processed data.

[0016] That is, after acquiring multiple sparse feature data, the present application can also convert the multiple sparse feature data to obtain multiple dense feature data, and the data volume of the dense feature data is much smaller than the data volume of the sparse feature data, and the efficiency of processing multiple dense feature data is higher than the efficiency of processing multiple sparse feature data.

[0017] In a second aspect, a data processing device is provided, comprising: an acquisition unit for acquiring multiple feature data; a nonlinear processing unit for automatically performing two different nonlinear combination processing on the multiple feature data to obtain two groups of processed data, wherein the two groups of processed data include a group of high-order data and a group of low-order data, the high-order data is related to a nonlinear combination of m feature data in the multiple feature data, and the low-order data is related to a nonlinear combination of n feature data in the multiple feature data, m≥3, m>n≥2; a prediction unit for determining predicted data based on multiple target data, the multiple target data including the two groups of processed data.

[0018] Optionally, the nonlinear processing unit includes: a high-order processing subunit, used to automatically perform high-order nonlinear combination processing on the multiple feature data to obtain the set of high-order data; and a low-order processing subunit, used to automatically perform low-order nonlinear combination processing on the multiple feature data to obtain the set of low-order data.

[0019] Optionally, the data processing device also includes: a linear processing unit, used to automatically perform linear combination processing on the multiple feature data to obtain a set of first auxiliary data, wherein the first auxiliary data is related to the linear combination of at least two feature data among the multiple feature data, and the multiple target data also include: the at least one first auxiliary data.

[0020] Optionally, the data processing device also includes: a processing module based on a factorization machine FM or a processing module based on a domain-aware decomposition machine FFM, the FM-based processing module includes: the low-order processing subunit and the linear processing unit; the FFM-based processing module includes: the low-order processing subunit and the linear processing unit.

[0021] Optionally, the data processing device also includes: a determination unit, which determines x empirical feature data in combination with expert experience and the multiple feature data, x≥1, and the x empirical feature data include: data obtained from the multiple feature data in combination with expert experience, and / or data obtained by performing nonlinear combination processing on at least two of the multiple feature data in combination with expert experience; an experience processing unit, which is used to automatically process the x empirical feature data to obtain a set of second auxiliary data, and the multiple target data also include: the set of second auxiliary data, wherein, when x=1, the second auxiliary data is related to one empirical feature data, and when x>1, the second auxiliary data is related to a linear combination of at least two of the x empirical feature data, and / or the second auxiliary data is related to a nonlinear combination of at least two of the x empirical feature data.

[0022] Optionally, the multiple feature data are multiple sparse feature data, and the data processing device also includes: a conversion unit, used to convert the multiple sparse feature data into multiple dense feature data; the nonlinear processing unit is also used to: automatically perform two different nonlinear combination processing on the multiple dense feature data to obtain the two groups of processed data.

[0023] In a third aspect, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the data processing method described in the first aspect is implemented.

[0024] In a fourth aspect, a computer program product is provided, which, when executed on a computer, enables the computer to execute the data processing method described in the first aspect.

[0025] In a fifth aspect, a data processing device is provided, comprising: at least one processor, at least one network interface, a memory and at least one bus, wherein the memory and the network interface are respectively connected to the processor via the bus; the processor is configured to execute instructions stored in the memory; the processor implements the data processing method provided by the first aspect or any possible implementation method of the first aspect by executing the instructions.

[0026] The technical effects obtained by the above-mentioned second to fifth aspects are similar to the technical effects obtained by the corresponding technical means in the above-mentioned first aspect, and this application will not repeat them here.

[0027] The beneficial effects of the technical solution provided by this application are:

[0028] In the related art, a group of processed data (including at least one high-order data) can be obtained by performing a nonlinear combination process (high-order nonlinear combination process) on multiple feature data; while the present application can automatically perform two different nonlinear combination processes (high-order nonlinear combination process and low-order nonlinear combination process) on multiple feature data and obtain two groups of processed data. That is, the amount of data of the two groups of processed data obtained in the present application is greater than the amount of data of the one group of processed data obtained in the related art, and accurate prediction needs to be based on a large amount of data to be realized, therefore, the accuracy of the predicted data obtained based on the two groups of processed data in the present application is higher than the accuracy of the predicted data obtained based on the one group of processed data in the related art. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A schematic diagram of an application scenario of a data processing method provided by an embodiment of the present invention;

[0030] Figure 2 A schematic diagram of the structure of a data processing device provided by an embodiment of the present invention;

[0031] Figure 3 A schematic diagram of the structure of a program module provided by an embodiment of the present invention;

[0032] Figure 4 A method flow chart of a data processing method provided by an embodiment of the present invention;

[0033] Figure 5 A schematic diagram of the structure of another data processing device provided by an embodiment of the present invention;

[0034] Figure 6 A schematic diagram of the structure of another data processing device provided by an embodiment of the present invention;

[0035] Figure 7 A schematic diagram of the structure of another data processing device provided by an embodiment of the present invention;

[0036] Figure 8 A schematic diagram of the structure of a data processing device provided in another embodiment of the present application. DETAILED DESCRIPTION

[0037] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0038] Figure 1 A schematic diagram of an application scenario of a data processing method provided by an embodiment of the present invention, such as Figure 1As shown, the server 1 can establish communication connections with multiple clients 2, and the server 1 can obtain data related to the client 2; the client 2 can also obtain data stored on the server from the server 1. The data processing device 0 is connected to the server 1, and the data processing device 0 can also be set on the server 1, and the data processing device 0 can process the data obtained by the server. For example, the data processing device 0 and the server 1, as well as the server 1 and the client 2 can be connected through a wired network or a wireless network.

[0039] It should be noted that the embodiments of the present invention involve the terms “linear combination correlation”, “nonlinear combination correlation”, “linear combination processing” and “nonlinear combination processing”. The embodiments of the present invention explain these terms here:

[0040] For example, when a plurality of feature data are combined and processed (linear combination processing or nonlinear combination processing), the plurality of feature data can be combined and processed according to a preset equation. When the preset equation is a linear equation, the combination processing of a plurality of feature data is also referred to as "linear combination processing". At this time, the data obtained by the combination processing is "related to the linear combination" of the plurality of feature data. When the preset equation is a nonlinear equation, the "combination processing" of a plurality of feature data is also referred to as "nonlinear combination processing". At this time, the data obtained by the combination processing is "related to the nonlinear combination" of the plurality of feature data. Among them, the linear equation is a multivariate linear equation, and the relationship between the independent variable and the dependent variable in the linear equation is a linear relationship, and the linear equation is represented as a straight line in a Cartesian coordinate system. The nonlinear equation is a multivariate multi-time equation, and the relationship between the independent variable and the dependent variable in the nonlinear equation is not a linear relationship, and the nonlinear equation is represented as a curve in a Cartesian coordinate system.

[0041] For example, if the multiple feature data include: feature data a1 and feature data a2, then performing linear combination processing on the multiple feature data can obtain data y1=d1*a1+d2*a2, wherein d1 is the weight of feature data a1, d2 is the weight of feature data a2, and y1 is related to the linear combination of feature data a1 and feature data a2; performing nonlinear combination processing on the multiple feature data can obtain data y2=a1*a2, and y2 is related to the nonlinear combination of feature data a1 and feature data a2. It should be noted that the embodiment of the present invention only takes y1=d1*a1+d2*a2 and y2=a1*a2 as examples to illustrate the linear combination processing and nonlinear combination processing of multiple feature data. In actual applications, the data obtained by the linear combination processing may not be d1*a1+d2*a2, and the data obtained by the nonlinear combination processing may not be a1*a2, and the embodiment of the present invention does not limit this.

[0042] Figure 2A schematic diagram of the structure of a data processing device provided in an embodiment of the present invention. Optionally, the data processing device may be Figure 1 The data processing device shown. Figure 2 As shown, the data processing device may include: at least one processor 10221 (e.g., a central processing unit), at least one network interface 10222, a memory 10223, and at least one bus 10224. The bus 10224 may be used to realize the connection and communication between the processor, the network interface, and the memory; the memory 10223 and the network interface 10222 are respectively connected to the processor 10221 through the bus 10224. The processor 10221 is used to execute executable modules stored in the memory 10223, such as computer programs. The memory 10223 may include a high-speed random access memory (Random Access Memory, RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. The communication connection between the data processing device and at least one other device is realized through at least one network interface 10222 (wired or wireless). In some embodiments, the memory 10223 stores a program module 10225, and the program module 10225 can be executed by the processor 10221 to realize the following. Figure 4 The data processing method shown.

[0043] For example, Figure 3 As shown, the program modules may include:

[0044] The acquisition module 01 is used to acquire a plurality of feature data, where the plurality of feature data is a plurality of sparse feature data.

[0045] The conversion module 02 is used to convert a plurality of sparse feature data into a plurality of dense feature data;

[0046] Nonlinear processing module ( Figure 3 Not marked in the figure), which is used to automatically perform two different nonlinear combination processing on multiple feature data to obtain two groups of processed data, wherein the two groups of processed data include a group of high-order data and a group of low-order data, the high-order data is related to the nonlinear combination of m feature data in the multiple feature data, and the low-order data is related to the nonlinear combination of n feature data in the multiple feature data, m≥3, m>n≥2; the nonlinear processing module is also used to: automatically perform two different nonlinear combination processing on multiple dense feature data to obtain two groups of processed data.

[0047] The linear processing module 04 is used to automatically perform linear combination processing on multiple feature data to obtain a set of first auxiliary data, where the first auxiliary data is related to the linear combination of at least two feature data in the multiple feature data. The multiple target data also includes: at least one first auxiliary data.

[0048] Prediction module 05, used to determine prediction data according to a plurality of target data, the plurality of target data including two sets of processed data;

[0049] Optionally, the nonlinear processing module may include: a high-order processing submodule 031 and a low-order processing submodule 032. The high-order processing submodule 031 can be used to automatically perform high-order nonlinear combination processing on multiple feature data to obtain a set of high-order data; the low-order processing submodule 032 can be used to automatically perform low-order nonlinear combination processing on multiple feature data to obtain a set of low-order data.

[0050] Optionally, the program module may also include: a processing module based on a factorization machine (FM) or a processing module based on a field-aware factorization machine (FFM), the FM-based processing module includes: a low-order processing sub-module 032 and a linear processing module 04; the FFM-based processing module includes: a low-order processing sub-module 032 and a linear processing module 04.

[0051] Optionally, the program module may also include: a determination module and an experience processing module ( Figure 3 (not shown in the figure), the determination module can be used to combine expert experience and multiple feature data to determine x empirical feature data, x≥1, the x empirical feature data include: data obtained from multiple feature data in combination with expert experience, and / or, data obtained by performing nonlinear combination processing on at least two feature data in the multiple feature data in combination with expert experience; the experience processing module can be used to automatically process the x empirical feature data to obtain a set of second auxiliary data, and the multiple target data also include: a set of second auxiliary data, wherein, when x=1, the second auxiliary data is related to one empirical feature data, and when x>1, the second auxiliary data is related to a linear combination of at least two empirical feature data in the x empirical feature data, and / or, the second auxiliary data is related to a nonlinear combination of at least two empirical feature data in the x empirical feature data.

[0052] Figure 4 A method flow chart of a data processing method provided by an embodiment of the present invention, the data processing method can be used for Figure 2 The data processing device shown, such as Figure 4 As shown, the data processing method may include:

[0053] Step 401: Acquire multiple sparse feature data.

[0054] When people analyze dynamically changing things (such as the number of visits to a website, the number of downloads of a file, the click-through rate of an app icon in an app store, or the click-through rate of a video on a video website), they can usually obtain multiple sparse feature data of the things.

[0055] For example, multiple application icons are displayed in the client of the application store. When the user needs to download an application, he can click the icon of the application to be downloaded in the multiple applications to enter the download interface of the application and decide whether to download the application. Therefore, whether the user will download the application is related to whether the user will click the application icon. The server of the application store can determine the recommendation priority of the application based on the click-through rate (CTR) of the application icon, assign a higher priority to applications with high click-through rates, and correspondingly display the icons of these applications in the application store first, so as to recommend applications that meet user needs to users in a timely and accurate manner, and improve user stickiness. In this scenario, the prediction of the click-through rate of the application icon is particularly important. When the click-through rate of the application icon needs to be predicted, the multiple sparse feature data obtained by the data processing device in step 401 can be the sparse feature data of the application, wherein the sparse feature data of the application includes: attribute data of the user and attribute data of the application. The attribute data of the user may include attribute data such as the user's identification, the user's gender, the user's age, the user's occupation, the user's income, the user's hobbies, and the user's education. The attribute data of the application may include: the identification of the application, the name of the application, the developer of the application, the type of the application, the size of the installation package of the application, the label of the application, the score of the application, the comments of the application, etc. For example, in the sparse feature data of a certain application: the user identification is Xiao Ming, the user's gender is male, the user's age is 20 years old, the name of the application is F1, and the type of the application is a chat application, the above sparse feature data obtained in step 401 can be arranged according to certain rules, such as "Xiao Ming, male, 20, F1, chat".

[0056] For another example, when a user logs in to a website through a terminal and searches for a keyword in a website that can download files, the website can control the terminal to display multiple file entries related to the keyword based on the keyword, and the user can click on the entry of a file to download the file according to the needs. For example, files that are downloaded multiple times usually meet the download needs of multiple users. Therefore, the website can determine the recommended priority of the file based on the download volume of the file, assign a higher priority to files with high download volume, and the corresponding control terminal preferentially displays these file entries. When people need to analyze the download volume of a certain file, they can obtain multiple sparse feature data of the file. For example, the multiple sparse feature data of the file may include: the name of the file, the size of the file, and the label of the file. For example, in the sparse feature data of a certain file: the name of the file is F2, the size of the file is 2 megabytes, and the label of the file is technology. The above sparse feature data obtained in step 401 can be arranged according to certain rules, such as "F2, 2 megabytes, technology".

[0057] like Figure 3 As shown, the program module in the data processing device may include an acquisition module, and in step 401, the data processing device may acquire a plurality of sparse feature data through the acquisition module.

[0058] Step 402: convert multiple sparse feature data to obtain multiple dense feature data.

[0059] After acquiring multiple sparse feature data with a large amount of data, the data processing device can convert the multiple sparse feature data to obtain multiple dense feature data with a smaller amount of data, and the data amount of the dense feature data is much smaller than the data amount of the sparse feature data. The efficiency of the subsequent data processing device in processing the multiple dense feature data is higher than the efficiency of the data processing device in processing the multiple sparse feature data.

[0060] like Figure 3 As shown, the program module in the data processing device may further include a conversion module. In step 402, the data processing device may convert the multiple sparse feature data acquired by the acquisition module through the conversion module to obtain multiple dense feature data.

[0061] Step 403: automatically perform high-order nonlinear combination processing on multiple dense feature data to obtain a set of high-order data.

[0062] After acquiring multiple dense feature data, the data processing device can perform high-order nonlinear combination processing on the input multiple dense feature data to obtain a group of high-order data, each high-order data is related to the nonlinear combination of m feature data in the multiple dense feature data. Wherein, m is any integer greater than or equal to 3, that is, a group of high-order data can include: multiple high-order data related to the nonlinear combination of at least three feature data. For example, m is 5, 6 or 7, at this time, the group of high-order data includes: high-order data related to the nonlinear combination of five feature data, high-order data related to the nonlinear combination of six feature data, and high-order data related to the nonlinear combination of seven feature data.

[0063] Optionally, the program module in the data processing device may further include a nonlinear processing module, which includes: a high-order processing submodule, through which the data processing device can automatically perform high-order nonlinear combination processing on multiple dense feature data in step 403 to obtain a set of high-order data. For example, the high-order processing submodule may be: a processing submodule based on a deep neural network (DNN), a processing submodule based on a factorization-machine supported neural network (FNN), or a processing submodule based on a product-based neural network (PNN).

[0064] In the first possible implementation, Figure 3 As shown, the high-order processing submodule may be a DNN-based processing submodule, and the DNN-based processing submodule may include a plurality of cascaded first processing elements 0311. The DNN-based processing submodule includes a total of H+1 layers of first processing elements ( Figure 3Two layers of first processing elements 0311 are schematically shown in the figure. The first processing element of the first layer is connected to the conversion module 02 and the first processing element of the non-first layer; the first processing element of the i-th layer is connected to the first processing element of the i-1th layer and the first processing element of the i+1th layer; the first processing element of the H+1th layer is connected to the first processing element of the Hth level and the prediction module 05 in the data processing device for finally determining the prediction data, i is an integer greater than 1 and less than H+1, and H is an integer greater than or equal to 1. The conversion module 02 can input the multiple dense feature data to the first processing element of the first layer, and the first processing element 021 of each layer can perform nonlinear combination processing on the input data, and input the processed data to the first processing element 021 of the next layer. Each first processing element of the H+1th level can output a high-order data to the prediction module 05 after performing nonlinear combination processing on the input data, and each high-order data is related to the nonlinear combination of m feature data among the multiple dense feature data acquired by the data processing device.

[0065] It should be noted that among the multiple cascaded first processing elements, any two connected first processing elements have an edge for connecting the two first processing elements, and the edge has a preset weight; the edge between the first processing element of the H+1th layer and the prediction module and the edge between the first processing element of the first layer and the conversion module also have a preset weight. The data processed by each first processing element are: data calculated by the data input to the first processing element through the edge and the weight on the edge.

[0066] In a second possible implementation, the high-order processing submodule may be a DNN-based processing submodule, which may include multiple cascaded second processing elements. Assuming that the DNN-based processing submodule includes a total of H+1 layers of second processing elements, the high-order processing submodule automatically performs high-order nonlinear combination processing on multiple dense feature data to obtain a set of high-order data: Among them, σ(x) is the activation function, a H is a vector whose elements are the data processed by the second processing element of the Hth level, b H+1 is the offset term of the second processing element at the H+1th level, W H+1 is a matrix whose elements are the weights of the edges between the second processing element of the Hth level and the second processing element of the H+1th level. For example, the activation function may be a sigmoid function (also called an S-shaped function), a rectified linear unit (relu) function, or a hyperbolic tangent (tanh) function.

[0067] The second processing unit of the Hth layer includes m HThe second processing element, and m H The data obtained after processing by each of the second processing elements are all real numbers. The second processing elements of the H+1th layer include m H+1 The second processing element, and m H+1 The data obtained after processing by each of the second processing elements are all real numbers. The weights of the edges connecting the second processing element of the Hth layer and the second processing element of the H+1th layer include m H ×m H+1 real number. l is an integer greater than or equal to 1 and less than or equal to H, a l is a vector whose elements are the data processed by the second processing element of each layer l, b l+1 is the offset term of the second processing element in the l+1th layer, W l+1 is a vector whose elements are the weights of each edge between the second processing element of the lth layer and the second processing element of the l+1th layer, l+1 is a vector whose elements are the data processed by the second processing element of each l+1th layer, a l+1 =σ(W l+1 a l +b l+1 ).

[0068] Step 404: automatically perform low-order nonlinear combination processing on multiple dense feature data to obtain a set of low-order data.

[0069] After acquiring a plurality of dense feature data, the data processing device can also perform low-order nonlinear combination processing on the input multiple dense feature data to obtain a group of low-order data, each low-order data is related to the nonlinear combination of n feature data in the plurality of dense feature data. Wherein n is any integer greater than or equal to 2 and less than m. For example, m is 5, 6 or 7, in which case n can be 2, 3 or 4, and the group of low-order data includes: low-order data related to the nonlinear combination of two feature data, low-order data related to the nonlinear combination of three feature data, and low-order data related to the nonlinear combination of four feature data.

[0070] Step 405: Automatically perform linear combination processing on multiple dense feature data to obtain a set of first auxiliary data.

[0071] After acquiring a plurality of dense feature data, the data processing device can also perform linear combination processing on the input plurality of dense feature data to obtain a set of first auxiliary data, each of which is related to a linear combination of at least two feature data in the plurality of dense feature data. For example, a set of first auxiliary data may include: first auxiliary data related to a linear combination of two feature data, first auxiliary data related to a linear combination of three feature data, and first auxiliary data related to a linear combination of four feature data.

[0072] It should be noted that in step 404, the data processing device can automatically perform low-order nonlinear combination processing on multiple dense feature data through a low-order processing submodule in the data processing device to obtain a set of low-order data. In step 405, the data processing device can automatically perform linear combination processing on multiple dense feature data through a linear processing module in the data processing device to obtain a set of first auxiliary data.

[0073] Furthermore, Figure 4 The program module in the data processing device shown may also include a FM-based processing module or a FFM-based processing module, the FM-based processing module includes: a low-order processing submodule 032 and a linear processing module 04, and the FFM-based processing module includes: a low-order processing submodule 032 and a linear processing module 04. That is, when executing steps 404 and 405, the data processing device can directly perform low-order nonlinear combination processing and linear combination on multiple dense feature data through the FM-based processing module or the FFM-based processing module to obtain a set of low-order data and a set of first auxiliary data.

[0074] In a first possible implementation, the data processing device can directly perform low-order nonlinear combination processing and linear combination on multiple dense feature data through a FM-based processing module to obtain a set of low-order data and a set of first auxiliary data. The data obtained by the FM-based processing module can be expressed as: Should It includes a set of low-order data and a set of first auxiliary data. The set of low-order data is The first auxiliary data set is

[0075] Where n is the number of multiple dense feature data obtained by the acquisition module, V a is the implicit vector of the ath feature data in the multiple dense feature data, 1≤a≤n-1, and V a To have a specified dimension, V a Each element in V is a real number. b is the implicit vector of the bth feature data in the multiple dense feature data, a+1≤b≤n, and V b With the specified dimension, V b Each element in X is a real number. i,a is the ath feature data among the multiple dense feature data, X i,b is the bth feature data among the multiple dense feature data, X i,j is the jth feature data among the multiple dense feature data, 1≤j≤n, w j is the weight of the j-th feature data.

[0076] In a second possible implementation, the data processing device can directly perform low-order nonlinear combination processing and linear combination on multiple dense feature data through an FFM-based processing module to obtain a set of low-order data and a set of first auxiliary data. The data obtained by the FFM-based processing module can be expressed as: It includes a set of low-order data and a set of first auxiliary data. The set of low-order data is The first auxiliary data set is

[0077] Among them, the function f a represents the domain where the ath feature data in the multiple dense feature data is located, and the function f b represents the domain where the bth feature data in the multiple dense feature data is input, is the ath feature data pair f in the multiple dense feature data b The implicit vector of To have the specified dimensions, Every element in is a real number, is the bth feature data pair f in the multiple dense feature data a The implicit vector of With the specified dimensions, Every element in is a real number.

[0078] Step 406: Determine prediction data according to a plurality of target data, where the plurality of target data includes a set of low-order data, a set of high-order data, and a set of first auxiliary data.

[0079] After obtaining a set of low-order data, a set of high-order data, and a set of first auxiliary data, the data processing device can determine the predicted data according to the multiple target data including the set of low-order data, the set of high-order data, and the set of first auxiliary data. Figure 3 The program module in the data processing device may further include a prediction module 05. In step 406, the data processing device may determine prediction data through the prediction module 05. Optionally, the prediction data determined by the prediction module may be:

[0080] For example, when the multiple sparse feature data acquired in step 401 are sparse feature data of an application (including user attribute data and application attribute data), the prediction data obtained by the data processing device in step 406 can be used to characterize the click rate of the icon of the application. Figure 1As shown, server 1 (the server of the application store) can analyze the user and each application in the application store through the data processing device 0 to determine the click rate of the user on the icon of each application in the application store, and sort multiple applications according to the click rate of each application icon, and send the corresponding application store interface information to client 2 (the client of the application store). Client 2 displays the application store interface based on the information, and the icons of the applications displayed in the interface are arranged in the order determined above.

[0081] That is, server 1 will re-sort all application icons for each user, so that the sorting of all applications in the client of the application store used by each user is related to the attribute characteristics of the user, so that the sorting of applications can better meet the needs of each user. For example, if the application store includes four applications, namely application Q1, application Q2, application Q3 and application Q4, the data processing device determines that the user has the highest click rate on the icon of application Q1 after processing the sparse feature data of the application, followed by the icon of application Q2, the icon of application Q4 and the icon of application Q3. The server can then arrange the icons of these four applications in the order of the icon of application Q1, the icon of application Q2, the icon of application Q3 and the icon of application Q4, so that the icons of these four applications in the application store interface displayed by the client are arranged in the order of the icon of application Q1, the icon of application Q2, the icon of application Q3 and the icon of application Q4.

[0082] Further, in order to improve the accuracy of the prediction data obtained by the data processing device, before the data processing device executes step 401, the server can obtain multiple training data, and can train the data processing device according to the multiple data through the training device to adjust the weights used by the data processing device in the process of processing data. For example, each training data may include: a group of sparse feature data and click data corresponding to the group of sparse feature data.

[0083] Taking user P and application Q as an example, a set of sparse feature data in a piece of training data may include: attribute data of user P and attribute data of application Q, and the click data in the piece of training data may include: the identifier of user P, the identifier of application Q, and indication data for indicating whether user P clicks on the icon of application Q. Optionally, the click data in the piece of training data may also include: the time when the piece of training data is counted, the weather conditions when the piece of training data is counted, the traffic conditions when the piece of training data is counted, the network connection conditions when the piece of training data is counted, and other contextual information when the piece of training data is counted. The identifier of user P may be the identity identification number (ID) of user P, and the identifier of application Q may also be the ID of application Q.

[0084] When training the data processing device, the training device can first input multiple groups of sparse feature data in multiple training data into the acquisition module in the data processing device. The data processing device can process each group of sparse feature data input in turn to obtain a prediction data, which is used to indicate whether the user will click on the icon of the application. The training device can also compare the prediction data obtained by the data processing device based on a group of sparse feature data with the click data corresponding to the group of sparse feature data, and determine whether the prediction data and the click data both indicate that the user will click on the icon of the application, or the user will not click on the icon of the application. If the prediction data and the click data do not simultaneously indicate that the user will click on the icon of the application, or the user will not click on the icon of the application, the training device needs to adjust the weights used in the process of processing data by the data processing device. After the data processing device is trained by the training device, the weights used by the data processing device in the process of processing data have been adjusted many times, and the data processing device can process multiple sparse feature data to obtain more accurate prediction data.

[0085] In addition, after obtaining a plurality of feature data, if the number of high-order data obtained by performing a high-order nonlinear combination process on a part of the feature data in the plurality of feature data is a first number, and the number of low-order data obtained by performing a low-order nonlinear combination process on another part of the feature data in the plurality of feature data is a second number, then the number of target data based on which the prediction data is finally determined is equal to the sum of the first number and the second number. If the plurality of data is subjected to a high-order nonlinear combination process, the number of high-order data obtained is greater than the first number, and the plurality of data is also subjected to a low-order nonlinear combination process, the number of low-order data obtained is greater than the second number, and the number of target data based on which the prediction data is finally determined is also greater than the sum of the first number and the second number. That is, when the plurality of feature data is subjected to a high-order nonlinear combination process and a low-order nonlinear combination process respectively, the number of target data based on which the prediction data is finally determined will be greater, and the accuracy of the prediction data will be higher. Therefore, in the embodiment of the present invention, all dense feature data obtained in step 402 are processed in steps 403, 404 and 405, so that there are more target data based on which the predicted data is determined in step 406, and the predicted data obtained in step 406 has higher accuracy.

[0086] Optional, Figure 3The program module in the data processing device shown may also include a determination module and an experience processing module. Before step 406, the data processing device may also determine x experience feature data by combining the expert experience and the plurality of dense feature data through the determination module, and automatically process the x experience feature data through the experience processing module to obtain a set of second auxiliary data. In step 406, the prediction module in the program module in the data processing device determines that the plurality of target data based on which the prediction data is based may also include: the set of second auxiliary data. That is, the plurality of target data may include: a set of high-order data, a set of low-order data, a set of first auxiliary data, and a set of second auxiliary data.

[0087] For example, x is an integer greater than or equal to 1, and the x empirical feature data include: data obtained from multiple dense feature data in combination with expert experience, and / or data obtained by performing nonlinear combination processing on at least two feature data in multiple dense feature data in combination with expert experience. Wherein, when x=1, the second auxiliary data is related to one empirical feature data, when x>1, the second auxiliary data is related to the linear combination of at least two empirical feature data in the x empirical feature data, and / or, the second auxiliary data is related to the nonlinear combination of at least two empirical feature data in the x empirical feature data.

[0088] Optionally, the "expert" mentioned in the embodiment of the present invention refers to a technician or multiple technicians in the field of computer technology, and the "expert" knows all the general technical knowledge in the field of computer technology, and can obtain the existing technology in the field of computer technology, and has conventional experimental means. The "expert experience" mentioned in the embodiment of the present invention is the "preset formula" determined by the "expert". The determination module combines the "expert experience" and multiple dense feature data to determine x empirical feature data, including: the determination module inputs multiple dense feature data as independent variables of the "preset formula" into the "preset formula", and the result obtained by the preset formula is x empirical feature data. For example, it is assumed that the multiple dense feature data are: data a1, data a2, data a3, data a4, data a5, data a6, data a7, data a8, data a9 and data a10, a total of 10 data. The determination module can combine expert experience to obtain data a1 from multiple dense feature data as an empirical feature data; or, the determination module can combine expert experience to perform nonlinear combination processing on data a9 and data a10 in multiple dense feature data to obtain an empirical data; or, the determination module obtains data a1 and uses data a1 as an empirical feature data, and combines expert experience to perform nonlinear combination processing on data a9 and data a10 to obtain another empirical data.

[0089] After the determination module obtains x empirical feature data, the empirical processing unit can also automatically process the x empirical feature data to obtain a set of second auxiliary data. It should be noted that when x=1, the second auxiliary data is related to one empirical feature data, and when x>1, the second auxiliary data is related to the linear combination and / or nonlinear combination of at least two of the x empirical feature data. For example, when x>1, the second auxiliary data is related to the linear combination of at least two of the x empirical feature data; or the second auxiliary data is related to the nonlinear combination of at least two of the x empirical feature data; or the second auxiliary data is simultaneously related to the linear combination and nonlinear combination of the at least two empirical feature data. Optionally, the empirical processing module can be a processing module based on logarithmic probability regression (LR), a processing module based on FM, a processing module based on FFM, or a processing module based on DNN.

[0090] It should be noted that the experience processing module needs to combine expert experience when processing multiple dense feature data, but the nonlinear processing module and the linear processing module do not need to combine expert experience when processing multiple dense feature data, and can automatically process the feature data.

[0091] In addition, in order to verify the accuracy of the predicted data obtained by the data processing device provided by the embodiment of the present invention, the embodiment of the present invention uses multiple training data and multiple groups of test data, and conducts multiple experiments on the data processing device provided by the embodiment of the present invention and the data processing device in the related art. The area under the receiver operating characteristic curve (ROC) curve (Area under the ROCcurve, AUC) and the loss function (Logloss) are used to compare the data processing device in the embodiment of the present invention with the data processing device in the related art. After comparison, it is known that the accuracy of the predicted data obtained by the data processing device provided by the embodiment of the present invention is higher than the predicted data obtained by the data processing device in the related art.

[0092] For example, when it is necessary to predict the click rate of the icon of application Q, based on the above data processing method, it can be known that the data processing device can obtain multiple sparse feature data of application Q, and convert the multiple sparse feature data to obtain multiple dense feature data of application Q. Then, the data processing device automatically performs high-order nonlinear combination processing, low-order nonlinear combination processing and linear combination processing on the multiple dense feature data of application Q respectively, and obtains a group of high-order data, a group of low-order data and a group of first auxiliary data. Furthermore, the data processing device can also determine x empirical feature data in combination with expert experience and multiple dense feature data of application Q; and automatically process the x empirical feature data to obtain a group of second auxiliary data. Finally, the data processing device determines the prediction data based on multiple target data including the group of high-order data, the group of low-order data, the group of first auxiliary data and the group of second auxiliary data. The prediction data can be used to characterize the click rate of the icon of application Q.

[0093] In summary, the embodiment of the present invention provides a data processing method that can automatically perform two different nonlinear combination processing (high-order nonlinear combination processing and low-order nonlinear combination processing) on ​​feature data and obtain two groups of processed data. In the related art, a group of processed data (including at least one high-order data) can be obtained by performing a nonlinear combination processing (high-order nonlinear combination processing) on ​​feature data. That is, the amount of data of the two groups of processed data obtained in the embodiment of the present invention is greater than the amount of data of the one group of processed data obtained in the related art, and accurate prediction needs to be based on a large amount of data to be achieved. Therefore, the accuracy of the predicted data obtained based on the two groups of processed data in the embodiment of the present invention is higher than the accuracy of the predicted data obtained based on the one group of processed data in the related art.

[0094] Furthermore, the target data based on which the prediction data is determined in the embodiment of the present invention may also include: a group of first auxiliary data and a group of second auxiliary data. That is, the target data in the embodiment of the present invention may include: a group of high-order data, a group of low-order data, a group of first auxiliary data and a group of second auxiliary data, a total of four groups of data, thereby further increasing the number of target data based on which the prediction data is finally determined, and further improving the accuracy of the prediction data.

[0095] Figure 5 A schematic diagram of the structure of another data processing device provided in an embodiment of the present invention, the data processing device can be Figure 1 The data processing device shown, such as Figure 5 As shown, the data processing device 50 may include:

[0096] An acquisition unit 501 is used to acquire a plurality of feature data;

[0097] A nonlinear processing unit 502 is used to automatically perform two different nonlinear combination processing on the multiple feature data to obtain two groups of processed data, wherein the two groups of processed data include a group of high-order data and a group of low-order data, the high-order data is related to the nonlinear combination of m feature data in the multiple feature data, and the low-order data is related to the nonlinear combination of n feature data in the multiple feature data, m≥3, m>n≥2;

[0098] The prediction unit 503 is used to determine prediction data according to a plurality of target data, where the plurality of target data includes two groups of processed data.

[0099] In summary, the embodiment of the present invention provides a data processing device, and the nonlinear processing unit can automatically perform two different nonlinear combination processing (high-order nonlinear combination processing and low-order nonlinear combination processing) on ​​the feature data and obtain two groups of processed data. In the related art, a group of processed data (including at least one high-order data) can be obtained by performing a nonlinear combination processing (high-order nonlinear combination processing) on ​​the feature data. That is, the amount of data of the two groups of processed data obtained in the embodiment of the present invention is greater than the amount of data of the one group of processed data obtained in the related art, and accurate prediction needs to be based on a large amount of data to be achieved. Therefore, the accuracy of the predicted data obtained based on the two groups of processed data in the embodiment of the present invention is higher than the accuracy of the predicted data obtained based on the one group of processed data in the related art.

[0100] Optional, please continue to refer to Figure 5 , the nonlinear processing unit 502 may include:

[0101] A high-order processing subunit 5021 is used to automatically perform high-order nonlinear combination processing on multiple feature data to obtain a set of high-order data;

[0102] The low-order processing subunit 5022 is used to automatically perform low-order nonlinear combination processing on multiple feature data to obtain a set of low-order data.

[0103] Optional, Figure 6 A structural diagram of another data processing device provided by an embodiment of the present invention is as follows: Figure 6 As shown, in Figure 5 On the basis of, the data processing device 50 may further include:

[0104] The linear processing unit 504 is used to automatically perform linear combination processing on multiple feature data to obtain a set of first auxiliary data, where the first auxiliary data is related to the linear combination of at least two feature data in the multiple feature data. The multiple target data also include: at least one first auxiliary data.

[0105] Optionally, the data processing device also includes: a processing module based on a factorization machine FM or a processing module based on a domain-aware decomposition machine FFM, wherein the processing module based on FM includes: a low-order processing sub-unit and a linear processing unit, and the processing module based on FFM includes: a low-order processing sub-unit and a linear processing unit.

[0106] Optional, Figure 7 A structural diagram of another data processing device provided by an embodiment of the present invention is as follows: Figure 7 As shown, in Figure 5 On the basis of, the data processing device 50 may further include:

[0107] The determination unit 505 determines x empirical feature data in combination with the expert experience and the plurality of feature data, where x≥1, and the x empirical feature data includes: data obtained from the plurality of feature data in combination with the expert experience, and / or data obtained by performing nonlinear combination processing on at least two of the plurality of feature data in combination with the expert experience;

[0108] The experience processing unit 506 is used to automatically process x experience feature data to obtain a set of second auxiliary data. The multiple target data also include: a set of second auxiliary data, wherein, when x=1, the second auxiliary data is related to one experience feature data, when x>1, the second auxiliary data is related to a linear combination of at least two of the x experience feature data, and / or, the second auxiliary data is related to a nonlinear combination of at least two of the x experience feature data.

[0109] The multiple feature data are multiple sparse feature data. Figure 8 A structural diagram of a data processing device provided in another embodiment of the present application is shown in FIG. Figure 8 As shown, in Figure 5 On the basis of, the data processing device 50 may further include:

[0110] A conversion unit 507, used to convert a plurality of sparse feature data into a plurality of dense feature data;

[0111] The nonlinear processing unit 502 can also be used to automatically perform two different nonlinear combination processes on multiple dense feature data to obtain two groups of processed data.

[0112] In summary, the embodiment of the present invention provides a data processing device, and the nonlinear processing unit can automatically perform two different nonlinear combination processing (high-order nonlinear combination processing and low-order nonlinear combination processing) on ​​the feature data and obtain two groups of processed data. In the related art, a group of processed data (including at least one high-order data) can be obtained by performing a nonlinear combination processing (high-order nonlinear combination processing) on ​​the feature data. That is, the amount of data of the two groups of processed data obtained in the embodiment of the present invention is greater than the amount of data of the one group of processed data obtained in the related art, and accurate prediction needs to be based on a large amount of data to be achieved. Therefore, the accuracy of the predicted data obtained based on the two groups of processed data in the embodiment of the present invention is higher than the accuracy of the predicted data obtained based on the one group of processed data in the related art.

[0113] Furthermore, the target data based on which the prediction data is determined in the embodiment of the present invention may also include: a group of first auxiliary data obtained by the linear processing unit and a group of second auxiliary data obtained by the experience processing unit. That is, the target data in the embodiment of the present invention may include: a group of high-order data, a group of low-order data, a group of first auxiliary data and a group of second auxiliary data, a total of four groups of data, thereby further increasing the number of target data based on which the prediction data is finally determined, and further improving the accuracy of the prediction data.

[0114] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product, which includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server, data center, etc. that contains one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium, or a semiconductor medium (e.g., a solid-state hard disk), etc.

[0115] In the embodiments of the present invention, the term "and / or" is only a description of the association relationship of the associated objects, indicating that there may be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0116] It should be noted that the method embodiments provided in the embodiments of the present invention can be referenced with the corresponding device embodiments, and the embodiments of the present invention are not limited to this. The order of the steps of the method embodiments provided in the embodiments of the present invention can be appropriately adjusted, and the steps can be increased or decreased accordingly according to the circumstances. Any technician familiar with the technical field can easily think of the changed methods within the technical scope disclosed in the embodiments of the present invention, which should be covered within the protection scope of the embodiments of the present invention, so they will not be repeated. The serial numbers of the embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

[0117] The above description is only an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A data processing method, characterized in that: The method comprises: Acquire a plurality of feature data, wherein the plurality of feature data includes feature data of one or more things among a website, a file, an application, and a video; Automatically performing two different nonlinear combination processing on the multiple feature data to obtain two groups of processed data, wherein the two groups of processed data include a group of high-order data and a group of low-order data, the high-order data is related to the nonlinear combination of m feature data in the multiple feature data, and the low-order data is related to the nonlinear combination of n feature data in the multiple feature data, m≥3, m>n≥2; Prediction data is determined based on a plurality of target data, the plurality of target data including the two sets of processed data.

2. The method according to claim 1, characterized in that The automatically performing two different nonlinear combination processes on the plurality of feature data to obtain two groups of processed data includes: Automatically performing high-order nonlinear combination processing on the plurality of feature data to obtain the set of high-order data; Automatically perform low-order nonlinear combination processing on the plurality of feature data to obtain the set of low-order data.

3. The method according to claim 1 or 2, characterized in that: After acquiring the plurality of feature data, the method further includes: Automatically perform linear combination processing on the multiple feature data to obtain a set of first auxiliary data, where the first auxiliary data is related to a linear combination of at least two feature data in the multiple feature data, and the multiple target data also includes: the at least one first auxiliary data.

4. The method according to claim 3, characterized in that: The step of automatically performing low-order nonlinear combination processing on the plurality of feature data to obtain the set of low-order data comprises: automatically performing low-order nonlinear combination processing on the plurality of feature data through a processing module based on a factor decomposition machine FM to obtain the set of low-order data; the step of automatically performing linear combination processing on the plurality of feature data to obtain a set of first auxiliary data comprises: automatically performing linear combination processing on the plurality of feature data through a processing module based on FM to obtain the set of first auxiliary data; Alternatively, the automatically performing low-order nonlinear combination processing on the multiple feature data to obtain the set of low-order data includes: automatically performing low-order nonlinear combination processing on the multiple feature data through a processing module based on a domain-aware decomposition machine (FFM) to obtain the set of low-order data; the automatically performing linear combination processing on the multiple feature data to obtain a set of first auxiliary data includes: automatically performing linear combination processing on the multiple feature data through a processing module based on the FFM to obtain the set of first auxiliary data.

5. The method according to claim 1 or 2, characterized in that: Before determining the prediction data according to the plurality of target data, the method further includes: Determine x empirical feature data in combination with the expert experience and the plurality of feature data, where x≥1, and the x empirical feature data include: data obtained from the plurality of feature data in combination with the expert experience, and / or data obtained by performing nonlinear combination processing on at least two feature data from the plurality of feature data in combination with the expert experience; The x empirical feature data are automatically processed to obtain a set of second auxiliary data, and the multiple target data also include: the set of second auxiliary data; wherein, when x=1, the second auxiliary data is related to one empirical feature data; when x>1, the second auxiliary data is related to a linear combination of at least two of the x empirical feature data, and / or, the second auxiliary data is related to a nonlinear combination of at least two of the x empirical feature data.

6. The method according to claim 1 or 2, characterized in that: The plurality of feature data are a plurality of sparse feature data, After acquiring the plurality of feature data, the method comprises: Converting the plurality of sparse feature data into a plurality of dense feature data; The automatically performing two different nonlinear combination processes on the plurality of feature data to obtain two groups of processed data includes: The two groups of processed data are automatically obtained by performing two different nonlinear combination processes on the plurality of dense feature data.

7. A data processing device, characterized in that: The data processing device comprises: An acquisition unit, configured to acquire a plurality of feature data, wherein the plurality of feature data includes feature data of one or more of a website, a file, an application, and a video; a nonlinear processing unit, used for automatically performing two different nonlinear combination processing on the plurality of feature data to obtain two groups of processed data, wherein the two groups of processed data include a group of high-order data and a group of low-order data, the high-order data is related to a nonlinear combination of m feature data in the plurality of feature data, and the low-order data is related to a nonlinear combination of n feature data in the plurality of feature data, m≥3, m>n≥2; The prediction unit is used to determine prediction data according to a plurality of target data, wherein the plurality of target data includes the two groups of processed data.

8. The data processing device according to claim 7, characterized in that: The nonlinear processing unit comprises: A high-order processing subunit, used for automatically performing high-order nonlinear combination processing on the plurality of feature data to obtain the set of high-order data; The low-order processing subunit is used to automatically perform low-order nonlinear combination processing on the multiple feature data to obtain the set of low-order data.

9. The data processing device according to claim 7 or 8, characterized in that: The data processing device further comprises: A linear processing unit is used to automatically perform linear combination processing on the multiple feature data to obtain a set of first auxiliary data, wherein the first auxiliary data is related to the linear combination of at least two feature data among the multiple feature data, and the multiple target data also includes: the at least one first auxiliary data.

10. The data processing device according to claim 9, characterized in that: The data processing device further comprises: a processing module based on a factorization machine FM or a processing module based on a domain-aware factorization machine FFM, The processing module based on the factorization machine FM comprises: a low-order processing subunit and the linear processing unit; The processing module based on the domain-aware factorization machine (FFM) includes: a low-order processing subunit and the linear processing unit.

11. The data processing device according to claim 7 or 8, characterized in that: The data processing device further comprises: A determination unit determines x empirical feature data in combination with the expert experience and the plurality of feature data, where x≥1, and the x empirical feature data include: data obtained from the plurality of feature data in combination with the expert experience, and / or data obtained by performing nonlinear combination processing on at least two feature data in the plurality of feature data in combination with the expert experience; An experience processing unit is used to automatically process the x experience feature data to obtain a set of second auxiliary data, and the multiple target data also include: the set of second auxiliary data; wherein, when x=1, the second auxiliary data is related to one experience feature data; when x>1, the second auxiliary data is related to a linear combination of at least two of the x experience feature data, and / or, the second auxiliary data is related to a nonlinear combination of at least two of the x experience feature data.

12. The data processing device according to claim 7 or 8, characterized in that: The plurality of feature data are a plurality of sparse feature data, and the data processing device further comprises: A conversion unit, used for converting the plurality of sparse feature data into a plurality of dense feature data; The nonlinear processing unit is further used for: The two groups of processed data are automatically obtained by performing two different nonlinear combination processes on the plurality of dense feature data.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the data processing method described in any one of claims 1 to 6 is implemented.

Images