A service data processing method, device and equipment, and a readable storage medium

Abstract

The application discloses a kind of service data processing method, device and equipment and readable storage medium, method includes: obtaining the N service data of service object, obtains the predicted outer layer browsing time length corresponding to N service data respectively;Predicted outer layer browsing time length refers to after service data is put into service object, the predicted browsing time length of service object for the business interface to be clicked of service data;According to the predicted outer layer browsing time length corresponding to N service data respectively, determine the time length sorting factor corresponding to N service data respectively;According to the basic sorting factor and time length sorting factor corresponding to N service data respectively, generate the fusion sorting factor corresponding to N service data respectively, according to N fusion sorting factor N service data is sorted, and sequence service data is obtained.Using the present application, when service data is sorted, the dimension of sorting can be enriched, and then the conversion rate of service data is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to a business data processing method and device, equipment and a readable storage medium. BACKGROUND

[0002] With the rapid development of computer technology, more and more users publish or obtain information through applications on electronic devices, and users creating business data can realize the putting of business data through the application of putting business data on electronic devices.

[0003] In general, in order to improve the conversion rate (that is, the ratio between the number of users who generate conversion behaviors such as ordering and purchasing through business data and the number of users who click the business data), the application platform will sort each business data before putting the business data into a user group, and then put the sorted business data into the user group in sequence. The existing scheme mainly sorts the business data based on the estimated click rate or the estimated conversion rate of the business data, and the sorting factor is relatively single, which may cause inaccurate sorting order, thereby causing the business data put into the business object to be unable to be effectively converted, and it is difficult to improve the conversion rate of the business data. SUMMARY

[0004] The embodiments of the present application provide a business data processing method, device, equipment and readable storage medium, which can enrich the sorting dimension when sorting the business data, and thereby improve the conversion rate of the business data.

[0005] The embodiments of the present application provide a business data processing method, device, equipment and readable storage medium, which can enrich the sorting dimension when sorting the business data, and thereby improve the conversion rate of the business data.

[0006] N business data corresponding to the business object are obtained, and the predicted outer layer browsing time corresponding to the N business data is obtained. The predicted outer layer browsing time refers to the predicted browsing time of the business object for the business interface to be clicked of the business data after the business data is put into the business object.

[0007] The time sorting factor corresponding to the N business data is determined according to the predicted outer layer browsing time corresponding to the N business data.

[0008] The fusion sorting factor corresponding to the N business data is generated according to the basic sorting factor and the time sorting factor corresponding to the N business data, the N business data is sorted according to the N fusion sorting factors, and the sequence business data is obtained.

[0009] The embodiments of the present application provide another business data processing method, which comprises the following steps:

[0010] obtain a sample business data feature corresponding to the sample business data, input the sample business data feature into a sample data prediction model, and output a sample predicted outer-layer browsing time length, a sample predicted inner-layer browsing time length, and a sample predicted click rate corresponding to the sample business data through the sample data prediction model; the sample predicted outer-layer browsing time length refers to a predicted browsing time length of a sample to-be-clicked business interface of the sample business data by the sample business object after the sample business data is delivered to the sample business object; the sample predicted click rate refers to a predicted probability of a click behavior of the sample to-be-clicked business interface by the sample business object; and the sample predicted inner-layer browsing time length refers to a predicted browsing time length of a sample to-be-converted business interface of the sample business data by the sample business object after the sample business object enters the sample to-be-converted business interface through the click behavior on the sample to-be-clicked business interface;

[0011] determine a target loss value of the sample business data according to the sample predicted outer-layer browsing time length, a real outer-layer browsing time length label corresponding to the sample business data, the sample predicted click rate, a real click behavior label corresponding to the sample business data, the sample predicted inner-layer browsing time length, and a real inner-layer browsing time length label corresponding to the sample business data;

[0012] adjust the sample data prediction model according to the target loss value to obtain a data prediction model; and the data prediction model is used to determine predicted outer-layer browsing time lengths respectively corresponding to N pieces of business data for a business object; and the predicted outer-layer browsing time lengths respectively corresponding to the N pieces of business data are used to sort the N pieces of business data to obtain sequence business data.

[0013] The application embodiment provides a business data processing device, which comprises:

[0014] a data acquisition module configured to acquire N pieces of business data for a business object;

[0015] a time length acquisition module configured to acquire predicted outer-layer browsing time lengths respectively corresponding to the N pieces of business data; the predicted outer-layer browsing time length refers to a predicted browsing time length of a to-be-clicked business interface of the business data by the business object after the business data is delivered to the business object;

[0016] a sorting factor determination module configured to determine time length sorting factors respectively corresponding to the N pieces of business data according to the predicted outer-layer browsing time lengths respectively corresponding to the N pieces of business data;

[0017] a factor fusion module configured to generate fusion sorting factors respectively corresponding to the N pieces of business data according to the basic sorting factors and the time length sorting factors respectively corresponding to the N pieces of business data;

[0018] a data sorting module configured to sort the N pieces of business data according to the N fusion sorting factors to obtain sequence business data.

[0019] In one embodiment, the N business data include business data M. i where i is a positive integer;

[0020] The duration acquisition module includes:

[0021] Feature input unit, used to acquire business data M i The corresponding business data features are input into the data prediction model. The data prediction model is trained based on the sample business data, including the predicted outer browsing time, the actual outer browsing time label, the predicted click-through rate, the actual click behavior label, the predicted inner browsing time, and the actual inner browsing time label. The predicted outer browsing time refers to the predicted browsing time of the sample business object on the sample business interface after the sample business data is delivered to the sample business object. The predicted click-through rate refers to the predicted probability that the sample business object will click on the sample business interface. The predicted inner browsing time refers to the predicted browsing time of the sample business object on the sample business interface after the sample business object clicks on the sample business interface and enters the sample conversion business interface of the sample business data.

[0022] The duration determination unit is used to determine the duration of business data M based on the data prediction model and business data characteristics. i The corresponding predicted outer browsing time.

[0023] In one embodiment, the N business data include business data M. i The predicted outer browsing duration includes business data M. i The corresponding predicted outer browsing duration T i where i is a positive integer;

[0024] The sorting factor determination module includes:

[0025] The parameter acquisition unit is used to obtain the predicted outer browsing duration T. i The duration calculation parameter;

[0026] The sorting factor determination unit is used to combine the duration calculation parameter with the predicted outer browsing duration T. i Perform calculations and processing to obtain business data M. i The corresponding duration sorting factor.

[0027] In one embodiment, the N business data include business data M. i where i is a positive integer;

[0028] The factor fusion module includes:

[0029] The basic factor determination unit is used to acquire business data M.i corresponding expected conversion unit, resource, predicted click rate, and predicted conversion rate;

[0030] The base factor determination unit is further configured to determine the business data M i corresponding real-time consumption resource, and the business data M i corresponding real-time consumption resource is determined as the business data M i corresponding base ranking factor;

[0031] The factor operation unit is configured to operate the business data M i corresponding duration ranking factor, and the business data M i corresponding base ranking factor, to obtain the business data M i corresponding fusion ranking factor.

[0032] In an embodiment, the business data processing apparatus further comprises:

[0033] The data placement module is configured to obtain a business demand placement amount, and sequentially obtain K pieces of business data from the sequence business data according to the business demand placement amount;

[0034] The data placement module is further configured to place the K pieces of business data to the business object.

[0035] In an embodiment, the business data processing apparatus further comprises:

[0036] The similar data recommendation module is configured to obtain real-time feedback data of the business object respectively for the K pieces of business data within a target time period;

[0037] The similar data recommendation module is further configured to determine positive feedback business data in the K pieces of business data according to the K pieces of real-time feedback data;

[0038] The similar data recommendation module is further configured to perform similar business data recommendation processing on the business object according to the positive feedback business data.

[0039] Embodiments of the present application provide another business data processing apparatus, comprising:

[0040] The sample feature input module is configured to obtain sample service data features corresponding to sample service data, input the sample service data features into the sample data prediction model, and output, by the sample data prediction model, a sample predicted outer-layer browsing time length, a sample predicted inner-layer browsing time length, and a sample predicted click rate corresponding to the sample service data. The sample predicted outer-layer browsing time length refers to a predicted browsing time length of a sample to-be-clicked service interface of the sample service data by a sample service object after the sample service data is delivered to the sample service object. The sample predicted click rate refers to a predicted probability of a click behavior of the sample to-be-clicked service interface by the sample service object. The sample predicted inner-layer browsing time length refers to a predicted browsing time length of a sample to-be-converted service interface of the sample service data by the sample service object after the sample service object enters the sample to-be-converted service interface by performing the click behavior on the sample to-be-clicked service interface.

[0041] The loss value determination module is configured to determine a target loss value of the sample service data according to the sample predicted outer-layer browsing time length, a real outer-layer browsing time length label corresponding to the sample service data, the sample predicted click rate, a real click behavior label corresponding to the sample service data, the sample predicted inner-layer browsing time length, and a real inner-layer browsing time length label corresponding to the sample service data.

[0042] The model adjustment module is configured to adjust the sample data prediction model according to the target loss value to obtain a data prediction model. The data prediction model is configured to determine predicted outer-layer browsing time lengths corresponding to N service data respectively for a service object. The predicted outer-layer browsing time lengths corresponding to the N service data respectively are configured to sort the N service data to obtain sequence service data.

[0043] In one embodiment, the loss value determination module comprises:

[0044] The loss value determination unit is configured to determine a first time length loss value corresponding to the sample service data according to the sample predicted outer-layer browsing time length and the real outer-layer browsing time length label.

[0045] The loss value determination unit is further configured to determine a second time length loss value corresponding to the sample service data according to the sample predicted click rate, the sample predicted inner-layer browsing time length, and the real inner-layer browsing time length label.

[0046] The loss value determination unit is further configured to determine a click loss value corresponding to the sample service data according to the sample predicted click rate and the real click behavior label.

[0047] The target loss value determination unit is configured to determine the target loss value of the sample service data according to the first time length loss value, the second time length loss value, and the click loss value.

[0048] In an embodiment, the loss value determination unit is further configured to obtain a first duration loss function, and obtain a logarithmic value corresponding to the real outer browsing duration label according to the first duration loss function;

[0049] The loss value determination unit is further configured to determine a first duration error value between the logarithmic value corresponding to the real outer browsing duration label and the sample predicted outer browsing duration according to the first duration loss function;

[0050] The loss value determination unit is further configured to determine the first duration error value as the first duration loss value corresponding to the sample service data.

[0051] In an embodiment, the loss value determination unit is further configured to determine a sample predicted delivery browsing duration corresponding to the sample service data according to the sample predicted click rate and the sample predicted inner browsing duration; the sample predicted delivery browsing duration refers to a predicted browsing duration of the sample service object for the sample to-be-converted service interface under the condition that the sample predicted click rate is used as a probability condition;

[0052] The loss value determination unit is further configured to obtain a second duration loss function, and obtain a logarithmic value corresponding to the real inner browsing duration label according to the second duration loss function;

[0053] The loss value determination unit is further configured to determine a second duration error value between the logarithmic value corresponding to the real inner browsing duration label and the sample predicted delivery browsing duration according to the first duration loss function;

[0054] The loss value determination unit is further configured to determine the second duration error value as the second duration loss value corresponding to the sample service data.

[0055] In an embodiment, the loss value determination unit is further configured to obtain a click loss function, and obtain a logarithmic value corresponding to the sample predicted click rate according to the click loss function;

[0056] The loss value determination unit is further configured to determine a click error value between the logarithmic value corresponding to the sample predicted click rate and the real click behavior label according to the click loss function;

[0057] The loss value determination unit is further configured to determine the click error value as a click loss value corresponding to the sample service data.

[0058] In an embodiment, the target loss value determination unit is further configured to obtain a first operation coefficient corresponding to the first duration loss value, a second operation coefficient corresponding to the second duration loss value, and a third operation coefficient corresponding to the click loss value;

[0059] The target loss value determination unit is further configured to perform operation processing on the first time length loss value and a first operation coefficient to obtain a first operation time length loss value.

[0060] The target loss value determination unit is further configured to perform operation processing on the second time length loss value and a second operation coefficient to obtain a second operation time length loss value.

[0061] The target loss value determination unit is further configured to perform operation processing on the click loss value and a third operation coefficient to obtain an operation click loss value.

[0062] The target loss value determination unit is further configured to perform operation processing on the first operation time length loss value, the second operation time length loss value, and the operation click loss value to obtain the target loss value of the sample business data.

[0063] An embodiment of the present application provides a computer device, including a processor and a memory.

[0064] The memory stores a computer program, and the computer program is executed by the processor to enable the processor to execute the method in the embodiments of the present application.

[0065] An embodiment of the present application provides a computer readable storage medium, which stores a computer program. The computer program includes program instructions, and the program instructions are executed by a processor to execute the method in the embodiments of the present application.

[0066] In an aspect of the present application, a computer program product or a computer program is provided, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to enable the computer device to execute the method provided in an aspect of the embodiments of the present application.

[0067] In this embodiment, after obtaining N business data for a business object, the predicted outer browsing duration corresponding to each of the N business data can be obtained (i.e., the estimated browsing duration of the business object for the clickable business interface of the business data after the business data is delivered to the business object). Subsequently, through the predicted outer browsing duration corresponding to each of the N business data, the duration ranking factor corresponding to each of the N business data can be determined. Based on the basic ranking factor and duration ranking factor corresponding to each of the N business data, the fusion ranking factor corresponding to each of the N business data can be generated. Based on the N fusion ranking factors, the N business data can be sorted to obtain the sequence business data. In other words, this application introduces the dimension of predicted outer browsing duration when sorting business data. Incorporating predicted outer browsing duration as a sorting factor enriches the sorting dimensions of the business data. Furthermore, since the browsing duration of a business object significantly impacts conversion rates—for example, a longer browsing duration may indicate that the business data better aligns with the business object's preferences, increasing the likelihood of conversion—incorporating browsing duration as a sorting factor enriches the sorting options. This allows for more accurate sorting of business data based on the business object's preferences, thereby significantly increasing the probability of conversion after the sorted business data is distributed to the business object, ultimately improving the conversion rate. In summary, this application, by introducing predicted outer browsing duration when sorting business data, enriches the sorting dimensions and improves the conversion rate of business data. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0069] Figure 1 This is a network architecture diagram provided in an embodiment of this application;

[0070] Figure 2 This application provides a schematic diagram of a scenario for sorting business data.

[0071] Figure 3 This application provides a flowchart illustrating a business data processing method according to an embodiment.

[0072] Figure 4 This is a schematic diagram of a scenario for presenting a business interface provided in an embodiment of this application;

[0073] Figure 5is a flow diagram of another service data method provided by an embodiment of the present application;

[0074] Figure 6 is an architecture diagram of a model training provided by an embodiment of the present application;

[0075] Figure 7 is a structural diagram of a service data processing apparatus provided by an embodiment of the present application;

[0076] Figure 8 is a structural diagram of a service data processing apparatus provided by an embodiment of the present application;

[0077] Figure 9 is a structural diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0078] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0079] Please refer to Figure 1 , Figure 1 is a network architecture diagram provided by an embodiment of the present application. As shown in Figure 1 , the network architecture can include a service server 1000 and a terminal device cluster, and the terminal device cluster can include one or more terminal devices, which will not be limited in number here. As shown in Figure 1 , the plurality of terminal devices can include terminal device 100a, terminal device 100b, terminal device 100c,..., and terminal device 100n; as shown in Figure 1 , the terminal device 100a, the terminal device 100b, the terminal device 100c,..., and the terminal device 100n can be respectively connected with the service server 1000 in a network, so that each terminal device can perform data interaction between the service server 1000 through the network connection.

[0080] As shown in Figure 1 , each terminal device can be installed with a target application, and when the target application runs in each terminal device, it can be respectively connected with Figure 1The business servers 1000 shown interact with each other to enable the business servers 1000 to receive business data from each terminal device. The target application can include an application that has the function of displaying data information such as text, images, audio, and video. For example, the application can be an application that supports business data placement, such as a multimedia application (e.g., a video application), an entertainment application (e.g., a game application), a social application, and the like. It should be understood that the business data in this application can be media data (e.g., advertising data), and the following will be described by taking advertising data as an example. The business server 1000 in this application can obtain advertising data created by a business creation user from the business creation object (the business creation object can refer to an application account of the user creating the advertising data in the application; the business creation object can also refer to a terminal device corresponding to the user creating the advertising data; the user creating the advertising data can also be referred to as a business creation user or an advertiser) according to the application. The business server 1000 in this application can sort the business data to obtain sequence business data, and then the business server 1000 can obtain one or more pieces of business data from the sequence business data in sequence and place them in a business object (which can refer to an application account of a user group in an application).

[0081] It should be understood that, taking advertising data as an example, for a certain piece of advertising data, there will be a to-be-clicked advertising interface and a to-be-converted advertising interface, and the to-be-converted advertising interface can refer to a landing page (which can be understood as a detail page of an article / good indicated by the advertising, which can provide a good conversion function), and the business object can place an order, purchase, or download the article through the to-be-converted advertising interface, that is, the business object can generate a conversion behavior through the to-be-converted advertising interface. When the advertising data is placed in a certain business object, the business object can enter the to-be-converted advertising interface through the triggering operation (by clicking the to-be-clicked advertising interface) of the to-be-clicked advertising interface, and then convert the advertising article through the to-be-converted advertising interface. In order to make the sorting of the business data more accurate, so as to more accurately find business data that is more in line with the preferences of the business object, and thus improve the conversion rate of the business object to the business data, the application can introduce the predicted outer browsing time of the business object to the business data (the predicted browsing time of the business object to the to-be-clicked business interface of the business data after the business data is placed in the business object, which is the estimated stay time of the to-be-clicked business interface) into the sorting of the business data.

[0082] Taking a business object as an example, assuming that the quantity of business data expected to be put into the business object is N (N is a positive integer), the business server 1000 in the present application can obtain the predicted outer layer browsing time length corresponding to each of the N pieces of business data; subsequently, the business server 1000 can determine the time length sorting factor corresponding to each of the N pieces of business data according to the predicted outer layer browsing time length corresponding to each of the N pieces of business data (one piece of business data can correspond to one predicted outer layer browsing time length, and the time length sorting factor corresponding to the piece of business data can be determined according to the predicted outer layer browsing time length of the piece of business data). Subsequently, the business server 1000 can obtain the basic sorting factor corresponding to each of the N pieces of business data, wherein the basic sorting factor can refer to the real-time put-in consumption resource corresponding to the piece of business data, which can refer to the resource that needs to be paid by the business creator (for example, the piece of business data is advertising data, and the business creator can refer to the advertiser) to the business put-in platform (that is, the platform for putting in the piece of business data, such as the target application described above) when putting in the piece of business data, that is, the resource consumed by the business creator when putting in the piece of business data. The real-time put-in consumption resource can be determined according to the expected conversion unit put-in resource, the predicted click rate, and the predicted conversion rate. The expected conversion unit put-in resource can be understood as the resource consumption of the estimated unit conversion number when the business creator puts in the piece of business data, for example, for advertising data, the expected conversion unit put-in resource can refer to the resource consumed by the advertiser when expecting one conversion. The click rate can refer to the ratio between the number of users who click on the piece of business data and the total put-in quantity (also referred to as the total exposure quantity) of the piece of business data, and the predicted click rate can refer to the estimated ratio before putting in the piece of business data. For example, if a piece of business data is put into 10 users, the total exposure quantity can be 10, and it is estimated that 6 people will have a click behavior, so the predicted click rate can be the ratio between 6 and 10: 0.6 (the representation can also be in the form of a percentage such as 60%, a fraction such as 6 / 10, etc.). The conversion rate can refer to the ratio between the number of users who have a conversion behavior on the piece of business data and the total click quantity (the number of users who have a click behavior on the piece of business data) of the piece of business data, and the predicted conversion rate can refer to the estimated ratio before putting in the piece of business data. For example, if a piece of business data is put into 10 users, it is estimated that 6 people will have a click behavior, and among the 6 people, it is estimated that 3 people will have a conversion behavior, so the predicted conversion rate can be the ratio between 3 and 6: 0.5 (the representation can also be in the form of a percentage such as 50%, a fraction such as 5 / 10, etc.).

[0083] Further, the business server 1000 can generate a fusion ranking factor corresponding to each of the N pieces of business data according to the basic ranking factor corresponding to each of the N pieces of business data and the time length ranking factor corresponding to each of the N pieces of business data. That is, for a certain piece of business data, a fusion ranking factor corresponding to the piece of business data can be determined according to the basic ranking factor corresponding to the piece of business data and the time length ranking factor corresponding to the piece of business data. Subsequently, the business server 1000 can sort the N pieces of business data according to the fusion ranking factors corresponding to the N pieces of business data to obtain a sequence of the N pieces of business data. Subsequently, the business server 1000 can select a target piece of business data (for example, a target piece of advertising data) from the sequence of the N pieces of business data and deliver the target piece of business data to a business object.

[0084] It should be understood that the browsing time length of a business object for a piece of business data can also reflect the interest degree of the business object for the piece of business data. For example, the longer the browsing time length of the business object for the piece of business data, the higher the interest degree of the business object for the piece of business data, and the greater the probability of conversion of the business object for the piece of business data. The application can predict the outer browsing time length of the business object for the piece of business data through a data prediction model, and determine a time length ranking factor according to the outer browsing time length to sort the pieces of business data, so that the arrangement order of the pieces of business data is more accurate. For example, in the order of descending ranking factors, the pieces of business data that are more in line with the preferences of the business object are arranged in the front after sorting by introducing the browsing time length. When delivering the pieces of business data, the pieces of business data arranged in the front can be preferentially delivered to the business object, because these pieces of business data are the pieces of business data that the business object is more interested in. When these pieces of business data are delivered to the business object, the business object is likely to click and watch and generate a conversion behavior, so that the conversion rate of the pieces of business data can be improved.

[0085] The embodiments of the application can select one terminal device as a target terminal device (the target terminal device can be a terminal device corresponding to a business creation user, or a terminal device corresponding to a business delivery platform) from a plurality of terminal devices. The terminal device can include a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart television, a smart speaker, a desktop computer, a smart watch, a vehicle-mounted device, and the like, which are smart terminals carrying multimedia data processing functions (for example, video data playing function, music data playing function), but are not limited thereto. For example, the terminal device 100a shown in FIG. 1 can be selected as the target terminal device, and the target application can be integrated into the target terminal device. At this time, the target terminal device can perform data interaction between the target application and the business server 1000. Figure 1

[0086] ​It can be understood that the method provided by the embodiments of the present application can be executed by a computer device, including but not limited to a terminal device or a service server. The service server can be a stand-alone physical server, a server cluster or a distributed system formed by multiple physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, and basic cloud computing services such as big data and artificial intelligence platforms.

[0087] The terminal device and the service server can be directly or indirectly connected through wired or wireless communication, which is not limited in the present application.

[0088] Optionally, it can be understood that the above computer device (such as the above service server 1000, terminal device 100a, terminal device 100b, etc.) can be a node in a distributed system, and the distributed system can be a blockchain system. The blockchain system can be a distributed system formed by connecting multiple nodes through network communication. The nodes can form a peer-to-peer (P2P) network, and the P2P protocol is an application layer protocol running on the transmission control protocol (TCP) protocol. In the distributed system, any form of computer device, such as a service server, a terminal device, or an electronic device, can become a node in the blockchain system by joining the peer-to-peer network. For ease of understanding, the concept of blockchain will be described below: blockchain is a new application mode of distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithm, mainly used for organizing data in chronological order and encrypting it into a ledger, making it tamper-proof and counterfeit-proof, while allowing data verification, storage, and updates. When a computer device is a blockchain node, due to the tamper-proof and counterfeit-proof properties of the blockchain, the data in the present application (such as the predicted outer browsing time and the predicted click rate for business data) can be made authentic and secure, so that the results obtained after processing the related business data based on these data are more reliable.

[0089] For ease of understanding, please refer to Figure 2 , Figure 2 The embodiments of the present application provide a scenario for sorting business data. As shown in Figure 2 , the terminal device 100a can be any terminal device in the embodiments corresponding to Figure 1 .

[0090] As shown in Figure 2The illustrated scenario is an example of a delivery scenario for delivering N pieces of advertising data (such as Figure 2 the illustrated advertising data set 200) to the business object a, where the advertising data set 200 can include business data 1, business data 2, business data 3, and business data 4, and each piece of advertising data in the advertising data set is data waiting to be delivered to the business object a, so each piece of advertising data can also be referred to as to-be-delivered advertising data.

[0091] As shown in Figure 2 , for the business data 1 of the business object a, the predicted outer-layer browsing duration of the business object a for the business data 1 (such as the predicted outer-layer browsing duration 1 shown in Figure 2 ) can be obtained, and the duration ordering factor corresponding to the business data 1 (such as the duration ordering factor 1 shown in Figure 2 ) can be determined according to the predicted outer-layer browsing duration 1; similarly, for the business data 2 of the business object a, the predicted outer-layer browsing duration of the business object a for the business data 2 (such as the predicted outer-layer browsing duration 2 shown in Figure 2 ) can be obtained, and the duration ordering factor corresponding to the business data 2 (such as the duration ordering factor 2 shown in Figure 2 ) can be determined according to the predicted outer-layer browsing duration 2. Similarly, the duration ordering factor 3 corresponding to the business data 3 and the duration ordering factor 4 corresponding to the business data 4 can be determined.

[0092] Further, the basic ordering factor corresponding to the business data 1 (for example, the basic ordering factor 1) can be obtained, and the fusion ordering factor corresponding to the business data 1 (such as the fusion ordering factor 1) can be generated according to the basic ordering factor 1 and the duration ordering factor 1 corresponding to the business data 1; similarly, the basic ordering factor corresponding to the business data 2 (for example, the basic ordering factor 2) can also be obtained, and the fusion ordering factor corresponding to the business data 2 (such as the fusion ordering factor 2) can be generated according to the basic ordering factor 2 and the duration ordering factor 2 corresponding to the business data 2. Similarly, the fusion ordering factor corresponding to the business data 3 (such as the fusion ordering factor 3) can be generated according to the basic ordering factor (such as the basic ordering factor 3) and the duration ordering factor 3 corresponding to the business data 3, and the fusion ordering factor corresponding to the business data 4 (such as the fusion ordering factor 4) can be generated according to the basic ordering factor (such as the basic ordering factor 4) and the duration ordering factor 4 corresponding to the business data 4. According to the fusion ordering factor 1, the fusion ordering factor 2, the fusion ordering factor 3, and the fusion ordering factor 4, the business data 1, the business data 2, the business data 3, and the business data 4 can be sorted to obtain the sequence business data {business data 1, business data 3, business data 2, business data 4} as shown in Figure 3 .

[0093] It can be understood that target service data can be selected from the sequence service data {service data 1, service data 3, service data 2, service data 4}, for example, the top 2 service data (i.e. service data 1 and service data 3) can be selected as target service data, and the target service data 1 and the target service data 3 can be put into the service object.

[0094] It should be understood that the browsing duration also reflects the degree of interest of the user in the service data, and then the embodiment of the application introduces the predicted outer layer browsing duration into the sorting of the service data, which can enrich the sorting dimension of the service data, can more accurately sort the service data based on the degree of interest of the service object in the service data, and then can select the service data that the service object is more interested in by optimization after sorting and put it into the service object, because the service data meets the preference of the service object, so the probability of the service object generating a conversion behavior can be improved, and the conversion rate of the service data can be improved.

[0095] Further, please refer to Figure 3 , Figure 1 The embodiment of the application provides a flowchart of a service data processing method. The service data processing method can be executed by a computer device, and the computer device can be a business server (such as the business server in the above Figure 1 corresponding embodiment), or can be any terminal device in a terminal device cluster (such as any terminal device in the terminal device cluster in the above Figure 3 corresponding embodiment). As shown in Figure 4 , the flow of the service data processing method can include at least the following steps S101-S103:

[0096] Step S101, obtaining N service data for a service object, and obtaining a predicted outer layer browsing duration corresponding to each of the N service data; the predicted outer layer browsing duration refers to the predicted browsing duration of a service object for a to-be-clicked service interface of the service data after the service data is put into the service object.

[0097] In the present application, the service data can refer to media data (such as advertising data), the delivery can refer to exposure, and the delivery of the service data to the service object is to expose the media data to the service object (or push to the service object). The service object can refer to the binding account of the service user running the target application (such as entertainment application, social application, video application, etc.) in the target application using the terminal device. The service user can log in to the target application using the binding account, and the target application can also determine whether the service user is logged in, obtain the relevant behavior data of the service user in the target application, etc. through the binding account. It should be understood that, taking the advertising data as an example, the above-mentioned target application such as entertainment application, social application, video application, etc. can be used as an advertising delivery platform, and the advertiser (also referred to as service creation object) creating the advertising data can deliver the advertising data to the advertising delivery platform. When delivering the advertising data, the service user group can be selected for targeted delivery, and different types of advertising data can be delivered to different service user groups in the target application, that is, different types of advertising data can be delivered to different service object sets (binding accounts corresponding to service user groups) in the target application.

[0098] It can be understood that before delivering the service data to a certain service object, N candidate service data (also referred to as a set of to-be-delivered service data) for the service object can be sorted first, and then the target service data is selected from the sequence service data in sequence and delivered to the service object. In the present application, the sorting of N service data can introduce a predicted outer layer browsing time to sort. The predicted outer layer browsing time can refer to the predicted browsing time of the to-be-clicked service interface of the service data by the service object after the service data is delivered to the service object.

[0099] It can be understood that for a certain service data (such as advertising data), there will be a to-be-clicked service interface and a to-be-converted service interface, and the to-be-converted service interface can refer to a landing page (which can be understood as a detail page of a promoted item (or commodity) indicated by the service data, which can provide an item conversion function). The service object can place an order, purchase, or download the item through the to-be-converted service interface, that is, the service object can generate a conversion behavior through the to-be-converted service interface. After the service data is put into a certain service object, the service object can enter the to-be-converted service interface through the triggering operation of the to-be-clicked service interface (by clicking on the to-be-converted service interface), and then convert the promoted item through the to-be-converted service interface. For example, taking the service data as advertising data as an example, after the advertising data is put into the service object, the to-be-clicked advertising interface is displayed first, and the service object can view the item to be promoted in the to-be-clicked interface. If the service object expects to purchase or download the item promoted in the advertising data, the service object can enter the to-be-converted advertising interface (which can be understood as an item detail page interface) through the triggering operation of the to-be-clicked interface. The service object can view the detail information of the item (such as the composition of the item, the attribute information of the item, the manufacturer of the item, etc.) in the to-be-converted advertising interface, and the service object can also purchase the item through the to-be-converted advertising interface (that is, generate a conversion behavior).

[0100] To facilitate understanding of the to-be-clicked service interface and the to-be-converted service interface, please refer to Figure 4 , Figure 4 is a scene schematic diagram provided by an embodiment of the present application. As Figure 4 indicated, taking the service data as advertising data as an example, when the advertising data containing the promoted item is put into the service object a, the interface presented to the service object a in advance can be the to-be-clicked advertising interface 4001, which can include the introduction of the promoted item (for example, the brand of the promoted item is brand A, and the promoted item is a Double Seventh limited gift box belonging to brand A). At the same time, as Figure 4As shown, the clickable advertisement interface 4001 may include an item display area P, which can display related items included in the promoted item. For example, Brand A's Qixi Festival limited edition gift box may include item 40a (such as foundation) and item 40b (such as face cream), and the item display area P can display icons for item 40a and item 40b. It can be understood that the clickable advertisement interface indicates the specific item being promoted by the advertisement data. The clickable advertisement interface 4001 provides a details view control, which allows business object A to view the details of the Qixi Festival limited edition gift box. In other words, business object A can enter the conversion-oriented advertisement interface through the details view control and view the details of the Qixi Festival limited edition gift box within the conversion-oriented advertisement interface.

[0101] For example, such as Figure 4 As shown, when business object A triggers the view details control, the terminal device can respond to this trigger and display the conversion-ready advertising interface 4002 (i.e., the item details interface, also known as the landing page) in the target application. This conversion-ready advertising interface 4002 can display details of all items included in the Qixi Festival limited edition gift box (e.g., item details for items 40a and 40b respectively). The business object can view the details of each item through the conversion-ready advertising interface 4002; simultaneously, as... Figure 4 As shown, the advertising interface 4002 to be converted can provide purchase controls (such as...). Figure 5 The "Buy Now" control and the "Exit" control are shown. Business object A can purchase an item using the "Buy Now" control or exit the pending conversion advertisement interface 4002 using the "Exit" control. It can be understood that if business object A purchases an item, it can be interpreted as business object A converting the advertisement data.

[0102] It can be known from the above that the service data can include a to-be-clicked service interface and a to-be-converted service interface, the service object can enter the to-be-converted service interface through a clicking behavior on the to-be-clicked service interface, and the predicted outer browsing time length in the present application can refer to an estimated value of a stay time length of the service object on the to-be-clicked service interface after the service data is put to the service object. For example, after the service data is put to the service object, the to-be-clicked service interface can be displayed first, the service object can browse the to-be-clicked service interface first, and then click to enter the to-be-converted service interface after browsing. Of course, the service object can also exit the to-be-clicked service interface after browsing and not click. Of course, the service object can also not browse the to-be-clicked service interface. That is, after the service data is put to the service object, the service object will have a browsing time length (which can be 0 or a value greater than 0) on the to-be-clicked service interface, and the present application can estimate the browsing time length of the service object on the to-be-clicked service interface before the service data is put, which can be referred to as a predicted outer browsing time length.

[0103] The estimation of the outer browsing time length of the service data can be realized by a data prediction model, that is, the predicted outer browsing time length corresponding to the service data can be determined by the data prediction model, so as to determine the predicted outer browsing time length of the N service data of the service object including the service data M i The specific implementation mode of the service data M i The corresponding service data features can be input into the data prediction model. The data prediction model is trained based on a sample predicted outer browsing time length, a real outer browsing time length label, a sample predicted click rate, a real click behavior label, a sample predicted inner browsing time length, and a real inner browsing time length label corresponding to sample service data. The sample predicted outer browsing time length refers to a predicted browsing time length of a sample to-be-clicked service interface of a sample service data by a sample service object after the sample service data is put to the sample service object. The sample predicted click rate refers to a predicted probability of a clicking behavior of the sample service object on the sample to-be-clicked service interface. The sample predicted inner browsing time length refers to a predicted browsing time length of a sample to-be-converted service interface of the sample service data by the sample service object after the sample service object enters the sample to-be-converted service interface through a clicking behavior on the sample to-be-clicked service interface. Then, the data prediction model and the service data features can be used to determine the predicted outer browsing time length of the service data M i .

[0104] The real outer-layer browsing duration label can be understood as the browsing duration of the sample business object on the sample to-be-clicked business interface after the sample business data is put into the business object. The sample prediction click rate can be understood as the estimated value of the probability of the sample business object clicking the sample to-be-clicked business interface. For example, for sample business data, the probability of the sample business object clicking the sample to-be-clicked business interface is estimated to be 0.7, and the probability of not clicking is 0.3. Therefore, the sample prediction click rate of the sample business object for the sample business data is 0.7. The real click behavior label can be understood as whether the sample business object actually clicks the sample to-be-clicked business interface. If the sample business object actually clicks the sample to-be-clicked business interface, the real click behavior label can be a label value indicating that the click behavior occurs (which can be referred to as a valid label value, for example, a numerical value 1). If the sample business object does not actually click the sample to-be-clicked business interface, the real click behavior label can be a label value indicating that the click behavior does not occur (which can be referred to as an invalid label value, for example, a numerical value 0).

[0105] As can be seen from the above, the business data can include a to-be-clicked business interface and a to-be-converted business interface. Therefore, the sample business data can also include a sample to-be-clicked business interface and a sample to-be-converted business interface. The sample business object can enter the sample to-be-converted business interface through the click behavior on the sample to-be-clicked business interface. The sample prediction inner-layer browsing duration in the present application can be understood as an estimated value of the stay duration of the sample business object on the to-be-converted business interface of the sample business data. For example, after the sample business data is put into the sample business object, the sample to-be-clicked business interface can be displayed first. The sample business object can browse the sample to-be-clicked business interface first and click the sample to-be-clicked business interface to enter the sample to-be-converted business interface. After the sample business object enters the sample to-be-converted business interface, the sample business object can also browse the sample to-be-converted business interface and decide whether to convert the promoted item. That is, after the sample business data is put into the business object, the sample business object will also have a browsing duration on the sample to-be-converted business interface after entering the sample to-be-converted business interface. The present application can estimate the browsing duration of the sample business object on the sample to-be-converted business interface after entering the sample to-be-converted business interface before the sample business data is put in. The estimated browsing duration can be referred to as the sample prediction inner-layer browsing duration.

[0106] It should be understood that, in order to improve the prediction accuracy of the data prediction model for the outer browsing duration, the application can train the initial data prediction model using sample business data, and when training the model, not only the sample predicted outer browsing duration corresponding to the sample business data and the real outer browsing duration label are used to train the model, but also the sample predicted outer browsing duration, the real outer browsing duration label, the sample predicted click rate, the real click behavior label, the sample predicted inner browsing duration and the real inner browsing duration label are introduced, and the initial data prediction model is trained through a multi-target training task to obtain a data prediction model. Through the trained data prediction model, the predicted outer browsing duration corresponding to each business data can be accurately estimated. The specific implementation of training the model based on the sample predicted outer browsing duration, the real outer browsing duration label, the sample predicted click rate, the real click behavior label, the sample predicted inner browsing duration and the real inner browsing duration label to obtain the data prediction model can be referred to the description of the corresponding embodiments in the subsequent Figure 5 .

[0107] Step S102, determining the duration ranking factor corresponding to each of the N business data according to the predicted outer browsing duration corresponding to each of the N business data.

[0108] In the application, the duration ranking factor corresponding to each business data can be determined according to the predicted outer browsing duration corresponding to each business data. Taking N business data including business data M i , the predicted outer browsing duration including the predicted outer browsing duration T i corresponding to the business data M i (i is a positive integer) as an example, the specific implementation of determining the duration ranking factor corresponding to the business data M i according to the predicted outer browsing duration T i corresponding to the business data M i may be: the duration operation parameter corresponding to the predicted outer browsing duration T i may be obtained; then, the duration operation parameter and the predicted outer browsing duration T i are operated to obtain the duration ranking factor corresponding to the business data M i .

[0109] For ease of understanding, please refer to formula (1) together, which can be used to represent the specific implementation of determining the duration ranking factor according to the predicted outer browsing duration:

[0110] duration quality =duration outer *γ formula (1)

[0111] wherein, duration quality may be used to represent the time length ordering factor corresponding to the certain service data, duration outer may be used to represent the predicted outer-layer browsing time length corresponding to the service data; γ may be used to represent a hyperparameter, and a default value of γ may be set as 0.1; when the predicted outer-layer browsing time length is greater, duration quality will also be greater, thereby enabling the service object to have a higher competitiveness (a higher ranking, a greater opportunity to be put into the service object) in the ranking of the service data when the service object stays longer in the service data.

[0112] In step S103, a fusion ranking factor corresponding to each of the N service data is generated according to the basic ranking factor and the time length ordering factor corresponding to each of the N service data, and the N service data are ranked according to the N fusion ranking factors to obtain the sequence service data.

[0113] In the present application, after the time length ordering factor corresponding to the service data is determined, the basic ranking factor corresponding to each service data can be obtained, and then the fusion ranking factor corresponding to each service data is determined according to the basic ranking factor and the time length ordering factor corresponding to each service data, so that the fusion ranking factor corresponding to each of the N service data (i.e., the N fusion ranking factors) can be obtained; subsequently, the N service data can be ranked according to the N fusion ranking factors to obtain the sequence service data. Taking the N service data including service data M i (i is a positive integer) as an example, the fusion ranking factor corresponding to service data M i may be determined according to the basic ranking factor and the time length ordering factor corresponding to service data M i The specific implementation manner of determining the fusion ranking factor corresponding to service data M i may include the following steps: the expected conversion unit delivery resource, the predicted click rate, and the predicted conversion rate corresponding to service data M i may be obtained; then, the real-time delivery consumption resource corresponding to service data M i may be determined according to the expected conversion unit delivery resource, the predicted click rate, and the predicted conversion rate; the real-time delivery consumption resource corresponding to service data M i may be used as the basic ranking factor corresponding to service data M i It should be understood that after the basic ranking factor corresponding to service data M i is determined, the time length ordering factor corresponding to service data M i may be operated with the basic ranking factor corresponding to service data M i , so that the fusion ranking factor corresponding to service data M

[0114] It can be understood that the expected conversion unit resource can be understood as the resource consumption expected by the business creation object when putting the business data (for example, the advertiser can be understood as the cost of an expected conversion, that is, the price bid by the advertiser, generally refers to the expected cost price of the advertiser for a conversion, that is, the cost of a conversion decided by the advertiser). The click rate can refer to the ratio between the number of users who click on the business data and the total amount of business data (also known as total exposure amount), and the predicted click rate can refer to the estimated ratio before the business data is put, and the predicted click rate for a single business object can refer to the estimated probability of the business object clicking on the business data. The conversion rate can refer to the ratio between the number of users who convert the business data and the total click amount (the number of users who click on the business data), and the predicted conversion rate can refer to the estimated ratio before the business data is put, and the predicted conversion rate for a single business object can refer to the estimated probability of the business object converting the business data. Taking the advertisement data as an example, by calculating the unit conversion unit resource, the predicted click rate and the predicted conversion rate, the real-time cost of the advertiser after multiple exposures (put) of the advertisement data can be obtained, which can also be called real-time put consumption resource. In the advertisement business, the real-time put consumption resource is mainly used for sorting the advertisement data. For ease of understanding, please refer to formula (2) for details:

[0115] eCPM = Bid x pCTR x pCVR Formula (2)

[0116] Wherein, eCPM in formula (2) can refer to the actual multiple exposure billing of a certain business data (such as per thousand exposure billing, that is, real-time put consumption resource); Bid can refer to the expected conversion unit resource of the business creation object (such as the advertiser) for the business data; pCTR can refer to the predicted click rate of the business data, and pCVR can refer to the predicted conversion rate of the business data. Generally, the real-time put consumption resource can be used as the basis for sorting the advertisement data.

[0117] From the above, after determining the corresponding basic sorting factor of the business data, the corresponding basic sorting factor of a certain business data can be calculated with the duration sorting factor to obtain the corresponding fusion sorting factor of the business data. For ease of understanding, please refer to formula (3) for details:

[0118] Ranking_score = eCPM + duration qualityFormula (3)

[0119] wherein the Ranking_score in Formula (3) can be used to represent a fusion ranking factor corresponding to the service data; the eCPM can be used to represent a basic ranking factor corresponding to the service data; the duration quality can be used to represent a time length ranking factor corresponding to the service data. It can be understood that the Ranking_score can also be referred to as a ranking score. In the advertising service, the ranking of each advertising data in the advertising queue can be determined according to the ranking score of the advertising data, wherein the higher the ranking score is, the higher the ranking in the advertising queue is, and the more opportunities the advertising data has to be put into the service object.

[0120] It can be understood that in the embodiments of the present application, when the basic ranking factor corresponding to the service data is determined, the predicted click rate corresponding to the service data can be determined by the above data prediction model. Since the above data prediction model is a multi-target task training model, it can also output the predicted click rate corresponding to the service data.

[0121] Optionally, it can be understood that after the N service data are ranked according to the N fusion ranking factors to obtain the sequence service data, the service demand delivery amount can be obtained, and the K service data can be obtained from the sequence service data in order according to the service demand delivery amount; then, the K service data can be put into the service object.

[0122] Optionally, it can be understood that the application can also obtain real-time feedback data (which can include positive feedback data and negative feedback data) of the business object to each business data after putting the K business data into the business object, and perform similar business data recommendation processing to the business object based on the real-time feedback data. The specific implementation mode can be: the real-time feedback data of the business object to the K business data can be obtained in the target time period; then, the positive feedback business data can be determined from the K business data according to the K real-time feedback data; and the similar business data recommendation processing can be performed to the business object according to the positive feedback business data. For example, the K business data includes business data a1, business data a2 and business data a3, after putting the business data a1, the business data a2 and the business data a3 into the business object, the real-time feedback data of the business object to the business data a3 and the business data a2 is positive feedback business data (such as positive feedback such as conversion, good comment on the business data, or sharing of the business data, etc.), and the real-time feedback data of the business object to the business data a1 is negative feedback business data (negative feedback such as short browsing time, and complaining about the business data, etc.). The business data a3 and the business data a2 can be regarded as positive feedback business data, the goods promoted by the business data a3 and the business data a2 can be obtained, and the similar goods (which can be called similar goods) can be obtained, the business data to which the similar goods belong can be regarded as similar business data, and the similar business data can be put into the business object. For example, after putting the K business data, if the target business object generates a conversion behavior to the business data A containing the business goods "whitening mask", the business data A can be regarded as positive feedback business data, which can indicate that the business object has interest in the goods "whitening mask", and the business object can be put into the business data containing similar goods (such as goods of the same type, such as whitening essence, whitening cream, etc.) according to the preference of the business object.

[0123] In the embodiment of the present application, after obtaining the N pieces of service data corresponding to the service object, the predicted outer layer browsing time corresponding to each of the N pieces of service data can be obtained (i.e. the estimated value of the browsing time of the service object for the service interface to be clicked on the service data after the service data is put into the service object). Then, the time sorting factor corresponding to each of the N pieces of service data can be determined through the predicted outer layer browsing time corresponding to each of the N pieces of service data. According to the basic sorting factor and the time sorting factor corresponding to each of the N pieces of service data, the fusion sorting factor corresponding to each of the N pieces of service data can be generated, and the N pieces of service data can be sorted according to the N fusion sorting factors to obtain the sequence service data. That is, when the service data is sorted, the dimension of the predicted outer layer browsing time is introduced, and the predicted outer layer browsing time is taken as a sorting factor to participate in the sorting of the service data, which can enrich the sorting dimension of the service data. At the same time, since the browsing time of the service object also has a great influence on whether the conversion occurs, for example, when the browsing time is longer, it can represent that the service data is more in line with the preferences of the service object and is more likely to occur conversion, so taking the browsing time as a sorting factor to participate in the sorting can enrich the sorting factors, and the service data can be sorted more accurately based on the preferences of the service object, and then the probability of the service object generating a conversion behavior can be improved after the sorted service data is put into the service object, thereby the conversion rate of the service data can be improved. In summary, the predicted outer layer browsing time can be introduced when the service data is sorted, which can enrich the sorting dimension and further improve the conversion rate of the service data.

[0124] Further, please refer to Figure 5 , Figure 3 is another flowchart of a service data method provided by the embodiment of the present application. The flow can correspond to the embodiment of the above Figure 5 corresponding embodiment, the sample data prediction model (also referred to as the initial data prediction model) is trained to obtain the data prediction model. As shown in Figure 6 , the flow can at least include the following steps S201-S203:

[0125] In step S201, sample service data features corresponding to sample service data are obtained, the sample service data features are input into a sample data prediction model, and sample predicted outer layer browsing time, sample predicted inner layer browsing time and sample predicted click rate corresponding to the sample service data are output by the sample data prediction model. The sample predicted outer layer browsing time refers to the predicted browsing time of a sample to-be-clicked service interface of the sample service data by a sample service object after the sample service data is put into the sample service object. The sample predicted click rate refers to the predicted probability of a click behavior of the sample to-be-clicked service interface by the sample service object. The sample predicted inner layer browsing time refers to the predicted browsing time of a sample to-be-converted service interface of the sample service data by the sample service object after the sample service object enters the sample to-be-converted service interface by performing a click behavior on the sample to-be-clicked service interface.

[0126] Specifically, the sample service data features of the sample service data can include service object dimension features (i.e., features of a user to be put, which can include user age, user gender, user interest and the like), service data dimension features (which can include service type, service number, item features included in the service data and the like), and service data environment state features (which can include time of user access to the service data, network state features and the like). After the sample service data features corresponding to the sample service data are input into the sample data prediction model, the sample predicted outer layer browsing time, the sample predicted inner layer browsing time and the sample predicted click rate corresponding to the sample service data can be output by the sample data prediction model.

[0127] In step S202, a target loss value of the sample service data is determined according to the sample predicted outer layer browsing time, a real outer layer browsing time label corresponding to the sample service data, the sample predicted click rate, a real click behavior label corresponding to the sample service data, the sample predicted inner layer browsing time and a real inner layer browsing time label corresponding to the sample service data.

[0128] Specifically, the specific implementation manner of determining the target loss value of the sample service data according to the sample predicted outer layer browsing time, the real outer layer browsing time label corresponding to the sample service data, the sample predicted click rate, the real click behavior label corresponding to the sample service data, the sample predicted inner layer browsing time and the real inner layer browsing time label corresponding to the sample service data can be that a first time loss value corresponding to the sample service data can be determined according to the sample predicted outer layer browsing time and the real outer layer browsing time label; a second time loss value corresponding to the sample service data can be determined according to the sample predicted click rate, the sample predicted inner layer browsing time and the real inner layer browsing time label; a click loss value corresponding to the sample service data can be determined according to the sample predicted click rate and the real click behavior label; and the target loss value of the sample service data can be determined according to the first time loss value, the second time loss value and the click loss value.

[0129] The specific implementation manner of determining the first time length loss value corresponding to the sample service data according to the sample predicted outer layer browsing time length and the real outer layer browsing time length label can be that a first time length loss function can be obtained, and a logarithmic value corresponding to the real outer layer browsing time length label can be obtained according to the first time length loss function; then, a first time length error value between the logarithmic value corresponding to the real outer layer browsing time length label and the sample predicted outer layer browsing time length can be determined according to the first time length loss function; and the first time length error value can be determined as the first time length loss value corresponding to the sample service data.

[0130] It should be understood that the first time length loss value can be understood as a loss value of outer layer browsing time length estimation, and the main target of training the sample data prediction model is to optimize the estimation accuracy of the outer layer browsing time length, that is, the estimation value of the real browsing time length of the service data is as large as possible, so that the real outer layer browsing time length label can be used as supervision information to train the sample data prediction model when training the model. For ease of understanding, see formula (4) together, formula (4) is a specific determination manner of the first time length loss value:

[0131] MSE1=∑ s∈outer (f(x,θ:θ1)-log(reality outer )) 2 Formula (4)

[0132] In formula (4), MSE1 can be used to represent the first time length loss value; reality outer can be used to represent the real outer layer browsing time length label of the sample service data, that is, the real browsing time length of the sample business object to the sample to-be-clicked business interface; (f(x, θ: θ1) can be used to represent the mathematical representation of the model related parameters of the estimated outer layer browsing time length, and x can be used to represent the sample service data features corresponding to the sample service data.

[0133] The specific implementation manner of determining the second time length loss value corresponding to the sample service data according to the sample predicted click rate, the sample predicted inner layer browsing time length and the real inner layer browsing time length label can be that the sample predicted delivery browsing time length corresponding to the sample service data is determined according to the sample predicted click rate and the sample predicted inner layer browsing time length, the sample predicted delivery browsing time length refers to the predicted browsing time length of the sample business object to the sample to-be-converted business interface under the condition that the sample predicted click rate is used as a probability; a second time length loss function is obtained, and a logarithmic value corresponding to the real inner layer browsing time length label is obtained according to the second time length loss function; a second time length error value between the logarithmic value corresponding to the real inner layer browsing time length label and the sample predicted delivery browsing time length is determined according to the first time length loss function; and the second time length error value is determined as the second time length loss value corresponding to the sample service data.

[0134] It should be understood that the second duration loss value can be understood as a loss value of inner layer browsing duration estimation. If the to-be-converted business interface is to be entered, a click behavior on the to-be-clicked business interface needs to be relied on first. The application can introduce a multiplication concept to determine the inner layer browsing duration estimation value after exposure (i.e., sample predicted delivery browsing duration). It should be understood that the sample predicted click rate can be understood as a click rate estimation value after exposure, and the sample predicted inner layer browsing duration can be understood as an estimation value of the browsing duration of the to-be-converted business interface after the click behavior occurs. Then, the inner layer browsing duration estimation value after exposure (i.e., the sample predicted delivery browsing duration, which can be understood as an inner layer browsing duration estimation value of the sample business data before the sample to-be-clicked business interface is clicked on the sample business object) can be determined by multiplying the sample predicted click rate and the sample predicted inner layer browsing duration.

[0135] It should be understood that the second duration loss value can be understood as a loss value of inner layer browsing duration estimation. The optimization goal of training the sample data prediction model is to optimize the estimation accuracy of the inner layer browsing duration, that is, the estimation value of the real browsing duration of the business data is as large as possible. Therefore, when training the model, the real inner layer browsing duration label can be used as supervision information to train the sample data prediction model. For ease of understanding, please refer to formula (5) together. Formula (5) is a specific determination method of the second duration loss value:

[0136] MSE2 =∑ s∈Inner (f(x, θ: θ2) * f(x, θ: θ3) - log(duration inner )) 2 Formula (5)

[0137] In formula (5), MSE2 can be used to represent the second duration loss value; f(x, θ: θ2) can be used to represent the mathematical representation of the click rate estimation; f(x, θ: θ3) can be used to represent the mathematical representation of the inner layer browsing duration estimation; duration inner can be used to represent the real inner layer browsing duration (i.e., the real browsing duration of the sample to-be-converted business interface after clicking).

[0138] In formula (5), MSE2 can be used to represent the second duration loss value; f(x, θ: θ2) can be used to represent the mathematical representation of the click rate estimation; f(x, θ: θ3) can be used to represent the mathematical representation of the inner layer browsing duration estimation; duration inner can be used to represent the real inner layer browsing duration (i.e., the real browsing duration of the sample to-be-converted business interface after clicking).

[0138] In formula (5), MSE2 can be used to represent the second duration loss value; f(x, θ: θ2) can be used to represent the mathematical representation of the click rate estimation; f(x, θ: θ3) can be used to represent the mathematical representation of the inner layer browsing duration estimation; duration inner can be used to represent the real inner layer browsing duration (i.e., the real browsing duration of the sample to-be-converted business interface after clicking).

[0139] It should be understood that the click loss value can be understood as a loss value of the click rate estimation, and the optimization target of the training sample data prediction model can also include optimizing the estimation accuracy of the click rate. Therefore, when training the model, the real click behavior label (which can be referred to as a click signal) can be used as supervision information to train the sample data prediction model. For ease of understanding, please refer to formula (6) for the specific determination method of the click loss value:

[0140] Cross_entropy = (1 - y) log (1 - f(x, θ: θ2)) - y log (f(x, θ: θ2)) Formula (6)

[0141] In formula (6), Cross_entropy can be used to represent the click loss value; y can be used to represent the click signal (y can be 1 when a click occurs after exposure, and y can be 0 when no click occurs after exposure); and f(x, θ: θ2) can be used to represent the mathematical representation of the click rate estimation.

[0142] It should be understood that in order to train the inner layer browsing time, the outer layer browsing time and the click rate to the model, the application can determine the target loss value for training and adjusting the sample data prediction model through the first time loss value, the second time loss value and the click loss value. The specific implementation method can be: the first operation coefficient corresponding to the first time loss value, the second operation coefficient corresponding to the second time loss value and the third operation coefficient corresponding to the click loss value can be obtained; then, the first time loss value and the first operation coefficient can be operated to obtain the first operation time loss value; the second time loss value and the second operation coefficient can be operated to obtain the second operation time loss value; the click loss value and the third operation coefficient can be operated to obtain the operation click loss value; then, the first operation time loss value, the second operation time loss value and the operation click loss value can be operated to obtain the target loss value of the sample service data.

[0143] For ease of understanding, taking the first operation coefficient, the second operation coefficient and the third operation coefficient as 1 as an example, please refer to formula (7) for the specific method of determining the target loss value:

[0144] Total loss = MSE1 + MSE2 + Cross_entropy Formula (7)

[0145] In formula (7), Total loss can be used to represent the target loss value. That is, the first time loss value, the second time loss value and the third time loss value can be added to obtain the target loss value.

[0146] In step S203, the sample data prediction model is adjusted according to the target loss value to obtain a data prediction model; the data prediction model is used to determine predicted outer browsing time lengths corresponding to N pieces of business data of the business object respectively; the predicted outer browsing time lengths corresponding to the N pieces of business data are used to sort the N pieces of business data to obtain sequence business data.

[0147] Specifically, the model parameters in the sample data prediction model can be adjusted based on the target loss value until a model convergence condition is met or a model iteration number is met, and a final data prediction model is obtained.

[0148] In the embodiments of the present application, when training the data prediction model, a joint training model of predicted click rate and browsing time length (including predicted outer browsing time length and predicted inner browsing time length) is adopted, and the outer browsing time length corresponding to the business object when a click occurs is used to train the data prediction model. Since the outer browsing time length reflects the interest degree of the business object (for example, the longer the outer browsing time length, the higher the interest degree, the greater the probability of a click, and the outer browsing time length is positively correlated with the click rate), the way of training the model by introducing the outer browsing time length can make the data prediction model more accurate and learn the real interest of the business object in a deep level, thereby better improving the accuracy of the click rate estimation. At the same time, since the inner browsing time length also reflects the interest degree of the business object (for example, the longer the inner browsing time length, the higher the interest degree, the greater the probability of conversion, and the inner browsing time length is positively correlated with the conversion rate), the embodiments of the present application can jointly train the model by using the inner browsing time length, the outer browsing time length and the click rate, and through the way of multi-target joint training model, the estimation accuracy of the browsing time length can be optimized, and the estimation accuracy of the click rate can also be optimized. The data prediction model obtained finally can more accurately determine the predicted click rate and the predicted outer browsing time length of the business data, and the predicted click rate and the predicted outer browsing time length can be used in the sorting of the business data to improve the accuracy of the sorting order, that is, the business data can be sorted more accurately based on the preferences of the business object, and then after the sorted business data is put into the business object, the probability of the business object generating a conversion behavior can be improved, thereby improving the conversion rate of the business data.

[0149] Further, please refer to Figure 6 , Figure 6 is an architecture diagram of a training model provided by the embodiments of the present application. As Figure 7As shown, the architecture can include a feature input module and a sample data prediction model, the feature input module can be used to input sample business data corresponding to sample business data features, the sample business data features can include business object dimension features, business data dimension features, environment features, etc., and in the feature input module, the features of each dimension can be fused (such as spliced) to obtain the final sample business data features. It should be understood that the sample business data features processed by the feature input module can be input into the sample data prediction model, which can include an outer layer browsing time estimation submodule, an inner layer browsing time estimation submodule, and a click rate estimation submodule. The outer layer browsing time estimation submodule can output a sample predicted outer layer browsing time, and the first time loss value can be determined by the sample predicted outer layer browsing time; the inner layer browsing time estimation submodule can output a sample predicted inner layer browsing time, and the sample predicted click rate can be output by the click rate estimation module, and the sample predicted inner layer browsing time and the sample predicted click rate can output the sample predicted delivery browsing time (i.e. the estimated value of the inner layer browsing time after exposure). According to the sample predicted delivery browsing time, the second time loss value can be determined, the click loss value can be determined by the sample predicted click rate, and the target loss value for training and adjusting the sample data prediction model can be determined according to the first time loss value, the second time loss value and the click loss value.

[0150] It can be understood that the outer layer browsing time and the click rate are positively correlated. Through experiments, it is found that as the outer layer browsing time increases, the real click rate also increases, and the estimated click rate also increases. However, the increase in the estimated click rate will be more and more different from the real click, that is, the deviation between the estimated click rate and the real click rate will increase as the outer layer browsing time increases. At the same time, the inner layer browsing time and the conversion rate are positively correlated. As the inner layer browsing time increases, the conversion rate also increases. Therefore, the present application can introduce the inner layer browsing time and the outer layer browsing time into the model training through multi-objective joint modeling, and use the time signal for modeling, which can optimize the click rate model and also optimize the estimation accuracy of the inner and outer layer browsing time. Through multi-objective joint modeling, the estimation accuracy of the predicted click rate, the predicted outer layer browsing time and the predicted inner layer browsing time can be modeled and optimized. In addition, the estimated browsing time (such as the predicted outer layer browsing time) can be introduced into the ranking as a ranking factor, which can optimize the user experience.

[0151] Further, please refer to Figure 7 , Figure 3is a structural schematic diagram of a service data processing apparatus provided by an embodiment of the present application. The service data processing apparatus can be a computer program (including program code) running in a computer device, for example, the service data processing apparatus is an application software; the service data processing apparatus can be used to execute the method shown in the following. Figure 7 As shown in the following, Figure 3 The service data processing apparatus 1 can include a data acquisition module 11, a time length acquisition module 12, an ordering factor determination module 13, a factor fusion module 14, and a data ordering module 15.

[0152] The data acquisition module 11 is configured to acquire N service data for a service object;

[0153] The time length acquisition module 12 is configured to acquire a predicted outer layer browsing time length corresponding to each of the N service data; the predicted outer layer browsing time length refers to a predicted browsing time length of a to-be-clicked service interface of the service data by the service object after the service data is put into the service object;

[0154] The ordering factor determination module 13 is configured to determine a time length ordering factor corresponding to each of the N service data according to the predicted outer layer browsing time length corresponding to each of the N service data;

[0155] The factor fusion module 14 is configured to generate a fusion ordering factor corresponding to each of the N service data according to the basic ordering factor and the time length ordering factor corresponding to each of the N service data;

[0156] The data ordering module 15 is configured to order the N service data according to the N fusion ordering factors to obtain sequence service data.

[0157] The specific implementation of the data acquisition module 11, the time length acquisition module 12, the ordering factor determination module 13, the factor fusion module 14, and the data ordering module 15 can refer to the description of steps S101-S103 in the above-mentioned corresponding embodiments, which will not be repeated here. Figure 3

[0158] In one embodiment, the N service data include service data M i , i is a positive integer;

[0159] The time length acquisition module 12 can include a feature input unit 121 and a time length determination unit 122.

[0160] The feature input unit 121 is configured to acquire service data M i ​The corresponding business data features are input into the data prediction model. The data prediction model is trained based on sample prediction outer browsing time length, real outer browsing time length label, sample prediction click rate, real click behavior label, sample prediction inner browsing time length, and real inner browsing time length label corresponding to sample business data. The sample prediction outer browsing time length refers to the predicted browsing time length of the sample business object for the sample to-be-clicked business interface of the sample business data after the sample business data is put into the sample business object. The sample prediction click rate refers to the predicted probability of the sample business object for the sample to-be-clicked business interface to occur a click behavior. The sample prediction inner browsing time length refers to the predicted browsing time length of the sample business object for the sample to-be-converted business interface after the sample business object enters the sample to-be-converted business interface of the sample business data by performing a click behavior on the sample to-be-clicked business interface.

[0161] The time length determination unit 122 is configured to determine the business data M i corresponding to the predicted outer browsing time length.

[0162] The specific implementation of the feature input unit 121 and the time length determination unit 122 can be referred to the description of step S101 in the above Figure 3 corresponding embodiment, which will not be repeated here.

[0163] In one embodiment, the N business data includes the business data M i , and the predicted outer browsing time length includes the predicted outer browsing time length T i corresponding to the business data M i , i is a positive integer.

[0164] The sorting factor determination module 13 can include a parameter acquisition unit 131 and a sorting factor determination unit 132.

[0165] The parameter acquisition unit 131 is configured to acquire the time length operation parameter for the predicted outer browsing time length T i .

[0166] The sorting factor determination unit 132 is configured to perform operation processing on the time length operation parameter and the predicted outer browsing time length T i to obtain the time length sorting factor corresponding to the business data M i .

[0167] The specific implementation of the parameter acquisition unit 131 and the sorting factor determination unit 132 can be referred to the description of step S102 in the above Figure 3 corresponding embodiment, which will not be repeated here.

[0168] In one embodiment, the N business data includes the business data Mi i is a positive integer;

[0169] The factor fusion module 14 can include a basic factor determination unit 141 and a factor operation unit 142.

[0170] The basic factor determination unit 141 is configured to obtain the service data M i corresponding to the expected conversion unit resource, the predicted click rate, and the predicted conversion rate;

[0171] The basic factor determination unit 141 is further configured to determine the service data M i corresponding to the real-time placement consumption resource, and determine the service data M i corresponding to the real-time placement consumption resource, as the service data M i corresponding to the basic ranking factor;

[0172] The factor operation unit 142 is configured to perform operation processing on the service data M i corresponding to the time length ranking factor, and the service data M i corresponding to the basic ranking factor, to obtain the service data M i corresponding to the fusion ranking factor.

[0173] The specific implementation of the basic factor determination unit 141 and the factor operation unit 142 can be referred to the description of step S103 in the above-mentioned Figure 3 corresponding embodiments, which will not be repeated here.

[0174] In one embodiment, the service data processing apparatus 1 can further include a data placement module 16.

[0175] The data placement module 16 is configured to obtain a service demand placement amount, and obtain K service data from the sequence service data in order according to the service demand placement amount.

[0176] The data placement module 16 is further configured to place the K service data to the service object.

[0177] The specific implementation of the data placement module 16 can be referred to the description of step S103 in the above-mentioned Figure 3 corresponding embodiments, which will not be repeated here.

[0178] In one embodiment, the service data processing apparatus 1 can further include a similar data recommendation module 17.

[0179] The similar data recommendation module 17 is configured to obtain real-time feedback data of the service object respectively for the K service data within a target time period.

[0180] The similar data recommendation module 17 is further configured to determine positive feedback service data from the K service data according to the K real-time feedback data.

[0181] The similar data recommendation module 17 is further configured to perform similar service data recommendation processing on the service object according to the positive feedback service data.

[0182] The specific implementation of the similar data recommendation module 17 can be referred to the description of step S103 in the above-mentioned Figure 8 corresponding embodiment, which will not be repeated here.

[0183] In the embodiment of the present application, after obtaining N service data corresponding to the service object, the predicted outer browsing time corresponding to each of the N service data (i.e. the estimated value of the browsing time of the service object for the service interface to be clicked on the service data after the service data is put into the service object) can be obtained. Then, the time sorting factor corresponding to each of the N service data can be determined through the predicted outer browsing time corresponding to each of the N service data. According to the basic sorting factor and the time sorting factor corresponding to each of the N service data, the fusion sorting factor corresponding to each of the N service data can be generated. According to the N fusion sorting factors, the N service data can be sorted to obtain the sequence service data. That is, when the service data is sorted, the present application introduces the dimension of the predicted outer browsing time, and the predicted outer browsing time is used as a sorting factor to participate in the sorting of the service data, which can enrich the sorting dimension of the service data. At the same time, since the browsing time of the service object also has a great influence on whether the conversion occurs, for example, when the browsing time is long, it can represent that the service data is more in line with the preferences of the service object and is more likely to occur conversion. Therefore, the browsing time is used as a sorting factor to participate in the sorting, which can enrich the sorting factors and more accurately sort the service data based on the preferences of the service object. Thus, after the sorted service data is put into the service object, the probability of the service object generating a conversion behavior can be improved, thereby improving the conversion rate of the service data. In summary, the present application can introduce the predicted outer browsing time when sorting the service data, which can enrich the sorting dimension and further improve the conversion rate of the service data.

[0184] Further, please refer to Figure 8 , Figure 4 is a structural schematic diagram of a service data processing device provided by an embodiment of the present application. The service data processing device can be a computer program (including program code) running in a computer device, for example, the service data processing device is an application software. The service data processing device can be used to execute the method shown in Figure 8 . As Figure 5As shown, the business data processing apparatus 2 can include a sample feature input module 21, a loss value determination module 22, and a model adjustment module 23.

[0185] The sample feature input module 21 is configured to obtain sample business data features corresponding to sample business data, input the sample business data features into a sample data prediction model, and output, by the sample data prediction model, a sample predicted outer-layer browsing time length, a sample predicted inner-layer browsing time length, and a sample predicted click rate corresponding to the sample business data. The sample predicted outer-layer browsing time length refers to a predicted browsing time length of a sample business object for a sample to-be-clicked business interface of the sample business data after the sample business data is launched to the sample business object. The sample predicted click rate refers to a predicted probability of a click behavior of the sample business object for the sample to-be-clicked business interface. The sample predicted inner-layer browsing time length refers to a predicted browsing time length of the sample business object for a sample to-be-converted business interface of the sample business data after the sample business object enters the sample to-be-converted business interface by performing a click behavior on the sample to-be-clicked business interface.

[0186] The loss value determination module 22 is configured to determine a target loss value of the sample business data according to the sample predicted outer-layer browsing time length, a real outer-layer browsing time length label corresponding to the sample business data, the sample predicted click rate, a real click behavior label corresponding to the sample business data, the sample predicted inner-layer browsing time length, and a real inner-layer browsing time length label corresponding to the sample business data.

[0187] The model adjustment module 23 is configured to adjust the sample data prediction model according to the target loss value to obtain a data prediction model. The data prediction model is configured to determine predicted outer-layer browsing time lengths corresponding to N pieces of business data for a business object. The predicted outer-layer browsing time lengths corresponding to the N pieces of business data are configured to sort the N pieces of business data to obtain sequence business data.

[0188] The specific implementation of the sample feature input module 21, the loss value determination module 22, and the model adjustment module 23 can be referred to the description of steps S201-S203 in the above-mentioned Figure 5 corresponding embodiments, which will not be described herein.

[0189] In one embodiment, the loss value determination module 22 can include a loss value determination unit 221 and a target loss value determination unit 222.

[0190] The loss value determination unit 221 is configured to determine a first time length loss value corresponding to the sample business data according to the sample predicted outer-layer browsing time length and the real outer-layer browsing time length label.

[0191] The loss value determination unit 221 is further configured to determine a second time length loss value corresponding to the sample service data according to the sample predicted click rate, the sample predicted inner layer browsing time length, and the real inner layer browsing time length label.

[0192] The loss value determination unit 221 is further configured to determine a click loss value corresponding to the sample service data according to the sample predicted click rate and the real click behavior label.

[0193] The target loss value determination unit 222 is configured to determine a target loss value of the sample service data according to the first time length loss value, the second time length loss value, and the click loss value.

[0194] The specific implementation of the loss value determination unit 221 and the target loss value determination unit 222 can be referred to the description of step S202 in the above Figure 9

[0195] In one embodiment, the loss value determination unit 221 is further configured to obtain a first time length loss function, and obtain a logarithmic value corresponding to the real outer layer browsing time length label according to the first time length loss function.

[0196] The loss value determination unit 221 is further configured to determine a first time length error value between the logarithmic value corresponding to the real outer layer browsing time length label and the sample predicted outer layer browsing time length according to the first time length loss function.

[0197] The loss value determination unit 221 is further configured to determine the first time length error value as the first time length loss value corresponding to the sample service data.

[0198] In one embodiment, the loss value determination unit 221 is further configured to determine a sample predicted delivery browsing time length corresponding to the sample service data according to the sample predicted click rate and the sample predicted inner layer browsing time length. The sample predicted delivery browsing time length refers to a predicted browsing time length of the sample service object for the sample to-be-converted service interface under the condition that the sample predicted click rate is used as a probability.

[0199] The loss value determination unit 221 is further configured to obtain a second time length loss function, and obtain a logarithmic value corresponding to the real inner layer browsing time length label according to the second time length loss function.

[0200] The loss value determination unit 221 is further configured to determine a second time length error value between the logarithmic value corresponding to the real inner layer browsing time length label and the sample predicted delivery browsing time length according to the first time length loss function.

[0201] The loss value determination unit 221 is further configured to determine the second time length error value as the second time length loss value corresponding to the sample service data. ​

[0202] In one embodiment, the loss value determination unit 221 is further specifically used to obtain the click loss function and obtain the logarithm of the sample predicted click rate according to the click loss function;

[0203] The loss value determination unit 221 is also specifically used to determine the click error value between the logarithm of the predicted click rate of the sample and the actual click behavior label according to the click loss function;

[0204] The loss value determination unit 221 is also specifically used to determine the click error value as the click loss value corresponding to the sample business data.

[0205] In one embodiment, the target loss value determination unit 222 is further specifically used to obtain the first calculation coefficient corresponding to the first duration loss value, the second calculation coefficient corresponding to the second duration loss value, and the third calculation coefficient corresponding to the click loss value;

[0206] The target loss value determination unit 222 is also specifically used to perform calculations on the first duration loss value and the first calculation coefficient to obtain the first calculation duration loss value;

[0207] The target loss value determination unit 222 is also specifically used to perform calculations on the second duration loss value and the second calculation coefficient to obtain the second calculation duration loss value;

[0208] The target loss value determination unit 222 is also specifically used to perform calculations on the click loss value and the third calculation coefficient to obtain the calculated click loss value;

[0209] The target loss value determination unit 222 is also specifically used to process the first computation time loss value, the second computation time loss value, and the computation click loss value to obtain the target loss value of the sample business data.

[0210] In this embodiment, a multi-objective joint modeling approach is used to incorporate the inner and outer browsing durations into the model training. The duration signals are used for modeling, optimizing both the click-through rate (CTR) model and the accuracy of predicting both inner and outer browsing durations. This multi-objective joint modeling approach can optimize the accuracy of predicting CTR, outer browsing duration, and inner browsing duration. Furthermore, the estimated browsing duration (such as the predicted outer browsing duration) can be incorporated into the ranking process as a ranking factor, thus optimizing the user experience.

[0211] Further, please see Figure 9 , Figure 9 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Figure 7 As shown above, Figure 8 The corresponding embodiment of the business data device 1 orFigure 9 The service data device 2 in the corresponding embodiment can be applied to the aforementioned computer device 8000. The computer device 8000 may include a processor 8001, a network interface 8004, and a memory 8005. Furthermore, the computer device 8000 also includes a user interface 8003 and at least one communication bus 8002. The communication bus 8002 is used to enable communication between these components. The user interface 8003 may include a display screen and a keyboard; optionally, the user interface 8003 may also include a standard wired interface or a wireless interface. The network interface 8004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 8005 may be high-speed RAM or non-volatile memory, such as at least one disk storage device. Optionally, the memory 8005 may also be at least one storage device located remotely from the aforementioned processor 8001. Figure 9 As shown, the memory 8005, which is a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a device control application program.

[0212] exist Figures 3 to 6 In the computer device 8000 shown, the network interface 8004 provides network communication functionality; the user interface 8003 is mainly used to provide an input interface for the user; and the processor 8001 can be used to call the device control application program stored in the memory 8005 to achieve:

[0213] Get N business data points for a business object, and get the predicted outer browsing duration for each of the N business data points; the predicted outer browsing duration refers to the predicted browsing duration of the business object for the clickable business interface of the business data after the business data is delivered to the business object.

[0214] Based on the predicted outer browsing time corresponding to each of the N business data points, determine the time ranking factor corresponding to each of the N business data points.

[0215] Based on the basic sorting factor and duration sorting factor corresponding to each of the N business data, generate the fusion sorting factor corresponding to each of the N business data. Sort the N business data according to the N fusion sorting factors to obtain the sequence business data.

[0216] Or achieve:

[0217] obtaining sample business data features corresponding to the sample business data, inputting the sample business data features into the sample data prediction model, and outputting, by the sample data prediction model, a sample predicted outer-layer browsing time length, a sample predicted inner-layer browsing time length, and a sample predicted click rate corresponding to the sample business data; the sample predicted outer-layer browsing time length refers to a predicted browsing time length of a sample to-be-clicked business interface of the sample business data by the sample business object after the sample business data is delivered to the sample business object; the sample predicted click rate refers to a predicted probability of a click behavior of the sample to-be-clicked business interface by the sample business object; and the sample predicted inner-layer browsing time length refers to a predicted browsing time length of a sample to-be-converted business interface of the sample business data by the sample business object after the sample business object enters the sample to-be-converted business interface by performing the click behavior on the sample to-be-clicked business interface;

[0218] determining a target loss value of the sample business data according to the sample predicted outer-layer browsing time length, a real outer-layer browsing time length label corresponding to the sample business data, the sample predicted click rate, a real click behavior label corresponding to the sample business data, the sample predicted inner-layer browsing time length, and a real inner-layer browsing time length label corresponding to the sample business data;

[0219] adjusting the sample data prediction model according to the target loss value to obtain a data prediction model; and the data prediction model is used to determine predicted outer-layer browsing time lengths corresponding to N business data respectively; and the predicted outer-layer browsing time lengths corresponding to the N business data respectively are used to sort the N business data to obtain sequence business data.

[0220] It should be understood that the computer device 8000 described in the embodiments of the present application can execute the foregoing descriptions of the business data processing method in the embodiments corresponding thereto. Figure 7 The foregoing descriptions of the business data processing method in the embodiments corresponding thereto can also be executed by the foregoing descriptions of the business data processing apparatus 1 in the embodiments corresponding thereto. Figure 8 The foregoing descriptions of the business data processing method in the embodiments corresponding thereto can also be executed by the foregoing descriptions of the business data processing apparatus 2 in the embodiments corresponding thereto. Figures 3 to 6 The foregoing descriptions of the business data processing method in the embodiments corresponding thereto can also be executed by the foregoing descriptions of the business data processing apparatus 2 in the embodiments corresponding thereto.

[0221] In addition, it should be noted that the embodiments of the present application also provide a computer readable storage medium, and the foregoing computer readable storage medium stores the computer program executed by the foregoing data processing computer device 1000, and the foregoing computer program includes program instructions, and when the foregoing processor executes the foregoing program instructions, the foregoing descriptions of the business data processing method in the embodiments corresponding thereto can be executed. Figure 1The description of the above business data processing method in the corresponding embodiment will not be repeated here. In addition, the description of the beneficial effects of using the same method will also not be repeated. For technical details not disclosed in the computer-readable storage medium embodiments involved in the present application, please refer to the description of the method embodiments of the present application.

[0222] The computer-readable storage medium can be an internal storage unit of the computer device, such as a hard disk or a memory of the computer device, provided by any of the foregoing embodiments. The computer-readable storage medium can also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the computer-readable storage medium can include both the internal storage unit and the external storage device of the computer device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0223] In one aspect of the present application, a computer program product or computer program is provided, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to enable the computer device to perform the method provided in one aspect of the embodiments of the present application.

[0224] The terms "first", "second", etc. in the specification and claims and drawings of the embodiments of the present application are used to distinguish different objects, and are not used to describe a specific order. In addition, the term "comprising" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product or equipment that includes a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units that are not listed, or can optionally include other steps or units inherent to the process, method, device, product or equipment.

[0225] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in the above description in a general manner. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0226] The method and related apparatus provided by the embodiments of the present application are described with reference to the method flowchart and / or structural schematic diagram provided by the embodiments of the present application. Each flow and / or block in the method flowchart and / or structural schematic diagram and the combination of the flows and / or blocks in the flowchart and / or block diagram can be realized by computer program instructions. The computer program instructions can be provided to a processor of a general purpose computer, a special purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate a device implemented in the flow Figure 1 The computer program instructions can also be stored in a computer readable memory capable of causing a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction devices that implement the functions specified in the flow Figure 1 The computer program instructions can also be stored in a computer readable memory capable of causing a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction devices that implement the functions specified in the flow Figure 1 The computer program instructions can also be stored in a computer readable memory capable of causing a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction devices that implement the functions specified in the flow Figure 1 The computer program instructions can also be stored in a computer readable memory capable of causing a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction devices that implement the functions specified in the flow ​ The computer program instructions can also be stored in a computer readable memory capable of causing a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction devices that implement the functions specified in the flow

[0227] The above disclosure is only the preferred embodiments of the present application, and of course cannot limit the scope of the rights of the present application, so the equivalent changes made by the claims of the present application still fall within the scope of the present application.

Claims

1. A service data processing method characterized by, The method comprises the following steps: acquiring N pieces of business data corresponding to a business object, and acquiring a predicted outer-layer browsing time length corresponding to each of the N pieces of business data; the predicted outer-layer browsing time length refers to a predicted browsing time length of a to-be-clicked business interface of the business data by the business object after the business data is put into the business object; determining a time length sorting factor corresponding to each of the N pieces of business data according to the predicted outer-layer browsing time length corresponding to each of the N pieces of business data; the time length sorting factor corresponding to each piece of business data is obtained by performing operation processing on a time length operation parameter of the business data and the predicted outer-layer browsing time length corresponding to the business data; and the time length sorting factors corresponding to the N pieces of business data are used for sorting the N pieces of business data; generating a fusion sorting factor corresponding to each of the N pieces of business data according to a basic sorting factor corresponding to each of the N pieces of business data and the time length sorting factor corresponding to each of the N pieces of business data; the basic sorting factor corresponding to each piece of business data refers to a real-time put-in consumption resource corresponding to the business data, and the real-time put-in consumption resource corresponding to the business data refers to a resource consumed in the process of putting in the business data; sorting the N pieces of business data according to the N fusion sorting factors to obtain sequence business data.

2. The method of claim 1, wherein, N said service data includes service data M i , i is a positive integer; The method further comprises the following steps: Obtaining the service data M i corresponding to the service data, and inputting the service data features into a data prediction model; the data prediction model is trained based on sample service data corresponding to sample predicted outer-layer browsing time, real outer-layer browsing time label, sample predicted click rate, real click behavior label, sample predicted inner-layer browsing time, and real inner-layer browsing time label; the sample predicted outer-layer browsing time refers to the predicted browsing time of the sample business object for the sample to-be-clicked business interface of the sample service data after the sample service data is put into the sample business object; the sample predicted click rate refers to the predicted probability of the sample business object for the sample to-be-clicked business interface to occur a click behavior; the sample predicted inner-layer browsing time refers to the predicted browsing time of the sample business object for the sample to-be-converted business interface of the sample service data after the sample business object enters the sample to-be-converted business interface by performing a click behavior on the sample to-be-clicked business interface. Determine the business data M through the data prediction model and the business data characteristics i The corresponding prediction outer layer browsing duration.

3. The method of claim 1, wherein, N pieces of the service data include service data M i The predicted outer-layer browsing duration includes the service data M i The corresponding predicted outer-layer browsing duration T i I is a positive integer; acquiring a business demand put-in amount, and acquiring K pieces of business data from the sequence business data in sequence according to the business demand put-in amount; acquiring a time length operation parameter for the predicted outer-layer browsing time length T i ; the predicted outer-layer browsing duration T i carries out operation processing to obtain the service data M i a corresponding duration ranking factor.

4. The method of claim 1, wherein, N said service data includes service data M i i is a positive integer; putting in the K pieces of business data into the business object. Obtaining the service data M i Corresponding expected conversion unit delivery resource, predicted click rate, and predicted conversion rate; determining the business data M from the desired conversion unit, the predicted click rate, and the predicted conversion rate i a corresponding real-time bid consumption resource, determining the business data M i a corresponding real-time bid consumption resource, determining the business data M i a corresponding base ranking factor The business data M i The corresponding time length sorting factor, and the business data M i The corresponding basic sorting factor is operated and processed to obtain the business data M i The corresponding fusion sorting factor.

5. The method of claim 1, wherein, The method further comprises the following steps: acquiring real-time feedback data of the business object corresponding to the K pieces of business data in a target time period; determining positive feedback business data from the K pieces of business data according to the K pieces of real-time feedback data; 6. The method of claim 5, wherein, performing similar business data recommendation processing on the business object according to the positive feedback business data. The method comprises the following steps: ​ ​ 7. A service data processing method characterized by comprising: ​ obtaining sample business data features corresponding to sample business data, inputting the sample business data features into a sample data prediction model, and outputting sample predicted outer layer browsing time, sample predicted inner layer browsing time and sample predicted click rate corresponding to the sample business data through the sample data prediction model; the sample predicted outer layer browsing time refers to the predicted browsing time of the sample business object for the sample to-be-clicked business interface of the sample business data after the sample business data is put into the sample business object; the sample predicted click rate refers to the predicted probability of the sample business object for the click behavior of the sample to-be-clicked business interface; the sample predicted inner layer browsing time refers to the predicted browsing time of the sample business object for the sample to-be-converted business interface of the sample business data after the sample business object enters the sample to-be-converted business interface through the click behavior of the sample to-be-clicked business interface; determining a target loss value of the sample business data according to the sample predicted outer layer browsing time, the real outer layer browsing time label corresponding to the sample business data, the sample predicted click rate, the real click behavior label corresponding to the sample business data, the sample predicted inner layer browsing time and the real inner layer browsing time label corresponding to the sample business data; adjusting the sample data prediction model according to the target loss value to obtain a data prediction model; the data prediction model is used to determine the predicted outer layer browsing time corresponding to N business data respectively; the predicted outer layer browsing time corresponding to N business data respectively is used to determine the time sorting factor corresponding to N business data respectively, the time sorting factor corresponding to each business data is obtained by operating the time length operation parameter of the business data with the predicted outer layer browsing time corresponding to the business data, and the time sorting factor corresponding to N business data respectively is used to sort N business data; the basic sorting factor and the time sorting factor corresponding to N business data respectively are used to generate the fusion sorting factor corresponding to N business data respectively, the basic sorting factor corresponding to each business data refers to the real-time put-in consumed resource corresponding to the business data, and the real-time put-in consumed resource corresponding to the business data refers to the resource consumed in the process of putting in the business data; N fusion sorting factors are used to sort N business data to obtain sequence business data.

8. The method of claim 7, wherein, The method for determining the target loss value of the sample business data according to the sample predicted outer layer browsing time, the real outer layer browsing time label corresponding to the sample business data, the sample predicted click rate, the real click behavior label corresponding to the sample business data, the sample predicted inner layer browsing time and the real inner layer browsing time label corresponding to the sample business data comprises: determining a first time length loss value corresponding to the sample business data according to the sample predicted outer layer browsing time and the real outer layer browsing time label; determine a first time length loss value corresponding to the sample service data according to the sample predicted outer layer browsing time length and the real outer layer browsing time length label; determine a click loss value corresponding to the sample service data according to the sample predicted click rate and the real click behavior label; determine a target loss value of the sample service data according to the first time length loss value, the second time length loss value and the click loss value.

9. The method of claim 8, wherein, The method further includes: obtaining a first time length loss function, and obtaining a logarithmic value corresponding to the real outer layer browsing time length label according to the first time length loss function; determine a first time length error value between the logarithmic value corresponding to the real outer layer browsing time length label and the sample predicted outer layer browsing time length according to the first time length loss function; determine the first time length error value as the first time length loss value corresponding to the sample service data. The method further includes: determining a sample predicted delivery browsing time length corresponding to the sample service data according to the sample predicted click rate and the sample predicted inner layer browsing time length; the sample predicted delivery browsing time length refers to a predicted browsing time length of the sample service object for the sample to-be-converted service interface under the condition that the sample predicted click rate is used as a probability condition; 10. The method of claim 8, wherein, obtain a second time length loss function, and obtain a logarithmic value corresponding to the real inner layer browsing time length label according to the second time length loss function; determine a second time length error value between the logarithmic value corresponding to the real inner layer browsing time length label and the sample predicted delivery browsing time length according to the first time length loss function; determine the second time length error value as the second time length loss value corresponding to the sample service data. The method further includes: obtaining a click loss function, and obtaining a logarithmic value corresponding to the sample predicted click rate according to the click loss function; determine a click error value between the logarithmic value corresponding to the sample predicted click rate and the real click behavior label according to the click loss function; 11. The method of claim 8, wherein, determine the click error value as the click loss value corresponding to the sample service data. The method further includes: obtaining a first operation coefficient corresponding to the first time length loss value, a second operation coefficient corresponding to the second time length loss value and a third operation coefficient corresponding to the click loss value; perform operation processing on the first time length loss value and the first operation coefficient to obtain a first operation time length loss value; perform operation processing on the second time length loss value and the second operation coefficient to obtain a second operation time length loss value; and 12. The method of claim 8, wherein, perform operation processing on the click loss value and the third operation coefficient to obtain a third operation click loss value. ​ ​ ​ The click loss value is processed with the third operation coefficient to obtain an operation click loss value; The first operation time loss value, the second operation time loss value, and the operation click loss value are processed to obtain a target loss value of the sample service data.

13. A service data processing apparatus characterized by comprising: The method comprises the following steps: The data acquisition module is configured to acquire N service data for a service object; The time length acquisition module is configured to acquire a predicted outer-layer browsing time length corresponding to each of the N service data; the predicted outer-layer browsing time length refers to a predicted browsing time length of a to-be-clicked service interface of the service object for the service data after the service data is delivered to the service object; a time length sorting factor corresponding to each of the service data is obtained by processing a time length operation parameter of the service data and the predicted outer-layer browsing time length corresponding to the service data; and the N service data correspond to N time length sorting factors, respectively. The sorting factor determination module is configured to determine the time length sorting factor corresponding to each of the N service data according to the predicted outer-layer browsing time length corresponding to each of the N service data. The factor fusion module is configured to generate a fusion sorting factor corresponding to each of the N service data according to a basic sorting factor corresponding to each of the N service data and the time length sorting factor corresponding to each of the N service data; the basic sorting factor corresponding to each of the service data refers to a real-time delivery consumption resource corresponding to the service data. The data sorting module is configured to sort the N service data according to the N fusion sorting factors to obtain sequence service data.

14. A computer device, comprising: The computer device comprises a processor, a memory, and a network interface. The processor is connected to the memory and the network interface; the network interface is configured to provide network communication functions; the memory is configured to store program codes; and the processor is configured to call the program codes to enable the computer device to execute the method of any one of claims 1 to 12. The computer program is stored in the computer readable storage medium and is adapted to be loaded by the processor to execute the method of any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that, The computer program product or the computer program comprises computer instructions stored in the computer readable storage medium, and the computer instructions are adapted to be read and executed by the processor to enable the computer device with the processor to execute the method of any one of claims 1 to 12.

16. A computer program product or computer program, characterised in that, ​

Images