Business data processing method and device and related equipment

Abstract

The embodiment of the invention discloses a business data processing method and device and related equipment, and can be applied to the technical field of data processing. The method comprises the following steps: acquiring payment transaction data; performing feature mapping processing on the time-class dimension information to obtain time mapping features; performing feature mapping processing on the numerical value type dimension information to obtain numerical value mapping features; performing word segmentation processing on the description class dimension information to obtain text segmented words, and performing feature conversion processing on the text segmented words to obtain description conversion features; performing heterogeneous conversion processing on the time mapping feature, the numerical value mapping feature and the description conversion feature by the first hierarchical network to obtain a heterogeneous feature data sequence; and based on the heterogeneous feature data sequence, determining a payment service feature sequence, and based on sequence features in the payment service feature sequence, performing object recognition on the object type of the service object to obtain an object recognition result. By adopting the embodiment of the invention, the accuracy of identifying the object type of the business object can be improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a business data processing method, apparatus, and related equipment. Background Art

[0002] Currently, when analyzing business data, the cumulative value of the business data collected by the business object (such as user A) over a period of time is used. This makes it difficult to mine and analyze the data correlation between the business data obtained by the business object during the period of time during data analysis. This not only reduces the accuracy of data analysis of the business data of the business object, but also reduces the accuracy of identifying the object type of the business object. Summary of the Invention

[0003] The embodiments of the present application provide a business data processing method, apparatus, and related equipment, which can construct heterogeneous conversion features based on information of each payment transaction in multiple dimensions, improve the accuracy of data analysis of transaction operations of business objects, and thereby help improve the accuracy of identifying object types of business objects.

[0004] On the one hand, an embodiment of the present application provides a method for processing business data, the method comprising:

[0005] Obtain payment transaction data for input into a target business model from a transaction sequence of a business object within a payment cycle; the target business model includes a first hierarchical network and a second hierarchical network; the payment transaction data is determined based on payment transactions initiated by the business object within the first business cycle; the payment transaction data includes time dimension information, numerical dimension information, and descriptive dimension information; the first business cycle is a business cycle within the payment cycle;

[0006] Inputting the time dimension information into the first hierarchical network, and performing feature mapping processing on the time dimension information by the first hierarchical network to obtain a time mapping feature corresponding to the time dimension information;

[0007] Inputting the numerical dimension information into the first hierarchical network, and performing feature mapping processing on the numerical dimension information by the first hierarchical network to obtain numerical mapping features corresponding to the numerical dimension information;

[0008] Inputting the description class dimension information into the first hierarchical network, the first hierarchical network performs word segmentation processing on the description class dimension information to obtain text segmentation of the description class dimension information, and performing feature conversion processing on the text segmentation to obtain description conversion features corresponding to the description class dimension information;

[0009] The first hierarchical network performs heterogeneous conversion processing on the time mapping feature, the value mapping feature, and the description conversion feature to obtain a heterogeneous feature data sequence of the business object within the first type of business cycle;

[0010] Based on the heterogeneous feature data sequence, the payment business feature sequence of the business object within the payment cycle is determined, and the payment business feature sequence is input into the second hierarchical network. The second hierarchical network performs object recognition on the object type of the business object based on the sequence features in the payment business feature sequence to obtain an object recognition result.

[0011] On the one hand, an embodiment of the present application provides a service data processing device, the device comprising:

[0012] A transaction data acquisition module is configured to acquire payment transaction data for input into a target business model from a transaction sequence of a business object within a payment cycle; the target business model includes a first hierarchical network and a second hierarchical network; the payment transaction data is determined based on payment transactions initiated by the business object within the first business cycle; the payment transaction data includes time dimension information, numerical dimension information, and descriptive dimension information; the first business cycle is a business cycle within the payment cycle;

[0013] A time mapping module is used to input the time dimension information into the first hierarchical network, and the first hierarchical network performs feature mapping processing on the time dimension information to obtain a time mapping feature corresponding to the time dimension information;

[0014] A numerical mapping module is used to input the numerical dimension information into the first hierarchical network, and the first hierarchical network performs feature mapping processing on the numerical dimension information to obtain numerical mapping features corresponding to the numerical dimension information;

[0015] A description conversion module is used to input the description class dimension information into the first hierarchical network, and the first hierarchical network performs word segmentation processing on the description class dimension information to obtain text segmentation of the description class dimension information, and performs feature conversion processing on the text segmentation to obtain description conversion features corresponding to the description class dimension information;

[0016] a heterogeneous conversion module, configured to perform heterogeneous conversion processing on the time mapping feature, the value mapping feature, and the description conversion feature by the first hierarchical network, to obtain a heterogeneous feature data sequence of the business object within the first type of business cycle;

[0017] The object recognition module is used to determine the payment business feature sequence of the business object within the payment cycle based on the heterogeneous feature data sequence, input the payment business feature sequence into the second hierarchical network, and the second hierarchical network performs object recognition on the object type of the business object based on the sequence features in the payment business feature sequence to obtain the object recognition result.

[0018] Wherein, the payment cycle includes N business cycles, the N business cycles include business cycle i, i is a positive integer less than or equal to N, business cycle i belongs to the first type of business cycle, the heterogeneous feature data sequence includes a first feature data sequence corresponding to business cycle i, and the first feature data sequence is determined by the heterogeneous conversion feature determined based on the payment transactions in business cycle i;

[0019] Wherein, the object recognition module includes: a payment feature sequence unit;

[0020] The payment feature sequence unit is used to perform feature fusion processing on the heterogeneous conversion features in the first feature data sequence corresponding to the business cycle i to obtain the periodic transaction features corresponding to the business cycle i, until the periodic transaction features of each business cycle in N business cycles are obtained. Based on the periodic transaction features of each business cycle obtained, the payment business feature sequence of the business object within the payment cycle is determined.

[0021] Among them, the payment feature sequence unit includes: fusion feature unit, cycle time feature unit, feature splicing unit;

[0022] a fusion feature unit configured to perform feature fusion processing on the heterogeneous conversion features in the first feature data sequence by the first hierarchical network to obtain a fusion feature corresponding to the service cycle i;

[0023] A cycle time feature unit is used to obtain cycle time information of service cycle i, input the cycle time information into the first layered network, and perform feature conversion processing on the cycle time information by the first layered network to obtain the cycle time feature corresponding to service cycle i;

[0024] The feature splicing unit is used to perform period feature splicing processing on the fusion feature and the period time feature to obtain the period transaction feature corresponding to the business period i.

[0025] The N business cycles also include business cycle j, where j is a positive integer less than or equal to N. If j is different from i, business cycle j belongs to a second type of business cycle, which is different from the first type of business cycle. The second type of business cycle is a business cycle in which no payment transaction occurs for the business object.

[0026] Wherein, the payment feature sequence unit further includes: a cycle filling unit;

[0027] Cycle filling unit, specifically used for:

[0028] Obtaining a cycle filling feature associated with the second type of business cycle, and determining the cycle filling feature as a cycle transaction feature of business cycle j;

[0029] Based on the periodic transaction characteristics of business cycle j and the periodic transaction characteristics of business cycle i, the periodic transaction characteristics of each of the N business cycles are determined.

[0030] Wherein, the payment feature sequence unit includes: a feature arrangement unit;

[0031] The feature arrangement unit is used to obtain the cycle time sequence of N business cycles, arrange the cycle transaction features of each business cycle based on the cycle time sequence, and obtain the payment business feature sequence of the business object in the payment cycle.

[0032] The number of payment transactions in the first business cycle is M, and the M payment transactions include payment transaction p, where p is a positive integer less than or equal to M;

[0033] Wherein, the heterogeneous conversion module includes: a first heterogeneous feature sequence unit;

[0034] The first heterogeneous feature sequence unit is used to perform heterogeneous conversion processing on the time mapping feature corresponding to the payment transaction p, the numerical mapping feature corresponding to the payment transaction p, and the description conversion feature corresponding to the payment transaction p by the first hierarchical network to obtain the heterogeneous conversion feature corresponding to the payment transaction p, until the heterogeneous conversion feature of each payment transaction in the M payment transactions is obtained, and then determine the heterogeneous feature data sequence of the business object within the first type of business cycle based on the obtained heterogeneous conversion feature of each payment transaction.

[0035] The first type of business cycle includes W business sub-cycles, where W is a positive integer; the W business sub-cycles include the first type of sub-cycle, and the first type of sub-cycle is a business sub-cycle in which the business object initiates a payment transaction; and the payment transaction within the first type of sub-cycle is the first payment transaction.

[0036] The heterogeneous conversion module includes: a sub-period feature unit and a second heterogeneous feature sequence unit;

[0037] a sub-period feature unit configured to, when the first hierarchical network determines the time mapping feature of the first payment transaction, the numerical mapping feature of the first payment transaction, and the description conversion feature of the first payment transaction, perform heterogeneous conversion processing based on the time mapping feature of the first payment transaction, the numerical mapping feature of the first payment transaction, and the description conversion feature of the first payment transaction, to obtain a heterogeneous conversion feature corresponding to the first type of sub-period;

[0038] The second heterogeneous feature sequence unit determines the heterogeneous feature data sequence of the business object in the first type of business cycle based on the heterogeneous conversion features corresponding to the first type of sub-cycle.

[0039] The W business sub-cycles also include a second type of sub-cycle that is different from the first type of sub-cycle. The second type of sub-cycle is a business sub-cycle in which no payment transaction occurs for the business object.

[0040] Wherein, the second heterogeneous feature sequence unit includes: a sub-period feature filling unit;

[0041] Sub-period feature filling unit, used for:

[0042] Obtaining a sub-period filling feature associated with the second type of sub-period, and determining the sub-period filling feature as a heterogeneous conversion feature corresponding to the second type of sub-period;

[0043] Based on the heterogeneous conversion characteristics corresponding to the first type of sub-period and the heterogeneous conversion characteristics corresponding to the second type of sub-period, a heterogeneous characteristic data sequence of the business object in the first type of business period is determined.

[0044] Wherein, the second heterogeneous feature sequence unit includes: a sub-period feature arrangement unit;

[0045] The sub-period feature arrangement unit is specifically used for:

[0046] Obtaining the sub-cycle time sequence of the W service sub-cycles, determining the sub-cycle time sequence of the first type of sub-cycle as the first sub-cycle time sequence, and determining the sub-cycle time sequence of the second type of sub-cycle as the second sub-cycle time sequence;

[0047] Based on the first sub-cycle time sequence and the second sub-cycle time sequence, the heterogeneous conversion features corresponding to the first type of sub-cycle and the heterogeneous conversion features corresponding to the second type of sub-cycle are used to determine the heterogeneous feature data sequence of the business object in the first type of business cycle.

[0048] The payment cycle includes N business cycles, and the sequence features in the payment business feature sequence include the periodic transaction features corresponding to each business cycle in the N business cycles;

[0049] Among them, the object recognition module includes: a feature aggregation unit and a feature recognition unit;

[0050] a feature aggregation unit configured to perform feature aggregation processing based on the periodic transaction features corresponding to the N business periods by the second hierarchical network to obtain object payment features of the business object;

[0051] The feature recognition unit is used to perform object recognition on the object type of the business object based on the object payment feature by the second layered network to obtain an object recognition result.

[0052] Wherein, the object type of the business object includes a first object type;

[0053] Wherein, the business data processing device further includes: a payment management module;

[0054] The payment management module is used to perform object payment management on the business object when the object identification result indicates that the business object is of the first object type.

[0055] The business data processing device further includes: a sample data acquisition module, a sample identification module, and a parameter adjustment module;

[0056] A sample data acquisition module is used to acquire sample data for training the initial business model; the sample data is determined based on sample labels and sample payment data, where the sample payment data refers to payment data in a sample transaction sequence of a sample object within a sample payment cycle; the sample label is used to indicate the reference object type of the sample object;

[0057] The sample identification module is used to input the sample data into the initial business model, and the initial business model performs object identification on the object type of the sample object to obtain a sample identification result; the sample identification result is used to indicate the identification object type of the sample object;

[0058] The parameter adjustment module is used to perform parameter adjustment processing on the model parameters of the initial business model based on the reference object type and the identification object type, and determine the target business model based on the initial business model after the parameter adjustment processing.

[0059] The description dimension information includes first description information and second description information;

[0060] The description conversion module includes: a word segmentation processing unit;

[0061] The word segmentation processing unit is specifically used for:

[0062] Obtaining a first text length threshold associated with the first description information and a second text length threshold associated with the second description information;

[0063] Based on a first text length threshold, the first description information is subjected to length normalization processing to obtain a first standard text corresponding to the first description information;

[0064] Based on the second text length threshold, the second description information is subjected to length normalization processing to obtain a second standard text corresponding to the second description information;

[0065] The first hierarchical network performs word segmentation processing on the first standard text and the second standard text to obtain text word segmentation describing class dimension information.

[0066] On the one hand, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. The computer program is suitable for being loaded and executed by a processor so that a computer device having the processor executes the method provided by the embodiment of the present application.

[0067] On one hand, an embodiment of the present application provides a computer program product, characterized in that it includes a computer program / instruction, and when the computer program / instruction is executed by a processor, the method provided by the embodiment of the present application is implemented.

[0068] By adopting the embodiments of the present application, the transaction sequence within the payment cycle can be obtained, and the payment cycle can be divided into finer granularity, such as obtaining a first type of business cycle by division, and then the corresponding features can be determined based on the information of the payment transactions in the first type of business cycle in multiple dimensions, that is, the time mapping feature is determined based on the time dimension information, the numerical mapping feature is determined based on the numerical dimension information, and the description conversion feature is determined based on the description dimension information, and then the features of each dimension can be heterogeneously converted to further obtain the heterogeneous feature data sequence corresponding to the first type of business cycle, so that the heterogeneous feature data sequence of the payment transactions within the fine-grained business cycle can be obtained, which is used to characterize the payment behavior characteristics of the business object within a fine-grained time period. In this way, the information on multiple dimensions of the payment transaction can be effectively fused, which can better characterize the user's transaction behavior, and the payment transactions within the payment cycle can be divided into finer granularity, so that more abundant potential laws of transaction behavior can be excavated, which helps to improve the accuracy of identifying the object type of the business object. Furthermore, based on the heterogeneous feature data sequence corresponding to the fine-grained business cycle, the payment business feature sequence corresponding to the entire payment cycle can be determined to characterize the characteristics of the payment behavior of the business object throughout the payment cycle. This can help to explore and analyze the operational correlation between the business object when performing different transaction operations within the fine-grained time, thereby improving the accuracy of data analysis of the business object's transaction operations, and helping to improve the accuracy of identifying the object type of the business object. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0070] Figure 1 This is a structural diagram of a data processing system provided in an embodiment of the present application;

[0071] Figure 2 This is a scenario diagram of a business data processing method provided by an embodiment of the present application;

[0072] Figure 3 This is a flowchart of a business data processing method provided by an embodiment of the present application;

[0073] Figure 4 This is a schematic diagram of determining a heterogeneous feature data sequence provided by an embodiment of the present application;

[0074] Figure 5 This is another schematic diagram of determining an isomerization conversion characteristic sequence provided by an embodiment of the present application;

[0075] Figure 6 This is a schematic diagram of determining a payment service feature sequence provided by an embodiment of the present application;

[0076] Figure 7 This is a schematic diagram of an object type identification process provided by an embodiment of the present application;

[0077] Figure 8 This is a flowchart of a business data processing method provided by an embodiment of the present application;

[0078] Figure 9 This is a flowchart of a business model training process provided by an embodiment of the present application;

[0079] Figure 10 This is a schematic diagram of the structure of a business data processing device provided in an embodiment of the present application;

[0080] Figure 11 It is a structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0081] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0082] See Figure 1 , Figure 1 This is a structural diagram of a data processing system provided in an embodiment of the present application. Figure 1 As shown, the data processing system may include terminal devices (such as device 11a, device 12a, device 13a) and server 100a. It is understood that Figure 1The number of terminal devices and servers in the example is merely illustrative; any number of terminal devices and servers may be used depending on implementation requirements. Terminal devices (e.g., device 11a, device 12a, and device 13a) may communicate with servers via a network (i.e., a medium providing a communication link via a wired or wireless communication link or fiber optic cable, etc.) to transmit data.

[0083] It is understood that a client can be running on a terminal device (such as device 12a), and the client can be a program that provides local services to a user (also known as a business object or operation object). Server 100a can be the server corresponding to the client, and the server 100a can run a program for providing resources, service data, and other services to the client. It is understood that the client running on the terminal device can also be called an application client, a business client, etc. For example, the client running on the terminal device can be a client for providing payment services, and the user can then initiate a payment transaction (also known as a payment service) on the terminal device.

[0084] It is understood that terminal devices (such as device 12a) may include, but are not limited to, mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, aircraft, smart speakers, smart home appliances, etc., and are not limited here. The server 100a can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms, and are not limited here.

[0085] It is understandable that the embodiments of the present application can be applied to a computer device, which can be the above-mentioned Figure 1 Servers in the Figure 1 A data processing device that processes business data within a data processing system in a data processing system. The data processing device can be a server or a terminal device, wherein the terminal device may include but is not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, aircraft, smart speakers, smart home appliances, etc., without limitation here. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms, without limitation here.

[0086] See Figure 2 , Figure 2 This is a scenario diagram of a business data processing method provided by an embodiment of the present application. Figure 2 As shown, the transaction sequence (also called business sequence) of a business object (such as a user initiating a payment transaction) within a certain period of time (i.e., a payment cycle) can be obtained. The payment cycle can be divided into one or more business cycles. Here, a payment cycle including a payment transaction initiated by a business object is taken as an example. For example, business cycle T1 (such as Figure 2 Taking 201a in FIG. 1 as an example, the process of identifying the object type of a business object based on a payment transaction initiated by the business object is described.

[0087] The business object initiates a payment transaction in the business cycle T1, which is represented by payment transaction Y1 (e.g. Figure 2 As shown in 202a in the example, the time dimension information corresponding to the payment transaction Y1 can be determined based on the payment transaction of the payment transaction Y1 (such as Figure 2 203a in ), numerical dimension information (such as Figure 2 204a in the figure) and description class dimension information (such as Figure 2 (as shown in 205a in the figure). The time dimension information may include the transaction time of the payment transaction within the business cycle T1. For example, if the business cycle T1 is one day, the time dimension information may be the hour within the day. The numerical dimension information may be used to indicate the transaction value of the payment transaction, such as the transaction amount, number of transactions, and other information, which are not limited here. The descriptive dimension information may be used to describe the textual information corresponding to the transaction content of the payment transaction, such as information about the counterparty and the product of the transaction.

[0088] Furthermore, payment transactions initiated by business objects can be processed using target business model 200a. Target business model 200a can include a first hierarchical network 206a and a second hierarchical network 208a. The first hierarchical network can be a network for determining characteristics corresponding to payment transactions within a business cycle, and the second hierarchical network can be a network for identifying object types based on the characteristics corresponding to each business cycle.

[0089] Taking a payment transaction Y1 as an example, the time dimension information 203a can be input into the first hierarchical network 206a for feature mapping processing to obtain the time mapping feature 261a; the value dimension information 204a can be input into the first hierarchical network 206a for feature mapping processing to obtain the value mapping feature 262a; the value dimension information 205a can be input into the first hierarchical network 206a for feature conversion processing to obtain the description conversion feature 263a. Furthermore, based on the time mapping feature, value mapping feature and description conversion feature corresponding to the payment transaction Y1, the heterogeneous feature data sequence corresponding to the business cycle T1 (such as Figure 2 264a in FIG. It is understood that the heterogeneous feature data sequence may be a sequence corresponding to the features of payment transactions within a business cycle, i.e., a sequence determined by heterogeneous conversion features, used to characterize the underlying patterns of payment behavior of a business object within business cycle T1. The heterogeneous conversion features may be determined by the time mapping features, value mapping features, and description conversion features corresponding to a payment transaction (e.g., payment transaction Y1).

[0090] Furthermore, the payment service feature sequence corresponding to the service cycle T1 can be determined based on the heterogeneous feature data sequence 264a (eg Figure 2 (as shown in 207a in the figure). The payment service feature sequence may be a feature sequence corresponding to payment transactions within a payment cycle, used to characterize the underlying patterns of payment behavior of the business object within the payment cycle. It is understood that if a payment cycle includes multiple business cycles, a corresponding heterogeneous feature data sequence may be determined for each business cycle. Furthermore, based on the corresponding heterogeneous feature data sequence determined for each business cycle, the corresponding periodic transaction features of each business cycle may be determined, thereby determining a payment service feature sequence based on the corresponding periodic transaction features of each business cycle.

[0091] Furthermore, the payment service feature sequence 207a can be processed by the second hierarchical network 208a in the target service model to perform object recognition processing on the object type of the service object, thereby obtaining an object recognition result 209a. It is understood that the object recognition result 209a can be used to indicate the object type of the service object.

[0092] By adopting the embodiments of the present application, the transaction sequence within the payment cycle can be obtained, and the payment cycle can be divided into finer granularity, such as obtaining a first type of business cycle by division, and then the corresponding features can be determined based on the information of the payment transactions in the first type of business cycle in multiple dimensions, that is, the time mapping feature is determined based on the time dimension information, the numerical mapping feature is determined based on the numerical dimension information, and the description conversion feature is determined based on the description dimension information, and then the features of each dimension can be heterogeneously converted to further obtain the heterogeneous feature data sequence corresponding to the first type of business cycle, so that the heterogeneous feature data sequence of the payment transactions within the fine-grained business cycle can be obtained, which is used to characterize the payment behavior characteristics of the business object within a fine-grained time period. In this way, the information on multiple dimensions of the payment transaction can be effectively fused, which can better characterize the user's transaction behavior, and the payment transactions within the payment cycle can be divided into finer granularity, so that more abundant potential laws of transaction behavior can be excavated, which helps to improve the accuracy of identifying the object type of the business object. Furthermore, based on the heterogeneous feature data sequence corresponding to the fine-grained business cycle, the payment business feature sequence corresponding to the entire payment cycle can be determined to characterize the characteristics of the payment behavior of the business object throughout the payment cycle. This can help to explore and analyze the operational correlation between the business object when performing different transaction operations within the fine-grained time, thereby improving the accuracy of data analysis of the business object's transaction operations, and helping to improve the accuracy of identifying the object type of the business object.

[0093] It is understandable that by adopting the embodiment of the present application, the features of various heterogeneous data can be extracted through a deep neural network (i.e., the target business model), and effective feature fusion is performed to better characterize the user's transaction behavior. In addition, the present application first aggregates the features of transactions according to the business cycle (such as one day) through the structure of a hierarchical transformer, and then learns the feature time series representation (i.e., payment business feature sequence) between multiple business cycles (such as multiple days). This can well alleviate the problem of single sequence learning difficulties caused by the user's historical transaction sequence being too long, and can better extract the feature time series characteristics of business objects in a longer cycle (i.e., payment cycle), thereby improving the accuracy of identifying the object type of the business object. It can be understood that the target business model involved in the embodiments of the present application can be a hierarchical time series classification model for processing multiple heterogeneous data such as time series, text and numerical values. The text, numerical and time features of transactions within a single business cycle (such as one day) are extracted respectively, and effectively integrated to obtain an effective expression of the single-day features (that is, the sequence features in the payment business feature sequence, that is, the business cycle features). Then, the time series features of multiple business cycles (such as multiple days) are aggregated through the hierarchical transformer structure, and the potential risks and default probabilities of business objects are predicted end-to-end through learning, which effectively improves the effectiveness of the financial risk control model.

[0094] It is understood that the embodiments of the present application can be applied in a variety of scenarios. For example, when applied to the financial sector, the object type of a business object can be identified to determine whether the business object is a risky object, thereby effectively controlling the risk of the business object. Accurate risk control is the lifeline for ensuring the security of many types of financial products, and the effectiveness of risk control directly depends on the effectiveness of the features. It is understandable that online payments are now an important part of life. Whether shopping, dining, or traveling, these payment behaviors record the important consumption habits of business users and reflect their spending power and credit status. Therefore, by analyzing the payment transactions initiated by business users, it is possible to determine whether the business object is a risky object, thereby improving the effectiveness of financial risk control. Among them, payment transaction data is a typical heterogeneous data, which contains multiple data formats such as text, time series, and numerical values. Text data, through merchant and product descriptions, can provide the most granular information about consumption scenarios and purchase intentions. Time series data, by recording the time of user payments, can reveal users' consumption habits and behavior patterns. The numerical data reflects the user's consumption capacity and financial status through the consumption amount, thereby identifying whether the business object has a certain repayment ability, that is, identifying the object type of the business object as an object type with repayment ability or an object type without repayment ability. Therefore, the object type of the business object can be identified by using the payment transaction data of the business object, so that the object type of the business object can be identified with a high-precision business model using the embodiment of the present application. For example, the embodiment of the present application can be used to detect the security of its own payment account, and by analyzing the payment behavior of the business object, it can be determined whether the payment account of the business object is at risk (such as whether it is illegally initiated by other illegal objects), that is, the object type of the business object can be identified as a type in which the payment account is at risk or a type in which there is no risk.

[0095] It is understood that the embodiments of the present application can be applied to the field of artificial intelligence technology. For example, target business models for object recognition of business object types can be obtained through artificial intelligence training. Artificial intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to achieve optimal results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. Artificial intelligence is the study of the design principles and implementation methods of various intelligent machines, giving them the capabilities of perception, reasoning, and decision-making. Artificial intelligence technology is an interdisciplinary subject covering a wide range of fields, including both hardware-level and software-level technologies. Basic AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, pre-trained model technology, operation / interaction systems, and mechatronics. Pre-trained models, also known as large models or basic models, can be widely applied to downstream tasks in various AI fields after fine-tuning. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0096] It is understandable that it needs to be explained that this application can display a prompt interface, pop-up window or output voice prompt information before collecting the user's relevant data (such as the transaction data of the payment transaction initiated by the business object) and during the process of collecting the user's relevant data. The prompt interface, pop-up window or voice prompt information is used to remind the user that its relevant data is currently being collected, so that this application only starts to execute the relevant steps of obtaining the user's relevant data after obtaining the user's confirmation operation on the prompt interface or pop-up window. Otherwise (that is, when the user's confirmation operation on the prompt interface or pop-up window is not obtained), the relevant steps of obtaining the user's relevant data are terminated, that is, the user's relevant data is not obtained. In other words, all user data collected by this application are collected with the user's consent and authorization, and the collection, use and processing of relevant user data need to comply with the relevant laws, regulations and standards of the relevant region.

[0097] Further, see Figure 3 , Figure 3 This is a flow chart of a business data processing method provided by an embodiment of the present application. The method can be executed by a computer device, for example, the computer device is the above-mentioned Figure 1 The method may include at least the following steps S101 to S106.

[0098] S101. Obtain payment transaction data for input into a target business model from a transaction sequence of a business object within a payment cycle; the target business model includes a first hierarchical network and a second hierarchical network; the payment transaction data is determined based on payment transactions initiated by the business object within a first type of business cycle; the payment transaction data includes time dimension information, numerical dimension information, and descriptive dimension information; the first type of business cycle is a business cycle within a payment cycle.

[0099] The payment cycle can be the time period for obtaining payment transactions when performing object type identification. For example, the payment cycle can be 1 month (or 2 months, 6 months, etc., which are not limited here). Then, the payment transactions initiated by the business object within 1 month can be obtained to perform object type identification based on the payment transactions initiated within 1 month. It is understandable that the payment cycle can include one or more business cycles, and the business cycle can be a time period divided within the payment cycle. It is understandable that the cycle size of each business cycle included in the payment cycle is the same. For example, a business cycle can be 1 day, 2 days, etc., which are not limited here.

[0100] It is understandable that the business cycles included in the payment cycle may include a first type of business cycle, that is, the first type of business cycle is a business cycle in the payment cycle. The first type of business cycle can be a business cycle in which a business object initiates a payment transaction. In other words, the first type of business cycle can be a business cycle in which a payment transaction initiated by a business object exists. Optionally, the payment cycle may also include a second type of business cycle, and the second type of business cycle can be a business cycle in which a business object does not initiate a payment transaction. In other words, the second type of business cycle can be a business cycle in which a payment transaction initiated by a business object does not exist. It is understandable that within the payment cycle, there may be a second type of business cycle (that is, the payment cycle includes the first type of business cycle and the second type of business cycle), or there may not be a second type of business cycle (that is, the payment cycle only includes the first type of business cycle), and this is not limited here. If there is no second type of business cycle in the payment cycle, it means that the business object has initiated a payment transaction in each business cycle in the payment cycle.

[0101] It is understood that a transaction sequence can be a sequence of payment transactions initiated by a business object within a payment cycle, and can also be referred to as a business sequence. In other words, the transaction sequence can include at least one payment transaction, and the transaction time of at least one payment transaction in the transaction sequence falls within the payment cycle. It is understood that the transaction sequence can be arranged in chronological order of transaction time, such as from earliest to latest transaction time, or from latest to earliest transaction time, without limitation here. It is understood that in a transaction sequence, each transaction can be defined according to a specific data format. For example, each payment transaction in a transaction sequence can be represented as <user ID (i.e., object identifier of the business object), transaction time, merchant name, product name, transaction amount>, where the transaction time can be represented as <month, day of the month, week, day of the week, hour>.

[0102] Among them, the payment transaction data can be determined based on the payment transactions initiated by the business object within the first type of business cycle, and can also be called payment business data. In other words, the payment transaction data can be the transaction data of the payment transactions initiated by the business object within the first type of business cycle (can also be called the payment transactions within the first type of business cycle). It should be understood that the payment transaction data can be the transaction data of the payment transactions within a first type of business cycle, and the transaction data of the payment transactions within each first type of business cycle can be called payment transaction data. It can be understood that the number of payment transactions within a first type of business cycle (that is, the number of payment transactions initiated by the business object within a first type of business cycle) can be one or more, which is not limited here. It should be understood that the payment transaction corresponding to the payment transaction data can be all or part of the payment transactions in the transaction sequence corresponding to the payment cycle, which can be specifically determined based on the first type of business cycle in which the transaction time of the payment transaction is located, which is not limited here.

[0103] Payment transaction data includes time-related dimension information, numerical dimension information, and descriptive dimension information. The time-related dimension information can be used to indicate the transaction time of the payment transaction, such as the month, day, week, day, and hour of the month. The numerical dimension information can be used to indicate the transaction value of the payment transaction, such as the transaction amount and number of transactions, which are not limited here. The number of transactions here can be the number of payment transactions initiated against the counterparty. For example, if a payment transaction is the fifth payment made by a business subject to merchant A, the number of transactions is 5. The descriptive dimension information can be used to describe the textual information corresponding to the transaction content of the payment transaction, such as counterparty information and product information. The counterparty information can be the merchant name when the business subject purchases a product (e.g., a financial product, insurance product, daily necessities, clothing product, etc.), or the name of the transfer recipient when the business subject makes a transfer, which are not limited here. The product information of the transaction can be an introduction to the product when the business object purchases the product (it can also be information such as the title, detailed description, etc.), or it can be information such as the transfer notes when the business object transfers money. In addition, the product information can also be a product description text (i.e., text used to describe the product) obtained by voice recognition or image recognition based on a description video of the product. The description video can be a video used to describe the product, such as a merchant’s introduction video for the product, or a live screen recording video that introduces the product during a live broadcast, etc., which is not limited here. Alternatively, the product information can also be a product description text obtained by image recognition based on a description image. The description image can be text used to describe the product, such as an image used to introduce the product on a product introduction page.

[0104] It is understood that the target business model can be a model used to identify the object type of a business object. It is understood that the target business model can include a first hierarchical network and a second hierarchical network, wherein the first hierarchical network can be a network used to determine the characteristics corresponding to the payment transactions of each first-category business cycle, and the second hierarchical network can be a network that identifies the object type based on the characteristics corresponding to each business cycle. It is understood that the target business model can be obtained by training an initial business model based on a large amount of sample data. The target business model can be trained in an end-to-end manner, so that the target business model can directly identify the object type of the business object based on the payment transaction data within the payment cycle.

[0105] S102: Input the time dimension information into the first hierarchical network, and perform feature mapping processing on the time dimension information by the first hierarchical network to obtain time mapping features corresponding to the time dimension information.

[0106] The feature mapping process may be a process of mapping information (e.g., time-related dimension information) into corresponding features. The time mapping feature may be a feature obtained by performing feature mapping on the time-related dimension information. The time mapping feature may be represented as a vector or matrix of a certain dimension. For example, the dimension of the time mapping feature may be a K-dimensional feature vector.

[0107] It is understood that the feature mapping processing of the time dimension information can be performed by a time processing subnetwork in the first hierarchical network for processing the time dimension. The time processing subnetwork can be a subnetwork for performing feature mapping processing on the time dimension information. For example, the time processing subnetwork can be a 1*K fully connected network, so that the time dimension information (such as the number of hours in the transaction time of a payment transaction) is processed based on the fully connected network to obtain a K-dimensional numerical feature representation, namely the time mapping feature.

[0108] S103: Input the numerical dimension information into the first hierarchical network, and perform feature mapping processing on the numerical dimension information by the first hierarchical network to obtain numerical mapping features corresponding to the numerical dimension information.

[0109] Among them, the feature mapping processing can refer to the above description, and then the feature mapping processing can be performed on the numerical dimension information to obtain the numerical mapping feature corresponding to the numerical dimension information. The numerical mapping feature can be a feature obtained by performing feature mapping processing on the numerical dimension information. The numerical mapping feature can be represented as a vector or matrix of a certain dimension. For example, the dimension of the numerical mapping feature can be a K-dimensional feature vector. It can be understood that the feature dimension of the numerical mapping feature can be the same as or different from the feature dimension of the time mapping feature, and this is not limited here.

[0110] It is understood that the feature mapping processing of the numerical dimension information can be performed by a numerical processing subnetwork in the first hierarchical network for processing numerical dimensions. The numerical processing subnetwork can be a subnetwork for performing feature mapping processing on numerical dimension information. For example, the numerical processing subnetwork can be a 2*K fully connected network, so that the numerical dimension information (such as the transaction amount and transaction number of the payment transaction) is processed based on the fully connected network to obtain a K-dimensional numerical feature representation, i.e., the numerical mapping feature.

[0111] Optionally, before the numerical dimension information is processed by the numerical processing sub-network, the numerical dimension information can be normalized first. For example, if the numerical dimension information includes <transaction amount, number of transactions>, it can be normalized by mapping through a normalization function (such as a log function) to obtain normalized numerical dimension information (such as mapping both the transaction amount and the number of transactions to the range of 0-1), and then the normalized numerical dimension information is subjected to feature mapping processing by the numerical processing sub-network to obtain the numerical mapping features corresponding to the numerical dimension information. This can improve the standardization of the numerical values ​​so that the numerical processing sub-network can perform feature mapping processing more quickly, reduce the time consumed by feature mapping processing of the numerical dimension information, and thus improve the processing efficiency of the numerical mapping.

[0112] S104. Input the description class dimension information into the first hierarchical network. The first hierarchical network performs word segmentation processing on the description class dimension information to obtain text segmentation of the description class dimension information. The text segmentation is subjected to feature conversion processing to obtain description conversion features corresponding to the description class dimension information.

[0113] The description of the descriptive dimension information can refer to the above-mentioned related description and is not repeated here. It should be understood that the descriptive dimension information can be text information of the descriptive class associated with the payment transaction, and then the descriptive dimension information can be segmented to obtain the text segmentation of the descriptive dimension information, and the description conversion features corresponding to the descriptive dimension information can be determined based on the text segmentation.

[0114] The word segmentation process can be the process of segmenting a continuous text sequence (i.e., the descriptive dimension information) into textual segments according to certain specifications. The textual segmentations can be segmented into segments obtained by segmenting the text indicated by the descriptive dimension information. It is understood that the word segmentation process can be performed in a variety of ways, such as a dictionary-based approach, a deep learning-based approach, and so on, without limitation here. Here, a dictionary-based approach is used as an example to illustrate the process of performing word segmentation on the descriptive dimension information. A trained dictionary for word segmentation can be obtained. Specifically, the following method is used: A large amount of text containing merchant names and product names from user transactions that does not require annotation is obtained. A desired dictionary size R is defined (for example, R = 20,000 can be set by default, or adjusted based on the business needs). The text is then input into a common word segmentation model or tool (such as Sentencepiece) to obtain a word segmentation dictionary of size R. This word segmentation dictionary can then be used to segment the text indicated by the descriptive dimension information, obtaining the textual segmentations of the descriptive dimension information. For example, if the description dimension information is "wool double-sided long coat", the text segmentation obtained by segmenting the description dimension information can be: "wool", "double-sided", "wool", "long", and "coat".

[0115] Among them, the feature conversion process can be a process for converting the text segmentation of the description class dimension information into corresponding features. The description conversion feature can be a feature converted from the text segmentation of the description class dimension information, and the description conversion feature can be represented as a vector or matrix of a certain dimension. It is understandable that the feature dimension of the description conversion feature can be the same as or different from the feature dimension of the time mapping feature (or numerical mapping feature), and this is not limited here.

[0116] It can be understood that the feature conversion processing of the text segmentation of the description class dimension information can be processed by the description processing subnetwork in the first hierarchical network for processing the text segmentation of the description class dimension. The description processing subnetwork can be a subnetwork for performing feature mapping processing on the description class dimension information. For example, the description processing subnetwork can include a fully connected feature mapping layer and a transformer layer (a neural network), so that the text segmentation obtained above can be input into the fully connected feature mapping layer to obtain a mapping feature representation of F*D dimensions; further adopt the position embedding (a position representation mapping method) in Bert (a neural network) to obtain the position feature representation of F*D dimensions corresponding to the description class dimension information; the two features (i.e., the above-mentioned F*D-dimensional mapping feature representation and the F*D-dimensional position feature representation) are bit-wise added and input into the transformer layer, and the output of the transformer layer is averaged (a pooling method) to obtain the embedded feature representation of the text indicated by the entire description class dimension information, i.e., the description conversion feature.

[0117] It is understandable that before the description processing sub-network processes the text segmentation of the description-class dimensional information, the length of the text segmentation of the description-class dimensional information can also be normalized, and then the text segmentation after the length normalization process is subjected to feature conversion processing to obtain the description conversion feature corresponding to the description-class dimensional information, that is, the text segmentation after the length normalization process is input into the description processing sub-network for feature conversion processing to obtain the description conversion feature corresponding to the description-class dimensional information. The length normalization process can be performed to adjust the text length of the description-class dimensional information to a fixed length. Specifically, if the text length of the description-class dimensional information is greater than the fixed length, the text of the description-class dimensional information is truncated so that the length of the truncated text of the description-class dimensional information is equal to the fixed length. For example, the description-like dimension information is "wool coat, double-sided, long, houndstooth", and its text length is 11. If the fixed length required for length standardization is 10, then 1 word can be truncated from the description-like dimension information to obtain the length-standardized description-like dimension information: "wool coat, double-sided, long, houndstooth"; if the text length of the description-like dimension information is less than the fixed length, the position that is less than the fixed length can be padded (such as by 0, also known as 0 padding), so that the length of the text of the description-like dimension information after padding is equal to the fixed length. For example, the description-like dimension information is "wool coat, double-sided, long", and its text length is 8. If the fixed length required for length standardization is 10, then 2 0s can be padded into the description-like dimension information to obtain the length-standardized description-like dimension information: "wool coat, double-sided, long, 00".

[0118] It is understood that if the descriptive dimension information includes one type of descriptive information, such as a commodity name or a product name, then only the text length of that one type of descriptive information can be length-normalized. The descriptive information after length-normalization of the text length of that one type of descriptive information is the length-normalized descriptive dimension information. If the descriptive dimension information includes multiple types of descriptive information, such as a commodity name and a product name, then the text lengths of each type of descriptive information can be length-normalized separately, and then the length-normalized descriptive dimension information is determined based on the multiple types of descriptive information that have undergone length normalization.

[0119] Specifically, the description-class dimension information includes first description information and second description information; the first hierarchical network performs word segmentation processing on the description-class dimension information to obtain text segmentation of the description-class dimension information, which may include the following steps: obtaining a first text length threshold associated with the first description information and a second text length threshold associated with the second description information; based on the first text length threshold, performing length standardization processing on the first description information to obtain a first standard text corresponding to the first description information; based on the second text length threshold, performing length standardization processing on the second description information to obtain a second standard text corresponding to the second description information; the first hierarchical network performs word segmentation processing on the first standard text and the second standard text to obtain text segmentation of the description-class dimension information.

[0120] The first description information and the second description information may be different types of description information included in the description dimension information. For example, the first description information may be the merchant name associated with the payment transaction, and the second description information may be the product name associated with the payment transaction.

[0121] It is understandable that the first text length threshold can be a fixed length obtained by standardizing the length of the first description information. The second text length threshold can be a fixed length obtained by standardizing the length of the second description information. The first text length threshold and the second text length threshold can be the same or different, and are not limited here. The first text length threshold and the second text length threshold can be thresholded based on empirical values. If the first description information is a product name, a large number of product names can be obtained, and then an average text length can be determined based on the text lengths of the large number of product names, and the average text length is determined as the first text length threshold.

[0122] The first standard text may be a text obtained by length-normalizing the first description information. The second standard text may be a text obtained by length-normalizing the second description information. It should be understood that the process of length-normalizing the first description information and the second description information can refer to the above-mentioned related description and is not further described here.

[0123] It can be understood that the first hierarchical network performs word segmentation processing on the first standard text and the second standard text to obtain text word segmentations describing the class dimension information. The word segmentation processing can be performed based on the first standard text to obtain the first text word segmentation corresponding to the first standard text, and the word segmentation processing can be performed based on the second standard text to obtain the second text word segmentation corresponding to the second standard text. The first text word segmentation and the second text word segmentation are determined as text word segmentations describing the class dimension information. For example, the descriptive dimension information may include two types of descriptive information: merchant name and product name. For example, if the first descriptive information is the merchant name and the second descriptive information is the product name, then the merchant name may be truncated to a text with a length of L1 (if it is less than L1, it will be padded with 0), and the product name may be truncated to a text with a length of L2 (if it is less than L2, it will be padded with 0). L1 (i.e., the first text length threshold mentioned above) and L2 (i.e., the second text length threshold mentioned above) are both positive integers. L1 and L2 may be the same or different, and are not limited here. Then, the text with a length of L1 and the text with a length of L2 may be segmented. For example, by inputting the segmentation model corresponding to the segmentation dictionary mentioned above, a segmentation list with a length of L1+L2 is obtained, i.e., the text segmentation of the descriptive dimension information mentioned above.

[0124] It is understood that the first and second description information herein are merely exemplary descriptions. The description-type dimension information may also include more types of description information, such as third description information. The third description information may then be length-normalized. For details, please refer to the description of length normalization for the first and second description information above, which will not be elaborated here. The standardized third description information is then segmented to obtain corresponding segmented text, and the segmented text corresponding to the description-type dimension information is determined based on the segmented text corresponding to each type of description information.

[0125] S105: The first hierarchical network performs heterogeneous conversion processing on the time mapping feature, the value mapping feature, and the description conversion feature to obtain a heterogeneous feature data sequence of the business object within the first type of business cycle.

[0126] Among them, the heterogeneous conversion processing can be a process of converting time mapping features, numerical mapping features, and description conversion features to obtain heterogeneous conversion features for determining a heterogeneous feature data sequence. Among them, the heterogeneous feature data sequence can be a sequence corresponding to the features of payment transactions within the first type of business cycle, that is, a sequence determined by the heterogeneous conversion features, which is used to characterize the potential patterns of the payment behavior of the business object within the first type of business cycle. The heterogeneous conversion feature is determined based on the payment transactions of the business object within the first type of business cycle, and specifically can be a feature obtained by converting the time mapping features, numerical mapping features, and description conversion features corresponding to the payment transactions within the first type of business cycle. It can be understood that in the heterogeneous feature data sequence, at least one heterogeneous conversion feature can be included.

[0127] Optionally, for each payment transaction within the first type of business cycle, the corresponding heterogeneous conversion features can be determined. Specifically, the payment transaction features corresponding to each payment transaction (i.e., the collective term for time mapping features, numerical mapping features, and description conversion features) can be subjected to feature fusion processing to obtain the heterogeneous conversion features corresponding to each payment transaction, and then the heterogeneous feature data sequence can be determined based on the heterogeneous conversion features corresponding to each payment transaction.

[0128] Among them, the number of heterogeneous conversion features in the heterogeneous feature data sequence (also known as sequence length) can be determined by the number of payment transactions within the first type of business cycle, that is, the number of heterogeneous conversion features in the heterogeneous feature data sequence is consistent with the number of payment transactions within the first type of business cycle.

[0129] Alternatively, the number of heterogeneous conversion features in the heterogeneous feature data sequence (also known as the sequence length) may also be a preset number. That is, if the number of payment transactions within the first type of business cycle is greater than the preset number, some payment transactions may be eliminated from the payment transactions within the first type of business cycle, so as to determine the heterogeneous feature data sequence based on the heterogeneous conversion features corresponding to the remaining payment transactions after eliminating some payment transactions, thereby making the sequence length of the heterogeneous feature data sequence a preset number. The elimination of sub-payment transactions here may be based on the transaction amount involved in the payment transaction, such as giving priority to eliminating payment transactions with small transaction amounts, in other words, retaining the payment transactions corresponding to the preset number with the largest transaction amount, or, giving priority to eliminating transactions with later transaction times, that is, retaining the payment transactions corresponding to the preset number with the earliest transaction time. If the number of payment transactions within the first business cycle is less than the preset number, a certain number of heterogeneous conversion features (i.e., the difference between the preset number and the number of payment transactions within the first business cycle) can be determined based on the transaction filling feature. A heterogeneous feature data sequence can be determined based on the heterogeneous conversion features determined by the transaction filling feature and the heterogeneous conversion features corresponding to the payment transactions, such that the sequence length of the heterogeneous feature data sequence is the preset number. The transaction filling feature can be a feature used to fill the heterogeneous feature data sequence, such as a zero vector or a zero matrix. The transaction filling feature has the same feature dimension as the heterogeneous conversion feature of a single payment transaction. If the required heterogeneous conversion feature is T-dimensional, the transaction filling feature can be a T-dimensional zero vector or a zero matrix. For example, if the preset number of heterogeneous feature data sequences corresponding to the heterogeneous feature is 30 and the number of payment transactions within the first business cycle is 35, the 30 payment transactions with the largest transaction amounts can be screened out from the 35 payment transactions based on their transaction amounts (i.e., excluding the five payment transactions with the smallest transaction amounts). The heterogeneous feature data sequence can then be determined based on the heterogeneous conversion features of the screened 30 payment transactions with the largest transaction amounts. For example, if the preset number corresponding to the heterogeneous feature data sequence is 30, and the number of payment transactions in the first type of business cycle is 25, then (30-25=5) heterogeneous conversion features can be determined based on the transaction filling features, and then the heterogeneous feature data sequence can be determined based on the heterogeneous conversion features of the 25 payment transactions and the 5 heterogeneous conversion features determined based on the transaction filling features.

[0130] Specifically, the number of payment transactions within the first type of business cycle is M, and the M payment transactions include payment transaction p, where p is a positive integer less than or equal to M; then, the first hierarchical network performs heterogeneous conversion processing on the time mapping feature, the numerical mapping feature, and the description conversion feature to obtain a heterogeneous feature data sequence of the business object within the first type of business cycle, which may include the following steps: the first hierarchical network performs heterogeneous conversion processing on the time mapping feature corresponding to the payment transaction p, the numerical mapping feature corresponding to the payment transaction p, and the description conversion feature corresponding to the payment transaction p to obtain a heterogeneous conversion feature corresponding to the payment transaction p, until the heterogeneous conversion feature of each payment transaction in the M payment transactions is obtained, and then, based on the obtained heterogeneous conversion feature of each payment transaction, the heterogeneous feature data sequence of the business object within the first type of business cycle is determined.

[0131] Wherein, M is the number of payment transactions in the first type of business cycle, and M is a positive integer. Payment transaction p can be any payment transaction among the M payment transactions, and p is a positive integer less than or equal to M.

[0132] It should be understood that the heterogeneous conversion processing is performed here on the time mapping feature corresponding to the payment transaction p, the numerical mapping feature corresponding to the payment transaction p, and the description conversion feature corresponding to the payment transaction p. The time mapping feature, the numerical mapping feature, and the description conversion feature corresponding to the payment transaction p can be connected in series, such as by concat (a series connection method).

[0133] It can be understood that the heterogeneous conversion characteristics corresponding to each payment transaction in the M payment transactions can be determined by referring to the determination method of the heterogeneous conversion characteristics corresponding to the payment transaction p, so that the heterogeneous conversion characteristics of each payment transaction in the M payment transactions can be obtained.

[0134] It should be understood that based on the heterogeneous conversion characteristics of each payment transaction obtained, the heterogeneous characteristic data sequence of the business object in the first business cycle is determined. The heterogeneous conversion characteristics of each payment transaction can be arranged in the transaction time sequence of the M payment transactions to obtain the heterogeneous characteristic data sequence of the business object in the first business cycle. For example, the data can be arranged in order from early to late according to the transaction time, or in order from late to early according to the transaction time. There is no limitation here.

[0135] For example, see Figure 4 , Figure 4 This is a schematic diagram of determining a heterogeneous feature data sequence provided by an embodiment of the present application. Figure 4As shown, within a first-category business cycle, a business object initiates multiple payment transactions, such as payment transaction 1, payment transaction 2, ..., payment transaction a, and so on. It is understandable that a corresponding heterogeneous conversion feature can be determined for each payment transaction. For example, the payment data corresponding to payment transaction 1 includes time dimension information 1, value dimension information 1, and description dimension information 1. Time mapping feature 1 can be determined based on time dimension information 1, value mapping feature 1 can be determined based on value dimension information 1, and description conversion feature 1 can be determined based on description dimension information 1. Consequently, heterogeneous conversion feature 2 corresponding to payment transaction 1 can be determined based on time mapping feature 1, value mapping feature 1, and description conversion feature 1. Referring to the process for determining heterogeneous conversion feature 1 corresponding to payment transaction 1, heterogeneous conversion feature 2 corresponding to payment transaction 2, ..., heterogeneous conversion feature a corresponding to payment transaction a, and so on, heterogeneous conversion feature a can be determined. Furthermore, a heterogeneous feature data sequence can be determined based on the determined heterogeneous conversion features corresponding to each payment transaction. The specific method can refer to the above-mentioned relevant description. If the number of heterogeneous conversion features is greater than a certain threshold, the heterogeneous conversion features of some payment transactions are eliminated. If it is less than a certain threshold, it is filled in to obtain a heterogeneous conversion feature sequence. It will not be elaborated here.

[0136] Optionally, the first-type business cycle can be divided into multiple sub-cycles (also called business sub-cycles), so that the corresponding heterogeneous conversion characteristics can be determined for each sub-cycle. It can be understood that the number of sub-cycles divided in the first-type business cycle is multiple, and the cycle length of each sub-cycle remains consistent. For example, if a first-type business cycle is 1 day, it can be divided into 24 hours within 1 day, and each hour is a sub-cycle. Among them, if the business object initiates a payment transaction in a sub-cycle, the corresponding heterogeneous conversion characteristics can be determined based on the payment transaction in the sub-cycle, that is, the heterogeneous conversion characteristics corresponding to a sub-cycle can be determined based on the payment transaction in the sub-cycle; if the business object does not initiate a payment transaction in a sub-cycle, the corresponding heterogeneous conversion characteristics in the sub-cycle can be determined based on certain filling features (such as a 0 vector or 0 matrix consistent with the feature dimension of the heterogeneous conversion characteristics). It can be understood that by dividing the first-class business cycle into sub-cycles, it is possible to avoid the difficulty of processing the business model due to the excessive number of payment transactions within a business cycle (that is, the sequence of transactions corresponding to a business cycle is too long), so that the length of the transaction characteristics corresponding to a first-class business cycle is consistent with the number of sub-cycles, which can better represent the characteristics.

[0137] Specifically, the first type of business cycle includes W business sub-cycles, where W is a positive integer; the W business sub-cycles include the first type of sub-cycle, and the first type of sub-cycle is a business sub-cycle in which the business object initiates a payment transaction; the payment transaction in the first type of sub-cycle is the first payment transaction; the first hierarchical network performs heterogeneous conversion processing on the time mapping feature, the numerical mapping feature, and the description conversion feature to obtain a heterogeneous feature data sequence of the business object in the first type of business cycle, which may include the following steps: when the first hierarchical network determines the time mapping feature, the numerical mapping feature, and the description conversion feature of the first payment transaction, heterogeneous conversion processing is performed based on the time mapping feature, the numerical mapping feature, and the description conversion feature of the first payment transaction to obtain a heterogeneous conversion feature corresponding to the first type of sub-cycle; based on the heterogeneous conversion feature corresponding to the first type of sub-cycle, the heterogeneous feature data sequence of the business object in the first type of business cycle is determined.

[0138] Where W is the number of business sub-cycles divided within the first-category business cycle, and W is a positive integer. The W business sub-cycles may include a first-category sub-cycle, which is a business sub-cycle in which a business object initiates a payment transaction. In other words, if a business object initiates a payment transaction within a business sub-cycle, that business sub-cycle may be identified as a first-category sub-cycle.

[0139] The first payment transaction may be a payment transaction initiated by the business object within the first sub-cycle. In other words, the first payment transaction is a payment transaction within the first sub-cycle, i.e., a payment transaction with a transaction time within the first sub-cycle. It is understood that the number of payment transactions within a first sub-cycle may be one or more, and this is not limited here.

[0140] It should be understood that the method for determining the time mapping characteristics of the first payment transaction, the numerical mapping characteristics of the first payment transaction, and the description conversion characteristics of the first payment transaction can refer to the description in the above steps S102-S104 and will not be repeated here.

[0141] Among them, if the number of first payment transactions within a first-type sub-period is single, heterogeneous conversion processing can be performed based on the time mapping characteristics of the first payment transaction, the numerical mapping characteristics of the first payment transaction, and the description conversion characteristics of the first payment transaction to obtain the heterogeneous conversion characteristics corresponding to the single first payment transaction. For details, please refer to the relevant description of determining the heterogeneous conversion characteristics corresponding to the payment transaction p above, which will not be repeated here. Then, the heterogeneous conversion characteristics corresponding to the single first payment transaction are determined as the heterogeneous conversion characteristics corresponding to the first-type sub-period.

[0142] If there are multiple first payment transactions within the first sub-period, the time mapping characteristics, numerical mapping characteristics and description conversion characteristics of each first payment transaction can be determined, and then the heterogeneous conversion characteristics corresponding to each first payment transaction can be determined based on the time mapping characteristics, numerical mapping characteristics and description conversion characteristics of each first payment transaction (for details, please refer to the relevant description of determining the heterogeneous conversion characteristics corresponding to the payment transaction p above, which will not be repeated here), so that the heterogeneous conversion characteristics corresponding to each first payment transaction can be subjected to feature fusion processing (such as averaging the heterogeneous conversion characteristics corresponding to each first payment transaction, or bitwise addition, etc., which are not limited here) to obtain the heterogeneous conversion characteristics corresponding to the first sub-period.

[0143] Alternatively, if there are multiple first payment transactions within a first sub-period, a first target payment transaction can be determined from the multiple first payment transactions, and then, based on the time mapping characteristics, value mapping characteristics, and description conversion characteristics of the first target payment transaction, the heterogeneous conversion characteristics corresponding to the first target payment transaction can be determined, and then the heterogeneous conversion characteristics corresponding to the first target payment transaction can be determined as the heterogeneous conversion characteristics corresponding to the first sub-period. It should be understood that the first target payment transaction can be the first payment transaction used to determine the heterogeneous conversion characteristics of a first sub-period. Optionally, determining the first target payment transaction can be by randomly selecting a first payment transaction from the multiple first payment transactions within the first sub-period as the first target payment transaction; or, determining the first target payment transaction can be by selecting the first payment transaction with the largest transaction amount from the multiple first payment transactions within the first sub-period as the first target payment transaction, or selecting the first payment transaction with the earliest transaction time as the first target payment transaction. Optionally, multiple transactions with the same transaction counterparty (such as the recipient of a transfer transaction, a merchant of a payment service, etc.) can be aggregated at the granularity of a sub-period (such as an hour) to obtain a transaction, such as counting the amounts of multiple transactions to obtain the total transaction amount, counting the number of transactions with the same transaction counterparty, etc., so that the aggregated single transaction can be represented as <user id (such as the object identifier of the business object), the sub-period (such as an hour) to which the aggregated transaction belongs, the transaction counterparty (such as the transaction merchant name), a list of transaction product names, the total transaction amount, the number of transactions>, and then the first target payment transaction can be determined based on each transaction obtained after aggregation, such as randomly determining the first target payment transaction from the transactions obtained after aggregation, or, from the transactions obtained after aggregation, determining the aggregated transaction with the largest total transaction amount (that is, the payment transaction obtained after aggregation) as the first target payment transaction, and determining the aggregated transaction with the most transactions here as the first target payment transaction.

[0144] It is understood that within a first-type business cycle, there may be one or more first-type sub-cycles. For each first-type sub-cycle, the corresponding heterogeneous conversion characteristics can be determined with reference to the above description. Furthermore, based on the heterogeneous conversion characteristics corresponding to each first-type sub-cycle, the heterogeneous characteristic data sequence of the business object within the first-type business cycle can be determined. It is understood that when determining the heterogeneous characteristic data sequence based on the heterogeneous conversion characteristics corresponding to each first-type sub-cycle, the sub-cycle sequence corresponding to each first-type sub-cycle can be obtained. Furthermore, each first-type sub-cycle can be arranged based on the sub-cycle sequence corresponding to each first-type sub-cycle, thereby obtaining the heterogeneous characteristic data sequence corresponding to the first-type business cycle.

[0145] Optionally, within the first type of business cycle, a second type of sub-cycle may also be included. The second type of sub-cycle is different from the first type of sub-cycle. The second type of sub-cycle is a business sub-cycle in which the business object does not generate a payment transaction. In other words, if the business does not generate a payment transaction within a business sub-cycle, the business sub-cycle can be determined as a second type of sub-cycle. It can be understood that within the first type of business cycle, there may be a second type of sub-cycle (that is, the first type of business cycle includes the first type of sub-cycle and the second type of sub-cycle), or there may not be a second type of sub-cycle (that is, the first type of business cycle only includes the first type of sub-cycle), which is not limited here. If there is no second type of business cycle within the payment cycle, it means that the business object has initiated a payment transaction in each business cycle within the payment cycle. Furthermore, since there is no corresponding payment transaction in the second sub-period, there is no heterogeneous conversion feature of the corresponding payment transaction, and the heterogeneous conversion feature corresponding to the second sub-period can be determined by filling, and then the heterogeneous feature data sequence corresponding to the first business cycle is determined based on the heterogeneous conversion feature corresponding to the first sub-period and the heterogeneous conversion feature corresponding to the second sub-period. In this way, the number of heterogeneous conversion features in the heterogeneous feature data sequence can be kept consistent with the number of business sub-periods in the first business cycle. Moreover, when training the target business model, the number of heterogeneous conversion features in the heterogeneous feature data sequence is also the same, so that the application of the model and the model training can be kept consistent, which helps to improve the accuracy of the target business model, that is, it helps the target business model to more accurately identify the type of business object.

[0146] Specifically, the W business sub-cycles also include a second type of sub-cycle that is different from the first type of sub-cycle, and the second type of sub-cycle is a business sub-cycle in which the business object does not generate a payment transaction; then, based on the heterogeneous conversion characteristics corresponding to the first type of sub-cycle, determining the heterogeneous feature data sequence of the business object in the first type of business cycle can include the following steps: obtaining the sub-cycle filling characteristics associated with the second type of sub-cycle, and determining the sub-cycle filling characteristics as the heterogeneous conversion characteristics corresponding to the second type of sub-cycle; based on the heterogeneous conversion characteristics corresponding to the first type of sub-cycle and the heterogeneous conversion characteristics corresponding to the second type of sub-cycle, determining the heterogeneous feature data sequence of the business object in the first type of business cycle.

[0147] The introduction to the second type of sub-period is described above and is not repeated here. The sub-period filling feature associated with the second type of sub-period can be a feature corresponding to the second type of sub-period used to fill the heterogeneous feature data sequence, such as a 0 vector or a 0 matrix. The sub-period filling feature is consistent with the feature dimension of the heterogeneous conversion feature corresponding to the first type of sub-period. For example, if the heterogeneous conversion feature corresponding to the first type of sub-period is S-dimensional, the sub-period filling feature corresponding to the second type of sub-period can be an S-dimensional 0 vector or 0 matrix.

[0148] It can be understood that when determining the heterogeneous characteristic data sequence of the business object within the first type of business cycle based on the heterogeneous conversion characteristics corresponding to the first type of sub-cycle and the heterogeneous conversion characteristics corresponding to the second type of sub-cycle, the sub-cycle sequence corresponding to the first type of sub-cycle (also called sub-cycle time sequence) and the sub-cycle sequence corresponding to the second type of sub-cycle (also called sub-cycle time sequence) can be arranged to obtain a heterogeneous characteristic data sequence.

[0149] Specifically, determining the heterogeneous feature data sequence of the business object within the first type of business cycle based on the heterogeneous conversion features corresponding to the first type of sub-cycle and the heterogeneous conversion features corresponding to the second type of sub-cycle can include the following steps: obtaining the sub-cycle time sequence of W business sub-cycles, determining the sub-cycle time sequence of the first type of sub-cycle as the first sub-cycle time sequence, and determining the sub-cycle time sequence of the second type of sub-cycle as the second sub-cycle time sequence; based on the first sub-cycle time sequence and the second sub-cycle time sequence, determining the heterogeneous feature data sequence of the business object within the first type of business cycle for the heterogeneous conversion features corresponding to the first type of sub-cycle and the heterogeneous conversion features corresponding to the second type of sub-cycle.

[0150] The sub-cycle time sequence can be used to indicate the time sequence of the service sub-cycles. The first sub-cycle time sequence is the sub-cycle time sequence of the first type of sub-cycles, and the second sub-cycle time sequence is the sub-cycle time sequence of the second type of sub-cycles.

[0151] It can be understood that, based on the time sequence of the first sub-cycle and the time sequence of the second sub-cycle, the heterogeneous conversion features corresponding to the first type of sub-cycle and the heterogeneous conversion features corresponding to the second type of sub-cycle can be arranged from early to late to obtain a heterogeneous feature data sequence, or the heterogeneous conversion features corresponding to the first type of sub-cycle and the heterogeneous conversion features corresponding to the second type of sub-cycle can be arranged from late to early to obtain a heterogeneous feature data sequence, which is not limited here.

[0152] See Figure 5 , Figure 5 This is another schematic diagram of determining the characteristic sequence of isomerization conversion provided by the embodiment of the present application. Figure 5 As shown, a business cycle can be divided into multiple business sub-cycles, such as business sub-cycle 1, business sub-cycle 2, ..., business sub-cycle b, and so on. In business sub-cycle 1, the business object initiates payment transaction 1. The payment data corresponding to payment transaction 1 includes time dimension information 1, value dimension information 1, and description dimension information 1. Time mapping feature 1 can be determined based on time dimension information 1, value mapping feature 1 can be determined based on value dimension information 1, and description conversion feature 1 can be determined based on description dimension information 1. Thus, heterogeneous conversion feature 2 corresponding to payment transaction 1 can be determined based on time mapping feature 1, value mapping feature 1, and description conversion feature 1. In business sub-cycle 2, the business object does not initiate a payment transaction. Therefore, heterogeneous conversion feature 2 can be determined based on sub-cycle filling feature 2. Business sub-cycle b includes payment transaction b. Therefore, the corresponding heterogeneous conversion feature b can be determined using the same method as for determining heterogeneous conversion feature 1. By analogy, the heterogeneous conversion characteristics corresponding to each business sub-cycle can be determined, and thus the heterogeneous feature data sequence corresponding to the business cycle can be determined based on the heterogeneous conversion characteristics corresponding to each business sub-cycle. The specific determination process can refer to the above-mentioned relevant description and will not be repeated here.

[0153] S106. Based on the heterogeneous feature data sequence, determine the payment business feature sequence of the business object within the payment cycle, input the payment business feature sequence into the second hierarchical network, and have the second hierarchical network perform object recognition on the object type of the business object based on the sequence features in the payment business feature sequence to obtain an object recognition result.

[0154] The payment service characteristic sequence may be a characteristic sequence corresponding to payment transactions within a payment cycle, used to characterize the underlying patterns of payment behavior of the business subject within the payment cycle. It is understood that a first-category business cycle may correspond to a heterogeneous characteristic data sequence, and the heterogeneous characteristic data sequence corresponding to each first-category business cycle within the payment cycle may be determined. Thus, based on the heterogeneous characteristic data sequence corresponding to each first-category business cycle, the payment service characteristic sequence corresponding to the payment cycle may be determined.

[0155] Specifically, the payment cycle includes N business cycles, the N business cycles include business cycle i, i is a positive integer less than or equal to N, business cycle i belongs to the first type of business cycle, and the heterogeneous feature data sequence includes a first feature data sequence corresponding to business cycle i, and the first feature data sequence is determined by the heterogeneous conversion features determined based on the payment transactions within business cycle i; then, based on the heterogeneous feature data sequence, determining the payment business feature sequence of the business object within the payment cycle may include the following steps: performing feature fusion processing on the heterogeneous conversion features in the first feature data sequence corresponding to business cycle i to obtain the periodic transaction features corresponding to business cycle i, until the periodic transaction features of each business cycle in the N business cycles are obtained, and then determining the payment business feature sequence of the business object within the payment cycle based on the obtained periodic transaction features of each business cycle.

[0156] The payment cycle may include N business cycles, where N is a positive integer. For details, please refer to the above description and will not be repeated here. Business cycle i is any first-category business cycle within the N business cycles.

[0157] The first characteristic data sequence may be a heterogeneous characteristic data sequence corresponding to business cycle i. It should be understood that the first characteristic data sequence is determined by heterogeneous conversion features determined based on payment transactions within business cycle i. The specific determination method can be referred to the description of the steps for determining heterogeneous characteristic data sequences above and is not further elaborated here.

[0158] The periodic transaction feature may be a feature corresponding to a first-category business cycle, and may be obtained by performing feature fusion processing on heterogeneous conversion features in the first feature data sequence.

[0159] It is understandable that the feature fusion processing here refers to the process of fusing the heterogeneous conversion features in a heterogeneous feature data sequence (such as the first feature data sequence) into one feature (which can be called a fusion feature). It is understandable that the feature fusion processing here can be implemented by a heterogeneous feature fusion network in the target business model. The heterogeneous feature fusion network can be a network for fusing the heterogeneous conversion features in a heterogeneous feature data sequence (such as the first feature data sequence) into one feature. For example, the heterogeneous feature fusion network can include a transformer layer (a neural network), and the heterogeneous feature data sequence (such as the first feature data sequence) can be input into the transformer layer, and the output of the transformer is aggregated using average (a pooling method) to obtain the transaction feature representation of the business object within a first type of business cycle (such as a certain day), that is, the periodic transaction feature corresponding to the first type of business cycle.

[0160] Optionally, when performing feature fusion processing on the heterogeneous conversion features in the first feature data sequence corresponding to business cycle i to obtain the periodic transaction features corresponding to business cycle i, a feature (which may be referred to as a fused feature) obtained by performing feature fusion processing on the heterogeneous conversion features in the first feature data sequence corresponding to business cycle i may be used as the periodic transaction features corresponding to business cycle i. Optionally, when performing feature fusion processing on the heterogeneous conversion features in the first feature data sequence corresponding to business cycle i to obtain the periodic transaction features corresponding to business cycle i, a feature (which may be referred to as a fused feature) obtained by performing feature fusion processing on the heterogeneous conversion features in the first feature data sequence corresponding to business cycle i may be used, and the fused feature may be spliced ​​with a feature (such as a periodic time feature) determined based on the periodic time information of the first type of business cycle (such as the number of months and the day of the first type of business cycle), so that the spliced ​​feature may be used as the periodic transaction features corresponding to business cycle i.

[0161] Specifically, the heterogeneous conversion features in the first feature data sequence corresponding to the business cycle i are subjected to feature fusion processing to obtain the periodic transaction features corresponding to the business cycle i, which may include the following steps: the first hierarchical network performs feature fusion processing on the heterogeneous conversion features in the first feature data sequence to obtain the fusion features corresponding to the business cycle i; the cycle time information of the business cycle i is obtained, the cycle time information is input into the first hierarchical network, and the first hierarchical network performs feature conversion processing on the cycle time information to obtain the periodic time features corresponding to the business cycle i; the fusion features and the periodic time features are subjected to periodic feature splicing processing to obtain the periodic transaction features corresponding to the business cycle i.

[0162] As mentioned above, the fusion feature can be a feature obtained by performing feature fusion processing on the heterogeneous conversion features in the heterogeneous feature data sequence (such as the first feature data sequence). The method for performing feature fusion processing can refer to the relevant description above and will not be repeated here.

[0163] Among them, the cycle time information can be information used to indicate the time of the business cycle in the payment cycle. The cycle time information can be used to indicate the business cycle number (i.e., the order of the business cycles) within the payment cycle. For example, if the payment cycle is 6 months and each business cycle is 1 day, the cycle time information can include the number of days within the payment cycle that the business cycle is. The cycle time information can also include more detailed time information. For example, within the payment cycle, each business cycle can be divided into time periods of different levels. For example, if the payment cycle is 6 months and one business cycle is one day, each month can be divided into a first-level cycle and each week can be divided into a second-level cycle. Then, the cycle time information can include the first-level cycle number or the second-level cycle number of the business cycle. For example, if the payment cycle is 6 months and each business cycle is 1 day, the cycle time information can include the number of months (i.e., first-level cycles) and the number of days of the month that the business cycle is within the payment cycle. It can also include the number of weeks (i.e., second-level cycles) and the number of days of the week that the business cycle is within the payment cycle, etc., which are not limited here.

[0164] The feature conversion process can convert the cycle time information into the features required by the target business model. The cycle time feature can be a feature obtained by performing feature conversion on the cycle time information, and is used to characterize the temporal position of the first type of business cycle within the payment cycle. In some scenarios, the payment transactions of business objects can have periodic changes over months, weeks, etc. Therefore, combining the cycle time information with the fusion features can enable the target business model to discover richer potential patterns, thereby improving the accuracy of identifying the object type of the business object.

[0165] It is understood that the periodic feature splicing process can be a process for splicing the fused feature and the periodic time feature into a single feature, which can also be referred to as splicing, feature splicing, etc. It is understood that when splicing the two, the periodic feature splicing process can be performed based on a certain feature order, such as splicing the fused feature and the periodic time feature based on the order in which the fused feature comes before the periodic time feature, or splicing the fused feature and the periodic time feature based on the order in which the fused feature comes after the periodic time feature, which is not limited here.

[0166] Optionally, a payment cycle may also include a second type of business cycle. For a detailed description of the second type of business cycle, please refer to the above description and will not be repeated here. Since there are no corresponding payment transactions within the second type of business cycle, the periodic transaction characteristics of the second type of business cycle can be determined based on certain filler characteristics (such as a zero vector or zero matrix with the same characteristic dimension as the periodic transaction characteristics of the first type of business cycle).

[0167] Specifically, the N business cycles also include business cycle j, where j is a positive integer less than or equal to N, j is different from i, and business cycle j belongs to a second type of business cycle different from the first type of business cycle. The second type of business cycle is a business cycle in which the business object does not generate a payment transaction; then, the embodiment of the present application also includes the following steps: obtaining the cycle filling feature associated with the second type of business cycle, and determining the cycle filling feature as the cycle transaction feature of business cycle j; based on the cycle transaction feature of business cycle j and the cycle transaction feature of business cycle i, determining the cycle transaction feature of each business cycle in the N business cycles.

[0168] The business cycle j may be any second-category business cycle within the payment cycle. The cycle filling feature may be a feature used to determine the periodic transaction feature of the second-category business cycle (e.g., business cycle j). It should be understood that the feature dimensions of the periodic transaction feature are consistent with the feature dimensions of the periodic transaction feature of the first-category business cycle.

[0169] The cycle filling feature can be a zero matrix whose feature dimension is consistent with the feature dimension of the periodic transaction feature of the first type of business cycle. Optionally, if the above-mentioned method of determining the periodic transaction feature based on the fusion feature and the periodic time feature is adopted, the cycle filling feature can be obtained by splicing the periodic time feature based on the second type of business cycle (the specific determination method can refer to the relevant description of determining the periodic time feature of business cycle i above, which is not repeated here) and a zero vector or zero matrix with the same dimension as the fusion feature of the first type of business cycle, which is not limited here.

[0170] It is understood that, based on the above description, the periodic transaction features corresponding to each of the N business cycles within a payment cycle can be determined, and then a payment service feature sequence can be determined based on the periodic transaction features corresponding to each of the N business cycles. In other words, the payment service feature sequence can include the periodic transaction features corresponding to the N business cycles. It is understood that when determining the payment service feature sequence, the business cycles can also be arranged based on the chronological order of the periods to obtain the payment service feature sequence.

[0171] Specifically, based on the obtained periodic transaction characteristics of each business cycle, determining the payment business characteristic sequence of the business object within the payment cycle can include the following steps: obtaining the periodic time sequence of N business cycles, arranging the periodic transaction characteristics of each business cycle based on the periodic time sequence, and obtaining the payment business characteristic sequence of the business object within the payment cycle.

[0172] The cycle time sequence indicates the chronological order of the N business cycles, and each business cycle has a corresponding cycle time sequence. It is understood that the arrangement of the cycle transaction characteristics of each business cycle based on the cycle time sequence can be done from earliest to latest, or from latest to earliest. This is not detailed here.

[0173] The object identification result may be a result of identifying the object type of the business object. It should be understood that the object identification result may be used to indicate the object type of the business object. It should be understood that the object type may be a first object type or a second object type, where the first object type and the second object type are different. The first object type may be an object type used to indicate that the business object requires payment management, while the second object type may be an object type used to indicate that the business object does not require payment management.

[0174] It can be understood that when performing object identification based on the payment business feature sequence, the sequence features in the payment business feature sequence can be aggregated to obtain the object payment feature used for object identification, thereby performing object identification on the object type of the business object based on the object payment feature.

[0175] For example, see Figure 6 , Figure 6 This is a schematic diagram of determining a payment service feature sequence provided by an embodiment of the present application. Figure 6As shown, a payment cycle may include multiple business cycles, such as business cycle 1, business cycle 2, ..., business cycle n, and so on. Furthermore, based on payment transactions within business cycle 1, a heterogeneous feature data sequence 1 corresponding to business cycle 1 can be determined. The heterogeneous conversion features in heterogeneous feature data sequence 1 may include: feature a1, feature a2, ..., feature a3, etc. The specific process can refer to the description of step S105 above and is not repeated here. Furthermore, feature fusion processing can be performed based on the heterogeneous conversion features in heterogeneous feature data sequence 1 to obtain a fused feature, and then the periodic transaction feature 1 corresponding to business cycle 1 can be determined based on the fused feature. Determining the periodic transaction feature 1 corresponding to business cycle 1 based on the fused feature can be done by directly determining the fused feature 1 as the periodic transaction feature 1, or by concatenating the fused feature 1 with the periodic time features corresponding to business cycle 1 to obtain the periodic transaction feature 1. This is not repeated here. For business cycle n, the periodic transaction feature n corresponding to business cycle n can be determined by referring to the description of determining the periodic transaction feature 1 corresponding to business cycle 1. In addition, the payment cycle may also include business cycles in which the business object does not initiate a payment transaction (i.e., the second type of business cycle). In this case, cycle transaction feature 2 can be determined based on cycle filling feature 2. The specific determination process is described above and is not repeated here. Based on this, the cycle transaction feature corresponding to each business cycle can be determined, and then the payment service feature sequence can be determined based on the cycle transaction feature corresponding to each business cycle.

[0176] Specifically, the payment cycle includes N business cycles, and the sequence features in the payment business feature sequence include the periodic transaction features corresponding to each business cycle in the N business cycles; then, the second hierarchical network performs object identification on the object type of the business object based on the sequence features in the payment business feature sequence to obtain the object identification result, which can include the following steps: the second hierarchical network performs feature aggregation processing based on the periodic transaction features corresponding to the N business cycles to obtain the object payment features of the business object; the second hierarchical network performs object identification on the object type of the business object based on the object payment features to obtain the object identification result.

[0177] Among them, the sequence features in the payment service feature sequence may include the periodic transaction features corresponding to each business cycle in the above-mentioned N business cycles. The method for determining the periodic transaction features corresponding to each business cycle can refer to the above-mentioned relevant description and will not be repeated here.

[0178] It is understood that the feature aggregation process can be a process for determining object payment features for object identification based on periodic transaction features corresponding to N business cycles. It is understood that the feature aggregation process can be performed by a feature aggregation subnetwork in the second layered network for performing feature aggregation processing on the periodic transaction features corresponding to the N business cycles. The feature aggregation subnetwork can be a subnetwork for performing feature aggregation processing. For example, the feature aggregation subnetwork can include a transformer layer (a type of neural network).

[0179] Furthermore, after obtaining the object payment feature, object recognition is performed on the object type of the business object based on the object payment feature, and this can be processed by the object recognition subnetwork in the second hierarchical network for performing object recognition on the object payment feature. The object recognition subnetwork can be an object recognition subnetwork in the second hierarchical network for performing object recognition on the object payment feature. For example, the object recognition subnetwork can be a fully connected classification layer. It is understandable that when object recognition is performed on the object type of the business object based on the object payment feature, the probability of each object type (such as the first object type and the second object type) can be determined based on the object payment feature, and then the object type corresponding to the business object can be determined based on the determined probability of each object type to obtain an object recognition result. For example, the object type whose probability is greater than a certain threshold can be determined as the object type to which the business object belongs.

[0180] For example, the periodic transaction features corresponding to N business cycles can be input into a feature aggregation subnetwork (such as a transformer layer), and the output of the transformer layer can be aggregated using average (a pooling method) to obtain an embedded feature representation of the entire transaction sequence, namely, the object payment feature. When the periodic transaction features corresponding to N business cycles are input into a feature aggregation subnetwork (such as a transformer layer), the positional feature representation of the payment business feature sequence can be obtained using position embedding (a positional representation method) in Bert, and the two features are bitwise added and input into the transformer layer. Furthermore, the object payment feature is input into a fully connected classification layer (i.e., the object recognition subnetwork) for the final classification output to obtain the above-mentioned object recognition result.

[0181] It is understandable that after determining the object type of the business object (i.e., after obtaining the business object identification result), corresponding processing can be performed based on the object type of the business object. For example, in some payment credit review scenarios, if the object type of the business object is identified as an object type that does not have payment credit, it can be determined that the payment credit review of the business object has failed, so that the business object may not be provided with further transactions in the future. Conversely, if the object type of the business object is identified as an object type that has payment credit, it can be determined that the payment credit review of the business object has passed, so that the business object may be provided with further transactions in the future.

[0182] Optionally, the object type of the business object includes a first object type; then, the embodiment of the present application may further include the following steps: when the object identification result indicates that the business object is of the first object type, performing object payment management on the business object.

[0183] Among them, object payment management can be performed when a business object initiates a payment transaction. The first object type can be used to indicate the object type that needs to be managed for the business object. For example, in some scenarios, the first object type can be an object type that has risks in payment behavior, also known as a risk object, and then object payment management can be performed on the business object to prevent the business object from affecting its own or others' finances. For example, object payment management can be performed to limit the transaction amount of transactions initiated by the business object, that is, when the transaction amount is greater than a certain threshold, the business object does not have the authority to initiate a payment transaction with an amount greater than the certain threshold. For another example, object payment management can be performed to output a prompt message to remind the business object that there are risks in the payment transaction. For another example, the business object is prohibited from initiating certain types of payment services, such as prohibiting the business object from purchasing certain types of financial products.

[0184] Here, with diagrams, we explain the complete process of determining the object type of a business object. Figure 7 , Figure 7 This is a schematic diagram of an object type identification process provided by an embodiment of the present application. A payment cycle may include multiple business cycles, and then the cycle transaction characteristics corresponding to each business cycle can be determined for payment transactions within each business cycle. Here, the cycle transaction characteristics 1 corresponding to business cycle 1 (such as Figure 7 Taking 612a in FIG as an example, the periodic transaction characteristics of a single business cycle are described. In business cycle 1, the business object initiates multiple payment transactions. Each payment transaction may include multiple dimension information, such as description dimension information (such as Figure 7 601a in ), numerical dimension information (such as Figure 7 602a in FIG), time dimension information (such as Figure 7 603a in ). Then, the description-class dimension information 601a can be used to determine the corresponding description conversion feature 604a, the corresponding numerical mapping feature 605a can be determined based on the numerical dimension information 602a, and the corresponding time mapping feature 606a can be determined based on the time dimension information 603a. The specific determination process refers to the above-mentioned related description and will not be repeated here. Furthermore, for each payment transaction, a corresponding heterogeneous conversion feature can be determined, such as heterogeneous conversion feature 1, and then, referring to the above steps, a plurality of heterogeneous conversion features for determining a heterogeneous feature data sequence can be determined, such as heterogeneous conversion feature 1, heterogeneous conversion feature 2, heterogeneous conversion feature 3, etc., and then a heterogeneous feature data sequence (such as heterogeneous conversion feature 1, heterogeneous conversion feature 2, heterogeneous conversion feature 3, etc.) can be determined based on the determined heterogeneous conversion features. Figure 7 Further, based on sub-network 1 (such as Figure 7 608a in the figure), the heterogeneous conversion features in the heterogeneous feature data sequence 607a are subjected to feature fusion processing to obtain fused features (as shown in FIG. Figure 7 As shown in 610a in FIG. 1 ). The sub-network 1 may be the heterogeneous feature fusion network described above. The specific processing process thereof is described in the above related description and will not be described here. In addition, the cycle time information corresponding to the service cycle 1 may also be obtained (e.g. Figure 7 609a), for example, the cycle time information can be used to indicate the number of months and the day of the business cycle 1. Then, the cycle time feature 1 corresponding to the business cycle 1 can be determined based on the cycle time information. The specific determination process can refer to the above-mentioned related descriptions and will not be repeated here. Further, the cycle transaction feature 1 corresponding to the business cycle 1 can be determined based on the cycle time feature 1 and the fusion feature 1. The cycle transaction features corresponding to other business cycles within the payment cycle can be processed with reference to the above-mentioned processing process of determining the cycle transaction feature 1. For example, for business cycle 2, the corresponding cycle time feature 2 and fusion feature 2 can be determined, so as to determine the cycle transaction feature 2 based on the cycle time feature 2 and the fusion feature 2. For business cycle 3, the corresponding cycle time feature 3 and fusion feature 3 can be determined, so as to determine the cycle transaction feature 3 based on the cycle time feature 3 and the fusion feature 3. In addition, if the business objects in some business cycles do not initiate payment transactions, the cycle transaction features can be determined by filling in the features, that is, filling in by adding 0. Then, the payment business feature sequence (such as Figure 7 613a in FIG).

[0185] Furthermore, the payment service feature sequence can be input into sub-network 2 (eg Figure 7 614a in FIG), to perform feature aggregation processing on the payment service feature sequence through sub-network 2, and obtain the object payment feature (such as Figure 7 The sub-network 2 may be the feature aggregation sub-network mentioned above. The specific processing process thereof is described in the above related descriptions and will not be described here. Further, the object payment feature may be input into the sub-network 3 (e.g. Figure 7 616a in FIG), so as to perform object recognition processing on the object payment feature through sub-network 3 to obtain an object recognition result (such as Figure 7 617a in the figure), thereby identifying the object type of the business object. Since the characteristics of the payment transaction in multiple dimensions are input during the object recognition process, and the characteristics of each dimension are combined, the payment habits of the business object can be better characterized, thereby helping to improve the accuracy of feature recognition. In addition, when performing object recognition, the payment transactions in a long period of time are divided into multiple time periods, so that the characteristics of the payment transactions in each time period are aggregated into one feature, avoiding the need for the model to directly process an excessively long feature sequence, thereby enabling the model to discover more potential payment behavior patterns and improve processing efficiency.

[0186] By adopting the embodiments of the present application, the transaction sequence within the payment cycle can be obtained, and the payment cycle can be divided into finer granularity, such as obtaining a first type of business cycle by division, and then the corresponding features can be determined based on the information of the payment transactions in the first type of business cycle in multiple dimensions, that is, the time mapping feature is determined based on the time dimension information, the numerical mapping feature is determined based on the numerical dimension information, and the description conversion feature is determined based on the description dimension information, and then the features of each dimension can be heterogeneously converted to further obtain the heterogeneous feature data sequence corresponding to the first type of business cycle, so that the heterogeneous feature data sequence of the payment transactions within the fine-grained business cycle can be obtained, which is used to characterize the payment behavior characteristics of the business object within a fine-grained time period. In this way, the information on multiple dimensions of the payment transaction can be effectively fused, which can better characterize the user's transaction behavior, and the payment transactions within the payment cycle can be divided into finer granularity, so that more abundant potential laws of transaction behavior can be excavated, which helps to improve the accuracy of identifying the object type of the business object. Furthermore, based on the heterogeneous feature data sequence corresponding to the fine-grained business cycle, the payment business feature sequence corresponding to the entire payment cycle can be determined to characterize the characteristics of the payment behavior of the business object throughout the payment cycle. This can help to explore and analyze the operational correlation between the business object when performing different transaction operations within the fine-grained time, thereby improving the accuracy of data analysis of the business object's transaction operations, and helping to improve the accuracy of identifying the object type of the business object.

[0187] Further, see Figure 8 , Figure 8This is a flow chart of a business data processing method provided by an embodiment of the present application. The method can be executed by a computer device, for example, the computer device is the above-mentioned Figure 1 The method may include at least the following steps S801 to S809.

[0188] S201. Obtain sample data for training the initial business model; the sample data is determined based on sample labels and sample payment data, and the sample payment data refers to the payment data of the sample object in the sample transaction sequence within the sample payment cycle; the sample label is used to indicate the reference object type of the sample object.

[0189] The initial business model may be an untrained network for object recognition of the object type of the business object. It is understood that the model result of the initial business model is consistent with the model result of the target business model, and will not be described in detail here.

[0190] The sample data may be data used to train the initial business model. The sample data may include sample labels and sample payment data. It should be understood that the sample payment data refers to the payment data in a sample transaction sequence of a sample object within a sample payment cycle. The sample object may be the object used to collect sample data, and the sample object may have initiated one or more payment transactions. The sample payment cycle may be the time period required to obtain payment transactions for object type identification during model training. The sample payment cycle may be the same as or different from the payment cycle described above, and this is not limited here. However, it should be understood that the sample transaction sequence may be a sequence of payment transactions initiated by the sample object within the sample payment cycle. It should be understood that the sample payment cycle may also be divided into one or more sample business cycles. The description of the sample business cycles can refer to the relevant description of the business cycles described above. For example, if the sample payment cycle is one month, a sample business cycle within the sample payment cycle may be one day.

[0191] The sample label is used to indicate the reference object type of the sample object, and the reference object type is used to indicate the object type that the sample object is actually labeled as, so that the object type identified based on the model during the training process is close to the reference object type.

[0192] S202: Input the sample data into the initial business model, and the initial business model performs object recognition on the object type of the sample object to obtain a sample recognition result; the sample recognition result is used to indicate the recognized object type of the sample object.

[0193] It is understood that the sample recognition result can be used to indicate the object type of the sample object identified by the initial business model when processing the sample data. The sample recognition result can be used to indicate the identified object type of the sample object. In other words, the identified object type is used to indicate the object type of the sample object identified based on the initial business model.

[0194] Among them, the initial business model performs object recognition on the object type of the sample object to obtain the processing process of the sample recognition result. The relevant description of the above steps S102-S106 can be referred to and will not be repeated here. Specifically, the sample transaction sequence may include sample payment transactions initiated by the sample object within the sample business cycle, and the transaction data of the sample payment transaction includes time dimension information, numerical dimension information and description dimension information; the sample business cycle is the business cycle in the sample payment cycle; the initial business model may include a first initial hierarchical network and a second initial hierarchical network; furthermore, the time dimension information may be input into the first initial hierarchical network in the initial business model, and the first initial hierarchical network performs feature mapping processing on the time dimension information to obtain the time mapping feature corresponding to the time dimension information of the sample payment transaction; the numerical dimension information may be input into the first initial hierarchical network, and the first hierarchical network performs feature mapping processing on the numerical dimension information to obtain the numerical mapping feature corresponding to the numerical dimension information of the sample payment transaction; the description dimension information may be input into the first initial hierarchical network, and the first hierarchical network performs feature mapping processing on the numerical dimension information to obtain the numerical mapping feature corresponding to the numerical dimension information of the sample payment transaction; The information is input into the first initial hierarchical network. The first initial hierarchical network performs word segmentation processing on the description dimension information to obtain text segmentation of the description dimension information. The text segmentation is then subjected to feature conversion processing to obtain description conversion features corresponding to the description dimension information of the sample payment transaction. The first initial hierarchical network performs heterogeneous conversion processing on the time mapping features, the value mapping features, and the description conversion features to obtain a heterogeneous feature data sequence of the sample object within the sample business cycle. Based on the heterogeneous feature data sequence corresponding to the sample business cycle, the payment business feature sequence of the sample object within the sample payment cycle is determined. The payment business feature sequence corresponding to the sample payment cycle is input into the second initial hierarchical network. The second initial hierarchical network performs object recognition on the object type of the sample object based on the sequence features in the payment business feature sequence corresponding to the sample payment cycle to obtain a sample recognition result. The first initial hierarchical network is an untrained first hierarchical network, and the second initial hierarchical network is an untrained second hierarchical network. In other words, when the initial business model after parameter adjustment is determined as the target business model, the first initial hierarchical network after parameter adjustment can be determined as the first hierarchical network, and the second initial hierarchical network after parameter adjustment can be determined as the second hierarchical network. The process for determining the time mapping characteristics, value mapping characteristics, and description conversion characteristics of the sample payment transaction can refer to the relevant description of determining the time mapping characteristics, value mapping characteristics, and description conversion characteristics of the payment transaction initiated by the business object, which is not repeated here. The process for determining the heterogeneous feature data sequence corresponding to the sample business cycle can refer to the relevant description of determining the heterogeneous feature data sequence corresponding to the business cycle, and the process for determining the payment business feature sequence corresponding to the sample payment cycle can refer to the relevant description of determining the payment business feature sequence corresponding to the payment cycle, which is not repeated here.

[0195] S203 : Based on the reference object type and the identified object type, perform parameter adjustment processing on the model parameters of the initial business model, and determine the target business model based on the initial business model after the parameter adjustment processing.

[0196] It should be understood that the parameter adjustment process can be a process for adjusting the model parameters in the initial business model. It should be understood that the goal of adjusting the model parameters of the initial business model is to ensure that the recognized object type is the same as the reference object type. In other words, as the initial business model is trained, the initial business model can gradually and accurately identify the object type of the sample object.

[0197] It can be understood that the target business model is determined based on the initial business model after parameter adjustment processing. After the initial business model after parameter adjustment processing is tested and passed through test data, the target business model is determined based on the initial business model after parameter adjustment processing that passes the test.

[0198] It is understandable that when training the initial business model, a large amount of sample data can be collected for training, that is, the number of the above-mentioned sample data is multiple, and the initial business model is trained end-to-end to obtain a target business model with high accuracy.

[0199] S204. Obtain payment transaction data for input into a target business model from a transaction sequence of the business object within a payment cycle; the target business model includes a first hierarchical network and a second hierarchical network; the payment transaction data is determined based on payment transactions initiated by the business object within a first type of business cycle; the payment transaction data includes time dimension information, numerical dimension information, and descriptive dimension information; the first type of business cycle is a business cycle within a payment cycle.

[0200] S205 : Input the time dimension information into the first hierarchical network, and perform feature mapping processing on the time dimension information by the first hierarchical network to obtain time mapping features corresponding to the time dimension information.

[0201] S206: Input the numerical dimension information into the first hierarchical network, and perform feature mapping processing on the numerical dimension information by the first hierarchical network to obtain numerical mapping features corresponding to the numerical dimension information.

[0202] S207. Input the description class dimension information into the first hierarchical network. The first hierarchical network performs word segmentation processing on the description class dimension information to obtain text word segmentation of the description class dimension information. The text word segmentation is subjected to feature conversion processing to obtain description conversion features corresponding to the description class dimension information.

[0203] S208: The first hierarchical network performs heterogeneous conversion processing on the time mapping feature, the value mapping feature, and the description conversion feature to obtain a heterogeneous feature data sequence of the business object within the first type of business cycle.

[0204] S209. Based on the heterogeneous feature data sequence, determine the payment business feature sequence of the business object within the payment cycle, input the payment business feature sequence into the second hierarchical network, and have the second hierarchical network perform object recognition on the object type of the business object based on the sequence features in the payment business feature sequence to obtain an object recognition result.

[0205] Among them, steps S204 to S209 can refer to the relevant descriptions of the above steps S101 to S106, and are not repeated here.

[0206] See Figure 9 , Figure 9 This is a flow chart of a business model training process provided by an embodiment of the present application. First, sample data (such as Figure 9 As shown in S901 in the example, the sample data may include sample tags and sample payment data. For detailed introduction, please refer to the above description, which will not be repeated here. Figure 9 As shown in S902 in the example, it is understandable that when dividing the payment transactions in the sample transaction sequence, they can be divided according to business sub-cycles (such as hours), so as to determine the payment transactions in each business sub-cycle. Further, for each payment transaction, the description class dimension information can be converted into features through the initial business model (such as Figure 9 As shown in S903 in ), to obtain the corresponding description conversion features, the numerical dimension information is mapped by the initial business model (as shown in Figure 9 As shown in S904 in ), to obtain the corresponding numerical mapping features, the time dimension information is mapped by the initial business model (as shown in Figure 9 ), to obtain the corresponding time mapping features. Further, the features obtained by processing can be processed by the initial business model (as shown in S905 in FIG. Figure 9 As shown in S906 in the example, the identification object type of the sample object is obtained. The features obtained here are the description conversion features, data mapping features and time mapping features mentioned above. Further, the initial business model is trained according to the sample labels of the sample data (as shown in S906 in the example). Figure 9 As shown in S907 in the example, the model parameters of the initial business model can be adjusted so that the object type identified by the initial business model is consistent with the reference object type indicated by the sample label. Further, after the initial business model is trained, the target business model obtained by the training can be saved and applied (as shown in S907 in the example). Figure 9 (S908 in the example), this application can be used to identify the object type of a business object based on its transaction sequence. That is, in online applications, the transaction flow of a business object within a certain payment cycle is taken, and the same feature processing method used during training is used to input it into the trained classification model (i.e., the target business object). This can then determine the object type (e.g., whether the business object has default risk) based on the user's historical transaction behavior over multiple days. For example, in scenarios involving payment credit review, malicious users can be audited and intercepted, thereby reducing the default risk of financial services.

[0207] By adopting the embodiments of the present application, the transaction sequence within the payment cycle can be obtained, and the payment cycle can be divided into finer granularity, such as obtaining a first type of business cycle by division, and then the corresponding features can be determined based on the information of the payment transactions in the first type of business cycle in multiple dimensions, that is, the time mapping feature is determined based on the time dimension information, the numerical mapping feature is determined based on the numerical dimension information, and the description conversion feature is determined based on the description dimension information, and then the features of each dimension can be heterogeneously converted to further obtain the heterogeneous feature data sequence corresponding to the first type of business cycle, so that the heterogeneous feature data sequence of the payment transactions within the fine-grained business cycle can be obtained, which is used to characterize the payment behavior characteristics of the business object within a fine-grained time period. In this way, the information on multiple dimensions of the payment transaction can be effectively fused, which can better characterize the user's transaction behavior, and the payment transactions within the payment cycle can be divided into finer granularity, so that more abundant potential laws of transaction behavior can be excavated, which helps to improve the accuracy of identifying the object type of the business object. Furthermore, based on the heterogeneous feature data sequence corresponding to the fine-grained business cycle, the payment business feature sequence corresponding to the entire payment cycle can be determined to characterize the characteristics of the payment behavior of the business object throughout the payment cycle. This can help to explore and analyze the operational correlation between the business object when performing different transaction operations within the fine-grained time, thereby improving the accuracy of data analysis of the business object's transaction operations, and helping to improve the accuracy of identifying the object type of the business object.

[0208] See Figure 10 , Figure 10 This is a schematic diagram of the structure of a business data processing device provided by an embodiment of the present application. Figure 10 As shown, the service data processing device 1 may be a computer device (for example, Figure 1A computer program (including program code) of the server 100a) in the embodiment of the present application, for example, the business data processing device 1 is an application software; it can be understood that the business data processing device 1 can be used to execute the corresponding steps in the business data processing method provided in the embodiment of the present application. Figure 10 As shown, the business data processing device 1 may include: a transaction data acquisition module 11, a time mapping module 12, a value mapping module 13, a description conversion module 14, a heterogeneous conversion module 15, and an object recognition module 16;

[0209] The transaction data acquisition module 11 is configured to acquire payment transaction data for input into a target business model from a transaction sequence of a business object within a payment cycle. The target business model includes a first hierarchical network and a second hierarchical network. The payment transaction data is determined based on payment transactions initiated by the business object within the first business cycle. The payment transaction data includes time dimension information, numerical dimension information, and descriptive dimension information. The first business cycle is a business cycle within the payment cycle.

[0210] The time mapping module 12 is configured to input the time dimension information into the first hierarchical network, and perform feature mapping processing on the time dimension information by the first hierarchical network to obtain a time mapping feature corresponding to the time dimension information;

[0211] The numerical mapping module 13 is configured to input the numerical dimension information into the first hierarchical network, and perform feature mapping processing on the numerical dimension information by the first hierarchical network to obtain numerical mapping features corresponding to the numerical dimension information;

[0212] The description conversion module 14 is configured to input the description class dimension information into the first hierarchical network, perform word segmentation processing on the description class dimension information by the first hierarchical network to obtain text segmentation of the description class dimension information, and perform feature conversion processing on the text segmentation to obtain description conversion features corresponding to the description class dimension information;

[0213] The heterogeneous conversion module 15 is configured to perform heterogeneous conversion processing on the time mapping feature, the value mapping feature, and the description conversion feature by the first hierarchical network to obtain a heterogeneous feature data sequence of the business object within the first business cycle;

[0214] The object identification module 16 is used to determine the payment business feature sequence of the business object within the payment cycle based on the heterogeneous feature data sequence, input the payment business feature sequence into the second hierarchical network, and the second hierarchical network performs object identification on the object type of the business object based on the sequence features in the payment business feature sequence to obtain an object identification result.

[0215] Wherein, the payment cycle includes N business cycles, the N business cycles include business cycle i, i is a positive integer less than or equal to N, business cycle i belongs to the first type of business cycle, the heterogeneous feature data sequence includes a first feature data sequence corresponding to business cycle i, and the first feature data sequence is determined by the heterogeneous conversion feature determined based on the payment transactions in business cycle i;

[0216] The object recognition module 16 includes: a payment feature sequence unit 161;

[0217] The payment feature sequence unit 161 is used to perform feature fusion processing on the heterogeneous conversion features in the first feature data sequence corresponding to the business cycle i to obtain the periodic transaction features corresponding to the business cycle i, until the periodic transaction features of each business cycle in N business cycles are obtained. Based on the periodic transaction features of each business cycle obtained, the payment business feature sequence of the business object within the payment cycle is determined.

[0218] The processing of the payment feature sequence unit 161 can refer to the above Figure 3 The relevant descriptions in the embodiments are not repeated here.

[0219] The payment feature sequence unit 161 includes: a fusion feature unit 1611, a cycle time feature unit 1612, and a feature splicing unit 1613;

[0220] The feature fusion unit 1611 is configured to perform feature fusion processing on the heterogeneous conversion features in the first feature data sequence by the first hierarchical network to obtain a fused feature corresponding to the service cycle i;

[0221] The cycle time feature unit 1612 is configured to obtain cycle time information of service cycle i, input the cycle time information into the first hierarchical network, and perform feature conversion processing on the cycle time information by the first hierarchical network to obtain the cycle time feature corresponding to service cycle i;

[0222] The feature splicing unit 1613 is used to perform period feature splicing processing on the fusion feature and the period time feature to obtain the period transaction feature corresponding to the business period i.

[0223] The processing of the fusion feature unit 1611, the periodic time feature unit 1612, and the feature splicing unit 1613 can refer to the above Figure 3 The relevant descriptions in the embodiments are not repeated here.

[0224] The N business cycles also include business cycle j, where j is a positive integer less than or equal to N. If j is different from i, business cycle j belongs to a second type of business cycle, which is different from the first type of business cycle. The second type of business cycle is a business cycle in which no payment transaction occurs for the business object.

[0225] The payment feature sequence unit 161 further includes: a cycle filling unit 1614;

[0226] The period filling unit 1614 is specifically configured to:

[0227] Obtaining a cycle filling feature associated with the second type of business cycle, and determining the cycle filling feature as a cycle transaction feature of business cycle j;

[0228] Based on the periodic transaction characteristics of business cycle j and the periodic transaction characteristics of business cycle i, the periodic transaction characteristics of each of the N business cycles are determined.

[0229] The processing of the cycle filling unit 1614 can refer to the above Figure 3 The relevant descriptions in the embodiments are not repeated here.

[0230] The payment feature sequence unit 161 includes: a feature arrangement unit 1615;

[0231] The feature arrangement unit 1615 is configured to obtain a cycle time sequence of N business cycles, arrange the cycle transaction features of each business cycle based on the cycle time sequence, and obtain a payment business feature sequence of the business object within the payment cycle.

[0232] The processing of the feature arrangement unit 1615 can refer to the above Figure 3 The relevant descriptions in the embodiments are not repeated here.

[0233] The number of payment transactions in the first business cycle is M, and the M payment transactions include payment transaction p, where p is a positive integer less than or equal to M;

[0234] The heterogeneous conversion module 15 includes: a first heterogeneous feature sequence unit 151;

[0235] The first heterogeneous feature sequence unit 151 is used to perform heterogeneous conversion processing on the time mapping feature corresponding to the payment transaction p, the numerical mapping feature corresponding to the payment transaction p, and the description conversion feature corresponding to the payment transaction p by the first hierarchical network to obtain the heterogeneous conversion feature corresponding to the payment transaction p, until the heterogeneous conversion feature of each payment transaction in the M payment transactions is obtained, and then determine the heterogeneous feature data sequence of the business object within the first type of business cycle based on the obtained heterogeneous conversion feature of each payment transaction.

[0236] The processing process of the first heterogeneous feature sequence unit 151 can refer to the above Figure 3 The relevant descriptions in the embodiments are not repeated here.

[0237] The first type of business cycle includes W business sub-cycles, where W is a positive integer; the W business sub-cycles include the first type of sub-cycle, and the first type of sub-cycle is a business sub-cycle in which the business object initiates a payment transaction; and the payment transaction within the first type of sub-cycle is the first payment transaction.

[0238] The heterogeneous conversion module 15 includes: a sub-period feature unit 152 and a second heterogeneous feature sequence unit 153;

[0239] The sub-period feature unit 152 is configured to, when the first hierarchical network determines the time mapping feature, the value mapping feature, and the description conversion feature of the first payment transaction, perform heterogeneous conversion processing based on the time mapping feature, the value mapping feature, and the description conversion feature of the first payment transaction, to obtain a heterogeneous conversion feature corresponding to the first type of sub-period;

[0240] The second heterogeneous feature sequence unit 153 determines the heterogeneous feature data sequence of the business object in the first type of business cycle based on the heterogeneous conversion features corresponding to the first type of sub-cycle.

[0241] The processing of the sub-period feature unit 152 and the second heterogeneous feature sequence unit 153 can refer to the above Figure 3 The relevant descriptions in the embodiments are not repeated here.

[0242] The W business sub-cycles also include a second type of sub-cycle that is different from the first type of sub-cycle. The second type of sub-cycle is a business sub-cycle in which no payment transaction occurs for the business object.

[0243] The second heterogeneous feature sequence unit 153 includes: a sub-period feature filling unit 1531;

[0244] The sub-period feature filling unit 1531 is used to:

[0245] Obtaining a sub-period filling feature associated with the second type of sub-period, and determining the sub-period filling feature as a heterogeneous conversion feature corresponding to the second type of sub-period;

[0246] Based on the heterogeneous conversion characteristics corresponding to the first type of sub-period and the heterogeneous conversion characteristics corresponding to the second type of sub-period, a heterogeneous characteristic data sequence of the business object in the first type of business period is determined.

[0247] The processing of the sub-period feature arrangement unit 1531 can refer to the above Figure 3 The relevant descriptions in the embodiments are not repeated here.

[0248] The second heterogeneous feature sequence unit 153 includes: a sub-period feature arrangement unit 1532;

[0249] The sub-period feature arrangement unit 1532 is specifically configured to:

[0250] Obtaining the sub-cycle time sequence of the W service sub-cycles, determining the sub-cycle time sequence of the first type of sub-cycle as the first sub-cycle time sequence, and determining the sub-cycle time sequence of the second type of sub-cycle as the second sub-cycle time sequence;

[0251] Based on the first sub-cycle time sequence and the second sub-cycle time sequence, the heterogeneous conversion features corresponding to the first type of sub-cycle and the heterogeneous conversion features corresponding to the second type of sub-cycle are used to determine the heterogeneous feature data sequence of the business object in the first type of business cycle.

[0252] The processing of the sub-period feature arrangement unit 1532 can refer to the above Figure 3 The relevant descriptions in the embodiments are not repeated here.

[0253] The payment cycle includes N business cycles, and the sequence features in the payment business feature sequence include the periodic transaction features corresponding to each business cycle in the N business cycles;

[0254] The object recognition module 16 includes: a feature aggregation unit 162 and a feature recognition unit 163;

[0255] A feature aggregation unit 162 is configured to perform feature aggregation processing based on the periodic transaction features corresponding to the N business periods by the second hierarchical network to obtain object payment features of the business object;

[0256] The feature identification unit 163 is configured to perform object identification on the object type of the business object based on the object payment feature by the second hierarchical network to obtain an object identification result.

[0257] The processing of the feature aggregation unit 162 and the feature identification unit 163 can refer to the above Figure 3 The relevant descriptions in the embodiments are not repeated here.

[0258] Wherein, the object type of the business object includes a first object type;

[0259] The business data processing device 1 further includes: a payment management module 17;

[0260] The payment management module 17 is configured to perform object payment management on the business object when the object identification result indicates that the business object is of the first object type.

[0261] The business data processing device 1 further includes: a sample data acquisition module 18, a sample identification module 19, and a parameter adjustment module 20;

[0262] The sample data acquisition module 18 is used to acquire sample data for training the initial business model; the sample data is determined based on the sample label and sample payment data, where the sample payment data refers to the payment data of the sample object in the sample transaction sequence within the sample payment cycle; the sample label is used to indicate the reference object type of the sample object;

[0263] The sample identification module 19 is used to input the sample data into the initial business model, and the initial business model performs object identification on the object type of the sample object to obtain a sample identification result; the sample identification result is used to indicate the identification object type of the sample object;

[0264] The parameter adjustment module 20 is configured to perform parameter adjustment processing on the model parameters of the initial business model based on the reference object type and the identification object type, and determine the target business model based on the initial business model after the parameter adjustment processing.

[0265] The processing of the sample data acquisition module 18, the sample identification module 19, and the parameter adjustment module 20 can refer to the above Figure 8 The relevant descriptions in the embodiments are not repeated here.

[0266] The description dimension information includes first description information and second description information;

[0267] The description conversion module 14 includes: a word segmentation processing unit 141;

[0268] The word segmentation processing unit 141 is specifically used for:

[0269] Obtaining a first text length threshold associated with the first description information and a second text length threshold associated with the second description information;

[0270] Based on a first text length threshold, the first description information is subjected to length normalization processing to obtain a first standard text corresponding to the first description information;

[0271] Based on the second text length threshold, the second description information is subjected to length normalization processing to obtain a second standard text corresponding to the second description information;

[0272] The first hierarchical network performs word segmentation processing on the first standard text and the second standard text to obtain text word segmentation describing class dimension information.

[0273] The processing of the word segmentation processing unit 141 can refer to the above Figure 3 The relevant descriptions in the embodiments are not repeated here.

[0274] See Figure 11 , Figure 11This is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. Figure 11 As shown, the computer device 1000 may include: a processor 1001, a network interface 1004 and a memory 1005. In addition, the above-mentioned computer device 1000 may also include: a user interface 1003, and at least one communication bus 1002. The communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), a keyboard (Keyboard), and the user interface 1003 may optionally include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one disk memory. The memory 1005 may optionally also be at least one storage device located away from the aforementioned processor 1001. As Figure 11 As shown, the memory 1005 as a computer-readable storage medium may include an operating system, a network communication module, a user interface module, and a device control application.

[0275] In such Figure 11 In the illustrated computer device 1000, the network interface 1004 provides network communication functionality; the user interface 1003 primarily provides an interface for user input; and the processor 1001 is configured to invoke a device control application stored in the memory 1005 to execute the description of the service data processing method described in any of the corresponding embodiments above, which will not be repeated here. Furthermore, the description of the beneficial effects of employing the same method will not be repeated here.

[0276] In addition, it should be pointed out here that: the embodiment of the present application also provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program executed by the business data processing device 1 mentioned above, and the computer program includes program instructions. When the processor executes the program instructions, it can execute the description of the business data processing method in the above embodiment. Therefore, it will not be repeated here. In addition, the description of the beneficial effects of using the same method will not be repeated. For technical details not disclosed in the computer-readable storage medium embodiment involved in this application, please refer to the description of the method embodiment of this application.

[0277] The computer-readable storage medium may be the data processing device provided in any of the aforementioned embodiments or the internal storage unit of the computer device, such as the hard disk or memory of the computer device. The computer-readable storage medium may 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. equipped on the computer device. Furthermore, the computer-readable storage medium may also include both the internal storage unit of the computer device and an external storage 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 may also be used to temporarily store data that has been output or is to be output.

[0278] In addition, it should be noted that the embodiments of the present application also provide a computer program product or computer program, 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, so that the computer device performs the method provided by any of the corresponding embodiments above. In addition, the description of the beneficial effects of using the same method will not be repeated. For technical details not disclosed in the computer program product or computer program embodiments involved in this application, please refer to the description of the method embodiments of this application.

[0279] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program that has a predetermined function and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0280] The terms "first", "second", etc. in the description, claims, and drawings of the embodiments of the present application are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "comprising" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product, or device comprising a series of steps or units is not limited to the listed steps or modules, but may optionally include steps or modules not listed, or may optionally include other step units inherent to these processes, methods, apparatuses, products, or devices.

[0281] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0282] The above disclosure is only a preferred embodiment of the present application, and certainly cannot be used to limit the scope of rights of the present application. Therefore, equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. A business data processing method, characterized in that: The method comprises: Obtaining payment transaction data for input into a target business model from a transaction sequence of a business object within a payment cycle; the target business model includes a first hierarchical network and a second hierarchical network; the payment transaction data is determined based on payment transactions initiated by the business object within a first type of business cycle; the payment transaction data includes time dimension information, value dimension information, and description dimension information; the first type of business cycle is a business cycle within the payment cycle; Inputting the time dimension information into the first hierarchical network, and performing feature mapping processing on the time dimension information by the first hierarchical network to obtain a time mapping feature corresponding to the time dimension information; Inputting the numerical dimension information into the first hierarchical network, and performing feature mapping processing on the numerical dimension information by the first hierarchical network to obtain numerical mapping features corresponding to the numerical dimension information; Inputting the descriptive dimension information into the first hierarchical network, performing word segmentation processing on the descriptive dimension information by the first hierarchical network to obtain text segmentation of the descriptive dimension information, and performing feature conversion processing on the text segmentation to obtain description conversion features corresponding to the descriptive dimension information; The first hierarchical network performs heterogeneous conversion processing on the time mapping feature, the value mapping feature, and the description conversion feature to obtain a heterogeneous feature data sequence of the business object within the first type of business cycle; Based on the heterogeneous feature data sequence, the payment business feature sequence of the business object in the payment cycle is determined, and the payment business feature sequence is input into the second hierarchical network. The second hierarchical network performs object recognition on the object type of the business object based on the sequence features in the payment business feature sequence to obtain an object recognition result.

2. The method according to claim 1, characterized in that The payment cycle includes N business cycles, the N business cycles include business cycle i, where i is a positive integer less than or equal to N, and the business cycle i belongs to the first category of business cycles. The heterogeneous feature data sequence includes a first feature data sequence corresponding to the business cycle i, and the first feature data sequence is determined by a heterogeneous conversion feature determined based on payment transactions within the business cycle i. The determining, based on the heterogeneous feature data sequence, the payment service feature sequence of the business object within the payment cycle includes: Perform feature fusion processing on the heterogeneous conversion features in the first feature data sequence corresponding to the business cycle i to obtain the periodic transaction features corresponding to the business cycle i, until the periodic transaction features of each business cycle in the N business cycles are obtained. Based on the obtained periodic transaction features of each business cycle, determine the payment business feature sequence of the business object in the payment cycle.

3. The method according to claim 2, characterized in that The performing feature fusion processing on the heterogeneous conversion features in the first feature data sequence corresponding to the business cycle i to obtain the periodic transaction features corresponding to the business cycle i includes: The first hierarchical network performs feature fusion processing on the heterogeneous conversion features in the first feature data sequence to obtain a fused feature corresponding to the service cycle i; Obtaining cycle time information of the service cycle i, inputting the cycle time information into the first hierarchical network, and having the first hierarchical network perform feature conversion processing on the cycle time information to obtain a cycle time feature corresponding to the service cycle i; The fusion feature and the periodic time feature are subjected to periodic feature splicing processing to obtain the periodic transaction feature corresponding to the business period i.

4. The method according to claim 2, characterized in that The N business cycles also include a business cycle j, where j is a positive integer less than or equal to N, j is different from i, and the business cycle j belongs to a second type of business cycle different from the first type of business cycle, and the second type of business cycle is a business cycle in which no payment transaction occurs for the business object; The method further comprises: Acquire a cycle filling feature associated with the second type of business cycle, and determine the cycle filling feature as a cycle transaction feature of the business cycle j; Based on the periodic transaction characteristics of the service cycle j and the periodic transaction characteristics of the service cycle i, the periodic transaction characteristics of each of the N service cycles are determined.

5. The method according to claim 2, characterized in that The determining, based on the obtained periodic transaction characteristics of each business cycle, a payment service characteristic sequence of the business object in the payment cycle includes: The cycle time sequence of the N business cycles is obtained, and the cycle transaction characteristics of each business cycle are arranged based on the cycle time sequence to obtain a payment business feature sequence of the business object in the payment cycle.

6. The method according to claim 1, characterized in that The number of payment transactions in the first business cycle is M, and the M payment transactions include payment transaction p, where p is a positive integer less than or equal to M; The first hierarchical network performs heterogeneous conversion processing on the time mapping feature, the value mapping feature, and the description conversion feature to obtain a heterogeneous feature data sequence of the business object within the first business cycle, including: The first hierarchical network performs heterogeneous conversion processing on the time mapping feature corresponding to the payment transaction p, the numerical mapping feature corresponding to the payment transaction p, and the description conversion feature corresponding to the payment transaction p to obtain the heterogeneous conversion feature corresponding to the payment transaction p, until the heterogeneous conversion feature of each payment transaction in the M payment transactions is obtained. Based on the obtained heterogeneous conversion feature of each payment transaction, the heterogeneous feature data sequence of the business object within the first type of business cycle is determined.

7. The method according to claim 1, characterized in that The first type of business cycle includes W business sub-cycles, where W is a positive integer; the W business sub-cycles include a first type of sub-cycle, and the first type of sub-cycle is a business sub-cycle in which the business object initiates a payment transaction; and the payment transaction in the first type of sub-cycle is a first payment transaction; The first hierarchical network performs heterogeneous conversion processing on the time mapping feature, the value mapping feature, and the description conversion feature to obtain a heterogeneous feature data sequence of the business object within the first business cycle, including: When the first hierarchical network determines the time mapping feature of the first payment transaction, the value mapping feature of the first payment transaction, and the description conversion feature of the first payment transaction, heterogeneous conversion processing is performed based on the time mapping feature of the first payment transaction, the value mapping feature of the first payment transaction, and the description conversion feature of the first payment transaction to obtain a heterogeneous conversion feature corresponding to the first type of sub-period; Based on the heterogeneous conversion characteristics corresponding to the first type of sub-period, a heterogeneous characteristic data sequence of the business object in the first type of business period is determined.

8. The method according to claim 7, characterized in that The W business sub-cycles further include a second type of sub-cycle that is different from the first type of sub-cycle, and the second type of sub-cycle is a business sub-cycle in which no payment transaction occurs for the business object; The determining, based on the heterogeneous conversion characteristics corresponding to the first type of sub-period, the heterogeneous characteristic data sequence of the business object within the first type of business period includes: Acquire a sub-period filling feature associated with the second-type sub-period, and determine the sub-period filling feature as a heterogeneous conversion feature corresponding to the second-type sub-period; Based on the heterogeneous conversion characteristics corresponding to the first type of sub-period and the heterogeneous conversion characteristics corresponding to the second type of sub-period, a heterogeneous characteristic data sequence of the business object in the first type of business period is determined.

9. The method according to claim 8, characterized in that The determining, based on the heterogeneous conversion characteristics corresponding to the first type of sub-period and the heterogeneous conversion characteristics corresponding to the second type of sub-period, of the heterogeneous characteristic data sequence of the business object in the first type of business period includes: Obtaining a sub-cycle time sequence of the W service sub-cycles, determining the sub-cycle time sequence of the first type of sub-cycles as a first sub-cycle time sequence, and determining the sub-cycle time sequence of the second type of sub-cycles as a second sub-cycle time sequence; Based on the first sub-cycle time sequence and the second sub-cycle time sequence, the heterogeneous conversion features corresponding to the first type of sub-cycle and the heterogeneous conversion features corresponding to the second type of sub-cycle are used to determine the heterogeneous feature data sequence of the business object in the first type of business cycle.

10. The method according to claim 1, characterized in that The payment cycle includes N business cycles, and the sequence characteristics in the payment business characteristic sequence include periodic transaction characteristics corresponding to each business cycle in the N business cycles; The second hierarchical network performs object recognition on the object type of the business object based on the sequence feature in the payment business feature sequence to obtain an object recognition result, including: The second hierarchical network performs feature aggregation processing based on the periodic transaction features corresponding to the N business periods to obtain object payment features of the business object; The second hierarchical network performs object identification on the object type of the business object based on the object payment feature to obtain the object identification result.

11. The method according to claim 1, wherein The object type of the business object includes a first object type; and the method further includes: When the object identification result indicates that the business object is of the first object type, object payment management is performed on the business object.

12. The method according to claim 1, characterized in that The method further comprises: Acquire sample data for training an initial business model; the sample data is determined based on a sample label and sample payment data, wherein the sample payment data refers to payment data of a sample object in a sample transaction sequence within a sample payment period; the sample label is used to indicate a reference object type of the sample object; Inputting the sample data into the initial business model, and performing object recognition on the object type of the sample object by the initial business model to obtain a sample recognition result; the sample recognition result is used to indicate the recognized object type of the sample object; Based on the reference object type and the identified object type, parameter adjustment processing is performed on the model parameters of the initial business model, and the target business model is determined based on the initial business model after the parameter adjustment processing.

13. The method according to claim 1, wherein The description dimension information includes first description information and second description information; The first hierarchical network performs word segmentation processing on the description-class dimension information to obtain text word segmentation of the description-class dimension information, including: Obtaining a first text length threshold associated with the first description information and a second text length threshold associated with the second description information; Based on the first text length threshold, performing length normalization processing on the first description information to obtain a first standard text corresponding to the first description information; Based on the second text length threshold, performing length normalization processing on the second description information to obtain a second standard text corresponding to the second description information; The first hierarchical network performs word segmentation processing on the first standard text and the second standard text to obtain text word segmentation of the description class dimension information.

14. A business data processing device, characterized in that: The device comprises: A transaction data acquisition module is configured to acquire payment transaction data for input into a target business model from a transaction sequence of a business object within a payment cycle; the target business model includes a first hierarchical network and a second hierarchical network; the payment transaction data is determined based on payment transactions initiated by the business object within a first type of business cycle; the payment transaction data includes time dimension information, numerical dimension information, and descriptive dimension information; the first type of business cycle is a business cycle within a payment cycle; a time mapping module, configured to input the time dimension information into the first hierarchical network, and perform feature mapping processing on the time dimension information by the first hierarchical network to obtain a time mapping feature corresponding to the time dimension information; a numerical mapping module, configured to input the numerical dimension information into the first hierarchical network, and perform feature mapping processing on the numerical dimension information by the first hierarchical network to obtain numerical mapping features corresponding to the numerical dimension information; a description conversion module, configured to input the description-class dimensional information into the first hierarchical network, perform word segmentation processing on the description-class dimensional information by the first hierarchical network to obtain text segmentations of the description-class dimensional information, and perform feature conversion processing on the text segmentations to obtain description conversion features corresponding to the description-class dimensional information; a heterogeneous conversion module, configured to perform heterogeneous conversion processing on the time mapping feature, the value mapping feature, and the description conversion feature by the first hierarchical network to obtain a heterogeneous feature data sequence of the business object within the first type of business cycle; The object identification module is used to determine the payment business feature sequence of the business object in the payment cycle based on the heterogeneous feature data sequence, input the payment business feature sequence into the second hierarchical network, and the second hierarchical network performs object identification on the object type of the business object based on the sequence features in the payment business feature sequence to obtain an object identification result.

15. A computer device, characterized in that: including memory and processor; The memory is connected to the processor, the memory is used to store a computer program, and the processor is used to call the computer program so that the computer device executes the method according to any one of claims 1 to 13.

16. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which is suitable for being loaded and executed by a processor, so that a computer device having the processor executes the method according to any one of claims 1 to 13.

17. A computer program product, characterized in that The method comprises a computer program / instruction, which implements the method according to any one of claims 1 to 13 when executed by a processor.

Images