Data processing method, device, computer equipment and storage medium

Abstract

This application relates to a data processing method, apparatus, computer device, and storage medium. The method includes: determining a target resource transfer record corresponding to a target object; obtaining target resource transfer data corresponding to a target data dimension from the resource transfer data corresponding to the target resource transfer record; obtaining a target resource transfer vector corresponding to the target resource transfer data; the target resource transfer vector is obtained by training according to a training resource transfer data sequence corresponding to the target data dimension, and the training resource transfer data in the training resource transfer data sequence are arranged in the order of resource transfer; obtaining a target resource transfer feature corresponding to the target resource transfer record based on the target resource transfer vector; and performing object recognition on the target object based on the target resource transfer feature to obtain a target recognition result. Using this method can improve the accuracy of object recognition. The target recognition result in this application can be obtained by model recognition based on artificial intelligence.

Description

Technical Field

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

[0002] With the development of computer technologies and big data technologies, object recognition is required in many scenarios. For example, for an e-commerce platform, in order to understand the interests of the objects on the platform, the objects on the platform are recognized, so that interesting products can be pushed to the objects according to the recognition results. For example, for mobile payment, by recognizing the object, it can be determined whether there is an illegal transaction by the object, such as whether there is a suspicion of money laundering.

[0003] Currently, multiple methods can be used for object recognition. For example, an object can be recognized by using a neural network model based on artificial intelligence.

[0004] However, the current methods for recognizing objects have the situation that objects cannot be accurately recognized, resulting in low object recognition accuracy. Summary of the Invention

[0005] Based on this, in view of the above technical problems, it is necessary to provide a data processing method, apparatus, computer device, and storage medium that can improve the accuracy of object recognition.

[0006] A data processing method, the method includes: determining a target resource transfer record corresponding to a target object; obtaining target resource transfer data corresponding to a target data dimension from resource transfer data corresponding to the target resource transfer record; obtaining a target resource transfer vector corresponding to the target resource transfer data; the target resource transfer vector is trained according to a training resource transfer data sequence corresponding to the target data dimension, and the training resource transfer data in the training resource transfer data sequence are arranged in the resource transfer order of the training resource transfer records corresponding to the training objects; obtaining a target resource transfer feature corresponding to the target resource transfer record based on the target resource transfer vector; and performing object recognition on the target object based on the target resource transfer feature to obtain a target recognition result corresponding to the target object.

[0007] A data processing device, the device comprising: a target resource transfer record determination module for determining a target resource transfer record corresponding to a target object; a target resource transfer data acquisition module for acquiring target resource transfer data corresponding to a target data dimension from the resource transfer data corresponding to the target resource transfer record; a target resource transfer vector acquisition module for acquiring a target resource transfer vector corresponding to the target resource transfer data; the target resource transfer vector being trained according to a training resource transfer data sequence corresponding to the target data dimension, and the training resource transfer data in the training resource transfer data sequence being arranged in the resource transfer order of the training resource transfer record corresponding to the training object; a target resource transfer feature obtaining module for obtaining a target resource transfer feature corresponding to the target resource transfer record based on the target resource transfer vector; a target recognition result obtaining module for performing object recognition on the target object based on the target resource transfer feature to obtain a target recognition result corresponding to the target object.

[0008] In some embodiments, the target resource transfer vector is selected from a set of training resource transfer vectors; the device further comprises a training resource transfer vector set obtaining module, and the training resource transfer vector set obtaining module comprises: a training resource transfer record sequence acquisition unit for acquiring a training resource transfer record sequence corresponding to a training object, the training resource transfer record sequence comprising a plurality of training resource transfer records arranged in the resource transfer order of the training resource transfer record; a training resource transfer data acquisition unit for acquiring training resource transfer data corresponding to the target data dimension from the resource transfer data corresponding to the training resource transfer record; a training resource transfer data sequence obtaining unit for arranging each of the training resource transfer data in the resource transfer order of the training resource transfer record to obtain the training resource transfer data sequence; a training resource transfer vector set composition unit for performing resource transfer vector training according to the training resource transfer data sequence to obtain training resource transfer vectors corresponding to each of the training resource transfer data and forming the set of training resource transfer vectors.

[0009] In some embodiments, there are multiple training resource transfer data sequences, and each of the training resource transfer data sequences forms a data sequence set; the training resource transfer vector set forming unit is further configured to obtain current training resource transfer data from the training resource transfer data sequences, and obtain associated resource transfer data corresponding to the current training resource transfer data in the training resource transfer data sequences; input the current training resource transfer data into a vector determination model to be trained to obtain a predicted association probability between the current training resource transfer data and the associated resource transfer data; obtain a standard association probability between the current training resource transfer data and the associated resource transfer data based on the data sequence set; adjust model parameters of the vector determination model based on the difference between the standard association probability and the predicted association probability to obtain a trained vector determination model, and obtain training resource transfer vectors corresponding to each of the training resource transfer data based on the trained vector determination model.

[0010] In some embodiments, the target resource transfer feature obtaining module includes: a first resource transfer data obtaining unit configured to obtain first resource transfer data corresponding to a first data dimension from the resource transfer data corresponding to the target resource transfer record; a first resource transfer feature determining unit configured to determine a first resource transfer feature corresponding to the first resource transfer data; and a first target resource transfer feature obtaining unit configured to splice the target resource transfer vector and the first resource transfer feature to obtain a target resource transfer feature corresponding to the target resource transfer record.

[0011] In some embodiments, there are multiple target resource transfer records, and the target recognition result obtaining module includes: a feature sequence obtaining unit configured to arrange the target resource transfer features corresponding to the target resource transfer records according to the resource transfer order corresponding to the target resource transfer records to obtain a feature sequence; and a target recognition result obtaining unit configured to input the feature sequence into an object recognition sequence model for processing to obtain a target recognition result corresponding to the target object.

[0012] In some embodiments, the target resource transfer feature obtaining module includes: a sorting feature obtaining unit configured to obtain a sorting feature corresponding to the target resource transfer vector according to the resource transfer order corresponding to the target resource transfer record; and a second target resource transfer feature obtaining unit configured to fuse the target resource transfer vector and the sorting feature to obtain a target resource transfer feature corresponding to the target resource transfer record.

[0013] In some embodiments, the target recognition result obtaining unit is further configured to obtain a recognition task matrix corresponding to the recognition task network in the object recognition sequence model; based on the target resource transfer feature in the feature sequence and the recognition task matrix, obtain the feature attention degrees respectively corresponding to the respective target resource transfer features in the feature sequence; adjust the corresponding target resource transfer features in the feature sequence based on the feature attention degrees to obtain adjusted target resource transfer features, and the adjusted target resource transfer features form an adjusted feature sequence; and obtain the target recognition result corresponding to the target object based on the adjusted feature sequence.

[0014] In some embodiments, the target resource transfer record determination module is further configured to obtain a resource transfer record of the target object for resource transfer to an exchange object as the target resource transfer record; the target recognition result obtaining module is further configured to perform object recognition on the target object based on the target resource transfer feature to obtain a recognition result that the target object is an object with abnormal resource transfer.

[0015] In some embodiments, the apparatus further includes a target data dimension obtaining module, and the target data dimension obtaining module includes: a data dimension set forming unit configured to obtain each data dimension corresponding to the target resource transfer record to form a data dimension set; a target data dimension obtaining unit configured to select a data dimension that meets a dimension selection condition from the data dimension set as the target data dimension; and the dimension selection condition includes that the data quantity of the data dimension is greater than a quantity threshold.

[0016] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the following steps are implemented: determining a target resource transfer record corresponding to a target object; obtaining target resource transfer data corresponding to a target data dimension from resource transfer data corresponding to the target resource transfer record; obtaining a target resource transfer vector corresponding to the target resource transfer data; the target resource transfer vector is obtained by training according to a training resource transfer data sequence corresponding to the target data dimension, and the training resource transfer data in the training resource transfer data sequence are arranged in the resource transfer order of the training resource transfer record corresponding to the training object; obtaining a target resource transfer feature corresponding to the target resource transfer record based on the target resource transfer vector; and performing object recognition on the target object based on the target resource transfer feature to obtain a target recognition result corresponding to the target object.

[0017] A computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, the following steps are implemented: determining a target resource transfer record corresponding to a target object; obtaining target resource transfer data corresponding to a target data dimension from resource transfer data corresponding to the target resource transfer record; obtaining a target resource transfer vector corresponding to the target resource transfer data; the target resource transfer vector is obtained by training according to a training resource transfer data sequence corresponding to the target data dimension, and the training resource transfer data in the training resource transfer data sequence are arranged according to the resource transfer order of the training resource transfer record corresponding to the training object; obtaining a target resource transfer feature corresponding to the target resource transfer record based on the target resource transfer vector; performing object recognition on the target object based on the target resource transfer feature to obtain a target recognition result corresponding to the target object.

[0018] In some embodiments, a computer program product or a computer program is provided. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the above method embodiments.

[0019] In the above data processing method, device, computer device, and storage medium, a target resource transfer record corresponding to a target object is determined, target resource transfer data corresponding to a target data dimension is obtained from resource transfer data corresponding to the target resource transfer record, a target resource transfer vector corresponding to the target resource transfer data is obtained, a target resource transfer feature corresponding to the target resource transfer record is obtained based on the target resource transfer vector, object recognition is performed on the target object based on the target resource transfer feature to obtain a target recognition result corresponding to the target object. Since the target resource transfer vector is obtained by training according to a training resource transfer data sequence corresponding to the target data dimension, and the training resource transfer data in the training resource transfer data sequence are arranged according to the resource transfer order of the training resource transfer record corresponding to the training object, the vector representation corresponding to the resource transfer data can be determined based on the resource transfer order, the accuracy of the generated resource transfer vector is improved, and further the accuracy of object recognition is improved.

[0020] A data processing method, the method comprising: obtaining a training resource transfer record sequence corresponding to a training object, the training resource transfer record sequence including a plurality of training resource transfer records arranged in the resource transfer order of the training resource transfer records; obtaining training resource transfer data corresponding to a target data dimension from the resource transfer data corresponding to the training resource transfer records; arranging each of the training resource transfer data in the resource transfer order of the training resource transfer records to obtain a training resource transfer data sequence; performing resource transfer vector training according to the training resource transfer data sequence to obtain training resource transfer vectors corresponding to each of the training resource transfer data.

[0021] A data processing device, the device comprising: a training resource transfer record sequence obtaining module, configured to obtain a training resource transfer record sequence corresponding to a training object, the training resource transfer record sequence including a plurality of training resource transfer records arranged in the resource transfer order of the training resource transfer records; a training resource transfer data obtaining module, configured to obtain training resource transfer data corresponding to a target data dimension from the resource transfer data corresponding to the training resource transfer records; a training resource transfer data sequence obtaining module, configured to arrange each of the training resource transfer data in the resource transfer order of the training resource transfer records to obtain a training resource transfer data sequence; a training resource transfer vector obtaining module, configured to perform resource transfer vector training according to the training resource transfer data sequence to obtain training resource transfer vectors corresponding to each of the training resource transfer data.

[0022] In some embodiments, there are a plurality of the training resource transfer data sequences, and each of the training resource transfer data sequences forms a data sequence set; the training resource transfer vector obtaining module includes: an associated resource transfer data obtaining unit, configured to obtain a current training resource transfer data from the training resource transfer data sequence, and obtain associated resource transfer data corresponding to the current training resource transfer data in the training resource transfer data sequence; a predicted association probability obtaining unit, configured to input the current training resource transfer data into a vector determination model to be trained to obtain a predicted association probability between the current training resource transfer data and the associated resource transfer data; a standard association probability obtaining unit, configured to obtain a standard association probability between the current training resource transfer data and the associated resource transfer data based on the data sequence set; a training resource transfer vector obtaining unit, configured to adjust model parameters of the vector determination model based on a difference between the standard association probability and the predicted association probability to obtain a trained vector determination model, and obtain training resource transfer vectors corresponding to each of the training resource transfer data based on the trained vector determination model.

[0023] A computer device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the following steps are implemented: obtaining a training resource transfer record sequence corresponding to a training object, where the training resource transfer record sequence includes multiple training resource transfer records arranged in the resource transfer order of the training resource transfer records; obtaining training resource transfer data corresponding to a target data dimension from the resource transfer data corresponding to the training resource transfer records; arranging each of the training resource transfer data in the resource transfer order of the training resource transfer records to obtain a training resource transfer data sequence; performing resource transfer vector training according to the training resource transfer data sequence to obtain training resource transfer vectors corresponding to each of the training resource transfer data.

[0024] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the following steps are implemented: obtaining a training resource transfer record sequence corresponding to a training object, where the training resource transfer record sequence includes multiple training resource transfer records arranged in the resource transfer order of the training resource transfer records; obtaining training resource transfer data corresponding to a target data dimension from the resource transfer data corresponding to the training resource transfer records; arranging each of the training resource transfer data in the resource transfer order of the training resource transfer records to obtain a training resource transfer data sequence; performing resource transfer vector training according to the training resource transfer data sequence to obtain training resource transfer vectors corresponding to each of the training resource transfer data.

[0025] In some embodiments, a computer program product or a computer program is provided. The computer program product or the computer program includes computer instructions that are 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, causing the computer device to execute the steps in the above method embodiments.

[0026] The above data processing method, device, computer equipment and storage medium obtain a training resource transfer record sequence corresponding to a training object, obtain training resource transfer data corresponding to a target data dimension from the resource transfer data corresponding to the training resource transfer record, arrange each piece of training resource transfer data in the resource transfer order of the training resource transfer record to obtain a training resource transfer data sequence, and perform resource transfer vector training according to the training resource transfer data sequence to obtain training resource transfer vectors corresponding to each piece of training resource transfer data. Since the training resource transfer record sequence includes multiple training resource transfer records arranged in the resource transfer order of the training resource transfer record, the vector representation corresponding to the resource transfer data can be determined based on the resource transfer order, improving the accuracy of the generated resource transfer vectors and further improving the accuracy of object recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is an application environment diagram of the data processing method in some embodiments;

[0028] Figure 2 It is a flowchart of the data processing method in some embodiments;

[0029] Figure 3A It is a schematic diagram of the principle of generating a resource transfer vector in some embodiments;

[0030] Figure 3B It is a flowchart of the data processing method in some embodiments;

[0031] Figure 4 It is a schematic diagram of the principle of generating a training sample in some embodiments;

[0032] Figure 5 It is the structure of the cbow model in some embodiments;

[0033] Figure 6 It is the structure of the skip-gram model in some embodiments;

[0034] Figure 7 It is the structure diagram of the Transformer in some embodiments;

[0035] Figure 8 It is the structure diagram of the encoder in some embodiments;

[0036] Figure 9 It is a schematic diagram of the principle of calculating attention in some embodiments;

[0037] Figure 10 It is a schematic diagram of the principle of calculating attention in some embodiments;

[0038] Figure 11 It is a structural block diagram of the data processing device in some embodiments;

[0039] Figure 12 is a structural block diagram of a data processing device in some embodiments;

[0040] Figure 13 is an internal structure diagram of a computer device in some embodiments;

[0041] Figure 14 is an internal structure diagram of a computer device in some embodiments. Detailed implementation manners

[0042] In order to make the purpose, technical solutions and advantages of the present application clearer, the following further details the present application in combination with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0043] Artificial Intelligence (AI) is to use a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, and is a theory, method, technology and application system that perceives the environment, acquires knowledge and uses knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science. It attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a way similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines, so that the machines have the functions of perception, reasoning and decision-making.

[0044] Artificial intelligence technology is an interdisciplinary subject, involving a wide range of fields, including both hardware-level technologies and software-level technologies. Artificial intelligence basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technologies mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0045] Natural Language Processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that can achieve effective communication between humans and computers in natural language. Natural language processing is a science that integrates linguistics, computer science and mathematics. Therefore, the research in this field will involve natural language, that is, the language used by people in daily life, so it has a close connection with the research of linguistics. Natural language processing technologies usually include technologies such as text processing, semantic understanding, machine translation, robot question answering, and knowledge graphs.

[0046] Machine Learning (ML) is an interdisciplinary subject that involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve their own performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and learning from demonstrations.

[0047] With the research and progress of artificial intelligence technology, artificial intelligence technology has been studied and applied in multiple fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, driverless, autonomous driving, drones, robots, smart healthcare, smart customer service, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0048] The solution provided in the embodiments of this application relates to technologies such as machine learning in artificial intelligence, and is specifically described through the following embodiments:

[0049] The data processing method provided in this application can be applied to the Figure 1 application environment as shown. Among them, the terminal 102 communicates with the server 104 through the network.

[0050] Specifically, the server 104 can obtain a sequence of training resource transfer records corresponding to a training object. The sequence of training resource transfer records includes multiple training resource transfer records arranged in the resource transfer order of the training resource transfer records. Obtain the training resource transfer data corresponding to the target data dimension from the resource transfer data corresponding to the training resource transfer records, arrange each training resource transfer data in the resource transfer order of the training resource transfer records to obtain a sequence of training resource transfer data, and perform resource transfer vector training according to the sequence of training resource transfer data to obtain training resource transfer vectors corresponding to each training resource transfer data, and form a set of training resource transfer vectors. The terminal 102 can send an object recognition request for the target object to the server. The server 104 can respond to the object recognition request, determine the target resource transfer record corresponding to the target object, obtain the target resource transfer data corresponding to the target data dimension from the resource transfer data corresponding to the target resource transfer record, and obtain the target resource transfer vector corresponding to the target resource transfer data from the set of training resource transfer vectors. The server 104 can obtain the target resource transfer feature corresponding to the target resource transfer record based on the target resource transfer vector, perform object recognition on the target object based on the target resource transfer feature, and obtain the target recognition result corresponding to the target object. The server 104 can return the target recognition result to the terminal 102, and the terminal 102 can display the target recognition result.

[0051] Among them, the terminal 102 can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, and portable wearable devices. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers.

[0052] It can be understood that the above application scenario is only an example and does not constitute a limitation on the data processing method provided by the embodiments of the present application. The method provided by the embodiments of the present application can also be applied in other application scenarios. For example, the data processing method provided by the present application can be executed by the terminal 102. The terminal 102 can upload the obtained set of training resource transfer vectors and the target recognition result to the server 104. The server 104 can store the set of training resource transfer vectors and the target recognition result, or forward the set of training resource transfer vectors and the target recognition result to other terminal devices.

[0053] In some embodiments, as Figure 2 shown, a data processing method is provided. This method can be executed by the terminal or the server, or jointly executed by the terminal and the server. In the embodiments of the present application, taking this method applied to Figure 1 the server 104 as an example for illustration, it includes the following steps:

[0054] S202. Determine the target resource transfer record corresponding to the target object.

[0055] Herein, an object refers to the subject that needs to be identified. For example, it can be a user, such as a network user. A network user can be, for example, a registered user on an application or a network platform. An object can be represented by an account. One account represents one object. For example, an object can be represented by a bank card number, a stock account, or a fund account. The target object can be any object that has transferred resources. There can be multiple target objects.

[0056] A resource refers to a resource that exists in an electronic account and can circulate, such as a resource that can circulate on the Internet through an account. The account can include at least one of a bank card number or an Internet payment account number. Resources can include funds in the account, virtual red envelopes, game coins, or virtual items, etc. Resource transfer means transferring resources from one object account to another object account.

[0057] The resource transfer record is used to record relevant information about resource transfer. The resource transfer record can include at least one of the resource transfer time, the resource transfer account, or the resource transfer amount, etc. Among them, the resource transfer time refers to the time when the resources are transferred. The resource transfer account can include at least one of the resource transfer-out account or the resource transfer-in account. The resource transfer-out account refers to the account from which the resources are transferred, and the resource transfer-in account refers to the account into which the resources are transferred. The resource transfer amount refers to the amount of the transferred resources. The resource transfer record can be, for example, the order details corresponding to the order for purchasing goods.

[0058] Specifically, the server can store resource transfer records corresponding to multiple objects respectively. The multiple objects can include objects from multiple object sources. The object source can be determined according to the application used by the object to transfer resources. The object source can include at least one of a bank application, a social application, or an instant messaging application. For example, when an object transfers resources through bank application A, the object source can be determined as bank application A. The server can obtain multiple objects belonging to the same object source and use the multiple objects belonging to the same object source as respective target objects. The server can store multiple resource transfer records corresponding to the target object. Among them, the resource transfer record corresponding to the target object can include at least one of the resource transfer record with the resource transfer-out party being the target object or the resource transfer record with the resource transfer-in party being the target object. The resource transfer-out party refers to the object that transfers the resources, that is, the object to which the resource transfer-out account belongs. The resource transfer-in party refers to the object that receives the resources, that is, the object to which the resource transfer-in account belongs.

[0059] In some embodiments, the server may use each resource transfer record corresponding to the target object stored as each target resource transfer record, or may screen out the target resource transfer record from each resource transfer record corresponding to the target object. The server may obtain, according to the resource transfer time, the resource transfer records that meet the transfer time selection condition from each resource transfer record of the target object as the target resource transfer records corresponding to the target object. The transfer time selection condition may include at least one of that the time interval between the resource transfer time and the current time is less than the time interval threshold or the resource transfer time belongs to the transfer time range. Among them, the time interval threshold may be preset, for example, it may be 24 hours. The transfer time range may be a preset time range, for example, it may be "from January 1, 2020 to January 5, 2020".

[0060] In some embodiments, the server may obtain the resource transfer time from the resource transfer records corresponding to the target object, calculate the time interval between the current time and the resource transfer time as the resource time interval, compare the resource time interval with the time interval threshold, and when it is determined that the resource time interval is less than the time interval threshold, use the resource transfer record as the target resource transfer record corresponding to the target object.

[0061] In some embodiments, the server may obtain the resource transfer time from the resource transfer records corresponding to the target object. When it is determined that the resource transfer time belongs to the transfer time range, that is, when it is determined that the transfer time range includes the resource transfer time. For example, assuming the resource transfer time is "January 3, 2020", if the transfer time range is "from January 1, 2020 to January 5, 2020", it can be determined that the resource transfer time "January 3, 2020" belongs to the transfer time range "from January 1, 2020 to January 5, 2020". The server may use the resource transfer records whose resource transfer time belongs to the transfer time range as the target resource transfer records corresponding to the target object.

[0062] S204, obtain the target resource transfer data corresponding to the target data dimension from the resource transfer data corresponding to the target resource transfer record.

[0063] Among them, the target resource transfer record may include one or more resource transfer data, and "a plurality" means at least two. Resource transfer data refers to data related to resource transfer, and may include at least one of resource transfer amount, resource transfer time, resource transfer account, resource transfer location, resource transfer device, exchange rate, country type, resource transfer account type, resource transfer channel, resource transfer network address, financial institution type, resource transfer type, resource transfer time interval, resource balance, resource transfer merchant identifier, credit card label, credit card limit, or resource recipient information, etc. The resource transfer amount may be the amount of the transferred resource, for example, it may be 100 yuan in RMB. The resource transfer time may include a date or a moment. The date may be represented by year, month, day, or week, and the moment may be represented by hour, minute, or second. For example, the resource transfer time may be March 5, 2020. The resource transfer account may be, for example, "abc123dhf456sdsdk". The resource transfer location may be, for example, "ABC Province". The resource transfer device may be, for example, "a mobile phone of the shfjh model". The resource transfer account type refers to the type of the account for resource transfer. For example, it may be a payment account type, and the payment account type may be the type of a bank card, and may include at least one of a credit card, a debit card, or a stored-value card. The resource transfer channel may be, for example, a payment method, and may include a bank card or a payment application software. The resource transfer network address refers to the network address of the device for resource transfer. For example, it may be an IP (Internet Protocol) address. The financial institution type may be, for example, a bank type. The resource transfer type may be either belonging to payment or not belonging to payment. The resource transfer time interval refers to the time interval between two adjacent resource transfers. For example, if the first resource transfer is made on March 1, 2020, and the second resource transfer is made on March 2, 2020, then the resource transfer time interval is 1 day. The resource balance refers to the remaining amount of resources. For example, it may be the remaining balance. The resource transfer merchant identifier is used to uniquely identify a merchant. For example, it may be a merchant number. For example, if product of merchant B is purchased on shopping platform A, then shopping platform A is the resource transfer merchant. The credit card label may include a label belonging to a credit card or a label not belonging to a credit card. The resource recipient may also be referred to as a counterparty, and the resource recipient information may include at least one of gender, age, region, or network address.

[0064] A resource transfer record can correspond to multiple data dimensions. The resource transfer record can include resource transfer data corresponding to multiple data dimensions respectively. A data dimension refers to the dimension to which the data corresponds. The data dimensions corresponding to the resource transfer record can include at least one of the resource transfer amount, the resource transfer time, the resource transfer account, the resource transfer location, or the resource transfer device. The target data dimension can include at least one data dimension among the respective data dimensions corresponding to the resource transfer record. There can be multiple target data dimensions. The target data dimension can be preset or set according to needs. The target resource transfer data refers to the resource transfer data of the target data dimension in the target resource transfer record.

[0065] For example, the target resource transfer record is "resource transfer amount = 100 RMB, resource transfer time = March 5, 2020, resource transfer account = abc123dhf456sdsdk, resource transfer location = Province ABC, resource transfer device = mobile phone of shfjh model, merchant number = 213213". If the target data dimension is "merchant number", then the target resource transfer data is "213213". If the target data dimension is "resource transfer location", then the target resource transfer data is "Province ABC".

[0066] Specifically, the server can determine the respective data dimensions corresponding to the target resource transfer record, form a data dimension set, and can obtain at least one data dimension from the data dimension set as the target data dimension. For example, all the data dimensions in the data dimension set can be used as the target data dimension, or data dimensions that meet the dimension selection conditions can be selected from the data dimension set as the target data dimension. The dimension selection conditions can include at least one of the data quantity of the data dimension being greater than the quantity threshold or the data type corresponding to the data dimension being a specific data type.

[0067] The data quantity refers to the quantity of the data corresponding to the data dimension. The data of the data dimension "gender" are "male" and "female", so the data quantity is 2. The quantity threshold can be preset or calculated based on the data quantity corresponding to the data dimensions in the data dimension set. For example, for data dimensions for which the data quantity can be determined, such as when the data corresponding to the data dimension is countable, the data quantity corresponding to the data dimension can be obtained. When there are multiple data dimensions for which the data quantity can be determined, the average operation can be performed on the data quantities corresponding to the multiple data dimensions, and the result of the average operation can be used as the average quantity. A quantity coefficient can be obtained, and the product operation can be performed on the average quantity and the quantity coefficient, and the result of the product operation can be used as the quantity threshold. The quantity coefficient can be preset, for example, it can be 1.2.

[0068] The data type can include at least one of discrete type or continuous type. The discrete type can include at least one of discrete countable type or discrete uncountable type. Each data dimension corresponds to a data type, and the data types corresponding to each data dimension can be the same or different, that is, each data type can correspond to multiple data dimensions. The data corresponding to the discrete data dimension is discrete, and the data corresponding to the continuous data dimension is continuous. The discrete countable type means that the data corresponding to the data dimension is discrete and countable, and the discrete uncountable type means that the data corresponding to the data dimension is discrete and uncountable. For example, the data corresponding to the data dimension "amount" is continuous, so the data dimension "amount" is of continuous type. For example, the data corresponding to the data dimension "merchant number" and the data dimension "gender" is discrete, so the data dimension "merchant number" and the data dimension "gender" are of discrete type. Since the data of the data dimension "merchant number" is uncountable, that is, the number of all "merchant numbers" is unknown, the data dimension "merchant number" is of discrete uncountable type. The data of the data dimension "gender" is "male" or "female", and the data of the data dimension "gender" is countable, so the data dimension "gender" is of discrete countable type. A specific data type can be preset, for example, it can be a data dimension of discrete uncountable type.

[0069] In some embodiments, the server can obtain a target data dimension from the data dimension set according to the data type corresponding to the data dimension. For example, the server can obtain the data dimension corresponding to a specific data type from the data dimension set as the target data dimension.

[0070] S206, obtain a target resource transfer vector corresponding to the target resource transfer data; the target resource transfer vector is obtained by training according to the training resource transfer data sequence corresponding to the target data dimension, and the training resource transfer data in the training resource transfer data sequence is arranged according to the resource transfer order of the training resource transfer records corresponding to the training object.

[0071] Among them, the resource transfer vector is the vector representation form corresponding to the resource transfer data. The target resource transfer vector refers to the resource transfer vector corresponding to the target resource transfer data.

[0072] The training resource transfer data sequence includes multiple training resource transfer data. Each training resource transfer data is arranged according to the resource transfer order. The earlier the resource transfer order, the more forward the sorting of the training resource transfer data in the training resource transfer data sequence. The training resource transfer data is the resource transfer data for training. The training resource transfer data sequence may include a training resource transfer data sequence identical to the target resource transfer data. There may be multiple training resource transfer data sequences. For example, it may include training resource transfer data sequences corresponding to multiple training objects respectively. Each training resource transfer data in each training resource transfer data sequence belongs to the same training object. The training object may be the same as or different from the target object.

[0073] The resource transfer order refers to the order of transferring resources. The resource transfer order can be determined according to the resource transfer time. The earlier the resource transfer time, the earlier the resource transfer order. The later the resource transfer time, the later the resource transfer order. For example, if the resource transfer time of resource transfer record S1 is January 3, 2020, and the resource transfer time of resource transfer record S2 is January 5, 2020, then the resource transfer order of resource transfer record S1 is before the resource transfer order of resource transfer record S2.

[0074] Specifically, the server can use the training resource transfer data sequence for resource transfer vector training to obtain the training resource transfer vectors corresponding to each training resource transfer data respectively. The training resource transfer vector is the vector representation form corresponding to the training resource transfer data. The training resource transfer vectors can be grouped into a training resource transfer vector set. Since the training resource transfer data sequence includes a training resource transfer data sequence identical to the target resource transfer data, the target resource transfer vector corresponding to the target resource transfer vector can be obtained from the training resource transfer vector set.

[0075] In some embodiments, the server can obtain multiple training resource transfer records corresponding to the same training object, arrange each training resource transfer record according to the resource transfer order to obtain the training resource transfer record sequence corresponding to the training object. The training resource transfer record includes multiple training resource transfer data. There may be multiple training objects, so multiple training resource transfer record sequences corresponding to multiple training objects respectively can be obtained. Each training resource transfer record in each training resource transfer record sequence belongs to the same training object.

[0076] In some embodiments, the server may obtain training resource transfer data corresponding to the target data dimension from each training resource transfer record in the training resource transfer record sequence, and arrange the training resource transfer data corresponding to the training resource transfer records according to the sorting of the training resource transfer records in the training resource transfer record sequence, to obtain a training resource transfer data sequence. There may be multiple training resource transfer record sequences. For example, based on the training resource transfer record sequences of multiple different training objects, training resource transfer data sequences corresponding to each training object can be obtained. For example, each training resource transfer data sequence is used to train a vector determination model to obtain vector representation forms corresponding to each different training resource transfer data. The training resource transfer data sequence may include, for example Figure 3A "The merchant number sequence of object 1 = [a1, a2, a3, a1, a3, a2, a4]", "The merchant number sequence of object 2 = [a2, a4, a3, a1, a2, a2, a3]", "The merchant number sequence of object 3 = [a3, a1, a3, a2, a4, a2, a4]", and "The merchant number sequence of object 4 = [a4, a1, a3, a1, a3, a2, a4]", where a1, a2, a3, and a4 are different merchant numbers. By training the vector determination model, vector representation forms corresponding to a1, a2, a3, and a4 can be obtained.

[0077] S208. Obtain a target resource transfer feature corresponding to the target resource transfer record based on the target resource transfer vector.

[0078] Among them, the target resource transfer feature refers to the feature corresponding to the target resource transfer record. The target resource transfer feature may include the target resource transfer vector.

[0079] Specifically, the target resource transfer feature refers to the feature corresponding to the target resource transfer record. The server may use the target resource transfer vector as the target resource transfer feature corresponding to the target resource transfer record, or the server may obtain the target resource transfer feature based on the resource transfer data of the first data dimension in the target resource transfer record and the target resource transfer vector. The first data dimension is different from the target data dimension, and the first data dimension may include at least one of the data dimensions different from the target data dimension among the respective data dimensions corresponding to the target resource transfer record. There may be multiple first data dimensions. For example, all the data dimensions different from the target data dimension among the respective data dimensions corresponding to the target resource transfer record may be used as the first data dimension.

[0080] In some embodiments, the server may obtain the resource transfer data corresponding to the first data dimension from the target resource transfer record as the first resource transfer data, obtain the feature corresponding to the first resource transfer data as the first resource transfer feature, and obtain the target resource transfer feature corresponding to the target resource transfer record based on the target resource transfer vector and the first resource transfer feature. For example, the server may splice the target resource transfer vector and the first resource transfer feature to obtain the target resource transfer feature corresponding to the target resource transfer record. Splicing means connecting in sequence. Among them, the server may obtain the first resource transfer feature by normalizing the first resource transfer data or encoding the first resource transfer data. The encoding may be, for example, one-hot encoding.

[0081] In some embodiments, the server may obtain multiple target resource transfer records of the same target object, arrange each target resource transfer record according to the resource transfer order, and use the arranged sequence as the target resource transfer record sequence. The server may determine the arrangement feature corresponding to each target resource transfer record according to the sorting (resource transfer order) of the target resource transfer record in the target resource transfer record sequence, and obtain the target resource transfer feature corresponding to the target resource transfer record based on the sorting feature of the target resource transfer record and the corresponding target resource transfer vector. For example, the server may fuse the target resource transfer vector and the sorting feature, and use the fused result as the target resource transfer feature. Among them, the fusion may include at least one of splicing, adding, or multiplying. Splicing means connecting in sequence. For example, the number of feature values included in the sorting feature may be the same as the number of vector values included in the target resource transfer vector. The server may add the target resource transfer vector and the sorting feature, and use the added result as the target resource transfer feature. Among them, adding the target resource transfer vector and the sorting feature means adding the vector values in the target resource transfer vector and the feature values at the corresponding positions in the sorting feature.

[0082] In some embodiments, the server may splice based on the target resource transfer vector and the first resource transfer feature, use the spliced result as the initial resource transfer feature, fuse the initial resource transfer feature and the sorting feature, and use the fused feature as the target resource transfer feature. Among them, the number of feature values included in the sorting feature may be the same as the number of feature values included in the initial resource transfer feature.

[0083] S210. Perform object recognition on the target object based on the target resource transfer feature to obtain the target recognition result corresponding to the target object.

[0084] Among them, object recognition may include at least one of recognition of object value, recognition of object label, or recognition of abnormal objects. Recognition of object value is used to recognize the value of an object, and recognition of object label is used to recognize the label of an object. The target recognition result is the recognition result obtained by recognizing the target object.

[0085] Specifically, the server may input a single target resource transfer feature into the object recognition model for object recognition, or may input a sequence composed of multiple target resource transfer features into the object recognition model for object recognition. For example, the server may obtain multiple target resource transfer features corresponding to the same target object, arrange each target resource transfer feature according to the resource transfer order to obtain a feature sequence, and input the feature sequence into the object recognition model for object recognition.

[0086] Among them, the object recognition model is a model used to recognize an object. The object recognition model may be a neural network model based on deep learning and may be a trained model. For example, it may be a model trained using training resource transfer features. The training resource transfer features refer to the resource transfer features for training the model by the object. The process of obtaining the training resource transfer features may refer to the relevant steps of the target resource transfer features and will not be elaborated here.

[0087] In some embodiments, the object recognition model may be a trained object scoring model. The object scoring model is used to determine the score of an object. When the object scoring model is a model trained with each training resource transfer feature as a training sample respectively, the target resource transfer feature may be input into the object scoring model for object recognition to obtain a target recognition result including the object score corresponding to the target object. When the object scoring model is a model trained with a sequence formed by arranging multiple training resource transfer features in the resource transfer order as a training sample, the feature sequence may be input into the object scoring model for object recognition to obtain a target recognition result including the object score corresponding to the target object.

[0088] In some embodiments, the object recognition model may be a trained object label classification model. The object label classification model is used to generate object labels. When the object label classification model is a model trained with each training resource transfer feature as a training sample respectively, the target resource transfer feature may be input into the object label classification model for object recognition to obtain a target recognition result including the object label corresponding to the target object. When the object label classification model is a model trained with a sequence formed by arranging multiple training resource transfer features in the resource transfer order as a training sample, the feature sequence may be input into the object label classification model for object recognition to obtain a target recognition result including the object label corresponding to the target object.

[0089] In some embodiments, the object recognition model may be a trained object detection model. The object detection model is used to determine the object type, which can be any one of a resource transfer abnormal object or a resource transfer normal object. The abnormal object may be a malicious object with malicious behavior, such as a malicious object with malicious behaviors such as gambling, porn, fraud, money laundering, or remittance. When the object detection model is a model trained with each training resource transfer feature as a training sample, the target resource transfer feature can be input into the object detection model for object recognition to obtain a target recognition result including the object type corresponding to the target object. When the object detection model is a model trained with a sequence formed by arranging multiple training resource transfer features in the resource transfer order, the feature sequence can be input into the object detection model for object recognition to obtain a target recognition result including the object type corresponding to the target object. In the embodiments of the present application, the law of resource transfer by the object can be captured for object recognition. For example, when identifying a money laundering object, it can be recognized based on the order of the money laundering orders of the object, improving the recognition accuracy.

[0090] In the above data processing method, the target resource transfer record corresponding to the target object is determined, the target resource transfer data corresponding to the target data dimension is obtained from the resource transfer data corresponding to the target resource transfer record, the target resource transfer vector corresponding to the target resource transfer data is obtained, the target resource transfer feature corresponding to the target resource transfer record is obtained based on the target resource transfer vector, and the target object is recognized based on the target resource transfer feature to obtain the target recognition result corresponding to the target object. Since the target resource transfer vector is trained according to the training resource transfer data sequence corresponding to the target data dimension, and the training resource transfer data in the training resource transfer data sequence is arranged according to the resource transfer order of the training resource transfer record corresponding to the training object, the vector representation corresponding to the resource transfer data can be determined based on the resource transfer order, improving the accuracy of the generated resource transfer vector and further improving the accuracy of object recognition.

[0091] In some embodiments, the target resource transfer vector is selected from a set of training resource transfer vectors; the steps of obtaining the set of training resource transfer vectors include: obtaining a sequence of training resource transfer records corresponding to a training object, where the sequence of training resource transfer records includes multiple training resource transfer records arranged in the resource transfer order of the training resource transfer records; obtaining the training resource transfer data corresponding to the target data dimension from the resource transfer data corresponding to the training resource transfer records; arranging the respective training resource transfer data in the resource transfer order of the training resource transfer records to obtain a sequence of training resource transfer data; performing resource transfer vector training based on the sequence of training resource transfer data to obtain the training resource transfer vectors corresponding to the respective training resource transfer data, and forming a set of training resource transfer vectors.

[0092] Among them, the training resource transfer record is a resource transfer record used for resource transfer vector training. There can be multiple sequences of training resource transfer records. For example, it can include sequences of training resource transfer records corresponding to multiple training objects respectively, and the training resource transfer records in each sequence of training resource transfer records belong to the same training object. The earlier the training resource transfer record is sorted in the sequence of training resource transfer records, the earlier the training resource transfer data corresponding to the training resource transfer record is sorted in the sequence of training resource transfer data. The set of training resource transfer vectors includes the target resource transfer vector.

[0093] Specifically, the server can obtain a vector determination model to be trained, train the vector determination model to be trained based on the sequence of training resource transfer data to obtain a trained vector determination model, and obtain the training resource transfer vectors corresponding to the respective training resource transfer data based on the trained vector determination model. The vector determination model can be a neural network model based on artificial intelligence.

[0094] In some embodiments, the server can obtain the current training resource transfer data from the sequence of training resource transfer data, and obtain the associated resource transfer data corresponding to the current training resource transfer data in the sequence of training resource transfer data. The server can form a training sample with the current training resource transfer data and the associated resource transfer data corresponding to the current training resource transfer data, and use the training sample to train the vector determination model to obtain a trained vector determination model.

[0095] Among them, the current training resource transfer data can be any training resource transfer data in the training resource transfer data sequence. The associated resource transfer data can include each training resource transfer data in the training resource transfer data sequence whose data sorting interval from the current training resource transfer data is less than the sorting interval threshold. Data sorting refers to the sorting of training resource transfer data in the training resource transfer data sequence. The data sorting interval refers to the interval between the current data sorting and the non-current data sorting. The current data sorting refers to the sorting of the current training resource transfer data in the training resource transfer data sequence, and the non-current data sorting refers to the sorting of the training resource transfer data other than the current training resource transfer data in the training resource transfer data sequence in the training resource transfer data sequence. For example, if the current data sorting is 1 and the non-current data sorting is 3, then the data sorting interval is 2. If the current data sorting is 4 and the non-current data sorting is 1, then the data sorting interval is 3.

[0096] The sorting interval threshold can be preset. For example, it can be a fixed value such as 2 or 3, or it can be calculated based on the number of training resource transfer data included in the training resource transfer data sequence. For example, the number of training resource transfer data included in the training resource transfer data sequence can be counted to obtain the total number of data, a quantity factor can be obtained, and the quantity factor and the total number of data are multiplied, and the result of the multiplication operation is used as the sorting interval threshold. The quantity factor can be preset. For example, it can be 0.1 or 0.2.

[0097] In some embodiments, the vector determination model can include an input layer, a hidden layer, and an output layer. The hidden layer can be represented by a hidden matrix. The server can obtain the hidden matrix corresponding to the hidden layer of the trained vector determination model, and obtain the training resource transfer vectors corresponding to each training resource transfer data from the hidden matrix. For example, each column in the hidden matrix can be used as a training resource transfer vector.

[0098] In this embodiment, the training resource transfer record sequence corresponding to the training object is obtained, the training resource transfer data corresponding to the target data dimension is obtained from the resource transfer data corresponding to the training resource transfer record, and each training resource transfer data is arranged in the resource transfer order of the training resource transfer record to obtain the training resource transfer data sequence. Resource transfer vector training is performed according to the training resource transfer data sequence to obtain the training resource transfer vectors corresponding to each training resource transfer data, and a training resource transfer vector set is formed. Since the training resource transfer record sequence includes multiple training resource transfer records arranged in the resource transfer order of the training resource transfer record, the process of generating the resource transfer vector is affected by the resource transfer order, which improves the accuracy of the generated resource transfer vector and further improves the accuracy of object recognition.

[0099] In some embodiments, there are multiple training resource transfer data sequences, and each training resource transfer data sequence forms a data sequence set; training the resource transfer vector according to the training resource transfer data sequence to obtain the training resource transfer vector corresponding to each training resource transfer data includes: obtaining the current training resource transfer data from the training resource transfer data sequence, and obtaining the associated resource transfer data corresponding to the current training resource transfer data in the training resource transfer data sequence; inputting the current training resource transfer data into the vector determination model to be trained to obtain the predicted association probability between the current training resource transfer data and the associated resource transfer data; obtaining the standard association probability between the current training resource transfer data and the associated resource transfer data based on the data sequence set; adjusting the model parameters of the vector determination model based on the difference between the standard association probability and the predicted association probability to obtain the trained vector determination model, and obtaining the training resource transfer vector corresponding to each training resource transfer data based on the trained vector determination model.

[0100] Among them, there may be one or more associated resource transfer data corresponding to the current training resource transfer data. For example, the training resource transfer data adjacent to the current training resource transfer data forward and the training resource transfer data adjacent to the current training resource transfer data backward in the training resource transfer data sequence can be used as the associated resource transfer data corresponding to the current training resource transfer data.

[0101] The associated resource transfer data is the training resource transfer data at the surrounding positions of the current training resource transfer data in the training resource transfer data sequence. The predicted association probability is the probability that the resource transfer data at the surrounding position where the vector determination model predicts the associated resource transfer data is the associated resource transfer data. The surrounding positions may include multiple positions. For example, the surrounding positions may be each position whose distance from the current position is less than the distance threshold. The current position refers to the position of the current training resource transfer data in the training resource transfer data sequence. The distance threshold can be preset, for example, it can be 2. Among them, the position can be represented by sorting. The standard association probability is the probability that the resource transfer data at the surrounding position of the associated resource transfer data is the associated resource transfer data.

[0102] Specifically, the server can obtain the training resource transfer data sequences corresponding to multiple training objects respectively. The training resource transfer data in each training resource transfer data sequence is the resource transfer data of the target data dimension. The server can obtain different training resource transfer data from each training resource transfer data sequence, form a training resource transfer data set, perform vector conversion on each training resource transfer data in the training resource transfer data set, and generate the original vectors corresponding to each training resource transfer data respectively. For example, the original vectors can be generated by one-hot encoding. The server can perform resource transfer vector training based on each original vector to obtain the training resource transfer vectors corresponding to each training resource transfer data.

[0103] In some embodiments, the vector determination model can predict the surrounding resource transfer data of the current training resource transfer data to obtain the data distributions corresponding to each surrounding position of the current training resource transfer data. The data distributions corresponding to the surrounding positions can include the prediction probabilities corresponding to each training resource transfer data in the training resource transfer data set. The prediction probability corresponding to the training resource transfer data is used to represent the probability that this surrounding position is this training resource transfer data. The server can obtain the data distributions corresponding to the surrounding positions of each associated resource transfer data respectively, and the data distribution includes the prediction association probability.

[0104] In some embodiments, the server can determine the positions of the current training resource transfer data in each training resource transfer data sequence of the data sequence set to obtain each current position, and form a current position set. When the current training resource transfer data appears 2 times in a training resource transfer data sequence, there are 2 current positions corresponding to this training resource transfer data sequence. The server can determine the first surrounding positions corresponding to each current position from each training resource transfer data sequence, obtain the training resource transfer data at each first surrounding position, form a surrounding resource transfer data set, and count the probability of the same training resource transfer data appearing in the surrounding resource transfer data set as the standard association probability corresponding to this training resource transfer data. Among them, the first surrounding position can be any position in the surrounding positions corresponding to the current position, for example, it can be the position adjacent backward to the current position.

[0105] In some embodiments, the server may calculate the difference between the standard association probability and the predicted association probability to obtain the association probability difference, adjust the model parameters of the vector determination model, and obtain the adjusted model parameters, such that the association probability difference changes in a decreasing direction until the parameter adjustment stop condition is satisfied. The vector determination model corresponding to the model parameters that satisfy the parameter adjustment stop condition is determined as the trained vector determination model. The parameter adjustment stop condition may include at least one of the difference between the association probability differences of two adjacent times being less than the probability difference threshold, or the degree of change of the model parameters being less than the change degree threshold. The probability difference threshold and the change degree threshold may be preset. The server may obtain the hidden matrix corresponding to the hidden layer in the trained vector determination model, and obtain the training resource transfer vectors corresponding to the respective training resource transfer data according to the hidden matrix.

[0106] In this embodiment, the current training resource transfer data is obtained from the training resource transfer data sequence, and the associated resource transfer data corresponding to the current training resource transfer data in the training resource transfer data sequence is obtained. The current training resource transfer data is input into the vector determination model to be trained, and the predicted association probability between the current training resource transfer data and the associated resource transfer data is obtained. Based on the data sequence set composed of the respective training resource transfer data sequences, the standard association probability between the current training resource transfer data and the associated resource transfer data is obtained. The model parameters of the vector determination model are adjusted based on the difference between the standard association probability and the predicted association probability, and the trained vector determination model is obtained. The training resource transfer vectors corresponding to the respective training resource transfer data are obtained based on the trained vector determination model, improving the efficiency of generating resource transfer vectors.

[0107] In some embodiments, obtaining the target resource transfer feature corresponding to the target resource transfer record based on the target resource transfer vector includes: obtaining the first resource transfer data corresponding to the first data dimension from the resource transfer data corresponding to the target resource transfer record; determining the first resource transfer feature corresponding to the first resource transfer data; and splicing the target resource transfer vector and the first resource transfer feature to obtain the target resource transfer feature corresponding to the target resource transfer record.

[0108] Wherein, the first data dimension is a data dimension different from the target data dimension, and the data type corresponding to the first data dimension may be different from the data type corresponding to the target data dimension. For example, the data type corresponding to the target data dimension is discrete, and the data type corresponding to the first data dimension is continuous. The first resource transfer data is the resource transfer data corresponding to the first data dimension in the target resource transfer record. The first resource transfer feature is the feature corresponding to the first resource transfer data.

[0109] Specifically, the server can determine the first resource transfer feature corresponding to the first resource transfer data based on the first data dimension. For example, if the target data dimension is a discrete uncountable type, the first data dimension can be any one of a continuous type or a discrete countable type. When the first data dimension is a continuous type, the server can perform a normalization process on the first resource transfer data and use the result of the normalization process as the first resource transfer feature.

[0110] In some embodiments, there can be multiple target resource transfer records corresponding to the target object. The server can obtain the first resource transfer data of the first data dimension from each of the target resource transfer records to form a set of first resource transfer data corresponding to the target object. There can be multiple target objects. The server can obtain the sets of first resource transfer data corresponding to each target object respectively, perform a union operation on each set of first resource transfer data to obtain a union set, and obtain the different first resource transfer data from the union set, and arrange the different first resource transfer data obtained to obtain a first resource transfer data sequence. Among them, when arranging the first resource transfer data, it can be arranged in any order or arranged according to a preset arrangement method. The preset arrangement method can be, for example, sorting in ascending order of character length, and the character length can be represented by the number of characters included in the first resource transfer data.

[0111] In some embodiments, the server can obtain the first resource transfer features corresponding to the respective first resource transfer data in the first resource transfer data sequence based on the first resource transfer data sequence. The first resource transfer feature can include the feature values corresponding to the respective first resource transfer data in the first resource transfer data sequence. Specifically, the server can determine the current first resource transfer data from the first resource transfer data sequence, determine the feature value corresponding to the current first resource transfer data as the first preset value, and determine the feature values corresponding to the non-current first resource transfer data as the second preset value. The server can arrange the feature value of the current first resource transfer data (i.e., the first preset value) and the feature values corresponding to the non-current first resource transfer data (i.e., the second preset value) according to the sorting of the first resource transfer data in the first resource transfer data sequence, and use the arranged sequence as the first resource transfer feature corresponding to the current first resource transfer data. The earlier the sorting of the first resource transfer data in the first resource transfer feature sequence, the earlier the arrangement of the feature value corresponding to the first resource transfer data in the first resource transfer feature.

[0112] Among them, the current first resource transfer data can be any first resource transfer data in the first resource transfer data sequence. The non-current first resource transfer data includes each first resource transfer data other than the current first resource transfer data in the first resource transfer data sequence. The first preset value and the second preset value can be set as needed. The first preset value is different from the second preset value. For example, the first preset value can be 1, and the second preset value can be 0.

[0113] In some embodiments, the server can obtain the target resource transfer data corresponding to the target data dimension and the first resource transfer data corresponding to the first data dimension from the target resource transfer record, obtain the target resource transfer vector corresponding to the target resource transfer data, obtain the first resource transfer feature corresponding to the first resource transfer data, splice the target resource transfer vector and the first resource transfer feature, and use the result after splicing as the target resource transfer feature corresponding to the target resource transfer record. For example, the result after splicing can be used as the target resource transfer feature corresponding to the target resource transfer record. For example, assume that the target resource transfer record is "month = March, merchant number = 213213", the target data dimension is "merchant number", and the first data dimension is "month". Then "213213" is the target resource transfer data, and "March" is the first resource transfer data. If the target resource transfer vector corresponding to "213213" is vector A, and the first resource transfer feature corresponding to "March" is feature B, then the target resource transfer feature can be (vector A, feature B).

[0114] In this embodiment, the first resource transfer data corresponding to the first data dimension is obtained from the resource transfer data corresponding to the target resource transfer record, the first resource transfer feature corresponding to the first resource transfer data is determined, the target resource transfer vector and the first resource transfer feature are spliced, and the target resource transfer feature corresponding to the target resource transfer record is obtained, realizing the splicing of the features of different types of data in the target resource transfer record to obtain the target resource transfer feature, improving the richness of the target resource transfer feature, and thus making the target resource transfer feature better reflect the feature of the target resource transfer record.

[0115] In some embodiments, there are multiple target resource transfer records. The object recognition of the target object based on the target resource transfer feature to obtain the target recognition result corresponding to the target object includes: arranging the target resource transfer features corresponding to the target resource transfer records in the resource transfer order corresponding to the target resource transfer records to obtain a feature sequence; inputting the feature sequence into the object recognition sequence model for processing to obtain the target recognition result corresponding to the target object.

[0116] Among them, there can be multiple target resource transfer records, and the multiple target resource transfer records belong to the same target object. The resource transfer order can be determined according to the resource transfer time in the target resource transfer record. The earlier the resource transfer time, the higher the resource transfer order; the later the resource transfer time, the lower the resource transfer order.

[0117] Each target resource transfer feature in the feature sequence is the target resource transfer feature corresponding to the target resource transfer record of the same target object. The higher the resource transfer order, the higher the sorting of the target resource transfer feature in the feature sequence.

[0118] The object recognition sequence model is a type of object recognition model. The object recognition sequence model is used to process the feature sequence to obtain the target recognition result corresponding to the target object.

[0119] Specifically, the server obtains multiple target resource transfer records of the same target object, obtains the resource transfer time from each target resource transfer record, determines the resource transfer order corresponding to the target resource transfer record based on the resource transfer time, or the target resource transfer features respectively corresponding to each target resource transfer record, arranges the target resource transfer features corresponding to each target resource transfer record according to the resource transfer order, and uses the arranged sequence as the feature sequence.

[0120] In some embodiments, the object recognition sequence model can be a temporal behavior model based on CNN (Convolutional Neural Networks) or RNN (Recurrent Neural Network), or can be LSTM (Long Short-Term Memory) or transformer, etc.

[0121] In some embodiments, the object recognition sequence model may include an attention determination network. The attention determination network is used to determine the feature attention corresponding to each target resource transfer feature in the feature sequence. The server can input the feature sequence into the attention determination network of the object recognition sequence model for attention calculation to obtain the feature attention corresponding to each target resource transfer feature in the feature sequence, use the feature attention to adjust the target resource transfer feature, obtain the adjusted target resource transfer feature, arrange the adjusted target resource transfer features according to the sorting of the target resource transfer feature before adjustment in the feature sequence, use the arranged sequence as the adjusted feature sequence, and perform object recognition based on the adjusted feature sequence.

[0122] In this embodiment, there are multiple target resource transfer records. According to the resource transfer order corresponding to the target resource transfer records, the target resource transfer features corresponding to the target resource transfer records are arranged to obtain a feature sequence. The feature sequence is input into an object recognition sequence model for processing to obtain the target recognition result corresponding to the target object. Thus, the recognition result of the target object can be determined based on the ordered target resource transfer features, improving the accuracy of object recognition.

[0123] In some embodiments, obtaining the target resource transfer features corresponding to the target resource transfer records based on the target resource transfer vector includes: obtaining the sorting features corresponding to the target resource transfer vector according to the resource transfer order corresponding to the target resource transfer records; and fusing the target resource transfer vector with the sorting features to obtain the target resource transfer features corresponding to the target resource transfer records.

[0124] Specifically, the server can obtain multiple target resource transfer records of the same target object, arrange each target resource transfer record according to the resource transfer order, and use the arranged sequence as the target resource transfer record sequence. The server can determine the arrangement features corresponding to each target resource transfer record according to the sorting (resource transfer order) of the target resource transfer record in the target resource transfer record sequence, and obtain the target resource transfer features corresponding to the target resource transfer records based on the sorting features of the target resource transfer records and the corresponding target resource transfer vectors. For example, the server can fuse the target resource transfer vector with the sorting features and use the fused result as the target resource transfer features.

[0125] In some embodiments, the number of feature values included in the sorting features can be the same as the number of vector values included in the target resource transfer vector. The server can add the target resource transfer vector to the sorting features and use the added result as the target resource transfer features. Here, adding the target resource transfer vector to the sorting features means adding the vector values in the target resource transfer vector to the feature values at the corresponding positions in the sorting features.

[0126] In this embodiment, obtaining the sorting features corresponding to the target resource transfer vector according to the resource transfer order corresponding to the target resource transfer records, and fusing the target resource transfer vector with the sorting features to obtain the target resource transfer features corresponding to the target resource transfer records, so that the sorting information of the target resource transfer features is covered in the target resource transfer features. Thus, when performing object recognition based on the target resource transfer features, the sorting of the resource transfer records can be considered, improving the accuracy of object recognition.

[0127] In some embodiments, inputting the feature sequence into the object recognition sequence model for processing to obtain the target recognition result corresponding to the target object includes: obtaining the recognition task matrix corresponding to the recognition task network in the object recognition sequence model; based on the target resource transfer features in the feature sequence and the recognition task matrix, obtaining the feature attentions corresponding to the respective target resource transfer features in the feature sequence; adjusting the corresponding target resource transfer features in the feature sequence based on the feature attentions to obtain the adjusted target resource transfer features, and the adjusted target resource transfer features form the adjusted feature sequence; and obtaining the target recognition result corresponding to the target object based on the adjusted feature sequence.

[0128] Among them, the matrix values in the recognition task matrix can be the parameter values corresponding to the model parameters in the trained object recognition model. The model parameters refer to the variable parameters inside the object recognition model. For a neural network model, it can also be called the neural network weight. For example, the object recognition model may include a recognition task network, and the recognition task matrix can be a matrix composed of the parameter values of the respective model parameters of the recognition task network in the trained object recognition model. The attention determination network may include a recognition task network, a key network, and a value network.

[0129] Specifically, the server can perform a multiplication operation on the target resource transfer feature and the recognition task matrix, and use the result of the multiplication operation as the task vector, and obtain the feature attention corresponding to the target resource transfer feature based on the result of the multiplication operation. For example, the server can obtain the key matrix and the value matrix. The object recognition model may include a key network and a value network. The key matrix can be a matrix composed of the parameter values of the respective model parameters of the key network in the trained object recognition model, and the value matrix can be a matrix composed of the parameter values of the respective model parameters of the value network in the trained object recognition model. The server can perform a multiplication operation on the target resource transfer feature and the key matrix, and use the result of the multiplication operation as the key vector, perform an inner product operation on the task vector and the key vector to obtain the target inner product value corresponding to the target resource transfer feature, perform a multiplication operation on the target resource transfer feature and the value matrix, and use the result of the multiplication operation as the value vector corresponding to the target resource transfer feature, and obtain the feature attention corresponding to the target resource transfer feature based on the target inner product value and the value vector corresponding to the target resource transfer feature. For example, the multiplication operation can be performed on the target inner product value corresponding to the target resource transfer feature and the corresponding value vector, and based on the result of the multiplication operation. Among them, the target resource transfer feature can be a vector, and the feature attention can be a vector with the same dimension as the target resource transfer feature. The key vector, the task vector, and the value vector can be vectors with the same dimension as the target resource transfer feature.

[0130] In some embodiments, the step of obtaining the feature attention degrees respectively corresponding to the target resource transfer features in the feature sequence based on the target resource transfer features in the feature sequence and the recognition task matrix includes: obtaining the current resource transfer feature and the associated resource transfer feature corresponding to the current resource transfer feature from the feature sequence, performing a multiplication operation on the current resource transfer feature and the key matrix, using the result of the multiplication operation as the current key vector, performing a multiplication operation on the associated resource transfer feature and the key matrix, using the result of the multiplication operation as the associated key vector, performing a multiplication operation on the current resource transfer feature and the recognition task matrix, using the result of the multiplication operation as the current task vector, performing an inner product operation on the current task vector and the current key vector to obtain the current inner product value corresponding to the current resource transfer feature, performing an inner product operation on the current task vector and the associated key vector to obtain the associated inner product value corresponding to the current resource transfer feature, and obtaining the feature attention degree corresponding to the current resource transfer feature based on the current inner product value and the associated inner product value.

[0131] Among them, there can be multiple associated resource transfer features corresponding to the current resource transfer feature. For example, at least one of the features in the feature sequence other than the current resource transfer feature can be used as the associated resource transfer feature corresponding to the current resource transfer feature. For example, all the features in the feature sequence other than the current resource transfer feature can be used as the associated resource transfer feature corresponding to the current resource transfer feature. The associated resource transfer feature corresponding to the current resource transfer feature can also be preset. For example, the target resource transfer feature ranked first in the feature sequence can be used as the associated resource transfer feature corresponding to the current resource transfer feature, or it can be determined according to the current feature ranking of the current resource transfer feature in the feature sequence. The current feature ranking refers to the ranking of the current resource transfer feature in the feature sequence. For example, the target resource transfer feature ranked before the current feature ranking in the feature sequence can be used as the associated resource transfer feature corresponding to the current resource transfer feature, or the target resource transfer feature ranked after the current feature ranking in the feature sequence can be used as the associated resource transfer feature corresponding to the current resource transfer feature, or the target resource transfer feature with a difference in ranking from the current feature ranking less than the ranking difference threshold can be used as the associated resource transfer feature corresponding to the current resource transfer feature. The ranking difference threshold can be preset. For example, it can be 2. That is, when the current feature ranking is 3, the target resource transfer features ranked 1, 2, 4, and 5 in the feature sequence can be used as the associated resource transfer features corresponding to the current resource transfer feature.

[0132] In some embodiments, the step of obtaining the feature attention corresponding to the current resource transfer feature based on the current inner product value and the associated inner product value includes: performing a multiplication operation on the current resource transfer feature and the value matrix, taking the result of the multiplication operation as the current value vector, performing a multiplication operation on the associated resource transfer feature and the value matrix, taking the result of the multiplication operation as the associated value vector, performing a weighted calculation on the current value vector and the associated value vector based on the current inner product value and the associated inner product value, and taking the vector obtained by the weighted calculation as the feature attention corresponding to the current resource transfer feature.

[0133] In some embodiments, the step of adjusting the corresponding target resource transfer feature in the feature sequence based on the feature attention to obtain the adjusted target resource transfer feature includes: performing an addition operation on the feature attention corresponding to the target resource transfer feature and the target resource transfer feature, and taking the result of the addition operation as the adjusted target resource transfer feature.

[0134] In some embodiments, the object recognition sequence model may further include an object recognition network. The server may input the adjusted feature sequence into the object recognition network for object recognition. The object recognition network may classify the object based on the adjusted feature sequence to determine at least one of the object type, object label, or object value corresponding to the target object.

[0135] In this embodiment, the recognition task matrix corresponding to the recognition task network in the object recognition sequence model is obtained. Based on the target resource transfer features in the feature sequence and the recognition task matrix, the feature attention corresponding to each target resource transfer feature in the feature sequence is obtained. The corresponding target resource transfer feature in the feature sequence is adjusted based on the feature attention to obtain the adjusted target resource transfer feature. Each adjusted target resource transfer feature forms an adjusted feature sequence. Based on the adjusted feature sequence, the target recognition result corresponding to the target object is obtained, improving the object recognition accuracy.

[0136] In some embodiments, determining the target resource transfer record corresponding to the target object includes: obtaining the resource transfer record of the target object for resource transfer to the exchange object as the target resource transfer record; obtaining the target recognition result corresponding to the target object based on the target resource transfer feature includes: performing object recognition on the target object based on the target resource transfer feature to obtain the recognition result that the target object is a resource transfer abnormal object.

[0137] Among them, the exchange objects can be various objects that can be traded. The objects can be at least one of living things or inanimate things. Living things can be animals, such as cats or dogs. Inanimate things can be real things such as computers, or virtual things such as mobile phone traffic. The exchange object can be, for example, any one of the goods in a physical store or the goods displayed on a shopping website. The resource transfer abnormal object refers to a malicious object with malicious behavior in resource transfer, and can include at least one of malicious objects such as gambling, porn, fraud, money laundering or remittance.

[0138] Specifically, there can be one or more exchange objects, and "more" means at least two. The server can obtain the resource transfer records for each exchange object corresponding to the target object as the target resource transfer records. For example, the server can obtain the order details of the target object purchasing various electronic products as the target resource transfer records.

[0139] In some embodiments, the server can identify the target object based on a trained object detection model. For example, when the object detection model is a model trained with each training resource transfer feature as a training sample, the server can input the target resource transfer feature into the object detection model for object identification.

[0140] In some embodiments, the object detection model is a model trained with a sequence obtained by arranging multiple training resource transfer features in the resource transfer order. Among them, the multiple training resource transfer features belong to the same object. The server can obtain multiple target resource transfer features corresponding to the same target object, sort each target resource transfer feature according to the resource transfer order to obtain a feature sequence, and input the feature sequence into the trained object detection model for object identification. The identification result can include the object type, and the object type can be any one of a resource transfer abnormal object or a resource transfer normal object. When the object detection model determines that the target object is a resource transfer abnormal object according to the input feature sequence, it outputs the identification result that the target object is a resource transfer abnormal object.

[0141] In this embodiment, the resource transfer records for the target object to transfer resources to the exchange object are obtained as the target resource transfer records, and the target object is identified based on the target resource transfer features, and the identification result that the target object is a resource transfer abnormal object is obtained.

[0142] In some embodiments, the steps of obtaining the target data dimension include: obtaining each data dimension corresponding to the target resource transfer record to form a data dimension set; selecting the data dimensions that meet the dimension selection conditions from the data dimension set as the target data dimensions; the dimension selection conditions include that the number of data in the data dimension is greater than the quantity threshold.

[0143] Among them, the data quantity refers to the quantity of data corresponding to the data dimension. For the data dimension "gender", the data are "male" and "female", so the data quantity is 2. The quantity threshold can be preset or calculated based on the data quantity corresponding to the data dimension in the data dimension set. For example, for a data dimension with a determinable data quantity, such as when the data corresponding to the data dimension is countable, the data quantity corresponding to the data dimension can be obtained. When there are multiple data dimensions with determinable data quantities, the data quantities corresponding to these multiple data dimensions can be averaged, and the result of the averaging operation is used as the average quantity. A quantity coefficient can be obtained, and the product of the average quantity and the quantity coefficient is used as the quantity threshold. The quantity coefficient can be preset, for example, it can be 1.2.

[0144] In this embodiment, each data dimension corresponding to the target resource transfer record is obtained to form a data dimension set, and the data dimension that meets the dimension selection condition is selected from the data dimension set as the target data dimension. Since the dimension selection condition includes that the data quantity of the data dimension is greater than the quantity threshold, the data dimension with a larger data quantity can be used as the target data dimension.

[0145] In some embodiments, as Figure 3B shown, a data processing method is provided. Taking the method applied to the server 104 in Figure 1 as an example, the method includes the following steps: S302, obtaining a training resource transfer record sequence corresponding to a training object, where the training resource transfer record sequence includes multiple training resource transfer records arranged in the resource transfer order of the training resource transfer records; S304, obtaining the training resource transfer data corresponding to the target data dimension from the resource transfer data corresponding to the training resource transfer record; S306, arranging each training resource transfer data in the resource transfer order of the training resource transfer record to obtain a training resource transfer data sequence; S308, performing resource transfer vector training according to the training resource transfer data sequence to obtain the training resource transfer vectors corresponding to each training resource transfer data.

[0146] In the above data processing method, a training resource transfer record sequence corresponding to a training object is obtained, training resource transfer data corresponding to a target data dimension is obtained from the resource transfer data corresponding to the training resource transfer record, and each training resource transfer data is arranged in the resource transfer order of the training resource transfer record to obtain a training resource transfer data sequence. Resource transfer vector training is performed according to the training resource transfer data sequence to obtain training resource transfer vectors corresponding to each training resource transfer data. Since the training resource transfer record sequence includes multiple training resource transfer records arranged in the resource transfer order of the training resource transfer record, the vector representation corresponding to the resource transfer data can be determined based on the resource transfer order, improving the accuracy of the generated resource transfer vector and further improving the accuracy of object recognition.

[0147] In some embodiments, there are multiple training resource transfer data sequences, and each training resource transfer data sequence forms a data sequence set; performing resource transfer vector training according to the training resource transfer data sequence to obtain training resource transfer vectors corresponding to each training resource transfer data includes: obtaining the current training resource transfer data from the training resource transfer data sequence, and obtaining the associated resource transfer data corresponding to the current training resource transfer data in the training resource transfer data sequence; inputting the current training resource transfer data into the vector determination model to be trained to obtain the predicted association probability between the current training resource transfer data and the associated resource transfer data; obtaining the standard association probability between the current training resource transfer data and the associated resource transfer data based on the data sequence set; adjusting the model parameters of the vector determination model based on the difference between the standard association probability and the predicted association probability to obtain the trained vector determination model, and obtaining the training resource transfer vectors corresponding to each training resource transfer data based on the trained vector determination model.

[0148] In this embodiment, obtaining the current training resource transfer data from the training resource transfer data sequence, and obtaining the associated resource transfer data corresponding to the current training resource transfer data in the training resource transfer data sequence, inputting the current training resource transfer data into the vector determination model to be trained to obtain the predicted association probability between the current training resource transfer data and the associated resource transfer data, obtaining the standard association probability between the current training resource transfer data and the associated resource transfer data based on the data sequence set formed by each training resource transfer data sequence, adjusting the model parameters of the vector determination model based on the difference between the standard association probability and the predicted association probability to obtain the trained vector determination model, and obtaining the training resource transfer vectors corresponding to each training resource transfer data based on the trained vector determination model improves the efficiency of generating resource transfer vectors.

[0149] In some embodiments, a data processing method is provided, including the following steps:

[0150] 1. Obtain multiple transaction records of the training object as training transaction records.

[0151] Among them, the transaction records may include amount, whether it is a payment, payment method (balance, bank card, Yu'E Bao, etc.), transaction location, transaction device, transaction IP, bank type, time, date (week, day, month, year), card type, time since the last behavior, merchant number, remaining balance, whether it is a credit card, credit card limit, characteristics of the counterparty (gender, age, region, IP), etc.

[0152] 2. Obtain the training transaction data corresponding to the target data dimension from each training transaction record, and arrange each training transaction data according to the transaction time to obtain a training transaction data sequence; the target data dimension is a discrete and uncountable data dimension.

[0153] For example, the target data dimension can be "merchant number". If there are 6 training transaction records, namely training transaction record 1 = (transaction time = January 1, 2020, merchant number = merchant number A), training transaction record 2 = (transaction time = February 1, 2020, merchant number = merchant number B), training transaction record 3 = (transaction time = March 1, 2020, merchant number = merchant number C), training transaction record 4 = (transaction time = April 1, 2020, merchant number = merchant number D), training transaction record 5 = (transaction time = May 1, 2020, merchant number = merchant number A), training transaction record 6 = (transaction time = June 1, 2020, merchant number = merchant number B), then the training transaction data sequence = (merchant number A, merchant number B, merchant number C, merchant number D, merchant number A, merchant number B).

[0154] Among them, the target data dimension can be a discrete and uncountable data dimension, such as merchant number or transaction IP, etc.

[0155] 3. Obtain the current training transaction data and the associated transaction data corresponding to the current training transaction data from the training transaction data sequence, form a training sample by combining the current training transaction data and the associated transaction data, and train the vector determination model to obtain a trained vector determination model.

[0156] As Figure 4 shown, the current training transaction data is "merchant number A" ranked first, and the associated transaction data are "merchant number B" ranked second and "merchant number C" ranked third. Therefore, the "merchant number A" in the first place can be combined with the "merchant number B" in the second place to form a training sample, and the "merchant number A" in the first place can be combined with the "merchant number C" in the third place to form a training sample.

[0157] Among them, the vector determination model can be a model of word vector technology, and the word vector technology can include at least one of Elmo word vector technology or word2vec word vector technology. Word2vec is a way to obtain high-quality word vectors. Its main idea is that the context of a word can well express the semantics of the word, and it is a way to generate word vectors through unsupervised learning of text. Word2vec corresponds to the skip-gram model and the cbow model. Both the cbow model and the skip-gram model are single-layer neural networks, and the parameters of the neural network are the obtained word vectors, that is, the word vectors are the parameters of the neural network, and the process of neural network training is the process of learning word vectors. Among them, the cbow model knows the surrounding words and predicts the central word, and the skip-gram model knows the central word and predicts the surrounding words. As Figure 5 shown, it is the structure of the cbow model in some embodiments. In the figure, "merchant number A", "merchant number B", merchant number "C", "merchant number D" and "merchant number E" are elements in the training transaction data sequence = (merchant number A, merchant number B, merchant number C, merchant number D, merchant number E). Taking this sequence as a sentence and each element in the sequence as a word, it can be seen from the figure that "merchant number C" is the central word, and "merchant number A", "merchant number B", "merchant number D" and "merchant number E" are the surrounding words of "merchant number C". The central word "merchant number C" can be predicted through "merchant number A", "merchant number B", "merchant number D" and "merchant number E". As Figure 6 shown, it is the structure of the skip-gram model in some embodiments. It can be seen from the figure that the surrounding words "merchant number A", "merchant number B", "merchant number D" and "merchant number E" of "merchant number C" can be predicted through the central word "merchant number C".

[0158] 4. Obtain the training transaction vectors corresponding to each training transaction data based on the trained vector determination model, and form a training transaction vector set with the training transaction vectors.

[0159] Among them, each training transaction vector can be obtained according to the weight matrix corresponding to the hidden layer in the trained vector determination model. For example, each row in the hidden layer can be used as a training transaction vector.

[0160] 5. Obtain the target transaction record corresponding to the target object, obtain the transaction data corresponding to the target data dimension from the target transaction record as the target transaction data, obtain the target transaction vector corresponding to the target transaction data from the training transaction vector set, and determine the target transaction feature corresponding to the target transaction record based on the target transaction vector.

[0161] 6. Obtain the first transaction data of the first data dimension and the first transaction data of the second data dimension in the target transaction record. The first data dimension is a discrete countable data dimension, and the second data dimension is a continuous data dimension. Perform one-hot encoding on the first transaction data to obtain the first transaction feature, perform normalization processing on the second transaction data to obtain the second transaction feature, and splice the first transaction feature, the second transaction feature, and the target transaction vector to obtain the target transaction feature corresponding to the target transaction record.

[0162] Among them, the target data dimension can be, for example, the discrete uncountable "merchant number", the first data dimension can be, for example, the discrete countable "month", and the second data dimension can be, for example, the continuous "amount".

[0163] 7. Obtain the set of target transaction records corresponding to the target object, obtain the target transaction features corresponding to each target transaction record in the set of target transaction records, and arrange the target transaction features according to the transaction time to obtain a feature sequence.

[0164] Among them, the set of target transaction records includes multiple target transaction records, and each target transaction record in the set of target transaction records belongs to the target object. The feature sequence can be represented by a matrix. For example, 400 transaction records of the target object can be obtained, the target transaction features corresponding to these 400 transaction records are obtained, and a feature matrix of size N×400 is formed using the target transaction features. Each column in the feature matrix is a target transaction feature, and N is the number of feature values included in the target transaction feature. Among them, the target transaction features in the feature matrix are arranged according to the transaction time.

[0165] 8. Input the feature sequence into the object recognition sequence model for object recognition to obtain the target recognition result corresponding to the target object. The object recognition sequence model includes a recognition task network.

[0166] Among them, the object recognition sequence model can be, for example, a time-series behavior model based on CNN or RNN, or a model such as LSTM or Transformer. As Figure 7 shown, it is the structure diagram of the Transformer in some embodiments. The Transformer includes multiple encoders and multiple decoders. Among them, the encoder can include a self-attention layer and a feed-forward neural network. As Figure 8As shown, it is the structural diagram of an encoder in some embodiments. It can be seen from the figure that there is a residual connection around each sub-layer in the encoder, and following this is a normalization step, that is, the encoder includes a residual block and a layer normalization layer. Among them, there are many ways of normalization. The goal of normalization is to transform the input into data with a mean of 0 and a variance of 1. Normalization is performed before feeding the data into the activation function because it is not desired that the data falls into the saturation region of the activation function. The normalization can be batch normalization (BN). The main idea of BN is to normalize each batch of data in each layer. After normalizing the input data and feeding it into the network layer, the data output by the network layer may no longer be normalized data. As the data is continuously affected by the network layer, the deviation of the data becomes larger and larger. Therefore, when backpropagating, these large deviations need to be considered, so that a smaller learning rate is used to reduce gradient vanishing or gradient explosion.

[0167] The decoder also includes a self-attention layer and a feed-forward neural network, and there is an attention layer between the self-attention layer and the feed-forward neural network layer in the decoder. The attention layer helps the current node obtain the key content that needs to be focused on currently.

[0168] The parallel computing ability of the RNN series of models is poor. In the RNN model, the calculation at time T depends on the hidden layer calculation result at time T - 1, and the calculation at time T - 1 depends on the hidden layer calculation result at time T - 2, and so on, forming a sequence dependence relationship, which results in poor parallel computing ability of the RNN. The feature extraction ability of the Transformer is better than that of the RNN series of models.

[0169] The Attention mechanism is utilized in the Transformer. The Attention mechanism includes Self-attention and Multi-head attention. The attention probability distribution of the Self-attention mechanism comes from the input of the network itself, while the attention probability distribution in the traditional Attention mechanism comes from the outside. The Self-attention mechanism helps the current node not only focus on the current word, thus enabling it to obtain the semantic meaning of the context. The Multi-head attention mechanism endows the attention layer with multiple "representation subspaces". There are 8 sets of Q\K\V weight matrices in the Transformer, that is, 8 groups of weight matrices, and each group of weight matrices includes a Q, a K, and a V. Each set is randomly initialized. Among them, Q is the abbreviation of query, K is the abbreviation of key, and V is the abbreviation of value. The Q weight matrix can also be called the query matrix, the K weight matrix can also be called the key matrix, and the V weight matrix can also be called the value matrix.

[0170] The Attention mechanism can be understood as a sequence focusing method. The basic idea is to assign attention weights to the sequence and focus the attention on the most relevant sequence. The Attention mechanism can be understood as an addressing process. By giving a query vector Q related to the task, calculating the attention distribution with the Key and attaching it to the Value, the Attention Value is calculated. This process is actually an embodiment of the Attention mechanism alleviating the complexity of the neural network. Instead of inputting all the inputs into the neural network for calculation, some information related to the task is selected and input into the neural network, which is similar to the gating mechanism idea in the RNN.

[0171] In some embodiments, the calculation process of the Attention mechanism may include the following steps:

[0172] First, information input;

[0173] Specifically, input Q, K, and V into the model. Let X = [x1, x2,..., x n where x i represents the input vector, and i is greater than or equal to 1 and less than or equal to n;

[0174] Second, calculate the attention distribution α;

[0175] Specifically, calculate the relevance by calculating Q and K through dot product, and calculate the scores through softmax. Let Q = K = V = X, and calculate the attention weights through softmax,

[0176] α i = softmax(s(ki , q)) = softmax(s(x i , q)) (1).

[0177] Among them, α i is the attention probability distribution, s(x i , q) is the attention scoring mechanism. The commonly used attention scoring mechanisms are as follows:

[0178] Additive model: s(x i , q) = υ T tanh(Wx i + Uq) (2);

[0179] Dot product model:

[0180] Scaled dot model:

[0181] Bilinear model:

[0182] Among them, q is a query vector, υ, W, and U are learnable parameters, and d k is the input vector, that is, x i 's dimension.

[0183] Third, information weighted average.

[0184] Specifically, the attention distribution α i is used to explain the degree of attention received by the i-th piece of information when querying the context q. The attention distribution α i can be understood as the degree of attention received by the i-th input vector when given the query vector q related to the task.

[0185]

[0186] In some embodiments, the attention calculation formula can be (7). Among them, Q = W Q X, K = W K X, V = W V X. Q can also be called the query vector sequence, K can be called the key vector sequence, and V can also be called the value vector sequence. As Figure 9 shown, it shows the schematic diagram of calculating attention using formula (7). As Figure 10As shown, the process of calculating self-attention in some embodiments is presented. In the figure, Transaction Record 1 and Transaction Record 2 are two transaction records of the same user. x1 is the target transaction feature corresponding to "Transaction Record 1", x2 is the target transaction feature corresponding to "Transaction Record 2", q1 is the query vector corresponding to x1, q2 is the query vector corresponding to x2, k1 is the key vector corresponding to x1, k2 is the key vector corresponding to x2, v1 is the value vector corresponding to x1, v2 is the value vector corresponding to x2, z1 is the attention corresponding to x1, z2 is the attention corresponding to x2, W Q is the query matrix, and W K is the key matrix, and W V is the value matrix.

[0187] 9. Multiply the target transaction feature in the feature sequence by the recognition task matrix of the recognition task network to obtain the feature attention corresponding to the target transaction feature.

[0188] 10. Adjust the corresponding target transaction feature in the feature sequence based on the feature attention to obtain the adjusted target transaction feature. Form the adjusted feature sequence with each target transaction feature, and obtain the target recognition result corresponding to the target object based on the adjusted feature sequence.

[0189] In this embodiment, based on word2vec, the transaction data sequence can be regarded as a sentence, and each transaction data behavior in the transaction data sequence is regarded as a word. Word2vec is used to perform embedding encoding on discrete features that are difficult to one-hot encode, rather than discarding such features as in the usual approach. The data processing method provided in this application can adopt technical solutions such as an encoder-decoder architecture, introducing a multi-head self-attention mechanism, a residual module, and layer normalization based on the transformer model, enabling the model to better focus on sequence features that are more worthy of attention, effectively improving the classification performance of the model, and effectively reducing the problem of gradient disappearance existing in CNN and RNN models.

[0190] This application also provides an application scenario where the object recognition sequence model can be an object scoring model, and this application scenario applies the above data processing method. Specifically, the application of this data processing method in this application scenario is as follows:

[0191] Specifically, the server can determine the object to be scored, obtain multiple transaction records of the object to be scored as each scoring transaction record, generate scoring transaction features corresponding to each scoring transaction data by using the data processing method proposed in this application, arrange the scoring transaction features according to the transaction time to obtain a scoring feature sequence, and input the scoring feature sequence into the trained object scoring model to obtain the scoring result of the object to be scored. Thus, a better understanding and control of the object can be achieved based on the scoring result. The object scoring model refers to a model used to score an object and can be a neural network model based on artificial intelligence.

[0192] This application also provides another application scenario. The object recognition sequence model can be an object label classification model, and this application scenario applies the above data processing method. Specifically, the application of the data processing method in this application scenario is as follows:

[0193] Specifically, the server can obtain the object to be labeled as the labeled object, obtain multiple transaction records of the labeled object as each labeled transaction record, generate labeled transaction features corresponding to the labeled transaction record by using the data processing method proposed in this application, arrange the labeled transaction features according to the transaction time to obtain a labeled feature sequence, and input the labeled feature sequence into the trained object label classification model to obtain the object label corresponding to the labeled object. Thus, an object portrait can be generated using the object label, and content can be recommended to the object based on the object portrait to achieve personalized business recommendations and improve the accuracy and effect of content recommendations. The object label classification model refers to a model used to determine the object label and can be a neural network model based on artificial intelligence.

[0194] This application also provides another application scenario. The object recognition sequence model can be an object detection model, and this application scenario applies the above data processing method. Specifically, the application of the data processing method in this application scenario is as follows:

[0195] Specifically, the server can obtain the object to be detected to get the detected object, obtain the detected transaction record corresponding to the detected object, generate the detected transaction feature corresponding to the detected transaction record by using the data processing method provided in this application, arrange each detected transaction feature according to the transaction time to obtain a detected feature sequence, and input the detected feature sequence into the trained object detection model to obtain the object detection result corresponding to the detected object. Thus, malicious objects with abnormal behaviors can be determined based on the object detection result, which is beneficial to rectify the policies for objects with abnormal behaviors and maintain the security and stability of the platform. The abnormal behavior can include at least one of gambling, porn, fraud, money laundering, or remittance. The object detection model refers to a model used to detect an object and can be a neural network model based on artificial intelligence.

[0196] The present application further provides an application scenario. The object recognition sequence model may be an object classification model, and this application scenario applies the above data processing method. Specifically, the application of the data processing method in this application scenario is as follows:

[0197] Specifically, the server may obtain an object to be classified to obtain a classification object, obtain a classification transaction record corresponding to the classification object, use the data processing method provided by the present application to generate a classification transaction feature corresponding to the classification transaction record, arrange each classification transaction feature according to the transaction time to obtain a classification feature sequence, and input the classification feature sequence into the trained object classification model to obtain an object classification result corresponding to the classification object. The object classification model refers to a model used to classify objects and may be an artificial intelligence-based neural network model.

[0198] It should be understood that although Figure 2-10 the steps in the flowchart of Figure 2-10 are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover,

[0199] In some embodiments, as Figure 11 shown, a data processing device is provided. This device may adopt a software module or a hardware module, or a combination of both to become a part of a computer device. Specifically, the device includes: a target resource transfer record determination module 1102, a target resource transfer data acquisition module 1104, a target resource transfer vector acquisition module 1106, a target resource transfer feature obtaining module 1108, and a target recognition result obtaining module 1110, where:

[0200] The target resource transfer record determination module 1102 is used to determine a target resource transfer record corresponding to a target object.

[0201] The target resource transfer data acquisition module 1104 is used to obtain target resource transfer data corresponding to a target data dimension from the resource transfer data corresponding to the target resource transfer record.

[0202] The target resource transfer vector acquisition module 1106 is configured to acquire a target resource transfer vector corresponding to target resource transfer data. The target resource transfer vector is obtained by training according to a training resource transfer data sequence corresponding to a target data dimension, and the training resource transfer data in the training resource transfer data sequence is arranged according to the resource transfer order of the training resource transfer records corresponding to the training object.

[0203] The target resource transfer feature obtaining module 1108 is configured to obtain a target resource transfer feature corresponding to a target resource transfer record based on the target resource transfer vector.

[0204] The target recognition result obtaining module 1110 is configured to perform object recognition on a target object based on the target resource transfer feature to obtain a target recognition result corresponding to the target object.

[0205] In some embodiments, the target resource transfer vector is selected from a training resource transfer vector set; the apparatus further includes a training resource transfer vector set obtaining module, and the training resource transfer vector set obtaining module includes:

[0206] The training resource transfer record sequence acquisition unit is configured to acquire a training resource transfer record sequence corresponding to a training object, and the training resource transfer record sequence includes a plurality of training resource transfer records arranged according to the resource transfer order of the training resource transfer records.

[0207] The training resource transfer data acquisition unit is configured to acquire training resource transfer data corresponding to a target data dimension from the resource transfer data corresponding to the training resource transfer record.

[0208] The training resource transfer data sequence obtaining unit is configured to arrange each training resource transfer data according to the resource transfer order of the training resource transfer records to obtain a training resource transfer data sequence.

[0209] The training resource transfer vector set composition unit is configured to perform resource transfer vector training according to the training resource transfer data sequence to obtain training resource transfer vectors corresponding to each training resource transfer data, and form a training resource transfer vector set.

[0210] In some embodiments, there are multiple training resource transfer data sequences, and each training resource transfer data sequence forms a data sequence set; the training resource transfer vector set composition unit is further configured to obtain the current training resource transfer data from the training resource transfer data sequences, and obtain the associated resource transfer data corresponding to the current training resource transfer data in the training resource transfer data sequences; input the current training resource transfer data into the vector determination model to be trained to obtain the predicted association probability between the current training resource transfer data and the associated resource transfer data; obtain the standard association probability between the current training resource transfer data and the associated resource transfer data based on the data sequence set; adjust the model parameters of the vector determination model based on the difference between the standard association probability and the predicted association probability to obtain the trained vector determination model, and obtain the training resource transfer vectors corresponding to each training resource transfer data based on the trained vector determination model.

[0211] In some embodiments, the target resource transfer feature obtaining module 1108 includes:

[0212] The first resource transfer data obtaining unit is configured to obtain the first resource transfer data corresponding to the first data dimension from the resource transfer data corresponding to the target resource transfer record.

[0213] The first resource transfer feature determining unit is configured to determine the first resource transfer feature corresponding to the first resource transfer data.

[0214] The first target resource transfer feature obtaining unit is configured to splice the target resource transfer vector and the first resource transfer feature to obtain the target resource transfer feature corresponding to the target resource transfer record.

[0215] In some embodiments, there are multiple target resource transfer records, and the target recognition result obtaining module 1110 includes:

[0216] The feature sequence obtaining unit is configured to arrange the target resource transfer features corresponding to the target resource transfer records according to the resource transfer order corresponding to the target resource transfer records to obtain a feature sequence.

[0217] The target recognition result obtaining unit is configured to input the feature sequence into the object recognition sequence model for processing to obtain the target recognition result corresponding to the target object.

[0218] In some embodiments, the target resource transfer feature obtaining module 1108 includes:

[0219] The sorting feature obtaining unit is configured to obtain the sorting feature corresponding to the target resource transfer vector according to the resource transfer order corresponding to the target resource transfer record.

[0220] The second target resource transfer feature obtaining unit is configured to fuse the target resource transfer vector with the sorting feature to obtain the target resource transfer feature corresponding to the target resource transfer record.

[0221] In some embodiments, the target recognition result obtaining unit is further configured to obtain the recognition task matrix corresponding to the recognition task network in the object recognition sequence model; based on the target resource transfer features in the feature sequence and the recognition task matrix, obtain the feature attention degrees corresponding to the respective target resource transfer features in the feature sequence; adjust the corresponding target resource transfer features in the feature sequence based on the feature attention degrees to obtain the adjusted target resource transfer features, and the adjusted target resource transfer features form an adjusted feature sequence; and obtain the target recognition result corresponding to the target object based on the adjusted feature sequence.

[0222] In some embodiments, the target resource transfer record determining module 1102 is further configured to obtain the resource transfer record of the target object for resource transfer to the exchange object as the target resource transfer record; the target recognition result obtaining module 1110 is further configured to perform object recognition on the target object based on the target resource transfer features to obtain the recognition result that the target object is a resource transfer abnormal object.

[0223] In some embodiments, the apparatus further includes a target data dimension obtaining module, and the target data dimension obtaining module includes:

[0224] The data dimension set composing unit is configured to obtain the respective data dimensions corresponding to the target resource transfer record and compose a data dimension set.

[0225] The target data dimension obtaining unit is configured to select the data dimensions that meet the dimension selection condition from the data dimension set as the target data dimensions; the dimension selection condition includes that the data quantity of the data dimension is greater than the quantity threshold.

[0226] In some embodiments, as Figure 12 shown, a data processing apparatus is provided. The apparatus can be a software module or a hardware module, or a combination of the two to form a part of a computer device. The apparatus specifically includes: a training resource transfer record sequence obtaining module 1202, a training resource transfer data obtaining module 1204, a training resource transfer data sequence obtaining module 1206, and a training resource transfer vector obtaining module 1208, where:

[0227] The training resource transfer record sequence obtaining module 1202 is configured to obtain the training resource transfer record sequence of the training object, and the training resource transfer record sequence includes a plurality of training resource transfer records arranged in the resource transfer order of the training resource transfer records.

[0228] The training resource transfer data acquisition module 1204 is used to obtain the training resource transfer data corresponding to the target data dimension from the resource transfer data corresponding to the training resource transfer record.

[0229] The training resource transfer data sequence generation module 1206 is used to arrange each training resource transfer data in the resource transfer order of the training resource transfer record to obtain a training resource transfer data sequence.

[0230] The training resource transfer vector generation module 1208 is used to perform resource transfer vector training based on the training resource transfer data sequence to obtain the training resource transfer vectors corresponding to each training resource transfer data.

[0231] In some embodiments, there are multiple training resource transfer data sequences, and each training resource transfer data sequence forms a data sequence set; the training resource transfer vector generation module X08 includes:

[0232] The associated resource transfer data acquisition unit is used to obtain the current training resource transfer data from the training resource transfer data sequence and obtain the associated resource transfer data corresponding to the current training resource transfer data in the training resource transfer data sequence.

[0233] The predicted association probability generation unit is used to input the current training resource transfer data into the vector determination model to be trained to obtain the predicted association probability between the current training resource transfer data and the associated resource transfer data.

[0234] The standard association probability generation unit is used to obtain the standard association probability between the current training resource transfer data and the associated resource transfer data based on the data sequence set.

[0235] The training resource transfer vector generation unit is used to adjust the model parameters of the vector determination model based on the difference between the standard association probability and the predicted association probability to obtain the trained vector determination model, and obtain the training resource transfer vectors corresponding to each training resource transfer data based on the trained vector determination model.

[0236] For the specific limitations of the data processing device, reference can be made to the limitations on the data processing method in the above text, which will not be elaborated here. Each module in the above data processing device can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor of the computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0237] In some embodiments, a computer device is provided. The computer device can be a terminal, and its internal structure diagram can be as Figure 13As shown in the figure. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a data processing method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or buttons, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0238] In some embodiments, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 14 shown in the figure. The computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device is used to store data related to the data processing method. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a data processing method.

[0239] Those skilled in the art can understand that Figure 13 and 14 the structures shown in the figure are only block diagrams of some structures related to the solution of the present application, and do not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0240] In some embodiments, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, it implements the steps in the above method embodiments.

[0241] In some embodiments, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0242] In some embodiments, a computer program product or a computer program is provided. The computer program product or the computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the steps in the above method embodiments.

[0243] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it may include the processes of the above method embodiments. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application may include at least one of non-volatile and volatile memories. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0244] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0245] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. A data processing method, characterized in that, The method includes: Determining a target resource transfer record corresponding to a target object; Obtaining target resource transfer data corresponding to a target data dimension from the resource transfer data corresponding to the target resource transfer record; Obtaining a target resource transfer vector corresponding to the target resource transfer data; the target resource transfer vector is obtained by training according to a training resource transfer data sequence corresponding to the target data dimension, and the training resource transfer data in the training resource transfer data sequence is arranged according to the resource transfer order of the training resource transfer record corresponding to the training object; Obtaining a target resource transfer feature corresponding to the target resource transfer record based on the target resource transfer vector; Performing object recognition on the target object based on the target resource transfer feature to obtain a target recognition result corresponding to the target object; Among them, the target resource transfer vector is selected from a set of training resource transfer vectors, and the steps of generating the training resource transfer vectors in the set of training resource transfer vectors include: Obtaining current training resource transfer data from a training resource transfer data sequence, and obtaining associated resource transfer data corresponding to the current training resource transfer data in the training resource transfer data sequence; the associated resource transfer data is the training resource transfer data at the peripheral position of the current training resource transfer data in the training resource transfer data sequence; there are multiple training resource transfer data sequences, and each of the training resource transfer data sequences forms a data sequence set; Inputting the current training resource transfer data into a vector determination model to be trained to obtain a predicted association probability between the current training resource transfer data and the associated resource transfer data; the predicted association probability is the probability that the resource transfer data at the peripheral position where the associated resource transfer data predicted by the vector determination model is located is the associated resource transfer data; Obtaining a standard association probability between the current training resource transfer data and the associated resource transfer data based on the data sequence set; Adjusting the model parameters of the vector determination model based on the difference between the standard association probability and the predicted association probability to obtain a trained vector determination model, and obtaining training resource transfer vectors corresponding to each of the training resource transfer data based on the trained vector determination model.

2. The method according to claim 1, wherein The method further includes: Obtaining a sequence of training resource transfer records corresponding to a training object, where the sequence of training resource transfer records includes multiple training resource transfer records arranged according to the resource transfer order of the training resource transfer records; Obtaining training resource transfer data corresponding to the target data dimension from the resource transfer data corresponding to the training resource transfer record; Arranging each of the training resource transfer data according to the resource transfer order of the training resource transfer record to obtain the training resource transfer data sequence.

3. The method according to claim 1, wherein The obtaining the target resource transfer feature corresponding to the target resource transfer record based on the target resource transfer vector includes: Obtaining first resource transfer data corresponding to a first data dimension from the resource transfer data corresponding to the target resource transfer record; Determine the first resource transfer feature corresponding to the first resource transfer data; Concatenate the target resource transfer vector and the first resource transfer feature to obtain the target resource transfer feature corresponding to the target resource transfer record.

4. The method according to claim 1, wherein There are multiple target resource transfer records. The object recognition of the target object based on the target resource transfer feature to obtain the target recognition result corresponding to the target object includes: Arrange the target resource transfer features corresponding to the target resource transfer record according to the resource transfer order corresponding to the target resource transfer record to obtain a feature sequence; Input the feature sequence into an object recognition sequence model for processing to obtain the target recognition result corresponding to the target object.

5. The method according to claim 4, wherein The obtaining of the target resource transfer feature corresponding to the target resource transfer record based on the target resource transfer vector includes: According to the resource transfer order corresponding to the target resource transfer record, obtain the sorting feature corresponding to the target resource transfer vector; Fuse the target resource transfer vector and the sorting feature to obtain the target resource transfer feature corresponding to the target resource transfer record.

6. The method according to claim 4, characterized in that, The inputting of the feature sequence into an object recognition sequence model for processing to obtain the target recognition result corresponding to the target object includes: Obtain the recognition task matrix corresponding to the recognition task network in the object recognition sequence model; Based on the target resource transfer features in the feature sequence and the recognition task matrix, obtain the feature attention degrees corresponding to each of the target resource transfer features in the feature sequence; Adjust the target resource transfer features corresponding to the feature sequence based on the feature attention degrees to obtain the adjusted target resource transfer features, and the adjusted target resource transfer features form an adjusted feature sequence; Obtain the target recognition result corresponding to the target object based on the adjusted feature sequence.

7. The method according to claim 1, characterized in that, The determination of the target resource transfer record corresponding to the target object includes: Obtain the resource transfer record corresponding to the target object for resource transfer to the exchange object as the target resource transfer record; The object recognition of the target object based on the target resource transfer feature to obtain the target recognition result corresponding to the target object includes: Perform object recognition on the target object based on the target resource transfer feature to obtain the recognition result that the target object is a resource transfer abnormal object.

8. The method according to claim 1, wherein The steps of obtaining the target data dimension include: Obtain each data dimension corresponding to the target resource transfer record to form a data dimension set; Select the data dimensions that meet the dimension selection conditions from the data dimension set as the target data dimensions; the dimension selection conditions include that the number of data in the data dimension is greater than the quantity threshold.

9. A data processing method, characterized in that, The method includes: Obtain the training resource transfer record sequence corresponding to the training object, where the training resource transfer record sequence includes multiple training resource transfer records arranged according to the resource transfer order of the training resource transfer records; Obtain the training resource transfer data corresponding to the target data dimension from the resource transfer data corresponding to the training resource transfer record; Arrange each of the training resource transfer data in the resource transfer order of the training resource transfer record to obtain a training resource transfer data sequence; there are multiple training resource transfer data sequences, and each of the training resource transfer data sequences forms a data sequence set; Perform resource transfer vector training according to the training resource transfer data sequence to obtain a training resource transfer vector corresponding to each of the training resource transfer data; Obtain the current training resource transfer data from the training resource transfer data sequence, and obtain the associated resource transfer data corresponding to the current training resource transfer data in the training resource transfer data sequence; the associated resource transfer data is the training resource transfer data at the surrounding position of the current training resource transfer data in the training resource transfer data sequence; Input the current training resource transfer data into the vector determination model to be trained to obtain the predicted association probability between the current training resource transfer data and the associated resource transfer data; the predicted association probability is the probability that the resource transfer data at the surrounding position where the associated resource transfer data predicted by the vector determination model is located is the associated resource transfer data; Obtain the standard association probability between the current training resource transfer data and the associated resource transfer data based on the data sequence set; Adjust the model parameters of the vector determination model based on the difference between the standard association probability and the predicted association probability to obtain a trained vector determination model, and obtain a training resource transfer vector corresponding to each of the training resource transfer data based on the trained vector determination model.

10. A data processing device, characterized in that, The device includes: A target resource transfer record determination module for determining a target resource transfer record corresponding to a target object; A target resource transfer data acquisition module for acquiring target resource transfer data corresponding to a target data dimension from the resource transfer data corresponding to the target resource transfer record; A target resource transfer vector acquisition module for acquiring a target resource transfer vector corresponding to the target resource transfer data; the target resource transfer vector is trained according to the training resource transfer data sequence corresponding to the target data dimension, and the training resource transfer data in the training resource transfer data sequence is arranged in the resource transfer order of the training resource transfer record corresponding to the training object; A target resource transfer feature obtaining module for obtaining a target resource transfer feature corresponding to the target resource transfer record based on the target resource transfer vector; A target recognition result obtaining module for performing object recognition on the target object based on the target resource transfer feature to obtain a target recognition result corresponding to the target object; The target resource transfer vector is selected from a set of training resource transfer vectors; the apparatus further includes a module for obtaining the set of training resource transfer vectors, and the module for obtaining the set of training resource transfer vectors includes a unit for forming the set of training resource transfer vectors, configured to obtain current training resource transfer data from a training resource transfer data sequence, and obtain associated resource transfer data corresponding to the current training resource transfer data in the training resource transfer data sequence; the associated resource transfer data is the training resource transfer data at the surrounding positions of the current training resource transfer data in the training resource transfer data sequence; there are multiple training resource transfer data sequences, and each of the training resource transfer data sequences forms a data sequence set; input the current training resource transfer data into a vector determination model to be trained to obtain a predicted association probability between the current training resource transfer data and the associated resource transfer data; the predicted association probability is the probability that the resource transfer data at the surrounding position where the associated resource transfer data predicted by the vector determination model is located is the associated resource transfer data; obtain a standard association probability between the current training resource transfer data and the associated resource transfer data based on the data sequence set; adjust the model parameters of the vector determination model based on the difference between the standard association probability and the predicted association probability to obtain a trained vector determination model, and obtain training resource transfer vectors corresponding to each of the training resource transfer data based on the trained vector determination model.

11. The data processing device according to claim 10, wherein The module for obtaining the set of training resource transfer vectors further includes: a unit for obtaining a training resource transfer record sequence, configured to obtain a training resource transfer record sequence corresponding to a training object, where the training resource transfer record sequence includes multiple training resource transfer records arranged in the resource transfer order of the training resource transfer records; a unit for obtaining training resource transfer data, configured to obtain the training resource transfer data corresponding to the target data dimension from the resource transfer data corresponding to the training resource transfer record; a unit for obtaining a training resource transfer data sequence, configured to arrange each of the training resource transfer data in the resource transfer order of the training resource transfer records to obtain the training resource transfer data sequence.

12. The data processing device according to claim 10, characterized in that, The module for obtaining the target resource transfer feature includes: a first resource transfer data obtaining unit, configured to obtain first resource transfer data corresponding to a first data dimension from the resource transfer data corresponding to the target resource transfer record; a first resource transfer feature determination unit, configured to determine a first resource transfer feature corresponding to the first resource transfer data; a first target resource transfer feature obtaining unit, configured to splice the target resource transfer vector and the first resource transfer feature to obtain a target resource transfer feature corresponding to the target resource transfer record.

13. The data processing device according to claim 10, wherein There are multiple target resource transfer records, and the module for obtaining the target recognition result includes: a feature sequence obtaining unit, configured to arrange the target resource transfer features corresponding to the target resource transfer records in the resource transfer order corresponding to the target resource transfer records to obtain a feature sequence; A target recognition result obtaining unit, configured to input the feature sequence into an object recognition sequence model for processing to obtain a target recognition result corresponding to the target object.

14. The data processing device according to claim 13, wherein The target resource transfer feature obtaining module includes: A sorting feature obtaining unit, configured to obtain a sorting feature corresponding to the target resource transfer vector according to a resource transfer order corresponding to the target resource transfer record; A second target resource transfer feature obtaining unit, configured to fuse the target resource transfer vector with the sorting feature to obtain a target resource transfer feature corresponding to the target resource transfer record.

15. The data processing device according to claim 13, characterized in that, The target recognition result obtaining unit is further configured to obtain a recognition task matrix corresponding to a recognition task network in the object recognition sequence model; based on the target resource transfer feature in the feature sequence and the recognition task matrix, obtain feature attentions respectively corresponding to the respective target resource transfer features in the feature sequence; adjust the target resource transfer features corresponding to the feature sequence based on the feature attentions to obtain adjusted target resource transfer features, and the adjusted target resource transfer features form an adjusted feature sequence; obtain a target recognition result corresponding to the target object based on the adjusted feature sequence.

16. The data processing device according to claim 10, wherein The target resource transfer record determining module is further configured to obtain a resource transfer record corresponding to the target object for resource transfer to an exchange object as a target resource transfer record; the target recognition result obtaining module is further configured to perform object recognition on the target object based on the target resource transfer feature to obtain a recognition result that the target object is a resource transfer abnormal object.

17. The data processing device according to claim 10, characterized in that, The apparatus further includes a target data dimension obtaining module, and the target data dimension obtaining module includes: A data dimension set forming unit, configured to obtain respective data dimensions corresponding to the target resource transfer record to form a data dimension set; A target data dimension obtaining unit, configured to select data dimensions that meet a dimension selection condition from the data dimension set as the target data dimensions; the dimension selection condition includes that the data quantity of the data dimension is greater than a quantity threshold.

18. A data processing device, characterized in that, The apparatus includes: A training resource transfer record sequence obtaining module, configured to obtain a training resource transfer record sequence corresponding to a training object, where the training resource transfer record sequence includes a plurality of training resource transfer records arranged according to the resource transfer order of the training resource transfer records; A training resource transfer data obtaining module, configured to obtain training resource transfer data corresponding to the target data dimension from the resource transfer data corresponding to the training resource transfer record; A training resource transfer data sequence obtaining module, configured to arrange the respective training resource transfer data according to the resource transfer order of the training resource transfer records to obtain a training resource transfer data sequence; there are a plurality of the training resource transfer data sequences, and the respective training resource transfer data sequences form a data sequence set; A training resource transfer vector obtaining module, including: An associated resource transfer data acquisition unit, configured to acquire current training resource transfer data from the training resource transfer data sequence, and acquire associated resource transfer data corresponding to the current training resource transfer data in the training resource transfer data sequence; the associated resource transfer data is the training resource transfer data at the peripheral position of the current training resource transfer data in the training resource transfer data sequence; A prediction association probability obtaining unit, configured to input the current training resource transfer data into a vector determination model to be trained, and obtain a predicted association probability between the current training resource transfer data and the associated resource transfer data; the predicted association probability is the probability that the resource transfer data at the peripheral position where the associated resource transfer data predicted by the vector determination model is located is the associated resource transfer data; A standard association probability obtaining unit, configured to obtain a standard association probability between the current training resource transfer data and the associated resource transfer data based on the data sequence set; A trained resource transfer vector obtaining unit, configured to adjust the model parameters of the vector determination model based on the difference between the standard association probability and the predicted association probability to obtain a trained vector determination model, and obtain a trained resource transfer vector corresponding to each of the training resource transfer data based on the trained vector determination model.

19. A computer device includes a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the method according to any one of claims 1 to 9 is implemented.

20. A computer-readable storage medium stores a computer program, characterized in that, When the computer program is executed by the processor, the method according to any one of claims 1 to 9 is implemented.

21. A computer program product comprising computer instructions, characterized in that, When the computer instruction is executed by the processor, the method according to any one of claims 1 to 9 is implemented.

Images