Object recognition methods, apparatuses, computing devices, and storage media

Abstract

The application provides an object identification method and device, a computing device and a storage medium. The method comprises: receiving multi-dimension information of a target object, wherein the target object contains statistical characteristics, and the multi-dimension information comprises information of at least two dimensions; determining an associated object of the target object based on the multi-dimension information; acquiring a basic feature of the associated object; inputting the basic feature and the statistical characteristics into a pre-trained feature extraction model for identification, and outputting an identification result. The application can effectively detect the registration of malicious objects, save the cost of manual pre-audit, and prevent risks in advance.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and more particularly to an object recognition method, apparatus, computing device, and storage medium. Background Technology

[0002] The use of neural network models to verify newly registered merchants and determine their credit scores and transaction risk levels has seen significant development in recent years. Existing solutions mostly involve establishing blacklists for verification during the merchant registration phase, or using merchant registration information to develop strategies, assessing risk levels by checking for negative records of associated already registered merchants. However, the information available during merchant registration is less than during the transaction phase, making it difficult to develop effective strategies using information beyond the penalty records of associated entities. Furthermore, strategies relying solely on penalty records are insufficient to cover the large number of blacklisted merchants who can be linked to already registered merchants upon registration, hindering timely and effective control over the registration of blacklisted merchants. Summary of the Invention

[0003] In view of this, this application provides an object recognition method, apparatus, computing device, and storage medium.

[0004] According to a first aspect of this application, an object recognition method is provided, characterized by comprising: receiving multi-dimensional information of a target object, wherein the target object includes statistical features, and the multi-dimensional information includes information of at least two dimensions; determining associated objects of the target object based on the multi-dimensional information; obtaining basic features of the associated objects; inputting the basic features and the statistical features together into a pre-trained feature extraction model for recognition, and outputting a recognition result.

[0005] In some embodiments, determining the associated objects of the target object based on the multi-dimensional information includes: querying a database for objects that have the same information as the target object in at least one of the multi-dimensional dimensions, and using them as associated objects of the target object; wherein the database stores at least information about historical objects in the multi-dimensional dimensions.

[0006] In some embodiments, determining the associated objects of the target object based on the multi-dimensional information includes: obtaining basic information of each basic object in an object database, wherein the object database contains different basic objects and basic information corresponding to each basic object; calculating the vector distance between the target object and each basic object according to the multi-dimensional information and the basic information; and taking the basic objects whose vector distance is greater than or equal to a preset threshold as associated objects.

[0007] In some embodiments, the pre-trained feature extraction model includes an aggregation unit and a recognition unit. The step of inputting the basic features and the statistical features into the pre-trained feature extraction model for recognition and outputting the recognition result includes: using the aggregation unit to aggregate the basic features to obtain aggregated features; inputting the aggregated features and the statistical features into the recognition unit for recognition and outputting the recognition result.

[0008] In some embodiments, obtaining the basic characteristics of the associated object includes: concatenating vectors of the transaction characteristics, investigation characteristics, and business registration characteristics of the associated object to construct a feature sequence of the object, wherein the transaction characteristics include buyer gender, occurrence time, and occurrence method, the historical investigation characteristics include violation tags, and the business registration characteristics include registered capital, number of employees, and length of establishment.

[0009] In some embodiments, the database includes a graph database, and the statistical characteristics of the associated objects include one or more of the following: the number of associated objects, the average duration of their presence, the proportion of objects with transactions, and the proportion of objects marked as anomalous.

[0010] In some embodiments, the aggregation processing of the basic features to obtain aggregated features includes: inputting the basic features of each associated object of the target object into a trained LSTM model for aggregation processing, and outputting the aggregated features.

[0011] In some embodiments, the feature extraction model is trained based on a sample set, which includes the basic features of each associated object of the historical object and the labels associated with the historical object.

[0012] In some embodiments, the associated objects of the historical object are determined based on the following steps: receiving multi-dimensional information of the historical object, the multi-dimensional information including information in at least two dimensions; querying a database for objects that have the same information as the historical object in at least one of the multi-dimensional dimensions, and using them as candidate associated objects of the historical object; filtering the candidate associated objects based on the entry time of the historical object, and retaining the candidate associated objects before the entry time of the historical object as the associated objects of the historical object; wherein the database stores at least the information of the historical object in the multi-dimensional dimensions.

[0013] In some embodiments, the feature extraction model is trained through the following steps: acquiring historical multi-dimensional information of a historical object and historical labels associated with the historical object, wherein the historical object contains historical statistical features, and the historical multi-dimensional information includes historical information in at least two dimensions; determining historical associated objects of the historical object based on the historical multi-dimensional information; acquiring historical basic features of the historical associated objects, and importing the historical basic features and historical labels associated with the historical object into a preset recognition network, wherein the preset recognition network includes a preset aggregation unit and a preset recognition unit; using the preset aggregation unit to aggregate the historical basic features to obtain historical aggregated features; inputting the historical aggregated features and the historical statistical features into the preset recognition unit for recognition, and outputting a target recognition result; determining a target loss based on the target recognition result and historical labels associated with the historical object; if the target loss does not meet a preset condition, adjusting the network parameters of the preset recognition network according to the target loss, and updating the target loss based on the preset recognition network after adjusting the network parameters; if the target loss meets the preset condition, using the preset recognition network corresponding to the target loss meeting the preset condition as the feature extraction model.

[0014] In some embodiments, the step of inputting the basic features and the statistical features into a pre-trained feature extraction model for recognition and outputting the recognition result includes: inputting the basic features and the statistical features into a pre-trained feature extraction model to calculate the verification score of the target object; comparing the verification score with a predetermined threshold to output the recognition result of the target object.

[0015] According to another aspect of this application, an object recognition apparatus is provided. The apparatus includes: a receiving module configured to receive multi-dimensional information of a target object, wherein the target object includes statistical features, and the multi-dimensional information includes information in at least two dimensions; an associated object determination module configured to determine associated objects of the target object based on the multi-dimensional information; a basic feature acquisition module configured to acquire basic features of the associated objects; and a recognition module configured to input the basic features and the statistical features together into a pre-trained feature extraction model for recognition, and output a recognition result.

[0016] According to another aspect of this application, a computing device is provided. The computing device includes: a memory configured to store computer-executable instructions; and a processor configured to perform the method described in any of the embodiments of the foregoing object identification method when the computer-executable instructions are executed by the processor.

[0017] According to another aspect of this application, a computer-readable storage medium is provided that stores computer-executable instructions, which, when executed, perform the method described in any of the embodiments described in the foregoing object identification embodiments.

[0018] This application proposes an object recognition method, apparatus, computing device, and storage medium. The method utilizes multi-dimensional information of a target object (specifically, a target merchant) to determine its associated objects (specifically, associated merchants), obtaining the basic characteristics of these associated objects. These basic characteristics, along with statistical features, are input into a machine learning model for object recognition. By leveraging transaction information, penalty information, and company information of all associated objects, a comprehensive risk assessment of newly registered merchants is made to aid in penalty or screening decisions. After a new merchant registers and submits basic information, this method can quickly identify associated objects based on this information, extract the information, and use a trained model for recognition. Merchants with abnormal identification are rejected from registering. This effectively detects malicious registrations, saves on pre-registration manual review costs, and proactively prevents risks. Simultaneously, it improves the robustness and computational efficiency of object recognition calculations. Attached Figure Description

[0019] Embodiments of this application will now be described in more detail with reference to the accompanying drawings, wherein:

[0020] Figure 1 This application illustrates some scenarios for object recognition according to embodiments of this application;

[0021] Figure 2a This illustration shows a user interface during merchant registration on a platform prior to object identification, using an embodiment of this application.

[0022] Figure 2b This illustration shows a user interface for submitting qualification information on a merchant's onboarding platform prior to object identification, according to an embodiment of this application.

[0023] Figure 3 The diagram illustrates the architecture for extracting basic features of associated objects based on a merchant graph.

[0024] Figure 4 The diagram illustrates the specific construction of the basic features of the associated object;

[0025] Figures 5a-5d The diagram illustrates the principle of the Long Short-Term Memory (LSTM) model.

[0026] Figure 6 This diagram illustrates the overall architecture of a feature extraction model according to an embodiment of the present invention.

[0027] Figure 7 A flowchart illustrating the object recognition method is shown schematically.

[0028] Figure 8 The diagram schematically illustrates a device for object recognition; and

[0029] Figure 9 An example system is illustrated schematically, which includes an example computing device representing one or more systems and / or devices that can implement the various technologies described herein. Detailed Implementation

[0030] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. The described embodiments are merely some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0031] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is the technology of simulating human cognitive abilities through machines. AI is a comprehensive discipline, encompassing a wide range of fields, including perception, learning, reasoning, and decision-making capabilities, and involving both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly include computer vision, speech processing, natural language processing, and machine learning / deep learning. The core capability of AI is to make judgments or predictions based on given input. For example, in facial recognition applications, it can identify the person in an input photo. In medical diagnosis, it can determine the cause and nature of a disease based on input medical images.

[0032] In artificial intelligence software technology, machine learning is a crucial technique for endowing computers with intelligent characteristics. Machine learning is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, and many other disciplines. Machine learning specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning typically includes techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, and inductive learning.

[0033] Blockchain is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and cryptographic algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying platform, a platform product and service layer, and an application service layer.

[0034] The underlying blockchain platform can include processing modules such as user management, basic services, smart contracts, and operational monitoring. The user management module is responsible for managing the identity information of all blockchain participants, including maintaining public and private key generation (account management), key management, and maintaining the correspondence between user real identities and blockchain addresses (access management). Furthermore, under authorization, it monitors and audits transactions of certain real identities and provides risk control rule configuration (risk control audit). The basic services module is deployed on all blockchain node devices to verify the validity of business requests. After consensus is reached on valid requests, they are recorded in storage. For a new business request, the basic services first perform interface adaptation parsing and authentication (interface adaptation), and then encrypt the business information using a consensus algorithm (consensus management). After encryption, the data is transmitted completely and consistently to the shared ledger (network communication) and recorded and stored. The smart contract module is responsible for contract registration, issuance, triggering, and execution. Developers can define contract logic using a programming language and publish it to the blockchain (contract registration). According to the contract terms, the key or other events are invoked to trigger execution and complete the contract logic. It also provides functions for contract upgrades and cancellations. The operation monitoring module is mainly responsible for deployment, configuration modification, contract settings, cloud adaptation, and real-time status visualization during product launch, such as alarms, monitoring network conditions, and monitoring the health status of node devices.

[0035] The platform's product service layer provides basic capabilities and implementation frameworks for typical applications. Developers can leverage these basic capabilities, along with the characteristics of their business needs, to implement business logic on the blockchain. The application service layer provides blockchain-based application services to business stakeholders. For example, the object identification method disclosed in this application allows the data information involved to be stored on the blockchain.

[0036] To facilitate understanding of the embodiments of this application, several concepts will be briefly introduced below.

[0037] A knowledge graph is a series of various graphical representations that display the development process and structural relationships of knowledge. It uses visualization techniques to describe knowledge resources and their carriers, and to mine, analyze, construct, draw, and display knowledge and the relationships between them. A knowledge graph typically consists of nodes and edges; each node represents an "entity," and each edge represents a "relationship" between entities.

[0038] Recurrent Neural Networks (RNNs) are a type of recurrent neural network that takes sequence data as input, recursively processes the sequence, and connects all nodes (recurrent units) in a chain-like manner. RNNs solved the problem of handling variable-length word sequences. RNNs are well-suited for modeling sequence data, where the current output of a sequence is related to previous outputs. Specifically, the neural network memorizes previous information and applies it to the calculation of the current output. This means that the nodes in the hidden layers of the neural network are no longer disconnected but connected, and the input to the hidden layer includes not only the output of the input layer but also the hidden state output of the previous time step.

[0039] Long Short-Term Memory (LSTM) is a type of recurrent neural network designed to address the long-term dependency problem inherent in general RNNs. All RNNs have a chain-like structure of repeating neural network modules.

[0040] Logistic Regression (LR) is a generalized linear regression analysis model commonly used in data mining, automated disease diagnosis, and economic forecasting. The LR classifier is a linear classifier based on Support Vector Machines (SVMs). It uses a mapping function with real-valued independent variables and a dependent variable of [0, 1] to map the classifier's output to the probability that an object belongs to a predefined category. The dependent variable of the mapping function is positively correlated with the independent variables; that is, the dependent variable increases as the independent variables increase and decreases as the independent variables decrease. For example, the sigmoid function can be used as the mapping function. Linear classifiers are trained using input features extracted from a corresponding neural network model.

[0041] Figure 1 An application scenario 100 for object recognition according to some embodiments of this application is illustrated. In this application scenario, one or more user interfaces 101 communicate bidirectionally with one or more computing devices 108 via an intermediate device 105. A user 104 interacts with one or more user interfaces 101 to complete bidirectional communication with the computing device 108.

[0042] Optionally, one or more databases, such as one or more of a first database 110, a second database 120, or a third database 130, may exist to cooperate with the computing device 108 to implement functions. It should be understood that in some embodiments, one or more of these databases may be integrated into the computing device 108.

[0043] In some embodiments, the intermediate device 105 may include a network connection, such as a combination of a local area network (LAN), a wide area network (WAN), a personal area network (PAN), and / or a communication network such as the Internet. In this case, the computing device 108 may act as a server, and the user interface 101 may interact with one or more computing devices 108, for example, via a network, such as sending data to or receiving data from them. Each of the computing device 108 and the one or more user interfaces 101 may include at least one communication interface (not shown) capable of communicating through the intermediate device 105. Such a communication interface may be one or more of the following: any type of network interface (e.g., a network interface card (NIC)), a wired or wireless (such as an IEEE 802.11 wireless LAN (WLAN)) wireless interface, a Wi-MAX interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, or Bluetooth. TM Interfaces, near field communication (NFC) interfaces, etc. Further examples of communication interfaces are described elsewhere in this document.

[0044] In some embodiments, the intermediate device 105 may be a direct electrical connection, and the user interface 101 and one or more computing devices 108 may be integrated onto one or more terminal devices (not shown). The one or more terminal devices may be any type of computing device, including mobile computers (e.g., Microsoft® Surface® devices, personal digital assistants (PDAs), laptop computers, notebook computers, and devices such as the Apple iPad). TM Tablet computers, netbooks, etc.), mobile phones (e.g., cellular phones, smartphones such as Microsoft Windows® phones, Apple iPhones, and devices implementing Google® Android). TM Operating systems such as phones, Palm® devices, Blackberry® devices, etc.), and wearable devices (such as smartwatches, head-mounted devices, including smart glasses such as Google® Glass). TM(e.g., mobile devices, game consoles, smart TVs, etc.) or other types of mobile devices. In some embodiments, one or more terminal devices may also be fixed devices, such as desktop computers, game consoles, smart TVs, etc. Furthermore, when multiple terminal devices exist, the multiple terminal devices may be the same or different types of devices.

[0045] The terminal device may include a display screen (not shown) and a terminal application (not shown) that can be interacted with by the user through the display screen. The terminal application may be a local application, a web application, or a lightweight application (such as a mobile mini-program or WeChat mini-program). If the terminal application is a local application that requires installation, it can be installed on the terminal device. If the terminal application is a web application, it can be accessed through a browser. If the terminal application is a mini-program, it can be opened directly on the user's terminal without installation by searching for relevant information (such as the application's name) or scanning its graphic code (such as a barcode or QR code).

[0046] Figure 2a This document illustrates a user interface for merchant registration on a platform before object recognition, according to an embodiment of this application. The user interface shows the merchant registration process: First, the merchant needs to register an account. Typically, merchants register quickly using a mobile phone verification code or other universal account. Next, the merchant claims their store and submits qualification materials. In one example, qualification materials may include store information, business license, legal representative's ID card, legal representative's mobile phone number, etc. As understood by those skilled in the art, qualification materials are not limited to the above information. After submitting the qualification materials, the backend verifies the merchant's qualification materials using the object recognition method provided in this application and provides the review results to the applicant merchant after a predetermined time (e.g., 1-3 business days). If the verification result indicates good credit and low transaction risk, the platform will sign a cooperation agreement with the merchant. The platform can categorize these agreements into sales agreements and self-service agreements depending on the industry and region. The platform will contact the signed merchant after the predetermined time.

[0047] Figure 2b This illustration shows a user interface for submitting qualification information to a merchant's onboarding platform before object identification, according to an embodiment of this application. The user interface prompts the user to fill in the relevant necessary qualification information. Figure 2b In the example shown, the qualification-related information includes: mobile phone number, ID card number, bank card number, credit code number, shareholder representative's name, company full name, and contact email address. As understood by those skilled in the art, the qualification-related information is not limited to the above information and may also include other suitable information.

[0048] Figure 3The diagram 300 schematically illustrates the architecture for extracting basic characteristics of associated objects (merchants) based on a merchant graph. Compared to the information available during the transaction phase, less information is available during the onboarding phase (pre-onboarding) of a new merchant 301. The credit and risk scoring conducted on new merchant 301 during the onboarding phase is referred to as "pre-onboarding scoring." Here, information is extracted... Figure 3 The structured information on the merchant graph 302 is used to train the feature extraction model. This allows for a comprehensive assessment of the entry risk of the new merchant 301 by utilizing transaction information, penalty information, and company entity information of related merchants extracted from the merchant graph 302. Specifically, extracting structured information from the merchant graph 302 includes: selecting related objects of the new merchant 301 and searching for the basic features constituted by various characteristics of the related objects at the time of their entry into the merchant graph 302. Figure 3 Each circle in the vector represents a historical merchant, and its vector representation is... These are the basic characteristics of the historical merchants. In one embodiment, each historical merchant has screening characteristics, including the merchant's blacklisting records and tags, etc. (For example, blacklisted merchants can be marked as 1, and unblacklisted merchants can be marked as 0). In one embodiment, each historical merchant has transaction characteristics, including the gender ratio of transaction users, average age, number of blocked transactions, etc. In one embodiment, each historical merchant has business characteristics, including the size of the merchant's registered company, registered capital, market establishment, etc. Figure 3 The newly registered merchant 301 shown is identified by using the following association relationships: the contact person's mobile phone number, legal representative's ID card number, bank card number, unified credit code, shareholder representative, full name of the merchant, and contact email address provided during merchant registration. In other words, if a historical merchant shares the same information as the newly registered merchant 301 in a certain dimension (e.g., using the same mobile phone number), that merchant is identified as an associated object of the newly registered merchant 301. Similarly, merchants that share the same information as the newly registered merchant 301 in any one or more other dimensions such as legal representative's ID card number, bank card number, unified credit code, shareholder representative, full name of the merchant, and contact email address can also be identified as associated objects of the newly registered merchant 301. Next, the feature sequences of the identified associated objects are input into a pre-trained feature extraction model (in this paper, a model combining an aggregation unit (Long Short-Term Memory network) and a recognition unit (Logistic Regression network) is used as an example) to predict the pre-rating of the registered merchants, and the results are output after identifying abnormal objects based on the ratings.

[0049] Figure 4This diagram illustrates the specific construction of the basic feature sequence of associated objects. Here, we use the construction of the basic features of associated objects of a target object as an example. The associated objects of the target object are selected from the merchants already registered in the historical records. Merchants that have not been penalized so far are marked as "white merchants" (e.g., marked as 0), and merchants that have been penalized or blocked so far are marked as "black merchants" (e.g., marked as 1). Next, based on the basic information submitted by the target object upon registration and the registration time, the merchants associated with the target object at the time of registration are selected as the aforementioned associated objects, and the basic features of these associated objects at the time of the target object's registration are obtained. Here, the basic features of the associated objects at the time of the target object's registration include at least one or more of the following: transaction features, investigation features, and business registration features, all of which are features up to the time of the target object's registration. Transaction features refer to the merchant's transaction pairs and corresponding tags, such as the gender of the buyer corresponding to the transaction, the time of the transaction, the method of the transaction, etc. In one embodiment, for a specific type of transaction of an associated object, the proportion of that type of transaction amount to the total transaction amount is calculated as a transaction characteristic of the associated object. Screening characteristics: If a merchant has been penalized or has had transactions restricted, corresponding violation tags will be left. These violation tags are encoded as screening characteristics of the associated object. In one embodiment, one-hot encoding is used to encode these violation tags. Business registration characteristics: Basic business registration information of the merchant's company, such as registered capital, number of employees, and establishment date. Figure 4 Based on the above characteristics, the basic features of each associated object are constructed. The basic features in the figure include the feature vectors of each associated object, and the feature vector of each associated object includes transaction features, investigation features, and business registration features.

[0050] It is worth noting that the target object includes statistical characteristics. These characteristics may include the number of associated objects of the target object, the average onboarding time of associated objects, the proportion of associated objects with transactions, and the proportion of associated objects that have been penalized. In one example, the statistical characteristics of the target object's associated objects can be calculated based on the characteristics of each associated object. First, the number of associated objects of the target object is counted. The onboarding time of the target object's associated objects is calculated based on the onboarding time of each associated object. Specifically, the average onboarding time of the target object's associated objects is the ratio of the sum of the onboarding times of all associated objects to the number of associated objects. For example, if the target object has 5 associated objects, and the average onboarding times are 10 months, 12 months, 20 months, 16 months, and 2 months, then the average onboarding time of the associated objects is (10+12+20+16+2) / 5=12 months. The proportion of associated objects with transactions is calculated based on the proportion of merchants who have made transactions among the associated objects. Specifically, the proportion of associated objects with transactions is the ratio of the number of associated objects with transactions to the total number of associated objects. For example, if a target object has 5 associated objects, and 3 of them have transactions, then the transaction ratio of the associated objects is 60%. The penalty ratio of the target object's associated objects is calculated based on the proportion of penalized merchants in each associated object. Specifically, the penalty ratio of the target object's associated objects is the ratio of the number of penalized objects to the total number of associated objects. For example, if a target object has 5 associated objects, and 2 of them have been penalized, then the penalty ratio of the associated objects is 40%. As those skilled in the art will understand, the statistical characteristics of the target object are not limited to the characteristics described above.

[0051] This application uses historical merchants as samples, filters other merchants associated with a merchant in the graph database according to time, and retains only the associated objects before the merchant's entry time. Then, it uses the basic features of the associated objects before the merchant's entry time (including transaction features, investigation features, business registration features, etc.) and the statistical features of the target object as input to the feature extraction model. Here, the feature extraction model includes aggregation units and recognition units. In one embodiment, the aggregation unit has an LSTM network structure. In another embodiment, the recognition unit has an LR network structure. The number of associated objects is uncertain. The basic features of the associated objects are constructed as a variable-length sequence as input to the LSTM. The output of the last hidden layer of the LSTM is extracted, and this output is concatenated with the statistical features of the target object as input to the LR model for end-to-end training.

[0052] The Long Short-Term Memory (LSTM) model in this application's feature extraction model consists of K structurally identical units. When evaluating newly joined merchants, the input to each unit of this LSTM is a basic feature of one of the associated objects identified as the new merchant's associated objects. Figures 5a-5d The diagram illustrates the principle of one cell in a K-cell Long Short-Term Memory (LSTM) unit.

[0053] In each LSTM unit, the input sequence is processed sequentially through the forget gate, input gate, and output gate. Figure 5a The dashed box in the image schematically illustrates the forget gate in an LSTM model. During the forget gate processing phase, the forget gate determines what information needs to be discarded. The following formula... The output of the forget gate is specifically given. The forget gate reads the output vector of the previous layer (the previous LSTM unit). and the vector to be input this time In this plan, These are the basic features of the associated objects. Multiple inputs into the LSTM can feed the entire sequence of associated objects into the model. After passing through σ, a value between 0 and 1 is obtained and assigned to each... The number.

[0054]

[0055] here, The output of the forget gate is σ, where σ is the sigmoid function. The weight matrix assigned to this layer, This is the output vector of the previous layer. Let b be the basic feature of a related object in the sequence of related objects, i.e., the vector to be input in this case. f The bias vector set for this layer.

[0056] Figure 5b The input gate in the Long Short-Term Memory (LSTM) model is schematically illustrated with a dashed box in the `-c` section. During the input gate processing phase, its role is to determine how much new information is added to the new state. First, and The process involves using a sigmoid function, which determines which information needs to be updated.

[0057]

[0058] in, W represents the output from σ in the input gate. i This represents the weight matrix assigned to this layer, b i This is the bias vector set for this layer. The meanings are the same: σ is the sigmoid function. The weight matrix assigned to this layer, This is the output vector of the previous layer. This refers to the basic characteristics of an associated object in a sequence of associated objects.

[0059] Next, the tanh layer generates a vector, and the purpose of this stage is to determine the content to be updated.

[0060]

[0061] in, For candidate neural unit information used for updating, This represents the weight matrix assigned to this layer. It is the bias vector set for this layer, and tanh is the hyperbolic tangent function.

[0062] This represents the state of the unit at time t, and in this scheme, it can represent a higher-order representation of the merchants. In the next step, these two parts are multiplied to update the unit's state. and Multiply, discard unnecessary information, then add. Once new candidate values ​​are obtained, merchants can extract the relevant features of associated objects. That is, merchants can select the associated objects of historical merchants and search the graph for the feature sequence of associated objects formed by the various features of the associated objects when they joined.

[0063]

[0064] in, For the output of the forget gate, i t The output of the input gate. For candidate neural unit information, C t-1 It is the state of the previous unit.

[0065] Figure 5d The dashed box schematically illustrates the output gate in a Long Short-Term Memory (LSTM) model. During the output gate processing phase... and After a sigmoid function determines which information needs to be output, then... The tanh function is used to process the data, and then multiplied by the output of the sigmoid function to determine the output. The output gate combines the state of the LSTM cell with the current input and outputs the result to the next cell. Since this application does not consider the order of input, after the merchant sequence input is completed, the following steps are used: This represents the aggregated representation of the merchants who have joined the platform.

[0066]

[0067] in, Indicates the process The output of (i.e., the sigmoid function), This represents the weight matrix assigned to this layer. It is the bias vector set for this layer.

[0068]

[0069] in, This represents the output after processing by the tanh function, where tanh is the hyperbolic tangent function.

[0070] Figure 6 The diagram illustrates a feature extraction model based on graph database-driven association feature extraction according to an embodiment of the present invention. Figure 6 The Long Short-Term Memory (LSTM) model in the dataset consists of K structurally identical units. When evaluating newly registered merchants, the input to each unit of the LSTM is a feature of one of the associated objects identified as belonging to the new merchant. For example, in one embodiment, if K users are identified from historical users who share the same information as the new merchant across one or more dimensions (including the contact person's mobile phone number, legal representative's ID card, bank card number, unified credit code, shareholder representative, merchant's full name, and contact email address), these K users are identified as merchants associated with the new merchant. The features of these K associated objects are determined, and each associated object's features are input into one of the K units of the LSTM. The number of associated objects is uncertain; the associated object features are constructed as a variable-length sequence as input to the LSTM. Finally, the last hidden layer of the LSTM is extracted and concatenated with the association statistical features, serving as input to the LR model for end-to-end training.

[0071] During the training phase of this model, historical merchants are used as samples. Other merchants associated with a merchant in the graph database are filtered by time, retaining only the associated objects before the merchant's entry time. Then, the transaction features and screening features of the associated objects before the merchant's entry time are used as model input. Similar to the testing phase, the number of associated objects is uncertain during the training phase. The associated object features are constructed as variable-length sequences as input to the LSTM. Finally, the last hidden layer of the LSTM is extracted and concatenated with the association statistical features as input to the LR model for end-to-end training.

[0072] In this application, the LSTM module and the LR module together constitute the feature extraction model. The feature extraction model is trained based on a sample set, which includes the basic features of each associated object of the historical object and the labels associated with the historical object. Each associated object of the historical object is determined based on the following steps: receiving multi-dimensional information of the historical object, the multi-dimensional information including information of at least two dimensions; querying the database for objects that have the same information as the historical object in at least one of the multi-dimensional dimensions, as candidate associated objects of the historical object; filtering the candidate associated objects based on the entry time of the historical object, and retaining the candidate associated objects before the entry time of the historical object as the associated objects of the historical object; wherein the database stores at least the information of the historical object in the multi-dimensional dimensions. In one embodiment, the feature extraction model is trained through the following steps: acquiring historical multi-dimensional information of a historical object and historical labels associated with the historical object, wherein the historical object contains historical statistical features, and the historical multi-dimensional information includes historical information in at least two dimensions; determining historical associated objects of the historical object based on the historical multi-dimensional information; acquiring the historical basic features of the historical associated objects and inputting the historical basic features into an LSTM model for aggregation processing to obtain historical aggregated features; inputting the historical aggregated features and the historical statistical features into a preset recognition network for recognition and outputting target recognition results; determining target loss based on the target recognition results and historical labels associated with the historical object; if the target loss does not meet preset conditions, adjusting the network parameters of the preset recognition network according to the target loss, and updating the target loss based on the preset recognition network after adjusting the network parameters; if the target loss meets the preset conditions, using the preset recognition network corresponding to the target loss meeting the preset conditions as the feature extraction model.

[0073] Figure 7 The diagram illustrates a method 700 for object recognition, which is performed using a machine learning model based on the extraction of association features from a graph database. In step 701 of method 700, multi-dimensional information about the target object is first received. This target object includes statistical features, and the multi-dimensional information includes information from at least two dimensions. Specifically, in a risk assessment of merchant onboarding to a platform, the target object represents a newly onboarded merchant. The statistical features of the associated object include one or more of the following: the number of associated objects, the average onboarding duration, the proportion of transactions, and the proportion of objects marked as abnormal. In one embodiment, receiving multi-dimensional information about the target user includes receiving qualification-related information about the target user, including information on the dimensions of mobile phone number, ID card number, bank card number, credit code number, shareholder representative name, full name, and contact email address.

[0074] In step 702, associated objects of the target object are determined based on multi-dimensional information. In one embodiment, users with the same information as the target user in at least one of the multi-dimensional dimensions are queried in the database and designated as associated objects of the target object; wherein the database stores at least historical user information in the multi-dimensional dimensions. In one embodiment, the database includes a graph database, which stores at least historical user information in the multi-dimensional dimensions. In another embodiment, the database includes a graph database, which stores at least historical user information in the multi-dimensional and other dimensions, including transaction characteristics, investigation characteristics, and business registration characteristics of historical users. In another embodiment, basic information of each basic object in the object database is obtained, wherein the object database contains different basic objects and basic information corresponding to each basic object; the similarity between the target object and each basic object is calculated based on the multi-dimensional information and the basic information; basic objects with a similarity greater than or equal to a preset threshold are designated as associated objects. As understood by those skilled in the art, similarity can be achieved by calculating the distance between vectors, including but not limited to cosine distance or Euclidean distance between vectors.

[0075] In step 703, the basic characteristics of the associated objects are obtained. Based on the basic information submitted by these sample merchants upon registration and their registration time, the merchants associated with these merchants at the time of registration are filtered out, and various characteristics of these merchants at the time of registration are calculated. Here, the various characteristics at the time of registration include at least one or more of the following: transaction characteristics, investigation characteristics, and business registration characteristics. The above characteristics are all characteristics of the sample merchants up to the time of registration. Obtaining the basic characteristics of the associated objects includes: constructing the basic characteristics of the associated objects by concatenating the vectors of the transaction characteristics, investigation characteristics, and business registration characteristics of the associated objects. The transaction characteristics include statistical values ​​for buyer gender, transaction time, and transaction method; the historical investigation characteristics include statistical values ​​for violation tags; and the business registration characteristics include statistical values ​​for registered capital, number of employees, and length of establishment.

[0076] In step 704, the basic features and statistical features are input together into a pre-trained feature extraction model for recognition, and the recognition result is output. In one embodiment, the pre-trained feature extraction model includes an aggregation unit and a recognition unit. Inputting the basic features and statistical features together into the pre-trained feature extraction model for recognition and outputting the recognition result includes: using the aggregation unit to aggregate the basic features to obtain aggregated features; inputting the aggregated features and statistical features together into the recognition unit for recognition, and outputting the recognition result. In one embodiment, the pre-trained feature extraction model is a feature extraction model with variable-length sequences as input. In another embodiment, the aggregation unit is a Long Short-Term Memory (LSTM) model, and the recognition unit is a combination of LR models. In one embodiment, the pre-trained feature extraction model is trained based on a sample set, which includes feature sequences of each associated object of a historical user and labels associated with the historical user. In one embodiment, each associated object of a historical user is determined based on the following steps: finding users who have the same information as the historical user in at least one dimension of multiple dimensions, and using them as associated objects of the historical user; filtering the associated objects of the historical user by time, and retaining the associated objects before the historical user's entry time. The characteristic sequences of each associated object of a historical user include the transaction characteristics, investigation characteristics, and business registration characteristics of each associated object before the historical user's onboarding time. The statistical characteristics of the associated objects include one or more of the following: the number of associated objects based on the target user, the average onboarding time of the associated objects, the proportion of associated objects with transactions, and the proportion of associated objects subject to penalties. Tags associated with historical users include whether the historical user was marked as a "white user" or a "black user."

[0077] In one embodiment, the feature extraction model is trained through the following steps: acquiring historical multi-dimensional information of a historical object and historical labels associated with the historical object, wherein the historical object contains historical statistical features, and the historical multi-dimensional information includes historical information in at least two dimensions; determining historical associated objects of the historical object based on the historical multi-dimensional information; acquiring the historical basic features of the historical associated objects, and importing the historical basic features and the historical labels associated with the historical object into a preset recognition network, wherein the preset recognition network includes a preset aggregation unit and a preset recognition unit; and using the preset aggregation unit to aggregate the historical basic features to obtain historical aggregated features. Specifically, the historical basic features of each historical associated object of the historical object are input into an LSTM model for aggregation processing. Here, the Long Short-Term Memory (LSTM) model consists of K model units with identical structures, and the input of each model unit in the LSTM is the historical basic feature of one of the historical associated objects identified as the historical object. Because the number of historical associated objects is uncertain, the features of the historical associated objects are constructed as a variable-length sequence as the input of the LSTM. Specifically, the output features of the last hidden layer of the LSTM can be extracted at the end of each unit of the LSTM and aggregated as historical aggregated features. The historical aggregated features and historical statistical features are input into a preset recognition unit for recognition, and the target recognition result is output. Based on the target recognition result and the historical labels associated with the historical object, the target loss is determined. In one example, the target loss is defined by a target loss function, which can be the maximum likelihood function. As those skilled in the art will understand, other forms of target loss functions can also be used. If the target loss does not meet a preset condition, the network parameters of the preset recognition network are adjusted according to the target loss, and the target loss is updated based on the preset recognition network with the adjusted network parameters. The preset condition is the convergence of the target loss function, specifically the convergence of the maximum likelihood function. If the target loss meets the preset condition, the preset recognition network corresponding to the target loss meeting the preset condition is used as the feature extraction model. Finally, the aggregated features and associated statistical features output from the last hidden layer of the LSTM are concatenated and used as the input to the recognition unit LR model in the feature extraction model. A validation score is output from the LR model, which is a score between 0 and 1.

[0078] The process of object identification involves inputting the feature sequences and statistical features of associated objects into a pre-trained feature extraction model. This includes: calculating a verification score for the target user; the verification score, output from the recognition unit (LR model) in the feature extraction model, is a score between 0 and 1; comparing the score with a predetermined threshold to output the verification result for the target user; and finally, identifying abnormal objects based on the identification result. For example, if the output risk score is 0.68, and the predetermined threshold is 0.5, the user is deemed risky, and their application is rejected.

[0079] This application proposes a machine learning model-based object recognition method using graph database-driven feature extraction. This method constructs a merchant graph, extracts unstructured information from the graph, trains a machine learning model, and then uses transaction information, penalty information, and company information of all associated merchants to comprehensively assess the risk of newly registered merchants, ultimately assisting in decision-making regarding penalties or investigations. After a new merchant registers and submits basic information, the method can quickly identify associated objects in the graph based on this information, extract the information, and score them using the trained model. Merchants scoring above a certain threshold are rejected from registering. This effectively detects malicious merchant registrations, saves on manual pre-screening costs, and proactively prevents risks. Simultaneously, it improves computational robustness and efficiency.

[0080] Figure 8A schematic diagram illustrates a machine learning model object recognition device 800 based on graph database-based association feature extraction. The device includes: a receiving module 801, an association object determination module 802, a basic feature acquisition module 803, an aggregated feature acquisition module 804, and a recognition module 805. The receiving module 801 is configured to receive multi-dimensional information of a target object, wherein the target object includes statistical features, and the multi-dimensional information includes information from at least two dimensions. The association object determination module 802 is configured to determine the association objects of the target object based on the multi-dimensional information. The basic feature acquisition module 803 is configured to acquire the basic features of the association objects. The recognition module 804 is configured to input the basic features and statistical features into a pre-trained feature extraction model for recognition and output the recognition result. This application proposes a machine learning model object recognition device based on graph database-based association feature extraction, which constructs a merchant graph. By extracting unstructured information from the graph, a machine learning model is trained, and then the transaction information, penalty information, and company entity information of all associated merchants are used to comprehensively assess the risk of newly registered merchants, ultimately assisting in decision-making regarding penalties or investigations. After a new merchant registers and submits basic information, the device can quickly identify related objects in the database based on this information. After extracting the information, a trained model scores the merchants, rejecting those scoring above a certain threshold. This effectively detects malicious merchant registrations, saving on pre-registration manual review costs and proactively mitigating risks. Simultaneously, it improves computational robustness and efficiency.

[0081] Figure 9 An example system 900 is illustrated, including an example computing device 910 representing one or more systems and / or devices that can implement the technical solutions described in the various embodiments herein. The computing device 910 may be, for example, a server of a service provider, a device associated with a server, a system-on-a-chip, and / or any other suitable computing device or computing system. References above... Figure 8 The object recognition device 800, which uses a machine learning model based on graph database association feature extraction, can take the form of a computing device 910. Alternatively, the object recognition device 800 can be implemented as a computer program as an application 916.

[0082] The example computing device 910, as illustrated in Figure 9, includes a processing system 911 communicatively coupled to each other, one or more computer-readable media 912, and one or more I / O interfaces 913. Although not shown, the computing device 910 may also include a system bus or other data and command transfer system that couples the various components to each other. The system bus may include any or a combination of different bus architectures, such as a memory bus or memory controller, a peripheral bus, a universal serial bus, and / or a processor or local bus utilizing any of the various bus architectures. Various other examples, such as control and data lines, are also conceived.

[0083] Processing system 911 represents the functionality of performing one or more operations using hardware. Therefore, processing system 911 is illustrated as including hardware elements 914 that can be configured as processors, function blocks, etc. This may include other logic devices implemented in hardware as application-specific integrated circuits (ASICs) or formed using one or more semiconductors. Hardware element 914 is not limited by the materials in which it is formed or the processing mechanism employed therein. For example, a processor may consist of semiconductors and / or transistors (e.g., integrated circuits (ICs)). In such a context, processor-executable instructions may be electronically executable instructions.

[0084] Computer-readable medium 912 is illustrated as including memory / storage device 915. Memory / storage device 915 represents a memory / storage capacity associated with one or more computer-readable media. Memory / storage device 915 may include volatile media (such as random access memory (RAM)) and / or non-volatile media (such as read-only memory (ROM), flash memory, optical disk, magnetic disk, etc.). Memory / storage device 915 may include fixed media (e.g., RAM, ROM, fixed hard disk drive, etc.) and removable media (e.g., flash memory, removable hard disk drive, optical disk, etc.). Computer-readable medium 912 may be configured in various other ways as further described below.

[0085] One or more I / O interfaces 913 represent the functionality to allow users to input commands and information to the computing device 910 using various input devices and optionally also to present information to the user and / or other components or devices using various output devices. Examples of input devices include keyboards, cursor control devices (e.g., mice), microphones (e.g., for voice input), scanners, touch functionality (e.g., capacitive or other sensors configured to detect physical touch), cameras (e.g., capable of detecting non-touch-related motion as gestures using visible or invisible wavelengths (such as infrared frequencies), and so on. Examples of output devices include display devices (e.g., monitors or projectors), speakers, printers, network interface cards (NICs), haptic-responsive devices, and so on. Therefore, the computing device 910 can be configured to support user interaction in various ways as further described below.

[0086] The computing device 910 also includes an application 916. The application 916 may be, for example, referred to... Figure 8 The software instance of the training device 800 for the described translation model is implemented in combination with other elements in the computing device 910 to implement the techniques described herein.

[0087] This document describes various technologies within the general context of software and hardware components or program modules. Generally, these modules include routines, programs, objects, elements, components, data structures, etc., that perform specific tasks or implement specific abstract data types. As used herein, the terms "module," "function," and "component" generally refer to software, firmware, hardware, or a combination thereof. The technologies described herein are platform-independent, meaning they can be implemented on a variety of computing platforms with various processors.

[0088] Implementations of the described modules and technologies may be stored on or transmitted across some form of computer-readable medium. The computer-readable medium may include a variety of media accessible by the computing device 910. By way of example and not limitation, the computer-readable medium may include "computer-readable storage media" and "computer-readable signal media".

[0089] In contrast to simple signal transmission, carrier waves, or signals themselves, a "computer-readable storage medium" refers to a medium and / or device capable of persistently storing information, and / or a tangible storage device. Therefore, a computer-readable storage medium refers to a non-signal-bearing medium. Computer-readable storage media include hardware such as volatile and non-volatile, removable and non-removable media and / or storage devices implemented using methods or techniques suitable for storing information (such as computer-readable instructions, data structures, program modules, logic elements / circuits, or other data). Examples of computer-readable storage media may include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical storage devices, hard disks, cassette tapes, magnetic tapes, disk storage devices or other magnetic storage devices, or other storage devices, tangible media, or articles of art suitable for storing desired information and accessible by a computer.

[0090] "Computer-readable signal medium" refers to a signal-bearing medium configured to transmit instructions, such as via a network, to computing device 910. A signal medium typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal, such as a carrier wave, data signal, or other transmission mechanism. Signal media also include any information transmission medium. The term "modulated data signal" refers to a signal in which one or more of its characteristics are set or altered to encode information. By way of example and not limitation, communication media include wired media such as wired networks or direct connections, and wireless media such as acoustic, RF, infrared, and other wireless media.

[0091] As previously stated, hardware element 914 and computer-readable medium 912 represent instructions, modules, programmable device logic, and / or fixed device logic implemented in hardware, which in some embodiments may be used to implement at least some aspects of the techniques described herein. Hardware elements may include components of integrated circuits or systems-on-a-chip, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), complex programmable logic devices (CPLDs), and other implementations or other hardware devices in silicon. In this context, hardware elements may serve as processing devices for executing program tasks defined by instructions, modules, and / or logic embodied by the hardware element, and as hardware devices for storing instructions for execution, such as the previously described computer-readable storage medium.

[0092] The foregoing combinations can also be used to implement the various techniques and modules described herein. Therefore, software, hardware, or program modules and other program modules can be implemented as one or more instructions and / or logic embodied on some form of computer-readable storage medium and / or by one or more hardware elements 914. The computing device 910 can be configured to implement specific instructions and / or functions corresponding to the software and / or hardware modules. Thus, for example, by using the computer-readable storage medium and / or hardware elements 914 of a processing system, modules can be implemented at least partially in hardware as modules executable as software by the computing device 910. Instructions and / or functions can be executable / operable by one or more articles of art (e.g., one or more computing devices 910 and / or processing systems 911) to implement the techniques, modules, and examples described herein.

[0093] In various embodiments, the computing device 910 can be configured in various ways. For example, the computing device 910 can be implemented as a computer-type device, including personal computers, desktop computers, multi-screen computers, laptop computers, netbooks, etc. The computing device 910 can also be implemented as a mobile device, including mobile devices such as mobile phones, portable music players, portable gaming devices, tablet computers, multi-screen computers, etc. The computing device 910 can also be implemented as a television-type device, including devices with or connected to a generally large screen in a leisure viewing environment. These devices include televisions, set-top boxes, game consoles, etc.

[0094] The techniques described herein can be supported by these various configurations of computing device 910, and are not limited to specific examples of the techniques described herein. Functionality can also be implemented, wholly or partially, on the “cloud” 920 using distributed systems, such as through platform 922 as described below.

[0095] Cloud 920 includes and / or represents platform 922 for resource 924. Platform 922 abstracts the underlying functionality of the hardware (e.g., server) and software resources of cloud 920. Resource 924 may include other applications and / or data that can be used when performing computer processing on a server located remotely from computing device 910. Resource 924 may also include services provided via the Internet and / or via subscriber networks such as cellular or Wi-Fi networks.

[0096] Platform 922 can abstract resources and functions to connect computing device 910 to other computing devices. Platform 922 can also be used to abstract resource hierarchy to provide a corresponding level of hierarchy for any encountered needs for resource 924 implemented via platform 922. Therefore, in interconnect device embodiments, the implementation of the functions described herein can be distributed throughout system 900. For example, functions can be implemented partly on computing device 910 and partly through platform 922, which abstracts the functions of cloud 920.

[0097] It should be understood that, for clarity, embodiments of this application have been described with reference to different functional units. However, it will be apparent that, without departing from this application, the functionality of each functional unit may be implemented in a single unit, in multiple units, or as part of other functional units. For example, functionality described as being performed by a single unit may be performed by multiple different units. Therefore, references to specific functional units are considered merely as references to the appropriate units used to provide the described functionality, and not as indicating a strict logical or physical structure or organization. Thus, this application may be implemented in a single unit, or may be physically and functionally distributed among different units and circuits.

[0098] It will be understood that although the terms first, second, third, etc., may be used herein to describe various devices, elements, components, or parts, these devices, elements, components, or parts should not be limited by these terms. These terms are used only to distinguish one device, element, component, or part from another device, element, component, or part.

[0099] Although this application has been described in conjunction with some embodiments, it is not intended to be limited to the specific forms set forth herein. Rather, the scope of this application is limited only by the appended claims. Additionally, although individual features may be included in different claims, these may be advantageously combined, and inclusion in different claims does not imply that such a combination of features is not feasible and / or advantageous. The order of features in the claims does not imply that the features must be in any particular order of their operation. Furthermore, in the claims, the word "comprising" does not exclude other elements, and the terms "a" or "an" do not exclude a plurality. Reference numerals in the claims are provided only as explicit examples and should not be construed as limiting the scope of the claims in any way.

Claims

1. An object recognition method, characterized in that, include: Receive multi-dimensional information of a target object, wherein the target object contains statistical features, the multi-dimensional information includes information in at least two dimensions, the target object includes newly joined merchants, and the statistical features are calculated based on the features of each associated object of the target object; Based on the multi-dimensional information, the associated objects of the target object are determined, and the associated objects include objects selected from the merchants who have already settled in the historical records; Obtain the basic features of the associated object, and construct the basic features into a variable-length sequence, wherein the basic features are features up to the entry time of the target object; The basic features and statistical features are input together into a pre-trained feature extraction model for recognition, and the recognition result is output. The pre-trained feature extraction model includes an aggregation unit and a recognition unit, including: using the aggregation unit to aggregate the basic features to obtain aggregated features, the aggregation unit having an LSTM network structure; inputting the aggregated features and statistical features together into the recognition unit for recognition, and outputting the recognition result. Based on the identification results, abnormal objects are identified, and those abnormal target objects are denied entry.

2. The method as described in claim 1, wherein determining the associated objects of the target object based on the multi-dimensional information includes: Query the database for objects that have the same information as the target object in at least one of the multiple dimensions, and use them as associated objects of the target object; The database contains at least information about historical objects across these multiple dimensions.

3. The method as described in claim 1, wherein determining the associated objects of the target object based on the multi-dimensional information includes: Obtain basic information about each basic object in the object database, wherein the object database contains different basic objects and the basic information corresponding to each basic object; Based on the multi-dimensional information and the basic information, calculate the vector distance between the target object and each of the basic objects; The base objects whose vector distance is greater than or equal to a preset threshold are designated as associated objects.

4. The method as described in claim 1, wherein the identification unit has an LR network structure.

5. The method of claim 4, wherein the feature extraction model is trained based on a sample set, the sample set including the basic features of each associated object of the historical object and the labels associated with the historical object.

6. The method of claim 5, wherein the associated objects of the historical object are determined based on the following steps: Receive multi-dimensional information about the historical object, wherein the multi-dimensional information includes information in at least two dimensions; Query the database for objects that have the same information as the historical object in at least one of the multiple dimensions, and use them as candidate associated objects of the historical object; The candidate associated objects are filtered based on the entry time of the historical objects, and the candidate associated objects before the entry time of the historical objects are retained as the associated objects of the historical objects. The database contains at least information about historical objects across these multiple dimensions.

7. The method of claim 1, wherein the feature extraction model is trained by the following steps: Obtain multi-dimensional historical information of the historical object and historical tags associated with the historical object, wherein, The historical object contains historical statistical features, and the historical multi-dimensional information includes historical information in at least two dimensions. Based on the aforementioned multi-dimensional historical information, determine the historical associated objects of the historical object; The historical basic features of the historical associated object are obtained, and the historical basic features and the historical tags associated with the historical object are imported into a preset recognition network, wherein the preset recognition network includes a preset aggregation unit and a preset recognition unit; The historical basic features are aggregated using a preset aggregation unit to obtain historical aggregated features; The historical aggregation features and the historical statistical features are input into a preset recognition unit for recognition, and the target recognition result is output. Based on the target identification results and the historical tags associated with the historical objects, the target loss is determined; If the target loss does not meet the preset conditions, the network parameters of the preset identification network are adjusted according to the target loss, and the target loss is updated based on the preset identification network after the network parameters are adjusted. If the target loss satisfies the preset conditions, the preset recognition network corresponding to the target loss satisfying the preset conditions is used as the feature extraction model.

8. The method according to any one of claims 1-3, wherein the basic features and the statistical features are input together into a pre-trained feature extraction model for recognition, and the output recognition result includes: The basic features and statistical features are input together into a pre-trained feature extraction model to calculate the verification score of the target object; The verification score is compared with a predetermined threshold to output the identification result of the target object.

9. An object recognition device, characterized in that, include: The receiving module is configured to receive multi-dimensional information of a target object, wherein the target object contains statistical features, the multi-dimensional information includes information in at least two dimensions, the target object includes newly joined merchants, and the statistical features are calculated based on the features of each associated object of the target object; The associated object determination module is configured to determine the associated objects of the target object based on the multi-dimensional information, wherein the associated objects include objects selected from the merchants that have already joined the platform as recorded in the historical records; The basic feature acquisition module is configured to acquire the basic features of the associated object and construct the basic features into a variable-length sequence, wherein the basic features are features up to the entry time of the target object. The identification module is configured to input the basic features and the statistical features into a pre-trained feature extraction model for identification and output the identification result. The pre-trained feature extraction model includes an aggregation unit and an identification unit, comprising: using the aggregation unit to aggregate the basic features to obtain aggregated features, the aggregation unit having an LSTM network structure; inputting the aggregated features and the statistical features into the identification unit for identification and outputting the identification result; determining abnormal objects based on the identification result and rejecting the entry of abnormal target objects.

10. A computer-readable storage medium storing computer-executable instructions that, when executed, perform the method as described in any one of claims 1-8.

11. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

Images