CN115018513A

CN115018513A - Data inspection method, device, equipment and storage medium

Info

Publication number: CN115018513A
Application number: CN202210585690.6A
Authority: CN
Inventors: 彭云胜; 刘成; 潘奋全; 梅敏君; 罗纳; 王旭
Original assignee: Ping An Bank Co Ltd
Current assignee: Ping An Bank Co Ltd
Priority date: 2022-05-27
Filing date: 2022-05-27
Publication date: 2022-09-06

Abstract

The invention relates to the technical field of internet and discloses a data polling method, a device, equipment and a storage medium. The method comprises the following steps: determining a polling task of a merchant to be polled and geographical position data corresponding to the polling task based on the inquired merchant transaction data and merchant state data; selecting a polling terminal for executing a polling task based on the geographic position data, and sending the polling task to the polling terminal; acquiring image data uploaded by a merchant according to the determined polling item, and inputting the image data into an OCR recognition model for recognition to obtain target merchant data; inputting target merchant data into a neural network of a graph for feature extraction to obtain a target feature vector of the data; and inputting the target characteristic vector into a prediction layer of the classification model to perform risk prediction, so as to obtain the inspection result of the merchant. The invention improves the timeliness of the inspection work of the merchant and solves the technical problems of low accuracy and efficiency of merchant information acquisition by periodically generating the inspection work order and finishing automatic dispatch.

Description

Data inspection method, device, equipment and storage medium

Technical Field

The invention relates to the technical field of internet, in particular to a data polling method, a device, equipment and a storage medium.

Background

Compared with traditional payment methods such as cash transaction and bank transfer, the network payment has obvious advantages in the aspects of efficiency, convenience and the like. In a network payment scenario, a payment mechanism should have necessary technical means to ensure timeliness, accuracy and safety of a payment service, and therefore, the payment mechanism needs to design a reasonable wind control rule to realize risk control of the payment service.

At present, with the increase of the number of merchants, a payment mechanism providing services for the merchants needs to perform risk control on the merchants, and requires that a traditional bank merchant check business patrol a company to run off-line to a merchant store and collect paper materials of the merchants. So as to verify whether the merchant is in normal operation and whether the certificate is complete. The authenticity of patrolling and examining personnel, the efficiency of patrolling and examining can't be guaranteed, the human cost is also higher relatively. Therefore, how to ensure the authenticity of the routing inspection and simultaneously improve the routing inspection efficiency becomes a technical problem to be solved by technicians in the field.

Disclosure of Invention

The invention mainly aims to generate the polling work order periodically through the server and finish automatic distribution, thereby improving the timeliness of the polling work of the merchant and solving the technical problems of low accuracy and efficiency of merchant information acquisition.

The invention provides a data inspection method in a first aspect, which comprises the following steps: after determining the merchant to be patrolled, inquiring merchant transaction data and merchant state data of the merchant to be patrolled from a preset database; determining a polling task of the merchant to be polled and geographical position data corresponding to the polling task based on the merchant transaction data and the merchant state data; selecting an inspection terminal for executing the inspection task based on the geographic position data, and sending the inspection task to the inspection terminal; determining a to-be-patrolled item preset in the patrolling task, acquiring image data uploaded by the to-be-patrolled merchant according to the patrolling item, and inputting the image data into a preset OCR recognition model for recognition to obtain target merchant data of the to-be-patrolled merchant; inputting the target merchant data into a preset graph neural network for feature extraction to obtain a target feature vector of the target merchant data; and inputting the target characteristic vector into a prediction layer of a preset classification model to perform risk prediction, and obtaining a polling result of the merchant to be polled.

Optionally, in a first implementation manner of the first aspect of the present invention, before determining, based on the merchant transaction data and the merchant state data, a patrol task of the merchant to be patrolled and geographic location data corresponding to the patrol task, the method further includes: carrying out blacklist check on the merchant to be checked; if the to-be-inspected merchant is not in a preset blacklist, generating an inspection task corresponding to the to-be-inspected merchant; and if the merchant to be inspected is positioned in the blacklist, outputting an error prompt.

Optionally, in a second implementation manner of the first aspect of the present invention, the selecting, based on the geographic location data, an inspection terminal that executes the inspection task, and sending the inspection task to the inspection terminal includes: the task information of the inspection task is issued to all inspection terminals meeting set inspection conditions; and if terminal response information which is returned by the inspection terminal and is related to the task information is received, determining the inspection terminal corresponding to the earliest received terminal response information as a target terminal, and sending the inspection task to the target terminal.

Optionally, in a third implementation manner of the first aspect of the present invention, after the selecting an inspection terminal that executes the inspection task based on the geographic location data, and sending the inspection task to the inspection terminal, the method further includes: acquiring preset category risk representation information of a merchant to be patrolled; determining a risk quantification label associated with the merchant to be patrolled based on the risk characterization information; and determining the to-be-patrolled item of the to-be-patrolled merchant according to the risk quantification label.

Optionally, in a fourth implementation manner of the first aspect of the present invention, the inputting the image data into a preset OCR recognition model for recognition to obtain the target merchant data of the merchant to be patrolled includes: dividing the image data to obtain a plurality of image blocks; denoising the image blocks to obtain denoised image blocks corresponding to each image block; inputting the image blocks into a preset OCR recognition model for recognition respectively to obtain a recognition result of each image block; and obtaining target merchant data of the merchant to be patrolled according to the identification result of each image block.

Optionally, in a fifth implementation manner of the first aspect of the present invention, before the graph neural network includes a sampling layer and a convolution layer, and the inputting the target merchant data into a preset graph neural network for feature extraction to obtain a target feature vector of the target merchant data, the method further includes: obtaining sample training data, and taking the sample training data as a training set, wherein the sample training set comprises risk quantification labels corresponding to sample merchants; generating a sample adjacency matrix and a sample transaction attribute characteristic matrix of the sample heterogeneous relation network diagram based on the sample training data; inputting the sample adjacency matrix and the sample transaction attribute feature matrix into an initial graph neural network for feature extraction to obtain an initial target feature vector of the sample merchant; determining a loss function based on the risk identification category corresponding to the initial target feature vector and the cross entropy of the risk quantization label; determining a descent gradient based on the loss function; and updating the network parameters of the initial graph neural network based on the descending gradient and a preset learning rate to obtain an updated graph neural network.

Optionally, in a sixth implementation manner of the first aspect of the present invention, after predicting whether the merchant to be patrolled has a risk according to the target feature vector and obtaining a patrol result of the merchant to be patrolled, the method further includes: acquiring verification information corresponding to the inspection result; verifying the authenticity of the verification information, and verifying the verification information after the authenticity of the verification information passes verification; and when the verification of the verification information passes, reporting the inspection result to a background management system.

A second aspect of the present invention provides a data inspection apparatus, including: the query module is used for querying merchant transaction data and merchant state data of the merchant to be patrolled from a preset database after the merchant to be patrolled is determined; the first determining module is used for determining a polling task of the merchant to be polled and geographical position data corresponding to the polling task based on the merchant transaction data and the merchant state data; the sending module is used for selecting an inspection terminal for executing the inspection task based on the geographic position data and sending the inspection task to the inspection terminal; the identification module is used for determining a preset item to be patrolled in the patrol task, acquiring image data uploaded by the merchant to be patrolled according to the patrol item, and inputting the image data into a preset OCR identification model for identification to obtain target merchant data of the merchant to be patrolled; the first feature extraction module is used for inputting the target merchant data into a preset graph neural network for feature extraction to obtain a target feature vector of the target merchant data; and the prediction module is used for inputting the target characteristic vector into a prediction layer of a preset classification model to carry out risk prediction so as to obtain the inspection result of the merchant to be inspected.

Optionally, in a first implementation manner of the second aspect of the present invention, the data inspection device further includes: the checking module is used for carrying out blacklist checking on the merchant to be checked; the first generation module is used for generating an inspection task corresponding to the merchant to be inspected if the merchant to be inspected is not in a preset blacklist; and if the merchant to be inspected is positioned in the blacklist, outputting an error prompt.

Optionally, in a second implementation manner of the second aspect of the present invention, the sending module is specifically configured to: the task information of the inspection task is issued to all inspection terminals meeting set inspection conditions; and if terminal response information which is returned by the inspection terminal and is related to the task information is received, determining the inspection terminal corresponding to the earliest received terminal response information as a target terminal, and sending the inspection task to the target terminal.

Optionally, in a third implementation manner of the second aspect of the present invention, the data inspection device further includes: the first acquisition module is used for acquiring preset category risk representation information of a merchant to be patrolled; the first determining module is used for determining a risk quantification label associated with the merchant to be patrolled based on the risk characterization information; and determining the to-be-patrolled items of the to-be-patrolled merchant according to the risk quantification labels.

Optionally, in a fourth implementation manner of the second aspect of the present invention, the identification module includes: the segmentation unit is used for segmenting the image data to obtain a plurality of image blocks; the denoising unit is used for denoising the plurality of image blocks to obtain a denoised image block corresponding to each image block; the recognition unit is used for inputting the image blocks into a preset OCR recognition model for recognition respectively to obtain a recognition result of each image block; and obtaining target merchant data of the merchant to be patrolled according to the identification result of each image block.

Optionally, in a fifth implementation manner of the second aspect of the present invention, the data inspection device further includes: the second generation module is used for acquiring sample training data and taking the sample training data as a training set, wherein the sample training set comprises risk quantification labels corresponding to sample merchants; generating a sample adjacency matrix and a sample transaction attribute characteristic matrix of the sample heterogeneous relation network diagram based on the sample training data; the second feature extraction module is used for inputting the sample adjacency matrix and the sample transaction attribute feature matrix into an initial graph neural network for feature extraction to obtain an initial target feature vector of the sample merchant; a third determining module, configured to determine a loss function based on a risk identification category corresponding to the initial target feature vector and a cross entropy of the risk quantization label; determining a descent gradient based on the loss function; and the updating module is used for updating the network parameters of the initial graph neural network based on the descending gradient and a preset learning rate to obtain an updated graph neural network.

Optionally, in a sixth implementation manner of the second aspect of the present invention, the data inspection device further includes: the second acquisition module is used for acquiring verification information corresponding to the inspection result; the verification module is used for verifying the authenticity of the verification information, and verifying the verification information after the authenticity of the verification information passes verification; and when the verification of the verification information passes, reporting the inspection result to a background management system.

A third aspect of the present invention provides data inspection equipment, including: a memory having instructions stored therein and at least one processor, the memory and the at least one processor interconnected by a line;

the at least one processor invokes the instructions in the memory to cause the data inspection device to perform the steps of the data inspection method described above.

A fourth aspect of the present invention provides a computer-readable storage medium having stored therein instructions, which, when run on a computer, cause the computer to perform the steps of the data inspection method described above.

In the technical scheme provided by the invention, after a merchant to be inspected is determined, merchant transaction data and merchant state data of the merchant to be inspected are inquired from a preset database; determining a polling task of a merchant to be polled and geographical position data corresponding to the polling task based on merchant transaction data and merchant state data; sending the inspection task to a corresponding inspection terminal based on the geographic position data corresponding to the inspection task; determining a preset item to be patrolled in the patrolling task, acquiring image data uploaded by a merchant to be patrolled according to the patrolling item, inputting the image data into a preset OCR recognition model for recognition, and obtaining target merchant data of the merchant to be patrolled; inputting the target merchant data into a preset graph neural network for feature extraction to obtain a target feature vector of the target merchant data; and inputting the target characteristic vector into a prediction layer of a preset classification model for risk prediction, predicting whether the merchant to be inspected has risks according to the target characteristic vector, and obtaining an inspection result of the merchant to be inspected. The invention realizes the whole-line coverage of the merchant inspection through the whole-line online management of the merchant inspection process. The technical problem of low merchant routing inspection efficiency is solved.

Drawings

FIG. 1 is a schematic diagram of a first embodiment of a data inspection method provided by the present invention;

FIG. 2 is a diagram of a second embodiment of a data inspection method provided by the present invention;

FIG. 3 is a diagram of a data inspection method according to a third embodiment of the present invention;

FIG. 4 is a diagram of a fourth embodiment of a data inspection method according to the present invention;

FIG. 5 is a diagram of a fifth embodiment of a data inspection method according to the present invention;

FIG. 6 is a schematic diagram of a first embodiment of a data inspection device provided by the present invention;

FIG. 7 is a schematic diagram of a second embodiment of a data inspection device provided in accordance with the present invention;

fig. 8 is a schematic diagram of an embodiment of the data inspection equipment provided by the invention.

Detailed Description

According to the data inspection method, the data inspection device, the data inspection equipment and the data inspection storage medium, after a merchant to be inspected is determined, merchant transaction data and merchant state data of the merchant to be inspected are inquired from a preset database; determining a polling task of a merchant to be polled and geographical position data corresponding to the polling task based on merchant transaction data and merchant state data; sending the inspection task to a corresponding inspection terminal based on the geographic position data corresponding to the inspection task; determining a preset item to be patrolled in the patrolling task, acquiring image data uploaded by a merchant to be patrolled according to the patrolling item, inputting the image data into a preset OCR recognition model for recognition, and obtaining target merchant data of the merchant to be patrolled; inputting the target merchant data into a preset graph neural network for feature extraction to obtain a target feature vector of the target merchant data; and inputting the target characteristic vector into a prediction layer of a preset classification model to carry out risk prediction, and predicting whether the merchant to be inspected has risks according to the target characteristic vector to obtain an inspection result of the merchant to be inspected. The invention realizes the whole-line coverage of the merchant inspection through the whole-line online management of the merchant inspection process. The technical problem of low merchant routing inspection efficiency is solved.

The terms "first," "second," "third," "fourth," and the like in the description and in the claims, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It will be appreciated that the data so used may be interchanged under appropriate circumstances such that the embodiments described herein may be implemented in other sequences than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," or "having," and any variations thereof, are intended to cover non-exclusive inclusions, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

For convenience of understanding, a specific flow of the embodiment of the present invention is described below, and referring to fig. 1, a first embodiment of the data inspection method in the embodiment of the present invention includes:

101. after determining the merchant to be patrolled, inquiring merchant transaction data and merchant state data of the merchant to be patrolled from a preset database;

in this embodiment, the data before and after the transaction of the merchant to be inspected and the data when the transaction occurs are mainly collected, so as to subsequently judge the state of the transaction of the merchant to be inspected.

Optionally, data before and after the transaction of the to-be-inspected merchant and data when the transaction occurs can be acquired when the merchant conducts transaction each time, and the merchant data acquired in advance is called to make judgment when the merchant needs to be identified in the follow-up process.

For example, the merchant status data comprises at least one of merchant health status data and merchant location status data; the merchant health status data may include at least one of step number information, heart rate information, and body temperature information. For example, the merchant positioning status data may be Location Based Service (LBS) Based information.

102. Determining a polling task of a merchant to be polled and geographical position data corresponding to the polling task based on merchant transaction data and merchant state data;

in the embodiment, the online patrol inspection model of the order receiving merchant adopts online management of the patrol inspection work order, and the branch can formulate information such as branch patrol inspection frequency, patrol inspection timeliness and patrol inspection mechanisms according to self conditions to dynamically generate annual (quarterly/monthly) patrol inspection orders; the temporary routing inspection work order can be produced according to special branch scenes; and analyzing the transaction and the state dynamic of the merchant by combining the wind control system with the big data to generate a risk inspection work order.

103. Selecting an inspection terminal for executing an inspection task based on the geographic position data, and sending the inspection task to the inspection terminal;

in this embodiment, the process of the merchant polling is online and remotely performed, first, a polling task corresponding to the merchant to be polled is generated at the service end, and the polling task is distributed to a polling terminal, where the polling terminal is one of at least two polling clients, and the polling client is a preset terminal for executing the polling task and is held by a polling person who visits. The determination mode of the inspection terminal may be randomly determined among all the inspection clients, and a certain inspection client may be designated as the inspection terminal, or other determination modes may be applied, which is not limited in the embodiment of the present invention.

104. Determining a preset item to be patrolled in the patrolling task, acquiring image data uploaded by a merchant to be patrolled according to the patrolling item, inputting the image data into a preset OCR recognition model for recognition, and obtaining target merchant data of the merchant to be patrolled;

in the embodiment of the invention, for convenience of description, the to-be-inspected items include legal person pictures, unified social credit codes, business licenses, enterprise office addresses, the number of staff, enterprise scores, annual revenue amounts and annual revenue gains. The registration data under the project to be inspected can be added manually at the server, the registration data can be identified and added in an application form (full-time investigation application form) provided by a merchant to be inspected to a bank, and the name of the merchant to be inspected and the project to be inspected can be used as keywords together to search in a bank database, the Internet or a database of an industrial and commercial department to obtain the registration data. It should be noted that the step is not limited to adding the registration data under each item to be inspected.

In the embodiment, the patrol personnel can check the patrol work order after logging in by using the patrol small program and claim the patrol task; the inspection personnel need to go to the store of the commercial tenant in the inspection period and provide image data and online data information according to the requirement of the inspection worksheet. And acquiring the image data uploaded by the merchant to be patrolled according to the patrolling item, and inputting the image data into a preset OCR recognition model for recognition to obtain target merchant data of the merchant to be patrolled.

In this embodiment, the OCR (Optical Character Recognition) refers to a process of analyzing, recognizing and processing an image file of text data to obtain text and layout information. Generally, image information is acquired and stored in an image file by a scanner, a camera, electronic facsimile software, or the like, and then OCR software reads, analyzes the image file and extracts a character string therein by character recognition. Among them, in the OCR technology, the image preprocessing is usually to correct the imaging problem of the image. After an input text enters a computer through a scanner, because the thickness, the smoothness and the printing quality of paper can cause character distortion and generate interferences such as broken pen, adhesion, stain and the like, before character recognition, a character image with noise needs to be processed.

Specifically, text recognition is to recognize text contents based on text detection, and mainly to recognize what each character is. For a character image, extracting features, and discarding the features to a classifier, the classifier classifies the characters and tells you which character the features should be recognized. The design method of the classifier generally comprises the following steps: template matching, discriminant function, neural network classification, rule-based reasoning, etc. Before the actual recognition, the classifier is often trained, which is a process of supervised learning. There are also many mature classifiers, including SVM, CNN, etc. And converting the text information in the image into text information.

105. Inputting the target merchant data into a preset graph neural network for feature extraction to obtain a target feature vector of the target merchant data;

in this embodiment, the Graph neural network may include Graph SAmple and aggregation, and Graph is an inductive learning Graph neural network that updates target node feature information by sampling and feature aggregation of neighbor nodes. In a specific embodiment, the target feature information may be a target feature vector, a graph neural network is used to sample neighbor nodes of the target merchant node, and aggregation processing is performed on transaction attribute feature vectors of the neighbor nodes obtained by sampling, so as to obtain the target feature vector.

106. And inputting the target characteristic vector into a prediction layer of a preset classification model to perform risk prediction, and obtaining a polling result of the merchant to be polled.

In this embodiment, the classification model may include, but is not limited to, a neural network and a decision tree, and the risk identification category of the classification model may include: the merchant to be patrolled is a risk merchant or the merchant to be patrolled is a non-risk merchant. In a specific embodiment, the target characteristic information is input into the classification model to carry out merchant risk identification, and a first identification probability that the merchant to be inspected is identified as a risk merchant and a second identification probability that the merchant to be inspected is identified as a non-risk merchant are obtained. In practical applications, the sum of the first recognition probability and the second recognition probability may be 1.

Specifically, the first recognition probability is used as the risk recognition index data; when the first probability is larger than a preset threshold value, the risk identification category of the to-be-patrolled merchant node corresponding to the to-be-patrolled merchant is a risk merchant; and when the first probability is smaller than a preset threshold value, the risk identification category of the to-be-patrolled merchant node corresponding to the to-be-patrolled merchant is a non-risk merchant. In practical applications, the preset threshold may be set in combination with accuracy of merchant risk identification in practical applications.

In this embodiment, the target feature vector output by the aggregation layer is input into a prediction layer of a preset classification model, and the prediction layer predicts whether the commercial tenant has an illegal platform risk according to the risk feature vector.

Due to the fact that the time sequence information of the multi-modal risk data is considered when the commercial tenant carries out the illegal platform risk on the prediction line, the time sequence information in the multi-modal risk data sequence data is subjected to two-way fusion, and information among all the modes can be utilized and fused more fully. Thus avoiding the loss of risk information. Therefore, the accuracy of identifying whether the online merchants have illegal platform risks can be improved, so that the illegal platform risks of the online merchants can be identified to a greater extent, and further, the routing inspection of illegal transactions, illegal investment and financing and sale prohibition behaviors of the online merchants can be better completed, so that the supervision of the online merchants is optimized.

In the embodiment of the invention, after the merchant to be patrolled is determined, the merchant transaction data and the merchant state data of the merchant to be patrolled are inquired from the preset database; determining a polling task of a merchant to be polled and geographical position data corresponding to the polling task based on merchant transaction data and merchant state data; sending the inspection task to a corresponding inspection terminal based on the geographic position data corresponding to the inspection task; determining a preset item to be patrolled in the patrolling task, acquiring image data uploaded by a merchant to be patrolled according to the patrolling item, inputting the image data into a preset OCR recognition model for recognition, and obtaining target merchant data of the merchant to be patrolled; inputting the target merchant data into a preset graph neural network for feature extraction to obtain a target feature vector of the target merchant data; and inputting the target characteristic vector into a prediction layer of a preset classification model for risk prediction, predicting whether the merchant to be inspected has risks according to the target characteristic vector, and obtaining an inspection result of the merchant to be inspected. The invention realizes the whole-line coverage of the merchant inspection through the whole-line online management of the merchant inspection process. The technical problem of low merchant routing inspection efficiency is solved.

Referring to fig. 2, a second embodiment of the data inspection method according to the embodiment of the present invention includes:

201. after determining the merchant to be patrolled, inquiring merchant transaction data and merchant state data of the merchant to be patrolled from a preset database;

202. carrying out blacklist check on the merchant to be checked;

in this embodiment, before allocating the inspection task, the server may perform blacklist inspection on the to-be-inspected merchant, and specifically, may search the name of the to-be-inspected merchant in a preset blacklist as a search condition, where the blacklist may be a blacklist inside a bank, and may also be a credit-losing enterprise list or an operation exception list provided by a national enterprise information public system, and the like.

203. If the merchant to be inspected is not in the preset blacklist, generating an inspection task corresponding to the merchant to be inspected;

in this embodiment, if the name of the merchant to be inspected is not in the blacklist, it is verified that the inspection task corresponding to the merchant to be inspected is valid, so that the inspection task corresponding to the merchant to be inspected is generated at the server.

204. If the merchant to be inspected is in the blacklist, outputting an error prompt;

in this embodiment, if the name of the merchant to be inspected is in the blacklist, an error prompt is output for further inspection by relevant personnel in order to prevent waste of human resources caused by executing an invalid inspection task.

205. Determining a polling task of a merchant to be polled and geographical position data corresponding to the polling task based on merchant transaction data and merchant state data;

206. based on the geographic position data corresponding to the inspection task, task information of the inspection task is issued to all inspection terminals meeting set inspection conditions;

in the embodiment, the task information of the inspection task is issued to all inspection terminals meeting the set inspection conditions, different inspection grades can be specifically set for different inspection terminals, and the inspection conditions are set to specific grades; and a client state can be set for the inspection terminal in advance, the client state comprises a non-inspection state and an inspection state, and the inspection condition is set to be the inspection state of the client state. The task information includes, but is not limited to, the business name of the merchant to be patrolled and the geographic location of the merchant to be patrolled.

207. If terminal response information which is returned by the inspection terminal and is related to the task information is received, the inspection terminal corresponding to the earliest received terminal response information is determined as a target terminal, and the inspection task is sent to the target terminal;

in this embodiment, after the task information is issued to the inspection terminal satisfying the inspection condition, the inspection staff having the inspection terminal can determine whether to return the client response information to the server side through the inspection terminal according to the task information. And if the server receives the client response information of the polling terminal about the task information, determining the polling terminal corresponding to the client response information received earliest as the polling terminal, and distributing the polling task to the polling terminal.

208. Acquiring preset category risk representation information of a merchant to be patrolled;

in this embodiment, risk characterization information of a merchant to be patrolled is obtained, and the risk characterization information includes at least one of the following: merchant subject information, financial subject information, company product information, usage and priority, and policy support content. The merchant main information may include information such as a merchant registration area, merchant operation duration and the like, the financial subject information may include information such as merchant registration capital, merchant operation flow and the like, the company product information may include information such as a company product type, a product scale and the like, the usage and priority may include information such as a company product usage, a type of industry where the company product is located and the like, and the policy support content may include policy content such as national credit relaxation, tightening, support and the like.

The risk characterization information is used for carrying out risk classification and risk quantitative evaluation on the merchant to be patrolled preliminarily, and the risk characterization information can be part of items to be patrolled of the network credit business. The method for acquiring the risk characterization information can be specifically transmitted to the inspection server by the merchant to be inspected through an online data transmission mode, and can also be acquired from online data sources such as a national merchant information display system and the like by the inspection server.

209. Determining a risk quantification label associated with the merchant to be patrolled based on the risk characterization information;

in this embodiment, based on the acquired risk characterization information, the merchants to be patrolled are subjected to labeling classification, so that risk classification and risk quantitative evaluation of the merchants to be patrolled are realized, and at least one risk quantitative label for characterizing credit risk information of the merchants to be patrolled is obtained.

210. Determining a to-be-patrolled item of a to-be-patrolled merchant according to the risk quantification label;

in this embodiment, according to the determined at least one risk quantification tag, a to-be-patrolled item for performing online patrolling on the to-be-patrolled merchant is determined. The to-be-inspected item is an inspection information set used for loan risk assessment and credit line determination, and may specifically include multi-dimensional data of the to-be-inspected merchant, for example, may include merchant main information, asset liability information, financial subject information, merchant business information, upstream and downstream merchant information, and the like of the to-be-inspected merchant.

Specifically, the merchant main information may include, for example, judicial data, business data, merchant shareholder data, merchant human resource data and other information of the merchant to be inspected, the liability information may include, for example, merchant asset information, merchant liability information, loss rate asset characteristics and other information of the merchant to be inspected, the financial subject information may include, for example, information of a merchant asset table, a merchant profit table, a merchant liability table and other information, the merchant business information may include, for example, merchant order data, invoice data, running data, inventory amount and other information, and the upstream and downstream merchant information may include, for example, merchant information, order data, running amount and other information having an upstream and downstream business relationship with the merchant to be inspected.

211. Determining a preset item to be patrolled in the patrolling task, acquiring image data uploaded by a merchant to be patrolled according to the patrolling item, inputting the image data into a preset OCR recognition model for recognition, and obtaining target merchant data of the merchant to be patrolled;

212. inputting the target merchant data into a preset graph neural network for feature extraction to obtain a target feature vector of the target merchant data;

213. and inputting the target characteristic vector into a prediction layer of a preset classification model to carry out risk prediction, and obtaining a polling result of the merchant to be polled.

The steps 201-.

Referring to fig. 3, a third embodiment of the data inspection method according to the embodiment of the present invention includes:

301. after determining the merchant to be patrolled, inquiring merchant transaction data and merchant state data of the merchant to be patrolled from a preset database;

302. determining a polling task of a merchant to be polled and geographical position data corresponding to the polling task based on merchant transaction data and merchant state data;

303. sending the inspection task to a corresponding inspection terminal based on the geographic position data corresponding to the inspection task;

304. dividing image data to obtain a plurality of image blocks;

in this embodiment, online data information such as image data uploaded by a merchant is obtained, and the cutting mode of the image data may be transverse cutting, longitudinal cutting, or transverse and longitudinal combined cutting, which is not limited in this embodiment of the application. The cutting mode of the electronic equipment for the image data can comprise a merchant intervention cutting mode and a non-merchant intervention cutting mode, and in the merchant intervention cutting mode, the electronic equipment can display a cutting line on the image data by detecting a merchant instruction and cut the image data by taking the displayed cutting line as a basis; in the non-merchant intervention cutting mode, the electronic device may cut the image data according to a preset mode, where the preset mode may be an equal-division cutting mode or a non-equal-division cutting mode, and the embodiment of the present application is not limited.

Based on the above description, the cutting the image data to obtain a plurality of image blocks may include: determining a current cutting mode of the electronic equipment; when the current cutting mode is a merchant intervention cutting mode, outputting a cutting line selection interface for a merchant to select line characteristics of a cutting line, wherein the line characteristics can comprise line width, color and direction; displaying the cutting line on the image data according to the cutting track input by the merchant and the line characteristics of the cutting line determined by the merchant; cutting the image data according to the displayed cutting lines to obtain a plurality of image blocks; and when the current cutting mode is a non-merchant intervention cutting mode, equally cutting the image data for preset times to obtain a plurality of image blocks. By implementing the method, flexible cutting of image data can be realized.

305. Denoising the plurality of image blocks to obtain denoised image blocks corresponding to each image block;

in this embodiment, the algorithm for performing noise reduction processing on the plurality of image blocks may be an airspace pixel feature denoising algorithm, a transform domain denoising algorithm, or a BM3D denoising algorithm, which is not limited in this embodiment. The spatial domain pixel feature denoising algorithm is specific to random noise. Then what is random noise? Random noise is a signal that exhibits an indeterminate change in level or level compared to the true signal of the image, and the sum of all random noise signals is 0. The method based on spatial pixel characteristics is a method for acquiring a new central pixel value by analyzing the direct relation between the central pixel and the rest of adjacent pixels in a gray scale space in a window with a certain size.

In the present embodiment, noise is an important cause of image disturbance. An image may have various noises in practical application, and these noises may be generated in transmission, quantization, etc. The noise and signal relationship can be divided into three forms (f (x, y) represents a given original image, g (x, y) represents an image signal, and n (x, y) represents noise.)

1) Additive noise, which is independent of the input image signal, and the noisy image can be expressed as f (x, y) ═ g (x, y) + n (x, y), and channel noise and noise generated when the camera of the photoconductive camera scans the image are the same.

2) Multiplicative noise, which is related to the image signal, and noisy images can be represented as f (x, y) g (x, y) + n (x, y) g (x, y), noise when the flying spot scanner is scanning the image, coherent noise in the tv image, and grain noise in the film.

3) Quantization noise, which is irrelevant to the input image signal, is generated by the quantization error existing in the quantization process and then reflected to the receiving end.

Specifically, the salt-pepper noise is a light-dark dot noise between black and white generated by an image sensor, a transmission channel, a decoding process, and the like. Salt and pepper noise tends to be lifted by the image cut bow. The most common algorithm for removing impulse interference and salt and pepper noise is median filtering. The road surface image belongs to a structured light image, white noise and partial salt and pepper noise are eliminated by using a threshold segmentation method in an area segmentation technology, while the white noise and the salt and pepper noise cannot be filtered by using median filtering, and as a filtering template changes the real gray distribution of pixels in light bars when roaming in the image, negative influence is brought to the subsequent gravity center method thinning process.

306. Respectively inputting the image blocks into a preset OCR recognition model for recognition to obtain a recognition result of each image block;

in this embodiment, the main training process of the OCR recognition model includes obtaining a first type of image sample with label data; performing model training on a preset OCR (optical character recognition) model by using the first type of image sample to obtain an initial OCR model; performing OCR recognition on the second type image sample of the label-free data by using the initial OCR recognition model; generating label data of the second type image sample according to an OCR recognition result, and labeling the second type image sample according to the generated label data; and performing model training on the initial OCR recognition model by adopting the first type of image sample and the second type of image sample marked by the label to obtain a final OCR recognition model.

Specifically, the image recognition area refers to an image area containing information to be recognized, that is, a target area for performing OCR recognition on the first type image sample. The business data refers to data recorded in the image recognition area, and the data are target data for performing OCR recognition on the first type image sample. After the first-class image samples are set, the first-class image samples are required to be used for carrying out first training on the preset OCR recognition model, so that the preset OCR recognition model has certain OCR recognition capability. It should be noted that, in the embodiment of the present invention, a model structure of an OCR recognition model that is conventional in the OCR technical field may be adopted to construct the preset OCR recognition model.

The trained initial OCR recognition model can recognize the second type of image samples to be recognized to a certain degree, and the initial OCR recognition model is used for recognizing the second type of samples without label data to obtain the recognition result of the initial OCR recognition model. And generating label data of the second type of image samples according to the OCR recognition result, and labeling the second type of image samples according to the generated label data.

And training the initial OCR recognition model through the image sample with the artificially marked real information and the image sample of the recognition result of the initial OCR recognition model. The recognition result of the initial OCR recognition model may include the recognized location, category, specific content, etc., for example, the category of the real sample is the identification card, and the recognition result is the bank card, which is an obvious category recognition error, and for example, the location of the recognition result is the birth date column of the identification card, and the real location is also the birth date column of the identification card, i.e., the recognition location is correct.

307. Acquiring the data of the merchants to be patrolled according to the identification result of each image block;

in this embodiment, the electronic device may further record a location identifier of each image block in the image data, and the obtaining the identification result of the image data according to the identification result of each image block may include: and obtaining the identification result of the image data according to the identification result of each image block and the corresponding position identification, and obtaining the data of the merchant to be patrolled according to the identification result of each image block.

308. Inputting the target merchant data into a preset graph neural network for feature extraction to obtain a target feature vector of the target merchant data;

309. and inputting the target characteristic vector into a prediction layer of a preset classification model to perform risk prediction, and obtaining a polling result of the merchant to be polled.

The steps 301-.

Referring to fig. 4, a fourth embodiment of the data inspection method according to the embodiment of the present invention includes:

401. after determining the merchant to be inspected, inquiring merchant transaction data and merchant state data of the merchant to be inspected from a preset database;

402. determining a polling task of a merchant to be polled and geographical position data corresponding to the polling task based on merchant transaction data and merchant state data;

403. sending the inspection task to a corresponding inspection terminal based on the geographic position data corresponding to the inspection task;

404. determining a preset item to be patrolled in the patrolling task, acquiring image data uploaded by a merchant to be patrolled according to the patrolling item, inputting the image data into a preset OCR recognition model for recognition, and obtaining target merchant data of the merchant to be patrolled;

405. acquiring sample training data, and taking the sample training data as a training set, wherein the sample training set comprises risk quantification labels corresponding to sample merchants;

in this embodiment, before performing training and learning of the graph neural network, sample training data may be determined, and specifically, in this embodiment, the sample heterogeneous relationship network graph and the risk quantification label corresponding to the sample merchant node in the sample heterogeneous relationship network graph may be obtained as training data.

In this embodiment, sample transaction data of a sample transaction object corresponding to a sample active merchant may be obtained from a payment transaction log; and generating a sample heterogeneous relation network diagram of the sample active merchant based on the sample transaction data of the corresponding sample transaction object. The sample active merchant may be extracted by combining transaction conditions of a business system in practical application, and in an alternative embodiment, the sample active merchant may be a merchant having a transaction record within three years. Specifically, the transaction data is based on sample transaction data of the corresponding sample transaction object.

406. Generating a sample adjacency matrix and a sample transaction attribute characteristic matrix of the sample heterogeneous relation network diagram based on the sample training data;

in this embodiment, the transaction statistical information may include, but is not limited to, transaction times and transaction amounts of the corresponding node in multiple time windows, and the multiple time windows may be set in combination with transaction payment traffic and merchant risk identification requirements in practical applications. Specifically, the transaction statistical information corresponding to each transaction object, that is, the transaction statistical information corresponding to each node, may be obtained from the transaction data of the transaction object corresponding to the target merchant.

Specifically, the node transaction attribute feature matrix may be a matrix used for characterizing a plurality of transaction attribute features of each node in the heterogeneous relationship network graph, where the plurality of transaction attribute features are respectively used for characterizing transaction statistics of the corresponding node under a plurality of time windows, and the transaction statistics may include transaction times and transaction amount.

407. Inputting the sample adjacency matrix and the sample transaction attribute feature matrix into an initial graph neural network for feature extraction to obtain an initial target feature vector of a sample merchant;

in this embodiment, the node mapping file is generated based on the mapping relationship between the node identification information and the target sequence number. And generating an adjacency list of the heterogeneous relationship network graph based on the node mapping file and the adjacency information. The adjacency list can be a linked list structure for storing adjacency information of each node, feature extraction is carried out on the adjacency relation of the adjacency list, and a corresponding adjacency matrix is obtained based on the weight of edges between the nodes. Specifically, an adjacency matrix is generated based on adjacency information of the nodes, and feature extraction of node adjacency relation can be realized.

408. Determining a loss function based on the risk identification category corresponding to the initial target feature vector and the cross entropy of the risk quantization label, and determining a descending gradient based on the loss function;

in this embodiment, the initial target feature information may be input to the classification model to perform merchant risk identification, so as to obtain a risk identification category of the sample merchant node, calculate a cross entropy between the risk identification category of the sample merchant node and a risk category label, use the cross entropy as a loss function, and calculate the loss function based on a chain derivation rule, so as to obtain a descent gradient.

In this embodiment, the chain derivation algorithm, also called chain algorithm, is a derivation algorithm in calculus, is used to calculate a derivative of a complex function, and is a common method in derivation operation of calculus. The derivative of the complex function will be the product of the derivatives of the finite functions at the corresponding points that make up the complex, and will loop around the loop as a chain, hence the chain rule.

409. Updating network parameters of the initial graph neural network based on the descending gradient and a preset learning rate to obtain an updated graph neural network;

in this embodiment, the graph neural network is trained based on the Mini-Batch gradient descent Method (MBGD).

Specifically, performing loss function calculation on the initial graph neural network, and comparing a currently calculated loss function with a historical loss function stored in a preset database; when the difference value between the current calculated loss function and the historical loss function is smaller than or equal to a preset value, judging that the combined initial graph neural network is in a convergence state; and when the difference value between the current calculated loss function and the historical loss function is larger than the preset value, judging that the combined initial graph neural network is not in a convergence state, and updating the historical loss function stored in the preset database by using the current calculated loss function.

And when the initial graph neural network is not converged, sending the combined initial graph neural network to a client, further training the combined initial graph neural network through the characteristic vector in the client to obtain an updated initial graph neural network, and taking the updated initial graph neural network as a target graph neural network.

And if the convergence is successful, updating the model loss function of the graph neural network of the local hospital by using the converged loss function, and locally finishing the construction of the graph neural network by using the updated model loss function to obtain the graph neural network.

410. Inputting the target merchant data into a preset graph neural network for feature extraction to obtain a target feature vector of the target merchant data;

411. and inputting the target characteristic vector into a prediction layer of a preset classification model to perform risk prediction, and obtaining a polling result of the merchant to be polled.

The steps 401-.

Referring to fig. 5, a fifth embodiment of the data inspection method according to the embodiment of the present invention includes:

501. after determining the merchant to be patrolled, inquiring merchant transaction data and merchant state data of the merchant to be patrolled from a preset database;

502. determining a polling task of a merchant to be polled and geographical position data corresponding to the polling task based on merchant transaction data and merchant state data;

503. sending the inspection task to a corresponding inspection terminal based on the geographic position data corresponding to the inspection task;

504. determining a preset item to be patrolled in the patrolling task, acquiring image data uploaded by a merchant to be patrolled according to the patrolling item, inputting the image data into a preset OCR recognition model for recognition, and obtaining target merchant data of the merchant to be patrolled;

505. inputting the target merchant data into a preset graph neural network for feature extraction to obtain a target feature vector of the target merchant data;

506. inputting the target characteristic vector into a prediction layer of a preset classification model to perform risk prediction to obtain a polling result of a merchant to be polled;

507. acquiring verification information corresponding to the inspection result;

in this embodiment, after the terminal device receives the polling task, the polling task is displayed on a display interface of the terminal device. And (4) related workers perform inspection according to the inspection task displayed on the display interface, and operate on the display interface to finally obtain an inspection result. And confirming by at least two workers based on the inspection result to obtain at least two pieces of verification information. Verifying the authenticity of at least two pieces of verification information, exemplarily, after at least two workers confirm the inspection result and submit the inspection confirmed verification information through signing, face recognition, account authorization and other modes, the terminal device or the background management system verifies the signing, face recognition, account authorization and the like to verify whether the information is authentic and is matched with the corresponding workers.

508. Verifying the authenticity of the verification information, and verifying the verification information after the authenticity of the verification information passes verification;

in this embodiment, after the authenticity of the verification information is verified, the at least two pieces of verification information are verified under the condition that the at least two pieces of verification information are both true, whether the at least two pieces of verification information are generated for one inspection task is checked, if yes, the at least two pieces of verification information are considered to be passed through verification, and the inspection result is reported to a background management system under the condition that the at least two pieces of verification information are determined to be based on the inspection task, so that the whole process of executing and processing the inspection task is completed.

509. And reporting the inspection result to a background management system after the verification of the verification information is passed.

In the embodiment, the inspection result is generated in response to the operation on the inspection task; acquiring at least two pieces of verification information corresponding to the inspection result; verifying the authenticity of the verification information, and verifying at least two pieces of verification information after the authenticity of the verification information passes verification; and when the verification of at least two pieces of verification information passes, reporting the inspection result to a background management system. Therefore, the steps from the receiving of the routing inspection task, the execution of the routing inspection task to the generation of the routing inspection result can be completed, at least two pieces of verification information corresponding to the routing inspection result are further verified, the routing inspection result is reported to the background management system, and the execution and processing processes of the whole routing inspection task are completed.

Step 501-506 in the present embodiment is similar to step 101-106 in the first embodiment, and will not be described herein again.

In the above description of the data inspection method in the embodiment of the present invention, referring to fig. 6, the data inspection apparatus in the embodiment of the present invention is described below, where the first embodiment of the data inspection apparatus in the embodiment of the present invention includes:

the query module 601 is configured to query, from a preset database, merchant transaction data and merchant state data of a merchant to be patrolled after the merchant to be patrolled is determined;

a first determining module 602, configured to determine, based on the merchant transaction data and the merchant state data, an inspection task of the merchant to be inspected and geographic location data corresponding to the inspection task;

a sending module 603, configured to select, based on the geographic position data, an inspection terminal that executes the inspection task, and send the inspection task to the inspection terminal;

the identification module 604 is configured to determine a preset item to be inspected in the inspection task, acquire image data uploaded by the merchant to be inspected according to the inspection item, and input the image data into a preset OCR recognition model for identification, so as to obtain target merchant data of the merchant to be inspected;

a first feature extraction module 605, configured to input the target merchant data into a preset graph neural network for feature extraction, so as to obtain a target feature vector of the target merchant data;

and the prediction module 606 is configured to input the target feature vector into a prediction layer of a preset classification model to perform risk prediction, so as to obtain a polling result of the merchant to be polled.

Referring to fig. 7, a second embodiment of the data inspection apparatus according to the embodiment of the present invention specifically includes:

the query module 601 is configured to query, from a preset database, merchant transaction data and merchant state data of a merchant to be patrolled and examined after the merchant to be patrolled and examined is determined;

a first determining module 602, configured to determine, based on the merchant transaction data and the merchant state data, a polling task of the merchant to be polled and geographic location data corresponding to the polling task;

In this embodiment, the data inspection device further includes:

the checking module 607 is configured to perform blacklist checking on the merchant to be checked;

a first generating module 608, configured to generate an inspection task corresponding to the merchant to be inspected if the merchant to be inspected is not in a preset blacklist; and if the merchant to be inspected is positioned in the blacklist, outputting an error prompt.

In this embodiment, the sending module 603 is specifically configured to:

the task information of the inspection task is issued to all inspection terminals meeting set inspection conditions;

and if terminal response information which is returned by the inspection terminal and is related to the task information is received, determining the inspection terminal corresponding to the earliest received terminal response information as a target terminal, and sending the inspection task to the target terminal.

In this embodiment, the data inspection device further includes:

the first obtaining module 609 is configured to obtain preset category risk representation information of the merchant to be inspected;

a second determining module 610, configured to determine, based on the risk characterization information, a risk quantification tag associated with the merchant to be patrolled; and determining the to-be-patrolled item of the to-be-patrolled merchant according to the risk quantification label.

In this embodiment, the identifying module 604 includes:

a dividing unit 6041 configured to divide the image data into a plurality of image blocks;

a denoising unit 6042, configured to perform denoising processing on the multiple image blocks to obtain a denoised image block corresponding to each image block;

the recognition unit 6043 is configured to input the image blocks into a preset OCR recognition model for recognition, so as to obtain a recognition result of each image block; and obtaining target merchant data of the merchant to be patrolled according to the identification result of each image block.

In this embodiment, the data inspection device further includes:

a second generating module 611, configured to obtain sample training data, and use the sample training data as a training set, where the sample training set includes risk quantification labels corresponding to sample merchants; generating a sample adjacency matrix and a sample transaction attribute characteristic matrix of the sample heterogeneous relation network diagram based on the sample training data;

a second feature extraction module 612, configured to input the sample adjacency matrix and the sample transaction attribute feature matrix into an initial graph neural network for feature extraction, so as to obtain an initial target feature vector of the sample merchant;

a third determining module 613, configured to determine a loss function based on the cross entropy of the risk identification category corresponding to the initial target feature vector and the risk quantization label; determining a descent gradient based on the loss function;

an updating module 614, configured to update the network parameter of the initial graph neural network based on the descent gradient and a preset learning rate, to obtain an updated graph neural network.

In this embodiment, the data inspection device further includes:

a second obtaining module 615, configured to obtain verification information corresponding to the inspection result;

a verification module 616, configured to verify authenticity of the verification information, and when the authenticity of the verification information passes the verification, verify the verification information; and when the verification of the verification information passes, reporting the inspection result to a background management system.

Fig. 6 and 7 describe the data inspection device in the embodiment of the present invention in detail from the perspective of the modular functional entity, and the data inspection device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

Fig. 8 is a schematic structural diagram of a data inspection apparatus according to an embodiment of the present invention, where the data inspection apparatus 800 may have a relatively large difference due to different configurations or performances, and may include one or more processors (CPUs) 810 (e.g., one or more processors) and a memory 820, and one or more storage media 830 (e.g., one or more mass storage devices) storing an application 833 or data 832. Memory 820 and storage medium 830 may be, among other things, transient or persistent storage. The program stored on the storage medium 830 may include one or more modules (not shown), each of which may include a series of instructions operating on the data inspection device 800. Further, the processor 810 may be configured to communicate with the storage medium 830, and execute a series of instruction operations in the storage medium 830 on the data inspection device 800 to implement the steps of the data inspection method provided by the above-described method embodiments.

The data patrol device 800 may also include one or more power supplies 840, one or more wired or wireless network interfaces 850, one or more input-output interfaces 860, and/or one or more operating systems 831, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, and the like. It will be appreciated by those skilled in the art that the data inspection apparatus configuration shown in fig. 8 does not constitute a limitation of the data inspection apparatus provided herein, and may include more or fewer components than shown, or some components may be combined, or a different arrangement of components.

The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, having stored therein instructions, which, when executed on a computer, cause the computer to perform the steps of the data inspection method described above.

It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.

The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a read-only memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.

The above-mentioned embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims

1. A data inspection method is characterized by comprising the following steps:

after determining the merchant to be patrolled, inquiring merchant transaction data and merchant state data of the merchant to be patrolled from a preset database;

determining a polling task of the merchant to be polled and geographical position data corresponding to the polling task based on the merchant transaction data and the merchant state data;

selecting an inspection terminal for executing the inspection task based on the geographic position data, and sending the inspection task to the inspection terminal;

determining a preset item to be inspected in the inspection task, acquiring image data uploaded by the merchant to be inspected according to the inspection item, and inputting the image data into a preset OCR recognition model for recognition to obtain target merchant data of the merchant to be inspected;

inputting the target merchant data into a preset graph neural network for feature extraction to obtain a target feature vector of the target merchant data;

and inputting the target characteristic vector into a prediction layer of a preset classification model to perform risk prediction, and obtaining a polling result of the merchant to be polled.

2. The data inspection method according to claim 1, before determining the inspection task of the merchant to be inspected and the geographical location data corresponding to the inspection task based on the merchant transaction data and the merchant state data, further comprising:

carrying out blacklist check on the merchant to be checked;

if the to-be-inspected merchant is not in a preset blacklist, generating an inspection task corresponding to the to-be-inspected merchant;

and if the merchant to be inspected is positioned in the blacklist, outputting an error prompt.

3. The data inspection method according to claim 1, wherein the selecting the inspection terminal to execute the inspection task based on the geographic location data and sending the inspection task to the inspection terminal includes:

4. The data inspection method according to claim 3, wherein after the inspection terminal performing the inspection task is selected based on the geographic location data and the inspection task is sent to the inspection terminal, the method further comprises:

acquiring preset category risk representation information of a merchant to be patrolled;

determining a risk quantification label associated with the merchant to be patrolled based on the risk characterization information;

and determining the to-be-patrolled item of the to-be-patrolled merchant according to the risk quantification label.

5. The data inspection method according to claim 1, wherein the step of inputting the image data into a preset OCR recognition model for recognition to obtain target merchant data of the merchant to be inspected comprises the steps of:

dividing the image data to obtain a plurality of image blocks;

denoising the image blocks to obtain denoised image blocks corresponding to each image block;

inputting the image blocks into a preset OCR recognition model for recognition respectively to obtain a recognition result of each image block;

and obtaining target merchant data of the merchant to be patrolled according to the identification result of each image block.

6. The data inspection method according to claim 1, wherein before the graph neural network includes a sampling layer and a convolution layer, and the target merchant data is input into a preset graph neural network for feature extraction to obtain the target feature vector of the target merchant data, the method further includes:

obtaining sample training data, and taking the sample training data as a training set, wherein the sample training set comprises risk quantification labels corresponding to sample merchants;

generating a sample adjacency matrix and a sample transaction attribute characteristic matrix of the sample heterogeneous relation network diagram based on the sample training data;

inputting the sample adjacency matrix and the sample transaction attribute feature matrix into an initial graph neural network for feature extraction to obtain an initial target feature vector of the sample merchant;

determining a loss function based on the risk identification category corresponding to the initial target feature vector and the cross entropy of the risk quantization label;

determining a descent gradient based on the loss function;

and updating the network parameters of the initial graph neural network based on the descending gradient and a preset learning rate to obtain an updated graph neural network.

7. The data inspection method according to claim 1, wherein after predicting whether the merchant to be inspected has a risk according to the target feature vector and obtaining an inspection result of the merchant to be inspected, the method further comprises:

acquiring verification information corresponding to the inspection result;

verifying the authenticity of the verification information, and verifying the verification information after the authenticity of the verification information passes verification;

and when the verification of the verification information passes, reporting the inspection result to a background management system.

8. The utility model provides a data inspection device which characterized in that, data inspection device includes:

the query module is used for querying merchant transaction data and merchant state data of the merchant to be patrolled from a preset database after the merchant to be patrolled is determined;

the first determining module is used for determining a polling task of the merchant to be polled and geographical position data corresponding to the polling task based on the merchant transaction data and the merchant state data;

the sending module is used for selecting an inspection terminal for executing the inspection task based on the geographic position data and sending the inspection task to the inspection terminal;

the identification module is used for determining a preset item to be patrolled in the patrol task, acquiring image data uploaded by the merchant to be patrolled according to the patrol item, and inputting the image data into a preset OCR identification model for identification to obtain target merchant data of the merchant to be patrolled;

the first feature extraction module is used for inputting the target merchant data into a preset graph neural network for feature extraction to obtain a target feature vector of the target merchant data;

and the prediction module is used for inputting the target characteristic vector into a prediction layer of a preset classification model to carry out risk prediction so as to obtain the inspection result of the merchant to be inspected.

9. The utility model provides a data inspection equipment which characterized in that, data inspection equipment includes: a memory having instructions stored therein and at least one processor, the memory and the at least one processor interconnected by a line;

the at least one processor invokes the instructions in the memory to cause the data inspection device to perform the steps of the data inspection method of any one of claims 1-7.

10. A computer-readable storage medium, having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the data inspection method according to any one of claims 1-7.