Merchant data auditing method, device and equipment based on RPA robot, medium and program product

Abstract

The invention provides a merchant data auditing method, device and equipment based on an RPA robot, a medium and a program product, and relates to the technical field of artificial intelligence. The method comprises the following steps: receiving dynamically distributed account information, and obtaining to-be-audited merchant data according to the account information; identifying the to-be-audited merchant data by using a multi-modal large model, extracting key information of the to-be-audited merchant data, and comparing the key information with reference information to generate a merchant data comparison result; and generating a merchant data auditing result according to the merchant data comparison result. According to the embodiment of the invention, the defects of low merchant data auditing accuracy, high labor cost and long auditing time limit in the prior art are overcome, and efficient, rapid and accurate merchant data auditing by using the RPA robot is realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a merchant information review method, device, equipment, medium and program product based on an RPA robot. Background Art

[0002] In the traditional merchant information review process, financial institutions, payment platforms, and other institutions usually rely on manual review, which has the following pain points: Inefficiency: Manually verifying business licenses, ID cards, and other documents uploaded by merchants with the information retained by the system is time-consuming and labor-intensive. Especially in the case of high-concurrency applications, the review cycle is long, affecting the merchant's onboarding experience.

[0003] Lack of accuracy: Manual review is susceptible to subjective factors and may result in missed detections and misjudgments, such as failure to identify forged documents or inconsistent information, leading to compliance risks.

[0004] Data silo problem: Merchant information may be scattered across different systems (such as industrial and commercial registration, taxation, and internal bank databases). Manual cross-system queries are inefficient and difficult to synchronize the latest data in real time.

[0005] Security and permission management challenges: Directly granting system permissions to RPA robots may lead to data leakage or unauthorized operation risks. Therefore, a secure account allocation and access mechanism must be designed.

[0006] How to review merchant information is a technical problem that needs to be solved urgently. Summary of the Invention

[0007] The present invention provides a merchant information review method, device, equipment, medium and program product based on RPA robots, which are used to solve the defects of low merchant information review accuracy, high labor cost and long review time in the existing technology, and realize efficient, fast and accurate merchant information review by using RPA robots.

[0008] In a first aspect, the present invention provides a merchant information review method based on an RPA robot, which is applied to the RPA robot. The method includes the following steps: Receive dynamically assigned account information and obtain the merchant information to be reviewed based on the account information; wherein the account information is used to instruct the RPA robot to log in to the merchant information review system to obtain the merchant information to be reviewed, so as to review the merchant information to be reviewed; the account information at least includes an acquisition permission, and the acquisition permission represents the permission for the RPA robot to obtain the authorized merchant information to be reviewed; Using a multimodal large model to identify the merchant profile to be reviewed, extracting key information of the merchant profile to be reviewed, and comparing the key information with reference information to generate a merchant profile comparison result; wherein the multimodal large model is characterized as a merchant profile review model determined according to the type of the merchant profile to be reviewed, and the merchant profile review model instructs to review the merchant profile to be reviewed of the corresponding type; A merchant information review result is generated based on the merchant information comparison result.

[0009] Preferably, according to a merchant information review method based on an RPA robot provided by the present invention, the step of obtaining the merchant information to be reviewed based on the account information includes: Log in to the merchant information review system according to the account information; Obtaining a system interface, and sending access permission and access request information to the corresponding third-party merchant information system according to the system interface; wherein the system interface is characterized as an interface connecting the user information review system and each third-party merchant information system; When the third-party merchant information system approves the acquisition request information and the acquisition authority, the merchant information to be reviewed and fed back by the third-party merchant information system is received.

[0010] Preferably, according to a merchant profile review method based on an RPA robot provided by the present invention, the multimodal large model includes at least a first merchant profile review model and a second merchant profile review model; The multimodal large model is used to identify the merchant information to be reviewed and extract key information of the merchant information to be reviewed, including: Classifying the types of the merchant information to be reviewed from different third-party merchant information systems to obtain rule-based merchant information and non-rule-based merchant information; Identify the rule-based merchant profile using the first merchant profile review model to obtain rule-based key information; The second merchant information review model is used to identify the irregular merchant information to obtain irregular key information.

[0011] Preferably, according to the RPA robot-based merchant information review method provided by the present invention, the first merchant information review model at least includes a text information extraction module; The first merchant profile review model is used to identify the rule-based merchant profile and obtain key rule-based information, including: The text information extraction module is used to extract the text information of the rule-based merchant information to obtain the structured text field of the merchant certificate information.

[0012] Preferably, according to the RPA robot-based merchant information review method provided by the present invention, the second merchant information review model at least includes an image feature extraction module; The second merchant profile review model is used to identify the irregular merchant profile to obtain irregular key information, including: Normalizing the irregular merchant data to obtain a normalized merchant storefront photo image; The image feature extraction module is used to extract the feature vector of the normalized merchant storefront photo image to obtain the merchant storefront photo feature vector.

[0013] Preferably, according to a merchant information review method based on an RPA robot provided by the present invention, the step of generating reference information includes: Defining conditions for randomly generating abnormal images, and guiding the mutation processing of the preset original merchant storefront photo images according to the generation conditions to generate abnormal reference images; wherein the abnormal reference images are used to compare with the merchant storefront photo images in the irregular merchant data; Randomly modifying the preset merchant certificate information to generate an abnormal reference field; wherein the abnormal reference field is used to compare with the structured text field of the regular merchant information; The step of comparing the key information with the reference information to generate a merchant information comparison result includes: Extracting the feature vector of the abnormal reference image to obtain the feature vector of the reference door head photo; Calculating the cosine similarity between the feature vector of the merchant storefront photo and the feature vector of the reference storefront photo, and comparing the cosine similarity with a preset image threshold to obtain a storefront photo comparison result; and calculating the semantic similarity between the structured text field and the abnormal reference field, and comparing the semantic similarity with a preset semantic threshold to obtain a document comparison result; The merchant information comparison result is determined based on the storefront photo comparison result and the ID comparison result.

[0014] In a second aspect, the present invention further provides a merchant information review device based on an RPA robot, which is applied to the RPA robot, and the device includes: An acquisition module, configured to receive dynamically assigned account information and acquire merchant information to be reviewed based on the account information; wherein the account information is used to instruct the RPA robot to log in to the merchant information review system to acquire the merchant information to be reviewed, so as to review the merchant information to be reviewed; the account information at least includes an acquisition permission, which represents the permission for the RPA robot to acquire the authorized merchant information to be reviewed; a comparison module, configured to use a multimodal large model to identify the merchant profile to be reviewed, extract key information of the merchant profile to be reviewed, and compare the key information with reference information to generate a merchant profile comparison result; wherein the multimodal large model is characterized as a merchant profile review model determined according to the type of the merchant profile to be reviewed, and the merchant profile review model instructs to review the merchant profile to be reviewed of the corresponding type; The generating module is used to generate a merchant information review result according to the merchant information comparison result.

[0015] In a third aspect, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, a merchant information review method based on an RPA robot as described above is implemented.

[0016] In a fourth aspect, the present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements any of the above-mentioned merchant information review methods based on RPA robots.

[0017] In a fifth aspect, the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned merchant information review methods based on RPA robots.

[0018] The present invention provides a merchant information review method, device, equipment, medium and program product based on RPA robot, which receives dynamically allocated account information and obtains the merchant information to be reviewed based on the account information; wherein, the account information is used to instruct the RPA robot to log in to the merchant information review system to obtain the merchant information to be reviewed, so as to review the merchant information to be reviewed; the account information at least includes acquisition authority, and the acquisition authority is characterized as the authority of the RPA robot to obtain the authorized merchant information to be reviewed; the merchant information to be reviewed is identified by using a multimodal large model, key information of the merchant information to be reviewed is extracted, and the key information is compared with reference information to generate a merchant information comparison result; wherein, the multimodal large model is characterized as a merchant information review model determined according to the type of merchant information to be reviewed, and the merchant information review model instructs to review the merchant information to be reviewed of the corresponding type; based on the merchant information comparison result, a merchant information review result is generated. It is used to solve the defects of low merchant information review accuracy, high labor costs and long review time in existing technologies, and to achieve efficient, fast and accurate review of merchant information through the use of RPA robots. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 This is one of the flow charts of the merchant information review method based on the RPA robot provided by the present invention.

[0021] Figure 2 This is the second flow chart of the merchant information review method based on RPA robot provided by the present invention.

[0022] Figure 3 This is a structural diagram of the RPA robot-based merchant information review system provided by the present invention.

[0023] Figure 4 It is a structural diagram of the merchant information review device based on the RPA robot provided by the present invention.

[0024] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0025] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0026] The following combination Figure 1-Figure 5 The present invention describes a merchant information review method, device, equipment, medium and program product based on RPA robots, which are used to solve the defects of low merchant information review accuracy, high labor costs and long review time in the existing technology, and realize efficient, fast and accurate merchant information review by using RPA robots.

[0027] Figure 1 This is one of the flow charts of a merchant information review method based on RPA robots provided by the present invention. Figure 1 As shown, the method may include but is not limited to steps S100 to S300: S100, receiving dynamically assigned account information, and obtaining merchant information to be reviewed based on the account information; wherein the account information is used to instruct the RPA robot to log in to the merchant information review system to obtain the merchant information to be reviewed, so as to review the merchant information to be reviewed; the account information at least includes an acquisition permission, and the acquisition permission represents the permission for the RPA robot to obtain the authorized merchant information to be reviewed; S200, using a multimodal large model to identify the merchant profile to be reviewed, extracting key information of the merchant profile to be reviewed, and comparing the key information with reference information to generate a merchant profile comparison result; wherein the multimodal large model is characterized as a merchant profile review model determined according to the type of the merchant profile to be reviewed, and the merchant profile review model instructs to review the merchant profile of the corresponding type to be reviewed; S300: Generate a merchant information review result based on the merchant information comparison result.

[0028] In step S100 of some embodiments, dynamically assigned account information is received, and the merchant information to be reviewed is obtained based on the account information; wherein, the account information is used to instruct the RPA robot to log in to the merchant information review system to obtain the merchant information to be reviewed, so as to review the merchant information to be reviewed; the account information at least includes acquisition permissions, and the acquisition permissions are represented by the permissions of the RPA robot to obtain the authorized merchant information to be reviewed.

[0029] It is understandable that RPA (Robotic Process Automation), also known as "Robotic Process Automation" in Chinese, is like a digital employee of an enterprise. Without changing the original system, it can simulate manual operations through simple configuration to help enterprises or employees complete repetitive and monotonous process work.

[0030] LLM stands for Natural Language Processing (NLP). It is a branch of artificial intelligence and linguistics that studies how to enable computer programs to understand, interpret, and generate human language. NLP aims to improve the interaction between humans and computers, enabling computers to more efficiently process and analyze large amounts of natural language data.

[0031] Furthermore, a temporary virtual account is dynamically generated by the background account management module (such as an API service based on Spring Boot), and a time limit (such as valid for 1 hour) and operation scope restrictions (such as only allowing reading of specified fields) are set.

[0032] Account information is transmitted to the RPA client through an encrypted channel (such as HTTPS) to prevent it from being exposed in plain text.

[0033] The technical effect of this step is to generate a temporary account on demand, which will automatically expire after the task is completed, reducing the risk of account leakage.

[0034] Permission isolation: Use sandbox mechanisms (such as database views or middleware agents) to limit RPA robots' direct operations on core data, and only grant read permissions to necessary fields.

[0035] At the database level, through predefined read-only views or stored procedures, only the necessary fields such as the unified social credit code and legal person ID number of the merchant information waiting for review are exposed.

[0036] RPA robot operations require access to data through a permission proxy service (such as an API gateway based on Spring Security), blocking direct SQL queries.

[0037] RPA robots cannot modify core data, ensuring the compliance of the audit process, limiting the scope of data access, and complying with privacy protection requirements.

[0038] Furthermore, in some embodiments of the present invention, obtaining merchant information to be reviewed based on the account information includes: Log in to the merchant information review system according to the account information; Obtaining a system interface, and sending access permission and access request information to the corresponding third-party merchant information system according to the system interface; wherein the system interface is characterized as an interface connecting the user information review system and each third-party merchant information system; When the third-party merchant information system approves the acquisition request information and the acquisition authority, the merchant information to be reviewed and fed back by the third-party merchant information system is received.

[0039] It is understandable that the temporary account dynamically generated by the permission control module of the merchant information review system (such as the backend service based on Spring Boot) contains the following attributes: Timeliness: The account is only valid for a certain period of time (such as 1 hour) and will automatically become invalid after the expiration date.

[0040] Permission scope: Only the permission to read the merchant information pending review is granted (such as reading the unified social credit code, legal person information and other fields).

[0041] Transmit account information to the RPA client through an encrypted channel (such as HTTPS) to avoid the risk of plaintext leakage.

[0042] The RPA client (such as a Python script) uses the Selenium or Playwright library to simulate browser operations and enter account information to log in to the audit system.

[0043] After successful login, obtain the list of merchants awaiting review and the corresponding third-party data interface address through the system API.

[0044] Obtain the interface address, API key or token of each third-party merchant information system (such as the UnionPay Business UP system) from the merchant information review system.

[0045] Construct request parameters, including: Access permissions: used to declare data retrieval permissions (such as corporate credit inquiry permissions) to third-party merchant information systems, and to be allowed to obtain merchant information awaiting review.

[0046] Get request information: including the merchant's unique identifier (such as a unified social credit code), data type (such as a business license, legal person ID card), storefront image, etc.

[0047] Send GET / POST requests to third-party merchant information systems via HTTP / HTTPS protocols, supporting multiple data formats (such as JSON and XML).

[0048] Exception handling: If the third-party merchant information system returns "Insufficient Permissions" or "Interface Call Limit Exceeded", an error log will be recorded and a retry mechanism (such as an exponential backoff algorithm) will be triggered.

[0049] If the third-party merchant information system responds for a timeout, switch to the backup interface or mark it as "pending manual processing."

[0050] After the third-party merchant information system passes the review, the merchant information to be reviewed (such as a scanned copy of the business license, a photo of the legal representative's ID card, and business address information) is pushed through the callback URL or message queue (such as Kafka).

[0051] Store unstructured data (such as storefront photos) in object storage (such as OSS / MinIO) and structured data (such as JSON) in a MySQL database.

[0052] Furthermore, the integrity of the merchant information data received for review is checked (such as required field checking and data format verification).

[0053] Extract text information from images through OCR (such as Tesseract) and merge it with structured data into a unified format (such as a standard JSON template).

[0054] This step achieves secure, efficient, and standardized merchant information acquisition through a combination of dynamic account numbers, interface automation, and intelligent verification, while also providing powerful exception handling capabilities and scalability.

[0055] In step S200 of some embodiments, a multimodal big model is used to identify the merchant profile to be reviewed, key information of the merchant profile to be reviewed is extracted, and the key information is compared with reference information to generate a merchant profile comparison result; wherein, the multimodal big model is characterized as a merchant profile review model determined according to the type of the merchant profile to be reviewed, and the merchant profile review model indicates to review the merchant profile to be reviewed of the corresponding type.

[0056] It can be understood that the multimodal large model includes at least a first merchant information review model and a second merchant information review model.

[0057] The multimodal large model is used to identify the merchant information to be reviewed and extract key information of the merchant information to be reviewed, including: Classifying the types of the merchant information to be reviewed from different third-party merchant information systems to obtain rule-based merchant information and non-rule-based merchant information; Identify the rule-based merchant profile using the first merchant profile review model to obtain rule-based key information; The second merchant information review model is used to identify the irregular merchant information to obtain irregular key information.

[0058] It should be noted that the first merchant information review model is used to process rule-based merchant information and obtain rule-based key information.

[0059] The second merchant profile review model is used to identify irregular merchant profiles and obtain irregular key information. Furthermore, an NLP model (such as BERT) analyzes the text description of the merchant profile to be reviewed (such as file name and tags), combined with an image classification model (such as ResNet) to identify the document type (such as business license, legal representative ID card, storefront photo).

[0060] For example: If the data contains the "unified social credit code" field and tabular data, it is determined to be a business license; if it contains a face and ID number, it is determined to be a legal person ID card.

[0061] Based on the data type, load the pre-trained vertical domain model through the model routing service (such as Kubernetes container orchestration): Textual materials: Use Transformer-based OCR + semantic matching models (such as LayoutLM); Image data: Use multimodal models (such as CLIP+ConvNeXt) to extract visual features; Composite data: large models that combine text, images, and structured data (such as the Vision-Language Model).

[0062] Automatically match the optimal model for different data types (such as forms, documents, and pictures), and improve the accuracy of key information extraction to over 95%.

[0063] In some embodiments of the present invention, the first merchant profile review model includes at least a text information extraction module; The first merchant profile review model is used to identify the rule-based merchant profile and obtain key rule-based information, including: The text information extraction module is used to extract the text information of the rule-based merchant information to obtain the structured text field of the merchant certificate information.

[0064] Use OCR (such as Tesseract) to identify fields such as "Registered Capital" and "Business Scope" in business licenses, and use regular expressions to verify the format validity.

[0065] Use a named entity recognition (NER) model (such as BERT-BiLSTM-CRF) to extract entities such as legal person names and registered addresses.

[0066] Use optical character recognition (OCR) (such as Tesseract and PaddleOCR) to extract text from merchant information in scanned or image format.

[0067] Perform image preprocessing (such as grayscale, binarization, and perspective correction) on blurred or tilted text to improve recognition accuracy.

[0068] Example: Extract fields such as "name", "address", and "business scope" from a business license image.

[0069] Load the corresponding rule template (such as XML / JSON format) according to the data type, and define the fields to be extracted and their location / format rules.

[0070] Use regular expressions or a rule engine such as Drools to extract target fields from the OCR results.

[0071] In some embodiments, format verification (such as regular expression matching YYYY-MM-DD) is performed on structured fields (such as date and amount).

[0072] Example: Extract "name," "ID number," and "validity period" from a legal person's ID card, and use regular expressions to verify the ID number format (such as ^[1-9]\d{16}[0-9X]$).

[0073] In some embodiments, it also includes logical association verification of the extracted fields (for example, the "registered address" must be consistent with the "business address", and the "validity period" must be later than the current date).

[0074] Trigger an alarm mechanism for missing or abnormal fields (such as the prompt "Missing unified social credit code").

[0075] Example: If the "Establishment Date" in the business license is later than the "Revocation Date", it is marked as an exception.

[0076] The extracted fields are output in standard JSON format, and the field names are aligned with the database / business system (for example, credit_code corresponds to the "unified social credit code").

[0077] Clean special characters, spaces, and line breaks to ensure data standardization.

[0078] This solution uses a three-step process of OCR + rule engine + logical verification to achieve efficient and accurate structured processing of rule-based merchant information, significantly reducing labor costs and risk misjudgment rate, while providing high-quality standardized data input for subsequent risk control, data analysis and other links.

[0079] In some embodiments of the present invention, the second merchant profile review model includes at least an image feature extraction module; The second merchant profile review model is used to identify the irregular merchant profile to obtain irregular key information, including: Normalizing the irregular merchant data to obtain a normalized merchant storefront photo image; The image feature extraction module is used to extract the feature vector of the normalized merchant storefront photo image to obtain the merchant storefront photo feature vector.

[0080] It is understandable that an image classification model (such as a lightweight CNN) is used to determine whether the input data belongs to an "irregular category" (such as non-standard storefront photos, handwritten receipts, complex background images, etc.).

[0081] Image resizing: Scale storefront photos to a fixed resolution (e.g., 224×224 pixels) to maintain the correct proportions of key areas (e.g., signs, doors, and windows).

[0082] Color space normalization: Convert to a standard color space (such as RGB→HSV) and standardize brightness and contrast (such as histogram equalization).

[0083] Background denoising: Remove interfering backgrounds through Gaussian blurring or semantic segmentation models (such as UNet) to retain the main area.

[0084] In some embodiments, a pre-trained convolutional neural network (such as ResNet-50, EfficientNet) is used as a base model to extract deep semantic features.

[0085] Based on the characteristics of storefront photos, we fine-tune the model (such as freezing some layers and training only the fully connected layers) to enhance its sensitivity to key features such as "signboard text", "business status", and "environmental layout".

[0086] Multi-scale feature fusion: Introducing attention mechanisms (such as CBAM and SE-Block) to dynamically focus on important areas in the image (such as shop signs, entrances, and billboards).

[0087] Combine global features (overall image semantics) with local features (text, logo details) to generate a high-dimensional feature vector (such as a 1024-dimensional floating-point vector).

[0088] The extracted feature vectors are L2 normalized to eliminate the dimension effect and facilitate subsequent comparison.

[0089] Feature storage and indexing: Store feature vectors in a vector database (such as Faiss and Milvus) to support fast similarity retrieval.

[0090] Combine spatiotemporal information (such as shooting time and geographic location) to build a multimodal feature index to improve association efficiency.

[0091] This step converts irregular storefront photos into high-density feature vectors that can be understood by machines through the process of image normalization → deep feature extraction → vector standardization, achieving efficient and accurate merchant information review.

[0092] In some embodiments of the present invention, the step of generating the reference information includes: Defining conditions for randomly generating abnormal images, and guiding the mutation processing of the preset original merchant storefront photo images according to the generation conditions to generate abnormal reference images; wherein the abnormal reference images are used to compare with the merchant storefront photo images in the irregular merchant data; Randomly modifying the preset merchant certificate information to generate an abnormal reference field; wherein the abnormal reference field is used to compare with the structured text field of the regular merchant information; The step of comparing the key information with the reference information to generate a merchant information comparison result includes: Extracting the feature vector of the abnormal reference image to obtain the feature vector of the reference door head photo; Calculating the cosine similarity between the feature vector of the merchant storefront photo and the feature vector of the reference storefront photo, and comparing the cosine similarity with a preset image threshold to obtain a storefront photo comparison result; and calculating the semantic similarity between the structured text field and the abnormal reference field, and comparing the semantic similarity with a preset semantic threshold to obtain a document comparison result; The merchant information comparison result is determined based on the storefront photo comparison result and the ID comparison result.

[0093] It should be noted that the definition and generation conditions for abnormal reference images include: pre-setting various image interference types (such as brightness adjustment, noise addition, occlusion, rotation, filter replacement, etc.), and defining mutation intensity parameters (such as brightness offset range, noise ratio, and occlusion area size). Mutation operations are selected from a rule base through random sampling (such as a 30% probability of adding Gaussian noise and a 20% probability of simulating occlusion), ensuring that each generated abnormal image is unpredictable.

[0094] Use preset clean storefront photo samples (such as historical archived images of real merchants) to perform mutation processing to generate diverse abnormal reference images (such as blurred signs, obscured stores, and falsified decoration status).

[0095] The mutation processing may at least include but is not limited to applying a random mutation operation (such as adding 30% Gaussian noise to the sign text area to simulate camera blur).

[0096] Use GAN or StyleGAN to modify the image style (such as converting a daytime doorway photo into a nighttime effect).

[0097] Superimposing forged elements (such as photoshopping a fake business license hanging image).

[0098] The tampering rules for abnormal reference fields include at least: randomly replacing key fields (such as changing the "legal person's name" to a homophone), deleting required fields (such as the "unified social credit code"), and inserting invalid characters (such as adding "X" to the last digit of the ID number to change it to "Y").

[0099] Logical conflict rules: force the generation of contradictory data (such as "establishment date" is later than "cancellation date").

[0100] Use real merchant ID data (anonymized) or synthetic data, randomly modify fields according to rules (such as 20% probability of tampering with the "registered capital" value, 10% probability of replacing the "business scope" keyword), or inject correlation errors (such as inconsistency between "registered address" and "actual business address") to generate abnormal reference fields (such as "Zhang San" → "Zhang Sanfeng", "Pudong New Area, Shanghai" → "Chaoyang District, Beijing").

[0101] Furthermore, a pre-trained image feature extraction model (e.g., ResNet-50) is used to extract feature vectors (e.g., 512-dimensional floating-point vectors) from the storefront photo and the anomaly reference image. The feature vectors for the storefront photo and the reference photo are then normalized (L2 regularization) to eliminate dimensionality effects.

[0102] Similarity calculation: Calculate the cosine similarity between the image to be reviewed (the merchant storefront photo feature vector) and all abnormal reference images, and take the maximum value as the storefront photo similarity score.

[0103] Threshold judgment: If the similarity exceeds the preset threshold (such as 0.85), it will be marked as "image abnormality".

[0104] Furthermore, semantic similarity calculation: Use the BERT model to encode structured text fields (such as "business scope") and abnormal reference fields into semantic vectors.

[0105] The semantic similarity between the structured text field and the abnormal reference field is calculated. If the semantic similarity is lower than a threshold (eg, 0.6), it is determined to be a "semantic mismatch."

[0106] Logical verification: Check for missing required fields and contradictory fields (such as "Establishment Date > Current Date"), and directly mark them as abnormal.

[0107] In some embodiments, if either the cosine similarity or the semantic similarity of the storefront photo exceeds a threshold, it is determined to be "data abnormality", and a merchant data comparison result indicating abnormal merchant data for the merchant to be reviewed is obtained.

[0108] If both are abnormal or trigger a logical conflict, it will be judged as "high-risk fraud" and the merchant information comparison result of the merchant information to be reviewed will be abnormal.

[0109] Only when the cosine similarity of the storefront photo is less than the preset image threshold, the semantic similarity of the structured text field is less than the preset semantic threshold, and the logical check passes, the merchant information comparison result of the merchant information to be reviewed is obtained as normal.

[0110] It should be noted that the preset image threshold and preset semantic threshold are automatically adjusted based on historical data distribution (for example, using the percentile method, the preset image threshold is set to "mean + 3 standard deviations"). Stricter judgment rules are set separately for high-risk industries (such as finance and healthcare).

[0111] Some embodiments of the present invention also include using adversarial sample generation techniques (such as FGSM) to simulate malicious tampering and improve the model's sensitivity to subtle differences. Style transfer models (such as CycleGAN) are also introduced to generate anomalous samples of storefront photos across seasons and time periods.

[0112] Sentence embedding is used in conjunction with industry dictionaries (such as business registration terminology) to improve field matching accuracy. Structured fields (such as dates and amounts) are formatted and aligned before comparison (for example, "2023-01-01" and "2023 / 1 / 1" are considered identical). This step achieves accurate review of merchant profiles through a closed-loop process: multimodal anomaly reference generation → deep feature comparison → dynamic risk assessment. This technical approach balances fraud coverage, attack resistance, and review efficiency, making it suitable for scenarios requiring high-reliability review, such as finance, e-commerce, and government affairs. It also meets compliance requirements through explainable design.

[0113] In step S300 of some embodiments, a merchant information review result is generated based on the merchant information comparison result.

[0114] It is understandable that the comparison result of the door head photos is the result of whether the cosine similarity score exceeds a preset image threshold.

[0115] The document comparison result is the result of whether the semantic similarity score exceeds the preset semantic threshold and whether the logical verification status is normal.

[0116] Normal category: The storefront photo similarity is ≤ the preset image threshold and the ID card semantic similarity is ≤ the preset semantic threshold, and there is no logical conflict.

[0117] Abnormal category: The similarity of the storefront photo is greater than the preset image threshold, or the semantic similarity of the ID card is greater than the threshold, or there is a logical conflict.

[0118] The merchant information to be reviewed is divided into risk levels according to the merchant information comparison result, so as to generate a merchant information review result for reviewing the merchant information to be reviewed.

[0119] Low risk: Only a single indicator is abnormal (e.g., the similarity of the storefront photo slightly exceeds the threshold but the ID is normal).

[0120] Medium risk: Two indicators are abnormal but traceable (e.g., the storefront photo shows obvious signs of tampering but the ID information is consistent).

[0121] High risk: Multiple indicators are abnormal and unexplainable (e.g., the storefront photo is highly similar to multiple reference photos + the ID information is contradictory).

[0122] The rule engine decision steps include the following: Merchant information awaiting review is automatically approved: low risk and no manual review mark (such as merchants with a good historical credit history).

[0123] For merchant information with a medium risk level awaiting review (such as storefront photos with an 85% similarity but consistent IDs), manual review is required to improve the accuracy of the review.

[0124] High risk but requires manual confirmation (such as suspected system misjudgment).

[0125] Direct rejection: High risk and meets fraud characteristics (such as a match with the blacklist reference of >90% and ID information is stolen).

[0126] The RPA robot directly generates the audit conclusion: pass / reject / review, and attaches a risk level label.

[0127] In some embodiments, the merchant profile review results also include a review evidence chain, such as associated anomaly reference graphs, conflict fields, similarity scores, and other data for tracing.

[0128] Generate merchant information review results in JSON / XML format according to business requirements, including merchant ID, review status, risk details, etc.

[0129] In some embodiments, data is pushed to downstream systems (such as risk control platforms and merchant backends) via APIs. Email / SMS alerts are triggered for abnormal cases (e.g., high-risk fraud requiring urgent action).

[0130] Furthermore, medium-risk merchant profiles awaiting review are assigned to the reviewer's workstation, where they are provided with comparison evidence (e.g., difference heatmaps, field highlighting). After a manual decision is made, the review status is updated and the process is closed.

[0131] In some embodiments of the present invention, a weighted scoring model (such as a 40% weight for storefront photos and a 60% weight for ID cards) can also be used to comprehensively judge risks and avoid misjudgment in a single dimension.

[0132] For contradictory results (such as the image is normal but the ID is abnormal), cross-validation logic is triggered (such as checking the consistency between the text in the storefront photo and the address on the ID).

[0133] In some embodiments, a dynamic threshold calibration step is further included: Automatically calculate thresholds based on historical data distribution (e.g., image threshold = industry average + 3σ), reducing manual configuration costs.

[0134] Stricter thresholds are set separately for sensitive industries (such as finance and healthcare) (for example, the image threshold is raised from 0.85 to 0.92).

[0135] Furthermore, in some embodiments, abnormal fields are automatically highlighted (e.g., "Establishment Date" is later than "Deregistration Date"). Historical and currently submitted materials can be displayed side-by-side to quickly identify signs of tampering. This step, through a closed-loop process of multimodal data comparison, rule engine decision-making, and evidence chain output, achieves efficient, accurate, and traceable merchant information review.

[0136] The present invention can achieve at least the following technical effects through the above embodiments: The use of RPA technology is to introduce a digital workforce that can work efficiently and continuously 24 hours a day. RPA simulates human actions to avoid human operational errors and information errors, and improves the accuracy and compliance of work results. System upgrades often consume huge costs, but the effects are difficult to estimate. RPA is a low-risk, non-invasive means that can be quickly deployed and implemented without interfering with or changing existing enterprise systems. This method uses visual models and large language models to extract information from captured content.

[0137] Combine Figure 2 As shown, this invention uses the dual-engine architecture of "RPA dynamic account management + large model intelligent comparison" to achieve full automation and intelligence of merchant information review. Its core implementation logic can be divided into the following four levels: 1) Permission control layer: A dynamic temporary account allocation mechanism is adopted, and the system automatically generates virtual accounts with timeliness and operation scope restrictions for RPA use.

[0138] The permission sandbox isolates RPA's direct access to the core database, allowing only reading of specified fields (such as the unified social credit code, legal person ID number, etc.).

[0139] 2) Data collection layer: RPA robots automatically log in to the UnionPay Business UP system, CM system, Qichacha system, and other systems through authorized accounts to capture the latest merchant registration information.

[0140] 3) Intelligent Analysis Layer: Multimodal large-scale model processing: Regular image data (business licenses / ID cards / operating permits) → OCR extracts structured data + large-scale language model automatically compares. Non-regular image data (storefront photos) → Visual model identifies key information in the image + large-scale language model automatically compares.

[0141] 4) Decision output layer: Automatically generate audit reports and mark risk levels (passed / suspicious / rejected).

[0142] Automatically trigger the manual review process for high-risk cases and retain operation audit logs.

[0143] In some embodiments of the present invention, a risk scoring model is constructed based on the merchant information comparison results (such as field matching and abnormality type), and three levels of labels are divided into "passed", "suspicious" and "rejected".

[0144] Automatically generate audit reports in HTML / PDF format and record operation logs (such as account usage records and data modification traces) through MySQL stored procedures.

[0145] This step can achieve the following technical benefits: Accurate classification: High-risk cases (such as mismatched legal person ID numbers) automatically trigger manual review, reducing missed detection rates. Traceability: Complete audit logs meet regulatory compliance requirements.

[0146] For manual intervention in high-risk cases, the review system (Vue front-end) marks high-risk tickets and pushes them to the manual review queue (such as real-time notification via WebSocket).

[0147] The manual review interface supports side-by-side comparison (such as original data VS RPA-collected data) to assist in decision-making.

[0148] Human-machine collaboration can significantly improve audit efficiency: the proportion of manual intervention is less than 15%, and the overall audit efficiency is increased by 80%.

[0149] Further, combined Figure 3 As shown, the present invention also provides a merchant information review system based on RPA robots. The entire system is divided into three parts: the review system, the RPA client, and the large model system. The three can run independently on different servers under network interconnection.

[0150] Audit system: Currently, an audit system is built using Java backend + Vue frontend. This system can display merchant information through the existing database: such as ID card, business license, operating license, storefront photo and other information. It adopts front-end and back-end separation architecture: Front-end: Vue.js 3.x + Element UI component library + Axios HTTP library.

[0151] Backend: Spring Boot 2.7 + Spring Security + MyBatis Plus.

[0152] Database: MySQL 8.0 cluster (read-write separation) + Redis cache.

[0153] In the audit information system, it is necessary to develop relevant backend code to implement the functions of account creation and permission allocation. This includes the creation of administrator users, general review users, etc.

[0154] RPA client: The RPA client is actually an application (similar to QQ). It is currently written in an executable file using the Python framework. Clicking the executable file opens the RPA client.

[0155] In this system, the workflow executed by the RPA client can execute the steps required for manual review by operating the browser. Images of pending work orders are saved locally, and the large model is used to recognize the images. The recognition results are then compared with system information using the large model.

[0156] Furthermore, a microservice deployment architecture is adopted, which includes the following core components: Model inference cluster: GPU computing nodes (NVIDIA A10G / A100) managed by Kubernetes.

[0157] Model Serving Gateway: handles request routing, load balancing, and protocol conversion.

[0158] Training scheduling platform: manages model iterative training tasks.

[0159] Image processing pipeline: Storefront photo analysis: Use ConvNeXt to extract visual features and the CLIP model to calculate image-text similarity.

[0160] Document recognition: LayoutLMv3 handles the extraction of structured information from documents.

[0161] Anomaly detection: The Diffusion model generates anomaly reference images for comparison.

[0162] Text processing pipeline: key information extraction: Deepseek-based field extraction and logical verification.

[0163] The present invention provides a merchant information review method, device, equipment, medium and program product based on RPA robot, which receives dynamically allocated account information and obtains the merchant information to be reviewed based on the account information; wherein, the account information is used to instruct the RPA robot to log in to the merchant information review system to obtain the merchant information to be reviewed, so as to review the merchant information to be reviewed; the account information at least includes acquisition authority, and the acquisition authority is characterized as the authority of the RPA robot to obtain the authorized merchant information to be reviewed; the merchant information to be reviewed is identified by using a multimodal large model, key information of the merchant information to be reviewed is extracted, and the key information is compared with reference information to generate a merchant information comparison result; wherein, the multimodal large model is characterized as a merchant information review model determined according to the type of merchant information to be reviewed, and the merchant information review model instructs to review the merchant information to be reviewed of the corresponding type; based on the merchant information comparison result, a merchant information review result is generated. It is used to solve the defects of low merchant information review accuracy, high labor costs and long review time in existing technologies, and to achieve efficient, fast and accurate review of merchant information through the use of RPA robots.

[0164] The following describes the merchant information review device based on the RPA robot provided by the present invention. The merchant information review device based on the RPA robot described below and the merchant information review method based on the RPA robot described above can be referenced to each other.

[0165] like Figure 4 FIG2 is a schematic diagram of the structure of a merchant information review device based on an RPA robot provided by the present invention. The merchant information review device based on an RPA robot is characterized in that it is applied to the RPA robot and includes: An acquisition module 410 is configured to receive dynamically assigned account information and acquire merchant information to be reviewed based on the account information; wherein the account information is used to instruct the RPA robot to log in to the merchant information review system to acquire the merchant information to be reviewed, so as to review the merchant information to be reviewed; the account information includes at least an acquisition permission, which represents the permission for the RPA robot to acquire the authorized merchant information to be reviewed; A comparison module 420 is configured to use a multimodal macro model to identify the merchant profile to be reviewed, extract key information from the merchant profile to be reviewed, and compare the key information with reference information to generate a merchant profile comparison result; wherein the multimodal macro model is characterized as a merchant profile review model determined according to the type of the merchant profile to be reviewed, and the merchant profile review model instructs the review of the merchant profile to be reviewed of the corresponding type; The generating module 430 is configured to generate a merchant information review result according to the merchant information comparison result.

[0166] Preferably, the merchant information review device based on the RPA robot provided by the present invention is further configured to log in to the merchant information review system according to the account information; Obtaining a system interface, and sending access permission and access request information to the corresponding third-party merchant information system according to the system interface; wherein the system interface is characterized as an interface connecting the user information review system and each third-party merchant information system; When the third-party merchant information system approves the acquisition request information and the acquisition authority, the merchant information to be reviewed and fed back by the third-party merchant information system is received.

[0167] Preferably, the RPA robot-based merchant data review device provided by the present invention is further configured such that the multimodal large model includes at least a first merchant data review model and a second merchant data review model; The multimodal large model is used to identify the merchant information to be reviewed and extract key information of the merchant information to be reviewed, including: Classifying the types of the merchant information to be reviewed from different third-party merchant information systems to obtain rule-based merchant information and non-rule-based merchant information; Identify the rule-based merchant profile using the first merchant profile review model to obtain rule-based key information; The second merchant information review model is used to identify the irregular merchant information to obtain irregular key information.

[0168] Preferably, the merchant information review device based on the RPA robot provided by the present invention is further configured such that the first merchant information review model at least includes a text information extraction module; The first merchant profile review model is used to identify the rule-based merchant profile and obtain key rule-based information, including: The text information extraction module is used to extract the text information of the rule-based merchant information to obtain the structured text field of the merchant certificate information.

[0169] Preferably, the RPA robot-based merchant information review device provided by the present invention is further configured such that the second merchant information review model at least includes an image feature extraction module; The second merchant profile review model is used to identify the irregular merchant profile to obtain irregular key information, including: Normalizing the irregular merchant data to obtain a normalized merchant storefront photo image; The image feature extraction module is used to extract the feature vector of the normalized merchant storefront photo image to obtain the merchant storefront photo feature vector.

[0170] Preferably, the merchant information review device based on the RPA robot provided by the present invention is further configured to generate the reference information in the following steps: Defining conditions for randomly generating abnormal images, and guiding the mutation processing of the preset original merchant storefront photo images according to the generation conditions to generate abnormal reference images; wherein the abnormal reference images are used to compare with the merchant storefront photo images in the irregular merchant data; Randomly modifying the preset merchant certificate information to generate an abnormal reference field; wherein the abnormal reference field is used to compare with the structured text field of the regular merchant information; The step of comparing the key information with the reference information to generate a merchant information comparison result includes: Extracting the feature vector of the abnormal reference image to obtain the feature vector of the reference door head photo; Calculating the cosine similarity between the feature vector of the merchant storefront photo and the feature vector of the reference storefront photo, and comparing the cosine similarity with a preset image threshold to obtain a storefront photo comparison result; and calculating the semantic similarity between the structured text field and the abnormal reference field, and comparing the semantic similarity with a preset semantic threshold to obtain a document comparison result; The merchant information comparison result is determined based on the storefront photo comparison result and the ID comparison result.

[0171] The present invention provides a merchant information review method, device, equipment, medium and program product based on RPA robot, which receives dynamically allocated account information and obtains the merchant information to be reviewed based on the account information; wherein, the account information is used to instruct the RPA robot to log in to the merchant information review system to obtain the merchant information to be reviewed, so as to review the merchant information to be reviewed; the account information at least includes acquisition authority, and the acquisition authority is characterized as the authority of the RPA robot to obtain the authorized merchant information to be reviewed; the merchant information to be reviewed is identified by using a multimodal large model, key information of the merchant information to be reviewed is extracted, and the key information is compared with reference information to generate a merchant information comparison result; wherein, the multimodal large model is characterized as a merchant information review model determined according to the type of merchant information to be reviewed, and the merchant information review model instructs to review the merchant information to be reviewed of the corresponding type; based on the merchant information comparison result, a merchant information review result is generated. It is used to solve the defects of low merchant information review accuracy, high labor costs and long review time in existing technologies, and to achieve efficient, fast and accurate review of merchant information through the use of RPA robots.

[0172] Figure 5 An example of a physical structure diagram of an electronic device is shown below. Figure 5As shown, the electronic device may include: a processor (processor) 510, a communication interface (Communications Interface) 520, a memory (memory) 530 and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 can call the logic instructions in the memory 530 to execute a merchant profile review method based on an RPA robot, the method including: receiving dynamically allocated account information, and obtaining the merchant profile to be reviewed based on the account information; wherein the account information is used to instruct the RPA robot to log in to the merchant profile review system to obtain the merchant profile to be reviewed, so as to review the merchant profile to be reviewed; the account information at least includes acquisition permissions, and the acquisition permissions are characterized as the permissions of the RPA robot to obtain the authorized merchant profile to be reviewed; using a multimodal large model to identify the merchant profile to be reviewed, extracting key information of the merchant profile to be reviewed, and comparing the key information with reference information to generate a merchant profile comparison result; wherein the multimodal large model is characterized as a merchant profile review model determined according to the type of the merchant profile to be reviewed, and the merchant profile review model instructs to review the merchant profile to be reviewed of the corresponding type; generating a merchant profile review result based on the merchant profile comparison result.

[0173] Furthermore, the logic instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0174] On the other hand, the present invention also provides a computer program product, comprising a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the merchant profile review method based on the RPA robot provided by the above-mentioned methods, the method comprising: receiving dynamically allocated account information, and obtaining merchant profile to be reviewed based on the account information; wherein the account information is used to instruct the RPA robot to log in to a merchant profile review system to obtain the merchant profile to be reviewed, so as to review the merchant profile to be reviewed; the account information includes at least an acquisition permission, wherein the acquisition permission is characterized by the RPA robot's permission to obtain the authorized merchant profile to be reviewed; using a multimodal large model to identify the merchant profile to be reviewed, extracting key information of the merchant profile to be reviewed, and comparing the key information with reference information to generate a merchant profile comparison result; wherein the multimodal large model is characterized by a merchant profile review model determined according to the type of the merchant profile to be reviewed, the merchant profile review model instructing to review the merchant profile to be reviewed of the corresponding type; and generating a merchant profile review result based on the merchant profile comparison result.

[0175] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the merchant profile review method based on the RPA robot provided by the above-mentioned methods, the method comprising: receiving dynamically allocated account information, and obtaining the merchant profile to be reviewed based on the account information; wherein the account information is used to instruct the RPA robot to log in to the merchant profile review system to obtain the merchant profile to be reviewed, so as to review the merchant profile to be reviewed; the account information at least includes an acquisition permission, and the acquisition permission is characterized as the permission for the RPA robot to obtain the authorized merchant profile to be reviewed; using a multimodal large model to identify the merchant profile to be reviewed, extracting key information of the merchant profile to be reviewed, and comparing the key information with reference information to generate a merchant profile comparison result; wherein the multimodal large model is characterized as a merchant profile review model determined according to the type of the merchant profile to be reviewed, and the merchant profile review model instructs to review the merchant profile to be reviewed of the corresponding type; and generating a merchant profile review result based on the merchant profile comparison result.

[0176] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0177] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0178] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A merchant information review method based on RPA robot, characterized in that: Applied to an RPA robot, the method includes: Receive dynamically assigned account information and obtain the merchant information to be reviewed based on the account information; wherein the account information is used to instruct the RPA robot to log in to the merchant information review system to obtain the merchant information to be reviewed, so as to review the merchant information to be reviewed; the account information at least includes an acquisition permission, and the acquisition permission represents the permission for the RPA robot to obtain the authorized merchant information to be reviewed; Using a multimodal large model to identify the merchant profile to be reviewed, extracting key information of the merchant profile to be reviewed, and comparing the key information with reference information to generate a merchant profile comparison result; wherein the multimodal large model is characterized as a merchant profile review model determined according to the type of the merchant profile to be reviewed, and the merchant profile review model instructs to review the merchant profile to be reviewed of the corresponding type; A merchant information review result is generated based on the merchant information comparison result.

2. The merchant information review method based on RPA robot according to claim 1 is characterized in that: The obtaining of merchant information to be reviewed based on the account information includes: Log in to the merchant information review system according to the account information; Obtaining a system interface, and sending access permission and access request information to the corresponding third-party merchant information system according to the system interface; wherein the system interface is characterized as an interface connecting the user information review system and each third-party merchant information system; When the third-party merchant information system approves the acquisition request information and the acquisition authority, the merchant information to be reviewed and fed back by the third-party merchant information system is received.

3. The merchant information review method based on RPA robot according to claim 2 is characterized in that: The multimodal large model includes at least a first merchant information review model and a second merchant information review model; The multimodal large model is used to identify the merchant information to be reviewed and extract key information of the merchant information to be reviewed, including: Classifying the types of the merchant information to be reviewed from different third-party merchant information systems to obtain rule-based merchant information and non-rule-based merchant information; Identify the rule-based merchant profile using the first merchant profile review model to obtain rule-based key information; The second merchant information review model is used to identify the irregular merchant information to obtain irregular key information.

4. The merchant information review method based on RPA robot according to claim 3 is characterized in that: The first merchant information review model at least includes a text information extraction module; The first merchant profile review model is used to identify the rule-based merchant profile and obtain key rule-based information, including: The text information extraction module is used to extract the text information of the rule-based merchant information to obtain the structured text field of the merchant certificate information.

5. The merchant information review method based on RPA robot according to claim 4 is characterized in that: The second merchant information review model at least includes an image feature extraction module; The second merchant profile review model is used to identify the irregular merchant profile to obtain irregular key information, including: Normalizing the irregular merchant data to obtain a normalized merchant storefront photo image; The image feature extraction module is used to extract the feature vector of the normalized merchant storefront photo image to obtain the merchant storefront photo feature vector.

6. The merchant information review method based on RPA robot according to claim 5 is characterized in that: The step of generating the reference information includes: Defining conditions for randomly generating abnormal images, and guiding the mutation processing of the preset original merchant storefront photo images according to the generation conditions to generate abnormal reference images; wherein the abnormal reference images are used to compare with the merchant storefront photo images in the irregular merchant data; Randomly modifying the preset merchant certificate information to generate an abnormal reference field; wherein the abnormal reference field is used to compare with the structured text field of the regular merchant information; The step of comparing the key information with the reference information to generate a merchant information comparison result includes: Extracting the feature vector of the abnormal reference image to obtain the feature vector of the reference door head photo; Calculating the cosine similarity between the feature vector of the merchant storefront photo and the feature vector of the reference storefront photo, and comparing the cosine similarity with a preset image threshold to obtain a storefront photo comparison result; and calculating the semantic similarity between the structured text field and the abnormal reference field, and comparing the semantic similarity with a preset semantic threshold to obtain a document comparison result; The merchant information comparison result is determined based on the storefront photo comparison result and the ID comparison result.

7. A merchant information review device based on RPA robot, characterized in that: Applied to an RPA robot, the device includes: An acquisition module, configured to receive dynamically assigned account information and acquire merchant information to be reviewed based on the account information; wherein the account information is used to instruct the RPA robot to log in to the merchant information review system to acquire the merchant information to be reviewed, so as to review the merchant information to be reviewed; the account information at least includes an acquisition permission, which represents the permission for the RPA robot to acquire the authorized merchant information to be reviewed; a comparison module, configured to use a multimodal large model to identify the merchant profile to be reviewed, extract key information of the merchant profile to be reviewed, and compare the key information with reference information to generate a merchant profile comparison result; wherein the multimodal large model is characterized as a merchant profile review model determined according to the type of the merchant profile to be reviewed, and the merchant profile review model instructs to review the merchant profile to be reviewed of the corresponding type; The generating module is used to generate a merchant information review result according to the merchant information comparison result.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the merchant information review method based on the RPA robot as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the merchant information review method based on the RPA robot is implemented as described in any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the merchant information review method based on the RPA robot is implemented as described in any one of claims 1 to 6.

Images