A leasing factoring business bottom asset verification system and method

Abstract

The present application belongs to the technical field of financial risk control, and particularly relates to a leasing factoring business bottom asset verification system and method, which takes an RPA scheduling management module as the core, links RPA robots to collect double-source logistics information, converts the information into structured skeleton data through pretreatment, relies on an asset verification large language model to complete core entity extraction and double-source data difference analysis, quantifies logistics information difference, trajectory similarity and overall reliability through a preset algorithm, combines rules to complete risk level determination and generate a verification report. The present application can solve the problems of low verification efficiency, low automation degree and weak risk control capability in the leasing factoring bottom asset verification scheme of the prior art.

Description

Technical Field

[0001] This invention belongs to the field of financial risk control technology, and in particular relates to a system and method for verifying the underlying assets of leasing factoring business. Background Technology

[0002] Leasing factoring is a financial business model in which a leasing platform packages accounts receivable from C-end customer leasing consumption orders into asset packages and transfers them to financial institutions to conduct factoring financing. The authenticity of the underlying leasing consumption order logistics information is the core of risk control in this business, directly determining the validity of the accounts receivable assets. Currently, financial institutions mainly use manual sampling to verify the authenticity of logistics information in the process of verifying the underlying assets in leasing factoring, combined with traditional RPA tools for simple process automation. This approach has the following technical shortcomings and industry pain points: 1. Manual verification is inefficient and has limited coverage: Financial institutions can only verify logistics information pushed by the leasing platform. The manual sampling inspection requires manual login to the logistics company's official website and manual comparison of the consistency between the logistics information on the official website and the information pushed by the platform. The verification of a single order is time-consuming and cannot meet the needs of the rapidly increasing order volume of leasing and factoring business. The uninspected part has the risk control risks of false logistics information and false orders. 2. Lack of automated processes and high risk of business continuity: The lack of a unified RPA scheduling and management system means that the creation, execution and result feedback of verification tasks rely on manual intervention throughout the entire process. There is no automated task scheduling and execution capability, and errors are prone to occur in the manual operation process. Moreover, when there are personnel changes or work interruptions, the verification business cannot be carried out continuously, and business continuity cannot be guaranteed. 3. Traditional RPA applications have significant limitations: Existing RPA tools (such as Enterprise RPA and Online Banking RPA) can only automate basic processes such as account login and simple data collection. They do not have the ability to deeply integrate with artificial intelligence models and cannot perform semantic analysis, logical reasoning, and discrepancy verification on the collected logistics information. The core verification and analysis process still needs to be completed manually, and true end-to-end automation has not been achieved. 4. Lack of a standardized verification and evaluation system: Manual verification can only provide a binary result of "consistent / inconsistent", which cannot quantitatively evaluate the authenticity of logistics information, nor can it accurately locate information discrepancies (such as deviations in shipping location, gaps in logistics trajectory, and discrepancies in shipping time). The reference value of the verification results is limited, and it is not conducive to subsequent risk control analysis and risk classification.

[0003] Therefore, existing asset verification solutions for leasing factoring suffer from low verification efficiency, low automation, and weak risk control capabilities. Summary of the Invention

[0004] The technical problem solved by this invention is to provide a system and method for verifying the underlying assets of leasing factoring business, so as to solve the problems of low verification efficiency, low automation and weak risk control capabilities in the existing leasing factoring underlying asset verification schemes.

[0005] The basic solution provided by this invention is a system for verifying underlying assets in a leasing factoring business. The system includes an RPA scheduling and management module and an RPA robot acquisition module, a preprocessing module, a logistics information entity extraction module, a logistics information difference analysis module, an indicator evaluation module, a risk judgment module, and a report generation module connected to the RPA scheduling and management module. The RPA scheduling and management module is used to receive leasing factoring business trigger instructions, create underlying asset verification tasks, and schedule the collaborative work of each module. The RPA robot acquisition module receives the underlying asset verification task created by the RPA scheduling and management module, logs into the official websites of the logistics company and the leasing platform, and retrieves the corresponding logistics information of the logistics company and the leasing platform based on the content of the underlying asset verification task. The preprocessing module is used to preprocess all captured logistics information and convert it into structured data to obtain structured skeleton data of the logistics information. The logistics information entity extraction module is used to receive structured skeleton data of logistics information. It calls the BERT lightweight chemical control-level large language model, which is finely tuned for the leasing and factoring logistics information scenario, through local and cloud dual interfaces. It extracts core entity information from the structured skeleton data of logistics information by combining the entity extraction prompt word template library and core entity annotation rules. The logistics information difference analysis module is used to receive the structured skeleton data of logistics information preprocessed by logistics companies and leasing platforms, and call the preset asset verification big data model to compare the differences between the structured skeleton data of logistics information of leasing platforms and the structured skeleton data of logistics information of logistics companies to obtain difference information. The indicator evaluation module is used to call the preset logistics information difference degree algorithm, the preset logistics trajectory similarity scoring algorithm, and the preset logistics information overall credibility evaluation algorithm based on the extracted core entity information and difference information to generate the logistics information overall difference degree index, logistics trajectory similarity scoring index, and logistics information overall credibility index. The risk assessment module is used to determine the risk level based on the overall credibility index of logistics information and according to preset risk assessment rules. The report generation module is used to generate a logistics information verification report based on the results generated by the above modules.

[0006] Furthermore, the RPA robot acquisition module includes a task communication unit, an RPA automated execution engine, and a verification and encryption unit, wherein: After receiving the underlying asset verification task issued by the RPA scheduling and management module, it parses out the core parameters in the underlying asset verification task and passes them to the RPA automated execution engine, and feeds back the status of the task execution to the RPA scheduling and management module. The RPA automation execution engine schedules the web page interaction component to automatically navigate to the logistics query page based on the parsed logistics company's official website address. After logging into the target logistics company's official website, the logistics information details page is loaded. The engine also schedules the data scraping component to scrape data from the logistics information details page and obtain the logistics information of the logistics company corresponding to the target task. The RPA automation execution engine also schedules the web page interaction component to initialize a new browser, navigate to the logistics information query page of the rental platform's official website, and log in to the logistics information details page of the rental platform; and schedules the data crawling component to crawl data from the logistics information details page of the rental platform to obtain the logistics information of the rental platform corresponding to the target task; The verification and encryption unit is used to verify the integrity of the logistics information captured from the logistics company and the leasing platform, and to encrypt the data to generate a hash value.

[0007] Furthermore, the preprocessing module includes an encrypted data receiving unit, an unstructured data processing unit, a multi-source structured data conversion unit, and a skeleton data verification unit, wherein: The encrypted data receiving unit is used to receive logistics information from logistics companies and leasing platforms pushed by the RPA robot collection module through a dedicated encrypted interface, and to receive the hash value corresponding to the data and perform hash value verification on the received data. The unstructured data processing unit is used to read logistics information, perform fully automated cleaning according to preset data cleaning rules, and output standardized unstructured logistics information after cleaning. The multi-source structured conversion unit is used to receive the cleaned standardized unstructured logistics information from logistics companies and leasing platforms, respectively, and then match it to the preset logistics information structured template through preset field mapping rules. It also converts the values ​​of all fields into the preset data types of the template through a data type conversion algorithm, and performs digital processing on the logistics trajectory information, integrating the scattered trajectory nodes into an array format in chronological order, and finally generating structured skeleton data of the logistics information corresponding to the logistics companies and leasing platforms respectively. The skeleton data verification unit is used to perform integrity and standardization verification on the generated structured skeleton data.

[0008] Furthermore, the logistics information entity extraction module includes a skeleton data receiving unit, a large model calling unit, and an entity extraction unit, wherein: The skeleton data receiving unit is used to receive structured skeleton data of logistics information from logistics companies and leasing platforms that has passed preprocessing and verification. The large model calling unit has a built-in local calling interface and a cloud backup interface for the asset verification large language model, and pre-configures the parameter rules for calling the asset verification large language model; the asset verification large language model is a lightweight chemical control-level logic reasoning model based on BERT and finely tuned by structured data of leasing factoring logistics information and core entity annotation data. The entity extraction unit integrates a pre-set asset verification big language model, an entity extraction prompt word template library, and core entity annotation rules. It is used to convert the structured skeleton data into a format according to the rules of the prompt word template library, input it into the asset verification big language model, trigger the model's natural language understanding and entity extraction capabilities, extract the core entity information required for the verification of lease factoring assets from the structured skeleton data, and classify and store the entities in a structured manner according to the core entity annotation rules.

[0009] Furthermore, the logistics information difference analysis module includes a skeleton data receiving unit, a large model calling unit, and a difference analysis execution unit, wherein: The skeleton data receiving unit is connected to the preprocessing module to receive structured skeleton data of logistics information from logistics companies and leasing platforms that has passed preprocessing and verification. The large model calling unit has a built-in local calling interface and a cloud backup interface for the asset verification large language model, and pre-configured parameter rules for calling the asset verification large language model. The difference analysis execution unit integrates a pre-set asset verification big language model, a dedicated prompt word template library for logistics information difference analysis, and difference labeling rules. It is used to convert and structurally concatenate the paired structured skeleton data according to the rules of the prompt word template library, and input it into the asset verification big language model. This triggers the model's natural language understanding, logical comparison, and difference recognition capabilities. It extracts the core difference information required for the verification of lease and factoring assets from the paired data and classifies and stores it in a structured manner according to the difference labeling rules.

[0010] Furthermore, the indicator evaluation module includes a logistics information difference calculation unit, a logistics trajectory similarity calculation unit, and a logistics information overall credibility calculation unit, wherein: The logistics information difference calculation unit is used to quantify the overall degree of difference between the logistics information of the leasing platform and the logistics company based on the extracted core entity information and difference information. The core entity information includes the waybill number. Delivery time Shipping location Delivery location Logistics trajectory nodes The expression is:

[0011] in, Indicates the overall degree of difference in logistics information, with a value range of [value missing]. , , , , , These represent the waybill number respectively. Delivery time Shipping location Delivery location Logistics trajectory nodes The weights are such that the sum of all weights is 1; This indicates the difference in the waybill number. This indicates the difference in delivery time. This indicates the difference in shipping location. This indicates the difference in delivery location. Indicates the difference value between logistics trajectory nodes; The logistics trajectory similarity calculation unit is used to quantitatively score the consistency of logistics trajectories. The expression is:

[0012] in, This represents the similarity score of logistics trajectories, with a value range of... , This represents the difference value between logistics trajectory nodes. To assess the overall discrepancy in logistics information, This is the overall difference correction coefficient; The overall credibility calculation unit for logistics information, based on the calculation results of the logistics information difference calculation unit and the logistics trajectory similarity calculation unit, comprehensively outputs the overall credibility of the logistics information, expressed as:

[0013] in, The overall reliability of logistics information, with a value range of [value missing]. , To assess the overall discrepancy in logistics information, The similarity score of the logistics trajectory is normalized. ; This is the comprehensive correction factor.

[0014] Furthermore, in the risk assessment module, the risk level is determined according to the overall credibility index of logistics information and preset risk assessment rules, specifically as follows: Overall reliability of logistics information A basic risk level is obtained by comparing the risk level with a preset risk level. The basic risk level includes low risk, medium risk, and high risk, and is expressed as follows: Low risk: Medium risk: High risk: ; Based on the basic risk level, a rule-matching algorithm is used in conjunction with the overall difference in logistics information. Similarity of logistics trajectories A weighted adjustment is performed, and the adjustment rules are as follows: If the basic risk level is low, but the similarity of logistics trajectories is high. If so, it is revised to medium risk; If the basic risk level is medium risk, but the overall difference in logistics information If so, it is revised to high risk.

[0015] A method for verifying the underlying assets in a leasing factoring business, applied to the aforementioned system for verifying the underlying assets in a leasing factoring business, includes: S1: Receive the leasing factoring business trigger instruction and create an underlying asset verification task; S2: Receive the created underlying asset verification task, and log in to the official websites of logistics companies and leasing platforms based on the RPA robot. Based on the content of the underlying asset verification task, retrieve the corresponding logistics information of the logistics company and the logistics information of the leasing platform. S3: Preprocess all captured logistics information and convert it into structured data to obtain structured skeleton data of logistics information; S4: Invoke the preset asset verification big language model to extract core entity information from the structured skeleton data of logistics information; S5: Call the preset asset verification big language model to compare the differences between the structured skeleton data of the logistics information of the leasing platform and the structured skeleton data of the logistics information of the logistics company, and obtain the difference information. S6: Based on the extracted core entity information and difference information, call the preset logistics information difference degree algorithm, the preset logistics trajectory similarity scoring algorithm and the preset logistics information overall credibility evaluation algorithm to generate the logistics information overall difference degree index, the logistics trajectory similarity scoring index and the logistics information overall credibility index. S7: Determine the risk level according to the preset risk assessment rules based on the overall credibility index of logistics information; S8: Generate a logistics information verification report based on the above generation results.

[0016] The principle and advantages of this invention are as follows: This invention uses the RPA scheduling and management module as the core scheduling hub, linking the RPA robot acquisition module to automatically capture dual-source logistics information from logistics companies and leasing platforms. The preprocessing module converts the unstructured logistics information into standardized structured skeleton data. Relying on the asset verification big language model, the logistics information entity extraction module extracts core entity information, and the logistics information difference analysis module identifies the core difference information between the two-source data. Then, the indicator evaluation module quantifies the logistics information difference degree, trajectory similarity, and overall credibility based on a preset algorithm. The risk judgment module combines credibility indicators and correction rules to complete the asset verification risk level judgment. Finally, the report generation module integrates the results of the entire process to generate a verification report. Through the deep integration of RPA process automation and big language model artificial intelligence technology, end-to-end automation, quantitative analysis, and intelligent risk control of the underlying asset logistics information verification in leasing factoring business are realized.

[0017] The advantages are as follows: This invention effectively solves the technical problems of low efficiency, low automation, and weak risk control capabilities in traditional leasing and factoring asset verification through manual sampling. It achieves automated collection and preprocessing of all logistics information through RPA robots, replacing manual operation, significantly improving verification efficiency and coverage, and eliminating risk risks associated with uninspected portions. It utilizes a large language model for asset verification to achieve intelligent entity extraction and precise difference analysis of logistics information, overcoming the limitations of traditional RPA which can only perform basic operations. Through a pre-set algorithm, a standardized quantitative evaluation system is constructed to achieve multi-dimensional quantification of logistics information authenticity and intelligent risk level determination, accurately locating discrepancies and completing risk classification, thereby improving the reference value and risk control accuracy of verification results. Simultaneously, relying on a unified RPA scheduling and management module, the entire verification task process is automated, reducing manual intervention, lowering the error rate, and ensuring the continuity of verification business. This provides an efficient, intelligent, and quantifiable technical solution for risk control of underlying assets in leasing and factoring businesses. Attached Figure Description

[0018] Figure 1 This is a functional block diagram of an embodiment of the present invention; Figure 2 This is a flowchart of an embodiment of the present invention. Detailed Implementation

[0019] The following detailed description illustrates the specific implementation method: The basic implementation examples are as follows: Figure 1 As shown: A system for verifying underlying assets in a leasing factoring business includes an RPA scheduling and management module and an RPA robot data acquisition module, a preprocessing module, a logistics information entity extraction module, a logistics information discrepancy analysis module, an indicator evaluation module, a risk assessment module, and a report generation module connected to the RPA scheduling and management module, wherein: The RPA scheduling management module is used to receive leasing factoring business trigger instructions, create underlying asset verification tasks, and schedule various modules to work together. In this embodiment, the RPA scheduling management module is a physical core scheduling module that deeply integrates hardware and software. It is deployed on the hardware carrier of the financial institution's distributed scheduling server and is equipped with a task triggering and creation unit, a multi-module collaborative scheduling unit, a task monitoring and management unit, a permission and rule configuration unit, and a data interaction and storage unit. It establishes bidirectional data interaction with all functional modules of the system and external business systems (domestic factoring-leasing business system, operation platform) through standardized API interfaces.

[0020] The task triggering and creation unit of the RPA scheduling and management module is developed based on the HTTP / HTTPS encrypted transmission protocol. It establishes multiple triggering interfaces with external business systems (domestic factoring-leasing business systems and operation platforms), and supports three task creation modes: scheduled triggering, manual triggering, and business linkage automatic triggering. It is used to receive leasing and factoring business triggering instructions, parse the core parameters in the instructions such as asset package information, order range, and verification priority, automatically create underlying asset verification tasks, and complete task number allocation and task attribute labeling (such as priority, business scenario, and timeout threshold) according to preset rules. The multi-module collaborative scheduling unit, as the core scheduling and execution unit of the module, has a built-in task queue scheduling algorithm and module dependency mapping table. It is used to decompose the created verification task into an ordered sub-task of "data acquisition - preprocessing - entity extraction - difference analysis - indicator evaluation - risk judgment - report generation". According to the real-time load and task priority of each functional module, it dynamically allocates sub-tasks and issues scheduling instructions. At the same time, it maintains the execution sequence of sub-tasks to ensure that the subsequent module is triggered only after the preceding module is completed, so as to realize the automated collaborative linkage of multiple modules. The task monitoring and management unit integrates a real-time monitoring engine, an exception handling engine, and a task retry mechanism to monitor the entire lifecycle of verification tasks (creation, decomposition, allocation, execution, and completion) in real time. It collects task execution status, progress, and time consumption data from each module. For tasks with execution anomalies (such as module unresponsiveness, execution timeout, or abnormal results), the retry mechanism is automatically triggered (3 times by default, configurable). If a retry fails, an exception is marked and an alert is pushed. It also provides task query, manual intervention (pause / continue / terminate), and statistical analysis functions, supporting precise queries by task number, order number, execution status, and other dimensions. The permission and rule configuration unit provides a visual permission management interface and a scheduling rule configuration interface. It has a built-in role permission library, scheduling rule library, and timeout threshold library. It supports configuring differentiated operation permissions by "administrator-risk control specialist-operation personnel" (e.g., administrators can configure scheduling rules, while risk control specialists can only query tasks). It supports custom configuration of scheduling parameters such as task decomposition rules, module load thresholds, retry counts, and timeout time. All configuration changes are automatically synchronized to the scheduling server cache, and operation logs are recorded throughout the process to achieve configuration traceability. The data interaction and storage unit is used to receive the execution results and status data fed back by each functional module, and to associate and store them according to the task number to form a full-process scheduling log; at the same time, it pushes the overall progress and final results of the verification task to the external business system to provide data support for business decision-making; the stored data includes basic task information, scheduling instruction records, module execution feedback, exception handling records, etc., supports permanent traceability and auditing, and is synchronized to the blockchain evidence storage node to ensure that the data is tamper-proof.

[0021] The RPA robot data acquisition module receives the underlying asset verification task created by the RPA scheduling and management module, and logs into the official websites of the logistics company and the leasing platform. Based on the content of the underlying asset verification task, it retrieves the corresponding logistics information from the logistics company and the leasing platform. In this embodiment, the RPA robot data acquisition module is deployed on a Windows virtual machine / industrial computer hardware platform, equipped with a task communication unit, an RPA automation execution engine, web page interaction components, data capture components, and a verification and encryption unit. It establishes bidirectional data interaction with the RPA scheduling and management module through a standardized API interface, wherein: The task communication unit is developed based on the HTTP / HTTPS encrypted transmission protocol. After receiving the underlying asset verification task issued by the RPA scheduling and management module, it parses out the core parameters in the underlying asset verification task, including the order number, waybill number, logistics company website address, leasing platform website logistics query address, collection time threshold, and data storage path, and transmits them to the RPA automated execution engine. It also feeds back the status of the task execution to the RPA scheduling and management module. The RPA automation execution engine schedules the web page interaction component to automatically navigate to the logistics query page based on the parsed logistics company's official website address. After logging into the target logistics company's official website, the logistics information details page is loaded. The engine also schedules the data crawling component to crawl data from the logistics information details page to obtain the logistics information of the logistics company corresponding to the target task. Specifically, the RPA automation execution engine first schedules the web page interaction component to initialize the browser kernel, automatically navigate to the logistics query page based on the parsed logistics company's official website address, and start page loading monitoring. If the logistics tracking page requires account and password login, the web page interaction component uses an element locator to locate the username, password input fields, and login button, automatically fills in the pre-configured compliant login information, and completes identity verification by clicking the login button. If a CAPTCHA verification is encountered, the CAPTCHA automatic recognition module of the exception handling component calls a lightweight OCR recognition model based on CRNN+CTC. This model has been trained on over 100,000 sets of CAPTCHA images from the logistics industry, achieving a high recognition accuracy. The OCR recognition model completes automatic recognition and input of the verification code. If the recognition fails, it triggers 3 automatic retries. If the retries fail, it reports the collection anomaly to the RPA scheduling and management module. The RPA scheduling and management module marks the task as "abnormal and pending" and pushes it to the risk control personnel. After successful login, the web page interactive component locates the waybill number input box, automatically fills in the parsed waybill number, and clicks the query button to load the logistics information details page; The RPA automated execution engine schedules a data scraping component to scrape data from logistics information detail pages: For structured table-based logistics information, such as waybill information and shipment information, it directly extracts all fields from the table; for unstructured text-based logistics trajectories, such as "collected at XX time - delivered at XX time", it extracts the trajectory text by splitting it according to time nodes; for logistics node identifiers on the page, such as collection, transit, and delivery, it completes accurate extraction through a node data matching plugin.

[0022] The above describes the process by which the RPA automated execution engine retrieves logistics information from logistics companies. Similarly, the same method is used to retrieve logistics information from the official website of the leasing platform. The RPA automated execution engine also schedules the web page interaction component to initialize a new browser, navigate to the logistics information query page of the leasing platform's official website, and log in to the logistics information details page of the leasing platform; and schedules the data retrieval component to retrieve data from the logistics information details page of the leasing platform to obtain the logistics information of the leasing platform corresponding to the target task. The verification and encryption unit is used to verify the integrity of the logistics information from the logistics company and the leasing platform. Verification dimensions include whether the waybill number, shipping time, shipping location, delivery location, and at least one logistics trajectory node have been captured. If core information is missing, the data collection is deemed a failure. The unit then performs encryption processing to generate a data hash value. Specifically, the verification and encryption unit of the RPA robot collection module, equipped with a local temporary storage component, encrypts and temporarily stores the captured logistics information from the logistics company in the format of waybill number-data type-collection time, while simultaneously generating a data hash value to ensure the data has not been tampered with. Similarly, the local temporary storage component encrypts and temporarily stores the captured logistics information from the leasing platform in the format of order number-waybill number-data type-collection time, establishing a connection with the logistics information from the logistics company, and simultaneously generating a data hash value.

[0023] In this embodiment, all web page access and data crawling operations in the above-mentioned units comply with the target website's robots protocol and relevant laws and regulations on web crawling. By setting reasonable page access intervals and avoiding high-frequency requests, the compliance of the collection behavior is ensured. At the same time, the crawled non-public data is only used for verifying the underlying assets of leasing and factoring, and is not used for other purposes.

[0024] The preprocessing module is used to preprocess all captured logistics information and convert it into structured data to obtain structured skeleton data of the logistics information. In this embodiment, it includes an encrypted data receiving unit, an unstructured data processing unit, a multi-source structured conversion unit, and a skeleton data verification unit, wherein: The encrypted data receiving unit is used to receive logistics information from logistics companies and leasing platforms pushed by the RPA robot collection module through a dedicated encrypted interface, and to receive the hash value corresponding to the data and perform hash value verification on the received data; if the verification is successful, it will send a message to the RPA robot collection module that the data was successfully received; if the verification fails, it will send a message that the data was received abnormally and trigger a data retransmission mechanism; the received data is temporarily stored in the server's high-speed cache area to provide fast reading support for subsequent processing.

[0025] The unstructured data processing unit is used to read logistics information, perform fully automated data cleaning according to preset data cleaning rules, and output standardized unstructured logistics information after cleaning; specifically: For logistics information in webpage source code format, a text parser is used to extract the core plain text content and remove noisy data such as HTML tags and webpage advertisements; For logistics tracking information in plain text format, segment and split it according to time sequence, and remove meaningless interjections and modifiers; Use the waybill number + order number as a unique identifier to eliminate duplicate logistics information data; The missing core fields are standardized and marked. After cleaning, the cleaned standardized unstructured logistics information is generated and transmitted to the multi-source structured conversion unit.

[0026] The multi-source structured data conversion unit receives standardized unstructured logistics information from logistics companies and leasing platforms after cleaning. It then matches this information to preset logistics information structured templates using pre-defined field mapping rules. A data type conversion algorithm converts all field values ​​to the template's preset data types. The unit also digitizes the logistics trajectory information, integrating scattered trajectory nodes into an array format in chronological order. Finally, it generates structured skeleton data for the logistics information corresponding to the logistics companies and leasing platforms. The logistics information structured template includes four core fields: basic identifier field, logistics subject field, logistics spatiotemporal field, and logistics trajectory field. The specific content is as follows: Basic identification fields include order number, waybill number, and data source; logistics entity fields include shipper name, consignee name, and logistics company name; logistics spatiotemporal fields include shipping time, shipping location, delivery time, and delivery location; logistics trajectory fields include the number of trajectory nodes, the latest trajectory node, and complete trajectory information.

[0027] Next, through a data type conversion algorithm, the values ​​of all fields are converted into the data types preset by the template, such as converting text-type time into timestamps and non-standardized locations into the "province-city-district / county" format; The logistics trajectory information is then processed into an array, which integrates the scattered trajectory nodes into an array format according to the time sequence and matches them to the "complete trajectory information" field. After completing the matching and transformation of all fields, structured skeleton data of logistics information is generated. Each piece of logistics information corresponds to a unique structured skeleton data, and the logistics information of the logistics company and the leasing platform are linked through waybill number + order number.

[0028] The skeleton data verification unit is used to perform integrity and standardization checks on the generated structured skeleton data. Integrity checks include verifying whether core fields such as basic identifier fields, logistics subject fields, and logistics spatiotemporal fields are missing in the structured skeleton data. If core fields are missing, they are marked as verification anomalies. Standardization checks include verifying whether the data type and format of each field conform to preset rules, such as whether the timestamp format is correct, whether the location is at the "province-city-district / county" level, and whether the number is a positive integer. If the format does not conform, it is marked as verification anomalies. The validated structured skeleton data is pushed to the logistics information entity extraction module and the logistics information difference analysis module through a dedicated interface in a standardized JSON / CSV format.

[0029] The logistics information entity extraction module receives structured skeleton data of logistics information and calls a preset asset verification big data language model to extract core entity information from the structured skeleton data of logistics information. The logistics information entity extraction module includes a skeleton data receiving unit, a big data model calling unit, and an entity extraction unit, wherein: The skeleton data receiving unit is used to receive structured skeleton data of logistics information from logistics companies and leasing platforms that has passed preprocessing and verification. The large model calling unit has a built-in local calling interface and a cloud backup interface for the asset verification large language model, and pre-configured parameter rules for calling the asset verification large language model. Among them, the preset asset verification large language model is a lightweight, chemical control-level logical reasoning large model adapted to financial logistics text analysis. For example, it uses the BERT model as the base model and fine-tunes the training for leasing and factoring logistics information scenarios. The training dataset includes structured data of logistics information from various logistics companies, logistics push data from leasing platforms, and core entity annotation data for leasing and factoring asset verification. The model has accurate entity recognition and semantic understanding capabilities in the financial logistics field. The model deployment adopts a dual mode of local lightweight deployment + cloud cluster deployment, such as using existing model pruning for lightweighting. The local model meets the needs of fast inference, while the cloud model meets the needs of large-scale, high-concurrency entity extraction. The model calling interface is compatible with the Spring AI Alibaba open source project and is seamlessly integrated with the overall system technical architecture.

[0030] The parameter rules for the call are automatically configured based on the data source and data volume. For example, for single / small batch data, the local lightweight model is called, with the inference temperature set to 0.1 and the maximum generation length set to 512 to ensure the accuracy of the extraction results. For large batch / high-concurrency data, the cloud cluster model is called, with the inference temperature set to 0.2 and the maximum generation length set to 1024 to improve extraction efficiency while ensuring accuracy, and at the same time, the corresponding entity extraction prompt word template is loaded.

[0031] The entity extraction unit integrates a pre-defined asset verification language model, a prompt word template library for entity extraction, and core entity annotation rules. It converts structured skeleton data according to the rules of the prompt word template library, inputs it into the asset verification language model, triggers the model's natural language understanding and entity extraction capabilities, extracts the core entity information required for leasing and factoring asset verification from the structured skeleton data, and classifies and stores the entities according to the core entity annotation rules. The core entity extraction scope includes identifier entities, subject entities, spatiotemporal entities, and trajectory entities. Specific entities extracted from identifier entities include order numbers and waybill numbers. The entity annotation rule is to annotate the original characters completely, without abbreviations or modifications. Specific entities extracted from subject entities... The entity extraction includes the shipper's name, the consignee's name, and the logistics company's name. The entity labeling rule is to label the complete official name, excluding abbreviations and aliases. The specific entities extracted for spatiotemporal entities include the shipping time, shipping location, delivery time, delivery location, and trajectory node time. The entity labeling rule is to label the shipping and delivery times as timestamps, label the location in a three-level format of province-city-district / county, and label the trajectory node time in the same format as the main time. The specific entities extracted for trajectory entities include logistics trajectory nodes, the latest trajectory status, and the number of trajectory nodes. The entity labeling rule is to label the trajectory nodes in chronological order in the format of "node name + time", label the latest trajectory status as a standardized status (such as pickup, transit, delivery, and receipt), and label the number of trajectory nodes as a positive integer.

[0032] The pre-defined entity extraction prompt template library uses standardized instruction formats to clearly inform the large language model of the extraction task, extraction scope, entity classification rules, and output format. A core template example is shown below: "As the expert in extracting logistics information entities for leasing and factoring, please extract core entity information from the following structured skeleton data of logistics information. The extraction scope includes four categories of entities: identifier, subject, spatiotemporal, and trajectory. Strictly follow the format of "entity category - entity name - entity attribute" for output. Entity attributes must conform to the preset labeling rules. If there are no related entities, label them as "none". Do not add any additional descriptive information: {structured skeleton data of logistics information}".

[0033] Meanwhile, to address the differences in structured skeleton data between logistics companies and leasing platforms, dedicated prompt word templates are pre-set to achieve differentiated and accurate extraction.

[0034] The above describes the process of extracting logistics information entities. Subsequently, the analysis process of logistics information difference is performed. The logistics information difference analysis module receives the structured skeleton data of the logistics information after preprocessing by the logistics company and the leasing platform, and calls the preset asset verification big language model to compare the differences between the structured skeleton data of the leasing platform's logistics information and the structured skeleton data of the logistics company's logistics information to obtain the difference information. In the logistics information discrepancy analysis module, its unit composition shares some elements with the logistics information entity extraction module, such as the skeleton data receiving unit and the large model scheduling unit. The different unit is the discrepancy analysis execution unit, specifically: The skeleton data receiving unit is connected to the preprocessing module to receive structured skeleton data of logistics information from logistics companies and leasing platforms that has passed preprocessing and verification. The large model scheduling unit has built-in local calling interfaces and cloud backup interfaces for the asset verification large language model, and pre-configured parameter rules for calling the asset verification large language model. The called asset verification large language model is a lightweight, chemical control-level logic reasoning model that has been fine-tuned for scenarios involving differences in logistics information between leasing and factoring. Based on general financial logistics text analysis capabilities, it adds specialized capabilities such as accurate comparison of logistics information fields, time-by-time matching of trajectory nodes, and semantic consistency judgment of logistics status. The training dataset includes real logistics difference cases between logistics companies and leasing platforms, labeled data for comparing logistics information fields, and labeled data for matching logistics trajectory nodes. Model deployment adopts a combination of local lightweight dedicated deployment and cloud cluster backup deployment. The local model optimizes the inference logic for logistics difference analysis scenarios, reducing response latency. The cloud-based model meets the needs of large-scale, high-concurrency differential analysis, and the model calling interface is compatible with the Spring AI Alibaba open-source project, seamlessly integrating with the overall system technical architecture.

[0035] The difference analysis execution unit integrates a pre-set asset verification big language model, a dedicated prompt word template library for logistics information difference analysis, and difference annotation rules. It is used to convert and structurally concatenate paired structured skeleton data according to the rules of the prompt word template library, inputting it into the asset verification big language model. This triggers the model's natural language understanding, logical comparison, and difference recognition capabilities, extracting the core difference information required for leasing and factoring asset verification from the paired data, and classifying and structurally storing it according to the difference annotation rules. Specifically, the core difference analysis scope for logistics information includes basic identifiers, logistics spatiotemporal categories, and logistics trajectory categories. The specific analysis fields for basic identifiers include order number and waybill number. The difference annotation rule is to annotate "one..." For discrepancies, both the platform value and the logistics company value must be labeled when there is a discrepancy. Specific analysis fields for logistics time-space categories include shipping time, shipping location, current location, and estimated delivery time. The discrepancy labeling rule is to label "consistent / inconsistent / no data," with both platform and logistics company values ​​labeled when there is a discrepancy. Time-related discrepancies require labeling the time difference, while location-related discrepancies require labeling the geographical level deviation (province / city / district). Specific analysis fields for logistics trajectory categories include trajectory node time, trajectory node address, and trajectory node status. The discrepancy labeling rule is to label each node sequentially in chronological order as "consistent / inconsistent / no data for platform / no data for logistics company," with details of the platform and logistics company trajectories displayed when there is a discrepancy, and the total number of trajectory node discrepancies is also counted.

[0036] The pre-defined prompt word template library for logistics information difference analysis adopts an instructional and structured format, clearly informing the large language model of the difference analysis task, analysis scope, annotation rules, and output format. It forces the model to perform comparative analysis solely based on the input data, without adding additional explanations or subjective judgments. The core template is fixed in the following format, with only paired data variables being dynamically replaced: "You are a dedicated tool for analyzing the differences in logistics information in leasing and factoring businesses. Your sole task is to compare and analyze the differences between the structured skeleton data of logistics information input from the leasing platform and the structured skeleton data of logistics information input from the logistics company. The analysis scope includes basic identifiers (order number, waybill number), logistics time and space (delivery time, delivery location, current location, estimated delivery time), and logistics trajectory (trajectory node time, trajectory node address, trajectory node status)."

[0037] Output rules: The results were labeled with differences by category and field by field, and the results were only "consistent / inconsistent / no data"; When there are inconsistencies, the specific difference value must be marked: platform value = XXX, logistics company value = XXX; time-type fields should be marked with time difference, location-type fields should be marked with geographical level, and trajectory-type fields should be compared node by node; The trajectory-type fields are output as comparison results node by node in chronological order, and the total number of differences between trajectory nodes is counted. Only output the difference information, without adding any additional descriptions, explanations, or subjective judgments; The output format is a JSON string, and the fields are consistent with the preset difference analysis categories, with no nesting.

[0038] Logistics information for the rental platform: {Platform structured skeleton data} Logistics company logistics information: {structured skeleton data of the logistics company}.

[0039] The core entity information extracted by the logistics information entity extraction module and the difference information obtained by the logistics information difference analysis module are transmitted to the indicator evaluation module for indicator evaluation. Specifically: The indicator evaluation module is used to call preset logistics information difference degree algorithms, preset logistics trajectory similarity scoring algorithms, and preset logistics information overall credibility evaluation algorithms based on the extracted core entity information and difference information, to generate logistics information overall difference degree indicators, logistics trajectory similarity scoring indicators, and logistics information overall credibility indicators. In this embodiment, the indicator evaluation module includes a logistics information difference degree calculation unit, a logistics trajectory similarity calculation unit, and a logistics information overall credibility calculation unit, wherein: The logistics information difference calculation unit is used to quantify the overall degree of difference between the logistics information of the leasing platform and the logistics company based on the extracted core entity information and difference information. The core entity information includes the waybill number. Delivery time Shipping location Delivery location Logistics trajectory nodes The expression is:

[0040] in, Indicates the overall degree of difference in logistics information, with a value range of [value missing]. , , , , , These represent the waybill number respectively. Delivery time Shipping location Delivery location Logistics trajectory nodes The weights are such that the sum of all weights equals 1, for example... , , , , ; This indicates the difference in waybill numbers; if they match, then... If they are inconsistent ; This represents the difference in delivery time, where:

[0041] in, The default time is 24 hours. If the time exceeds 24 hours, then... ; The delivery time is the logistics information for the rental platform. This refers to the shipping time as per the logistics company's logistics information.

[0042] This indicates the difference in shipping location. The value represents the difference in delivery location, calculated based on the geographical level of matching. A value of 0 indicates a perfect match at the provincial / municipal / district level, 0.3 indicates a match at two levels, 0.6 indicates a match at the first level, and 1 indicates a complete mismatch. This represents the difference value between logistics trajectory nodes, where:

[0043] in, This refers to the number of track nodes shared by the leasing platform and the logistics company. , This refers to the total number of tracking nodes for leasing platforms and logistics companies.

[0044] The logistics trajectory similarity calculation unit is used to quantitatively score the consistency of logistics trajectories. The expression is:

[0045] in, This represents the similarity score of logistics trajectories, with a value range of... , This represents the difference value between logistics trajectory nodes. To assess the overall discrepancy in logistics information, The overall difference correction coefficient is preset to 0.3; The overall credibility calculation unit for logistics information, based on the calculation results of the logistics information difference calculation unit and the logistics trajectory similarity calculation unit, comprehensively outputs the overall credibility of the logistics information, expressed as:

[0046] in, The overall reliability of logistics information, with a value range of [value missing]. , To assess the overall discrepancy in logistics information, The similarity score of the logistics trajectory is normalized. ; The comprehensive correction coefficient is preset to 0.5 to achieve a balanced evaluation of overall difference and trajectory similarity.

[0047] The risk assessment module is used to determine the risk level based on the overall credibility index of logistics information according to preset risk assessment rules; specifically, determining the risk level based on the overall credibility index of logistics information according to preset risk assessment rules involves: Overall reliability of logistics information A basic risk level is obtained by comparing the risk level with a preset risk level. The basic risk level includes low risk, medium risk, and high risk, and is expressed as follows: Low risk: This indicates high accuracy of logistics information, no significant risk control risks to the underlying assets, and the pre-set disposal recommendation is to directly include them in the factoring asset package without manual intervention, upon verification; Medium risk: This indicates that there are some non-core discrepancies in the logistics information, and the underlying assets have potential risk control risks. The default handling suggestion is to manually review the reasons for the discrepancies. If the review is correct, the information will be verified. If the review is abnormal, it will be marked as high risk. High risk: This indicates that there are core discrepancies in the logistics information, the authenticity of the underlying assets is questionable, and there are significant risk control risks. The pre-set disposal recommendation is to directly reject the verification application, exclude it from the factoring asset package, issue a risk warning to the leasing platform, and verify other orders on the platform. In addition, the above risk classification also sets differentiated credibility thresholds based on order attributes, including: Order amount dimension: For large orders (such as the rental amount of a single piece of equipment) (10,000 yuan), raising the low-risk threshold to Raise the verification standards for large orders; for small orders (such as the rental amount of a single mobile phone)... (RMB), maintaining the default threshold to balance risk control accuracy and business efficiency; Platform rating: For high-quality partner platforms (risk control rating A), the medium-risk threshold will be lowered to [missing information]. Appropriately lower the verification standards; for risky cooperation platforms (risk control rating C), raise the low-risk threshold to [missing information]. Strengthen risk control and review; Based on the basic risk level, a rule-matching algorithm is used in conjunction with the overall difference in logistics information. Similarity of logistics trajectories A weighted adjustment is performed, and the adjustment rules are as follows: If the basic risk level is low, but the similarity of logistics trajectories is high. If so, it is revised to medium risk; If the basic risk level is medium risk, but the overall difference in logistics information If so, it is revised to high risk.

[0048] The final output includes risk level results, early warning information, full data of quantitative indicators, and basic order information. The output information is encapsulated in a standardized JSON format, with encapsulated fields including order number, waybill number, overall credibility of logistics information, risk level, early warning level, reason for early warning, preset handling suggestions, and judgment time.

[0049] The data generated above is used by the report generation module to generate a logistics information verification report.

[0050] Compared with existing RPA+large model financial risk control technologies, the substantial innovation of this invention lies in: 1. Design a dual-source logistics information structured skeleton data template for the scenario of verifying logistics information in leasing and factoring, so as to realize the standardized matching of data between logistics companies and leasing platforms and solve the problem of inconsistent data formats in existing technologies that cannot be directly compared; 2. A weighted logistics information difference algorithm is proposed, which sets differentiated weights for core entities such as waybill number and logistics trajectory, and quantifies differences more accurately than the binary comparison of existing technologies; 3. Design a credibility-based risk level correction rule, which combines the similarity of logistics trajectories and the overall difference to make a weighted correction to the basic risk level, thereby solving the problem of the single-mindedness of existing technology risk assessment; 4. Achieve deep integration of the compliant data collection mechanism of RPA robots with the scenario-based fine-tuning of large language models, which significantly improves the verification efficiency and accuracy compared to the simple splicing of existing technologies.

[0051] like Figure 2 As shown, in another embodiment of this example, a method for verifying the underlying assets of a lease factoring business is also included, applied to the aforementioned system for verifying the underlying assets of a lease factoring business, comprising: S1: Receive the leasing factoring business trigger instruction and create an underlying asset verification task; S2: Receive the created underlying asset verification task, and log in to the official websites of logistics companies and leasing platforms based on the RPA robot. Based on the content of the underlying asset verification task, retrieve the corresponding logistics information of the logistics company and the logistics information of the leasing platform. S3: Preprocess all captured logistics information and convert it into structured data to obtain structured skeleton data of logistics information; S4: Invoke the preset asset verification big language model to extract core entity information from the structured skeleton data of logistics information; S5: Call the preset asset verification big language model to compare the differences between the structured skeleton data of the logistics information of the leasing platform and the structured skeleton data of the logistics information of the logistics company, and obtain the difference information. S6: Based on the extracted core entity information and difference information, call the preset logistics information difference degree algorithm, the preset logistics trajectory similarity scoring algorithm and the preset logistics information overall credibility evaluation algorithm to generate the logistics information overall difference degree index, the logistics trajectory similarity scoring index and the logistics information overall credibility index. S7: Determine the risk level according to the preset risk assessment rules based on the overall credibility index of logistics information; S8: Generate a logistics information verification report based on the above generation results.

[0052] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A system for verifying underlying assets in a leasing factoring business, characterized in that: It includes an RPA scheduling and management module and an RPA robot acquisition module, a preprocessing module, a logistics information entity extraction module, a logistics information difference analysis module, an indicator evaluation module, a risk assessment module, and a report generation module connected to the RPA scheduling and management module. The RPA scheduling and management module is used to receive leasing factoring business trigger instructions, create underlying asset verification tasks, and schedule the collaborative work of each module. The RPA robot acquisition module receives the underlying asset verification task created by the RPA scheduling and management module, logs into the official websites of the logistics company and the leasing platform, and retrieves the corresponding logistics information of the logistics company and the leasing platform based on the content of the underlying asset verification task. The preprocessing module is used to preprocess all captured logistics information and convert it into structured data to obtain structured skeleton data of the logistics information. The logistics information entity extraction module is used to receive structured skeleton data of logistics information. It calls the BERT lightweight chemical control-level large language model, which is finely tuned for the leasing and factoring logistics information scenario, through local and cloud dual interfaces. It extracts core entity information from the structured skeleton data of logistics information by combining the entity extraction prompt word template library and core entity annotation rules. The logistics information difference analysis module is used to receive the structured skeleton data of logistics information preprocessed by logistics companies and leasing platforms, and call the preset asset verification big data model to compare the differences between the structured skeleton data of logistics information of leasing platforms and the structured skeleton data of logistics information of logistics companies to obtain difference information. The indicator evaluation module is used to call the preset logistics information difference degree algorithm, the preset logistics trajectory similarity scoring algorithm, and the preset logistics information overall credibility evaluation algorithm based on the extracted core entity information and difference information to generate the logistics information overall difference degree index, logistics trajectory similarity scoring index, and logistics information overall credibility index. The risk assessment module is used to determine the risk level based on the overall credibility index of logistics information and according to preset risk assessment rules. The report generation module is used to generate a logistics information verification report based on the results generated by the above modules.

2. The underlying asset verification system for leasing factoring business according to claim 1, characterized in that: The RPA robot data acquisition module includes a task communication unit, an RPA automated execution engine, and a verification and encryption unit, wherein: After receiving the underlying asset verification task issued by the RPA scheduling and management module, it parses out the core parameters in the underlying asset verification task and passes them to the RPA automated execution engine, and feeds back the status of the task execution to the RPA scheduling and management module. The RPA automation execution engine schedules the web page interaction component to automatically navigate to the logistics query page based on the parsed logistics company's official website address. After logging into the target logistics company's official website, the logistics information details page is loaded. The engine also schedules the data scraping component to scrape data from the logistics information details page and obtain the logistics information of the logistics company corresponding to the target task. The RPA automation execution engine also schedules the web page interaction component to initialize a new browser, navigate to the logistics information query page of the rental platform's official website, and log in to the logistics information details page of the rental platform; and schedules the data crawling component to crawl data from the logistics information details page of the rental platform to obtain the logistics information of the rental platform corresponding to the target task; The verification and encryption unit is used to verify the integrity of the logistics information captured from the logistics company and the leasing platform, and to encrypt the data to generate a hash value.

3. The underlying asset verification system for leasing factoring business according to claim 2, characterized in that: The preprocessing module includes an encrypted data receiving unit, an unstructured data processing unit, a multi-source structured data conversion unit, and a skeleton data verification unit, wherein: The encrypted data receiving unit is used to receive logistics information from logistics companies and leasing platforms pushed by the RPA robot collection module through a dedicated encrypted interface, and to receive the hash value corresponding to the data and perform hash value verification on the received data. The unstructured data processing unit is used to read logistics information, perform fully automated cleaning according to preset data cleaning rules, and output standardized unstructured logistics information after cleaning. The multi-source structured conversion unit is used to receive the cleaned standardized unstructured logistics information from logistics companies and leasing platforms, respectively, and then match it to the preset logistics information structured template through preset field mapping rules. It also converts the values ​​of all fields into the preset data types of the template through a data type conversion algorithm, and performs digital processing on the logistics trajectory information, integrating the scattered trajectory nodes into an array format in chronological order, and finally generating structured skeleton data of the logistics information corresponding to the logistics companies and leasing platforms respectively. The skeleton data verification unit is used to perform integrity and standardization verification on the generated structured skeleton data.

4. The underlying asset verification system for leasing factoring business according to claim 3, characterized in that: The logistics information entity extraction module includes a skeleton data receiving unit, a large model calling unit, and an entity extraction unit, wherein: The skeleton data receiving unit is used to receive structured skeleton data of logistics information from logistics companies and leasing platforms that has passed preprocessing and verification. The large model calling unit has a built-in local calling interface and a cloud backup interface for the asset verification large language model, and pre-configures the parameter rules for calling the asset verification large language model; the asset verification large language model is a lightweight chemical control-level logic reasoning model based on BERT and finely tuned by structured data of leasing factoring logistics information and core entity annotation data. The entity extraction unit integrates a pre-set asset verification big language model, an entity extraction prompt word template library, and core entity annotation rules. It is used to convert the structured skeleton data into a format according to the rules of the prompt word template library, input it into the asset verification big language model, trigger the model's natural language understanding and entity extraction capabilities, extract the core entity information required for the verification of lease factoring assets from the structured skeleton data, and classify and store the entities in a structured manner according to the core entity annotation rules.

5. The underlying asset verification system for leasing factoring business according to claim 4, characterized in that: The logistics information discrepancy analysis module includes a skeleton data receiving unit, a large model calling unit, and a discrepancy analysis execution unit, wherein: The skeleton data receiving unit is connected to the preprocessing module to receive structured skeleton data of logistics information from logistics companies and leasing platforms that has passed preprocessing and verification. The large model calling unit has a built-in local calling interface and a cloud backup interface for the asset verification large language model, and pre-configured parameter rules for calling the asset verification large language model. The difference analysis execution unit integrates a pre-set asset verification big language model, a dedicated prompt word template library for logistics information difference analysis, and difference labeling rules. It is used to convert and structurally concatenate the paired structured skeleton data according to the rules of the prompt word template library, and input it into the asset verification big language model. This triggers the model's natural language understanding, logical comparison, and difference recognition capabilities. It extracts the core difference information required for the verification of lease and factoring assets from the paired data and classifies and stores it in a structured manner according to the difference labeling rules.

6. The underlying asset verification system for leasing factoring business according to claim 5, characterized in that: The indicator evaluation module includes a logistics information difference calculation unit, a logistics trajectory similarity calculation unit, and a logistics information overall credibility calculation unit, wherein: The logistics information difference calculation unit is used to quantify the overall degree of difference between the logistics information of the leasing platform and the logistics company based on the extracted core entity information and difference information. The core entity information includes the waybill number. Delivery time Shipping location Delivery location Logistics trajectory nodes The expression is: in, Indicates the overall degree of difference in logistics information, with a value range of [value missing]. , , , , , These represent the waybill number respectively. Delivery time Shipping location Delivery location Logistics trajectory nodes The weights are such that the sum of all weights is 1; This indicates the difference in the waybill number. This indicates the difference in delivery time. This indicates the difference in shipping location. This indicates the difference in delivery location. Indicates the difference value between logistics trajectory nodes; The logistics trajectory similarity calculation unit is used to quantitatively score the consistency of logistics trajectories. The expression is: in, This represents the similarity score of logistics trajectories, with a value range of... , This represents the difference value between logistics trajectory nodes. To assess the overall discrepancy in logistics information, This is the overall difference correction coefficient; The overall credibility calculation unit for logistics information, based on the calculation results of the logistics information difference calculation unit and the logistics trajectory similarity calculation unit, comprehensively outputs the overall credibility of the logistics information, expressed as: in, The overall reliability of logistics information, with a value range of [value missing]. , To assess the overall discrepancy in logistics information, The similarity score of the logistics trajectory is normalized. ; This is the comprehensive correction factor.

7. The underlying asset verification system for leasing factoring business according to claim 6, characterized in that: In the risk assessment module, the risk level is determined according to the overall credibility index of logistics information and preset risk assessment rules, specifically as follows: Overall reliability of logistics information A basic risk level is obtained by comparing the risk level with a preset risk level. The basic risk level includes low risk, medium risk, and high risk, and is expressed as follows: Low risk: Medium risk: High risk: ; Based on the basic risk level, a rule-matching algorithm is used in conjunction with the overall difference in logistics information. Similarity of logistics trajectories A weighted adjustment is performed, and the adjustment rules are as follows: If the basic risk level is low, but the similarity of logistics trajectories is high. If so, it is revised to medium risk; If the basic risk level is medium risk, but the overall difference in logistics information If so, it is revised to high risk.

8. A method for verifying underlying assets in a leasing factoring business, applied to the underlying asset verification system for a leasing factoring business as described in any one of claims 1-7, characterized in that: include: S1: Receive the leasing factoring business trigger instruction and create an underlying asset verification task; S2: Receive the created underlying asset verification task, and log in to the official websites of logistics companies and leasing platforms based on the RPA robot. Based on the content of the underlying asset verification task, retrieve the corresponding logistics information of the logistics company and the logistics information of the leasing platform. S3: Preprocess all captured logistics information and convert it into structured data to obtain structured skeleton data of logistics information; S4: Invoke the preset asset verification big language model to extract core entity information from the structured skeleton data of logistics information; S5: Call the preset asset verification big language model to compare the differences between the structured skeleton data of the logistics information of the leasing platform and the structured skeleton data of the logistics information of the logistics company, and obtain the difference information. S6: Based on the extracted core entity information and difference information, call the preset logistics information difference degree algorithm, the preset logistics trajectory similarity scoring algorithm and the preset logistics information overall credibility evaluation algorithm to generate the logistics information overall difference degree index, the logistics trajectory similarity scoring index and the logistics information overall credibility index. S7: Determine the risk level according to the preset risk assessment rules based on the overall credibility index of logistics information; S8: Generate a logistics information verification report based on the above generation results.

Images