Ai-based expense reimbursement voucher information processing method and system

Abstract

The application provides an AI-based expense reimbursement voucher information processing method and system, and relates to the technical field of data processing.The application constructs an automatic processing flow integrating intelligent identification, dynamic mapping, automatic checking and closed-loop early warning, effectively solves the technical problems of low efficiency, high error rate, great compliance risk and system rigidity caused by manual processing of unstructured vouchers in the existing expense reimbursement system, realizes automation, precision, compliance and intelligence of expense reimbursement voucher processing, and reduces manual cost and financial risk control.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for processing expense reimbursement voucher information based on AI (Artificial Intelligence). Background Technology

[0002] In expense reimbursement management scenarios for companies such as financial leasing firms, with the widespread adoption of paperless office practices, employees typically need to submit electronic invoices and related payment vouchers online when submitting expense reports. These vouchers may include screenshots of payments made through third-party payment platforms like Alipay and WeChat, as well as menu images for entertainment expenses. These attachments are crucial for verifying the authenticity and compliance of the reimbursement process and require consistency checks against the invoice information.

[0003] Currently, mainstream corporate expense reimbursement systems have largely digitized the reimbursement process. However, the extraction of information from unstructured data such as payment screenshots and hospitality menus still relies on manual input, and the verification process lacks automated means. This results in low reimbursement efficiency, data errors, and insufficient compliance risk control, failing to meet the needs of enterprises for refined expense control management. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-based method and system for processing expense reimbursement voucher information, so as to alleviate the problems of low reimbursement efficiency, data error, and insufficient compliance risk control in the existing technology.

[0005] In a first aspect, the present invention provides an AI-based method for processing expense reimbursement voucher information, including: When a reimbursement voucher to be processed is obtained, the AI ​​recognition interface is called to identify the information in the voucher and obtain the target recognition information; the reimbursement voucher includes an image of a hospitality menu. Based on dynamically configured field mapping rules, target identification information is mapped to the fields corresponding to the reimbursement details to obtain target reimbursement details information; The system automatically verifies the target expense details, including consistency checks between menu amounts and invoice amounts, as well as compliance checks based on the list of prohibited items for expense reimbursement. When an abnormal event is detected during the processing of reimbursement vouchers, a corresponding warning message is triggered so that the relevant user can handle and review the abnormality, and the processing status of the reimbursement voucher is updated based on the feedback results of the warning message. The abnormal events include information identification abnormalities, amount matching abnormalities, or compliance abnormalities.

[0006] In an optional implementation, the reimbursement voucher also includes a payment voucher image; by calling an AI recognition interface to perform information recognition on the reimbursement voucher, target recognition information is obtained, including: Generate a voucher identifier and voucher type recognition request with reimbursement voucher; wherein, voucher type includes payment voucher corresponding to payment voucher image and hospitality menu corresponding to hospitality menu image; Send the recognition request to the AI ​​recognition interface to recognize the information of the corresponding voucher type on the reimbursement voucher; Receive the raw recognition information returned by the AI ​​recognition interface, and standardize the raw recognition information to obtain the target recognition information.

[0007] In an optional implementation, based on dynamically configured field mapping rules, the target identification information is mapped to the fields corresponding to the expense reimbursement details to obtain the target expense reimbursement detail information, including: Retrieve the mapping table of the current configuration. The mapping table defines the mapping relationship between the fields identified by AI and the fields corresponding to the expense reimbursement details. According to the mapping table, the target identification information is assigned to the corresponding field of the reimbursement details to obtain the target reimbursement details information. During the assignment process, for the payment time field in the reimbursement details, the original time information obtained by identification is converted into a standard date format before being assigned. If the conversion fails, the original time information is assigned as text.

[0008] In an optional implementation, the consistency verification between the menu amount and the invoice amount includes: Compare the menu amount in the target reimbursement details with the invoice amount in the reimbursement document associated with the reimbursement voucher; If the menu amount is less than the invoice amount, an amount matching error will be triggered, and the submission of the reimbursement document will be blocked until the consistency check passes.

[0009] In an optional implementation, compliance verification based on the list of prohibited reimbursement items includes: The system performs keyword matching between the names of dishes in the target reimbursement details and a pre-defined list of prohibited dishes; the list of prohibited dishes can be dynamically configured by the user. If a matching dish name is found in the target expense reimbursement details, a compliance exception will be triggered, and the submission of the expense reimbursement document associated with the expense reimbursement voucher will be blocked until the compliance verification is passed.

[0010] In an optional implementation, the corresponding warning information is triggered, including: Based on the type of abnormal event, generate early warning information with differentiated handling guidelines, and push it to the user end, the review end or the administrator end through at least one of the following channels: reimbursement system interface pop-up window, enterprise instant messaging tool message or email. Among them, the warning information corresponding to information recognition anomalies is used to guide users to manually fill in the missing information; the warning information corresponding to amount matching anomalies is used to guide users to supplement the explanation of the reason; and the warning information corresponding to compliance anomalies is used to notify the reviewers or administrators to review.

[0011] In an optional implementation, the AI-based method for processing expense reimbursement voucher information further includes: Based on a preset statistical period, statistical analysis is performed on preset operation and maintenance indicators; among which, operation and maintenance indicators include at least one of the following: information identification success rate, trigger frequency and handling time of different types of abnormal events, and automatic verification interception rate. The analysis results are displayed in the form of visual charts on the administrator's operation and maintenance dashboard to assist in system optimization or risk strategy adjustment.

[0012] Secondly, the present invention provides an AI-based expense reimbursement voucher information processing system, comprising: The recognition module is used to identify information on the reimbursement voucher by calling the AI ​​recognition interface when a reimbursement voucher to be processed is obtained, and to obtain the target recognition information; the reimbursement voucher includes an image of a hospitality menu. The mapping module is used to map target identification information to the corresponding fields of the reimbursement details based on dynamically configured field mapping rules, so as to obtain the target reimbursement details information. The verification module is used to automatically verify the target reimbursement details. The automatic verification includes consistency verification between menu amount and invoice amount and compliance verification based on the list of prohibited reimbursement dishes. The early warning module is used to trigger corresponding early warning information when an abnormal event is detected during the processing of reimbursement vouchers, so that the relevant users can handle and review the abnormality, and update the processing status of the reimbursement vouchers based on the feedback results of the early warning information; among them, abnormal events include information identification abnormality, amount matching abnormality, or compliance abnormality.

[0013] Thirdly, the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the AI-based expense reimbursement voucher information processing method of any of the foregoing embodiments.

[0014] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, performs the AI-based expense reimbursement voucher information processing method described in any of the foregoing embodiments.

[0015] The present invention provides an AI-based method and system for processing expense reimbursement vouchers. The method includes: when an expense reimbursement voucher to be processed is obtained, the voucher is identified by calling an AI recognition interface to obtain target identification information; wherein the expense reimbursement voucher includes an image of a hospitality menu; based on dynamically configured field mapping rules, the target identification information is mapped to the corresponding fields of the expense details to obtain target expense details information; the target expense details information is automatically verified, including consistency verification of the menu amount and the invoice amount, and compliance verification based on a list of prohibited reimbursement items; when an abnormal event is detected during the processing of the expense reimbursement voucher, a corresponding warning message is triggered to enable the relevant user to handle and review the abnormality, and the processing status of the expense reimbursement voucher is updated based on the feedback results of the warning message; wherein the abnormal event includes information identification abnormality, amount matching abnormality, or compliance abnormality. This constructs an automated processing workflow that integrates "intelligent recognition, dynamic mapping, automatic verification, and closed-loop early warning," effectively solving the technical problems of low efficiency, high error rate, high compliance risk, and system rigidity caused by the reliance on manual processing of unstructured vouchers in existing expense reimbursement systems. It achieves automation, accuracy, compliance, and intelligence in expense reimbursement voucher processing, reducing labor costs and controlling financial risks. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the structure of an AI-based expense reimbursement voucher information processing system provided in an embodiment of the present invention; Figure 2 A schematic diagram of the overall architecture of an AI-based expense reimbursement voucher information processing system provided in this embodiment of the invention; Figure 3 A flowchart illustrating an AI-based method for processing expense reimbursement voucher information, provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] In expense reimbursement management scenarios, existing technical solutions mainly have the following shortcomings: 1. Lack of automated identification of unstructured data, high reliance on manual intervention: Existing expense reimbursement systems generally lack the ability to automatically extract information from unstructured image data such as payment screenshots, printouts, or handwritten menus. Key information, such as payment amount, payee, dish name, and total menu cost, still needs to be manually entered by the person submitting the expense report. This not only increases the workload for employees but is also highly susceptible to data entry errors due to human negligence, thereby affecting the efficiency and accuracy of subsequent review processes.

[0020] 2. Insufficient cross-verification capabilities and weak compliance risk management: In the existing process, the logical consistency verification between payment voucher information and invoice information (such as amount and payee), and the comparison between menu amounts and invoice amounts, basically rely on manual verification by financial auditors, which cannot achieve real-time and automated verification. In particular, for specific consumption items that are prohibited from reimbursement according to the company's internal control policies (such as shark fin and bird's nest), there is a lack of a systematic automatic screening mechanism. Judgments are mainly made based on the experience and sense of responsibility of auditors, resulting in a high degree of blind spot in compliance risk management and posing potential financial and compliance risks to the company.

[0021] 3. The system architecture is rigid, resulting in poor business adaptability and maintainability. In some systems with basic recognition capabilities, the mapping between recognition results and expense reimbursement form fields is usually fixed and hard-coded. When business rules change and the correspondence between recognition fields and business fields needs to be added or adjusted, it often requires large-scale modification and release of the system source code, resulting in high system maintenance costs and an inability to flexibly and quickly respond to the dynamic needs of enterprises for refined expense control.

[0022] 4. Absence of abnormal handling procedures and lack of intelligent early warning and guidance: Faced with common anomalies such as illegible handwriting in handwritten menus leading to recognition failures, and discrepancies between menu amounts and invoice amounts, the existing system lacks standardized processing procedures and intelligent early warning mechanisms. Auditors need to spend a significant amount of time repeatedly communicating and confirming with the person requesting reimbursement, resulting in a cumbersome and time-consuming process that severely impacts overall reimbursement approval efficiency.

[0023] 5. Poor multi-device compatibility, resulting in shortcomings in user experience: Existing solutions often fail to adequately consider the user experience on mobile devices. When employees upload payment screenshots or menu images using mobile devices, they frequently encounter issues such as upload compatibility problems, unstable recognition functions, or poor interface adaptation, which affect the convenience and smoothness of the reimbursement process and fail to fully leverage the advantages of mobile office.

[0024] In summary, existing technologies suffer from low automation, weak verification capabilities, poor system flexibility, inefficient exception handling, and poor cross-terminal experience when processing unstructured expense reimbursement vouchers. These issues make it difficult to meet the urgent needs of modern enterprises for efficient, accurate, compliant, and intelligent management of expense reimbursement processes.

[0025] Based on this, embodiments of the present invention provide an AI-based method and system for processing expense reimbursement voucher information. By employing an AI-based scheme for identifying and verifying expense reimbursement voucher information, the following problems can be solved and alleviated: 1. Provide an AI-based multi-scenario voucher information recognition solution to automatically extract key information from payment screenshots, hospitality menus (printed / handwritten), etc., replacing manual data entry and improving reimbursement efficiency; 2. Construct an automated verification and compliance management system to achieve real-time comparison of identification information and invoice information, automatic screening of prohibited reimbursement items, and reduce the risk of illegal reimbursement; 3. Design a flexible field mapping mechanism to support dynamic configuration of identification fields and expense detail fields, adapt to the differentiated business needs of enterprises, and reduce system maintenance costs; 4. Improve the abnormal scenario handling process and intelligent early warning mechanism, trigger accurate early warnings for abnormal situations such as identification failure and amount mismatch, provide standardized handling guidelines, and improve review efficiency; 5. Achieve multi-terminal adaptation, ensure stable operation of attachment upload and AI recognition functions on mobile and PC (Personal Computer) terminals, and optimize user experience.

[0026] To facilitate understanding of this embodiment, a detailed description of an AI-based expense reimbursement voucher information processing system disclosed in this embodiment of the invention will be provided first.

[0027] This invention provides an AI-based expense reimbursement voucher information processing system, such as... Figure 1 As shown, the system mainly includes: The recognition module 101 is used to identify the information of the reimbursement voucher by calling the AI ​​recognition interface when the reimbursement voucher to be processed is obtained, and to obtain the target recognition information; wherein, the reimbursement voucher includes an image of a hospitality menu. The mapping module 102 is used to map the target identification information to the fields corresponding to the reimbursement details based on dynamically configured field mapping rules, so as to obtain the target reimbursement details information. The verification module 103 is used to automatically verify the target reimbursement details. The automatic verification includes consistency verification between the menu amount and the invoice amount, as well as compliance verification based on the list of prohibited reimbursement dishes. The early warning module 104 is used to trigger corresponding early warning information when an abnormal event is detected during the processing of reimbursement vouchers, so that the relevant users can handle and review the abnormality, and update the processing status of the reimbursement vouchers based on the feedback results of the early warning information; among them, abnormal events include information identification abnormality, amount matching abnormality, or compliance abnormality.

[0028] The AI-based expense reimbursement voucher information processing system provided in this invention constructs an automated processing flow integrating "intelligent recognition, dynamic mapping, automatic verification, and closed-loop early warning." It effectively solves the technical problems of low efficiency, high error rate, high compliance risk, and system rigidity caused by the reliance on manual processing of unstructured vouchers in existing expense reimbursement systems. It realizes the automation, accuracy, compliance, and intelligence of expense reimbursement voucher processing, reduces labor costs, and manages financial risks.

[0029] Optionally, the aforementioned reimbursement voucher also includes a payment voucher image (such as a payment screenshot). The aforementioned recognition module 101 is specifically used to: generate a recognition request for a voucher identifier and voucher type with the reimbursement voucher; wherein, the voucher type includes a payment voucher corresponding to the payment voucher image and a hospitality menu corresponding to the hospitality menu image; send the recognition request to the AI ​​recognition interface to perform information recognition on the reimbursement voucher under the corresponding voucher type; receive the original recognition information returned by the AI ​​recognition interface, and perform standardized processing on the original recognition information to obtain the target recognition information.

[0030] Specifically, the system generates a request with a unique ID (i.e., voucher identifier) ​​and voucher type for each uploaded voucher (payment screenshot or hospitality menu) and sends it to the AI ​​recognition interface. The AI ​​recognition interface then calls the corresponding recognition model based on the voucher type to intelligently recognize the reimbursement voucher corresponding to the voucher identifier and returns original recognition information such as payee, amount, and dishes. Furthermore, it performs standardization processing on the original recognition information, such as data format verification, redundant data removal, and key information extraction. For example, it converts the non-standard date "August 3, 2025" to "2025-08-03" and the amount "430" to the number "430.00", thereby obtaining structurally standardized and reliable target recognition information for subsequent dynamic field mapping and automated verification.

[0031] Optionally, the mapping module 102 is specifically used to: obtain the currently configured mapping relationship table, which defines the mapping relationship between the fields identified by AI and the fields corresponding to the expense reimbursement details; assign the target identification information to the fields corresponding to the expense reimbursement details according to the mapping relationship table to obtain the target expense reimbursement details information; wherein, during the assignment process, for the payment time field in the expense reimbursement details, the original time information obtained by AI is converted into a standard date format before being assigned; if the conversion fails, the original time information is assigned as text.

[0032] Specifically, the system queries a pre-set, dynamically adjustable mapping table and then automatically assigns the target identification information to the corresponding fields in the reimbursement details according to this mapping table, forming the target reimbursement details information. During this process, for special fields like "payment time," the system attempts intelligent processing: it prioritizes converting the original time identified by AI (e.g., "August 3, 2025") into a standard format (e.g., "2025-08-03") before filling it in; if the conversion fails (e.g., the format is unrecognizable), the original time text is directly filled in instead of being discarded or flagged as an error, thus ensuring the integrity of the information and the robustness of the process.

[0033] Optionally, when performing consistency verification between the menu amount and the invoice amount, the above-mentioned verification module 103 is specifically used to: compare the menu amount in the target reimbursement details with the invoice amount in the reimbursement document associated with the reimbursement voucher; if the menu amount is less than the invoice amount, an amount matching anomaly is triggered, and the submission operation of the reimbursement document is blocked until the consistency verification passes.

[0034] This embodiment provides an automated reimbursement amount verification and risk control mechanism. Specifically, when an employee submits a reimbursement claim, the system automatically compares two key amounts: one is the actual total consumption amount identified from the entertainment menu image (menu amount), and the other is the reimbursement amount entered by the employee on the reimbursement form based on the invoice entry (invoice amount). If the menu amount is less than the invoice amount (i.e., the actual consumption is less than the reimbursement amount), it is immediately identified as an abnormal situation of amount mismatch. At this time, the system will proactively intercept the reimbursement form, preventing it from being submitted to the next approval stage, and requiring the employee to provide supplementary explanations or modify the amount. Only when the amount comparison result becomes consistent or reasonable will the system allow the reimbursement form to continue circulating. This effectively prevents the risk of over-reimbursement due to errors in filling out the form or intentional falsification.

[0035] Optionally, when performing compliance verification based on the list of prohibited reimbursement dishes, the verification module 103 is specifically used to: match the dish names in the target reimbursement details with the preset list of prohibited reimbursement dishes using keywords; wherein, the list of prohibited reimbursement dishes supports dynamic configuration by the user; if there is a successfully matched dish name in the target reimbursement details, a compliance exception is triggered, and the submission operation of the reimbursement voucher associated with the reimbursement document is blocked until the compliance verification passes.

[0036] This invention provides an automated compliance risk screening mechanism. Specifically, before a user submits an expense report, the system identifies the names of dishes from the menu image and compares them with a preset, dynamically adjustable list of prohibited dishes (e.g., containing "shark fin," "bird's nest," and "cigarettes"). If a prohibited item from the list is found in the reimbursement list, the system immediately triggers a compliance anomaly and forcibly blocks the submission of the corresponding expense report, preventing it from entering the review process. This anomaly is only lifted after the user modifies the menu, removes the prohibited dish, or obtains special approval through a specific authorization process, allowing the process to continue. This achieves automatic and forceful interception of illegal spending, preventing financial and compliance risks for enterprises.

[0037] Furthermore, the aforementioned early warning module 104 is specifically used to: generate early warning information with differentiated handling guidelines based on the type of abnormal event, and push it to the user end, review end, or administrator end through at least one of the following channels: a pop-up window in the reimbursement system interface, a message from the enterprise instant messaging tool, or an email; wherein, the early warning information corresponding to information identification abnormality is used to guide the user to manually fill in the missing information; the early warning information corresponding to amount matching abnormality is used to guide the user to supplement the explanation of the reason; and the early warning information corresponding to compliance abnormality is used to notify the reviewer or administrator to review.

[0038] This invention provides a refined early warning and handling mechanism for anomalies in expense reimbursement processing. Specifically, when the system detects anomalies during AI identification, amount matching, or compliance verification, it automatically generates early warning information containing clear operational instructions based on the type of anomaly. This information is then precisely pushed to relevant personnel through various channels (such as pop-ups in the expense reimbursement system and WeChat / DingTalk messages). For example, if the anomaly is information identification (e.g., blurry menu text), the early warning will prompt the user to manually fill in the information; if the anomaly is amount matching (e.g., the menu amount is less than the invoice amount), the early warning will require the user to provide a further explanation; and if the anomaly is compliance (e.g., the inclusion of prohibited items for reimbursement), the early warning will directly notify the reviewer or administrator for review. This "categorized early warning, targeted push, and precise guidance" approach transforms different types of anomalies into standardized processing tasks, directly assigning them to the most suitable personnel, greatly improving the efficiency and standardization of anomaly handling.

[0039] Optionally, the system further includes a visualization module for: performing statistical analysis on preset operation and maintenance indicators based on a preset statistical period; wherein the operation and maintenance indicators include at least one of information identification success rate, trigger frequency and handling time of different types of abnormal events, and automatic verification interception rate; and displaying the analysis results in the form of visual charts on the administrator's operation and maintenance dashboard to assist in system optimization or risk strategy adjustment.

[0040] This invention provides system operation status monitoring and decision support functions. Specifically, the system automatically collects key operational data according to fixed statistical periods (such as daily, weekly, or monthly), including: the accuracy of AI recognition, the frequency and processing time of various anomalies (such as recognition failure, inconsistent amounts, and non-compliant dishes), and the number of orders successfully intercepted by automatic verification rules. The system transforms these statistical results into visualizations such as charts and trend graphs, and displays them centrally on an administrator-dedicated operational monitoring dashboard. This allows administrators to clearly understand the system's health, business risk points, and process bottlenecks, thereby enabling them to optimize AI models, adjust verification rules, or improve management strategies in a targeted manner, achieving continuous system improvement and refined risk control.

[0041] Optionally, the verification rules for the aforementioned automated verification can be executable rules automatically generated by parsing corporate financial system text using LLM (Large Language Model). During automated verification, when performing keyword matching, a weighted calculation method is used: the trigger counts of each verification rule are weighted and summed based on its matching weight to obtain a total risk score. The risk level corresponding to the total risk score determines whether the verification passes and whether an early warning is triggered. Based on this, the system also includes an update module for: updating the verification rules for automated verification based on a pre-established rule effectiveness decay model; wherein the rule effectiveness decay model records the metadata of each verification rule, including creation time, most recent trigger time, historical trigger frequency, and current matching weight, and updates the matching weights of the verification rules based on the metadata.

[0042] This invention can automatically reduce the weight of long-term untriggered verification rules and actively strengthen high-frequency abnormal rules through an update module. Specifically, long-term untriggered verification rules are monitored periodically. When a first rule is detected as not triggered within a preset monitoring period, its matching weight is reduced based on the most recent trigger time of the first rule. Optionally, an exponential decay model can be used to calculate the updated matching weight: W i(t) =max(W min W i(t_last) ×exp(-α×Δt)), where W i(t)Let W be the matching weight of the i-th rule after the update at the current time t. min W is the preset minimum matching weight. i(t_last) Let be the matching weight of the i-th rule at the most recent trigger (i.e., the matching weight before the update), α be the preset decay coefficient, 0 < α < 1, and Δt be the interval between the current time t and the most recent trigger time t_last.

[0043] When the second rule is matched in automated validation (e.g., successfully screening for "shark fin"), its trigger record (including the most recent trigger time and historical trigger frequency) is updated, and the matching weight of the second rule is increased based on the updated trigger record. Optionally, the updated matching weight can be determined by the following formula: W i(new) =min(W max W i(old) +β×(1+γ×F recent ), where W i(new) Let W be the matching weight updated for the i-th rule. max W is the preset maximum matching weight. i(old) Let F be the matching weight before the i-th rule is updated, β be the preset enhancement coefficient (β>0), γ be the preset frequency gain coefficient (γ>0), and F be the matching weight before the i-th rule is updated. recent This refers to the historical trigger frequency. The historical trigger frequency can be the number of triggers within the most recent time window (such as the past 30 days).

[0044] Optionally, the aforementioned verification module 103 is further configured to: acquire multi-source associated data related to reimbursement vouchers, the multi-source associated data including at least: merchant names and transaction times identified from payment voucher images (such as payment screenshots), employee travel application and trajectory data, and historical food purchase price data from the enterprise's procurement system; construct a fraud risk feature vector based on the multi-source associated data and input it into a pre-trained fraud identification model, the fraud identification model outputting fraud behavior identification results, the fraud behavior identification results including at least one of the following: fake merchant risk, time-based fraud risk, and abnormal food price risk; when the identified fraud behavior risk exceeds a preset risk threshold, trigger the highest level of compliance anomaly warning.

[0045] In practice, the first step is to collect multi-source related data, including: 1) Voucher identification data: Extracting merchant name A, transaction time T1, and transaction amount M1 from payment screenshots; extracting the menu and its unit price from the hospitality menu. 2) Employee behavior data: Linking the travel applications of employees who submitted expense reports to obtain their planned travel location L_plan, planned time window T_plan, and actual geographical location trajectory L_actual obtained through the company's mobile office APP / access control system. 3) Enterprise system data: Calling the procurement system interface to query the historical purchase unit price P_history of the same or similar dishes in contracts signed with merchant name A or similar merchants.

[0046] Secondly, feature construction is carried out on multi-source associated data, including: 1) Fake merchant risk features: Feature 1 (geographical location contradiction feature), which can calculate the matching degree between the registered address / usual address of merchant A and the actual geographical location L_actual of the employee before and after the transaction time T1 (e.g., calling the map API to calculate the distance D between the two). If the matching degree is low (e.g., D>50 kilometers) and the employee has no evidence of movement, the geographical location contradiction score is high; Feature 2 (merchant existence feature), which can be combined with the third-party enterprise information query API to verify whether the business registration status of merchant name A is abnormal. 2) Time-based fraud risk features: Feature 3 (time window deviation feature), which can calculate whether the transaction time T1 is within the employee's planned travel time window T_plan. If it is completely outside T_plan, the risk is high; Feature 4 (abnormal behavior density feature), which can analyze the density of the employee's expense records in a short period of time before and after T1. Abnormally dense consumption may be fraudulent. 3) Abnormal Price Risk Characteristics of Dishes: Feature 5 (Unit Price Deviation Characteristic) can calculate the deviation rate (P_current-P_history) / P_history between the unit price P_current and the historical purchase unit price P_history for each dish in the menu, and take the average deviation rate or the maximum deviation rate of the entire menu as the feature; Feature 6 (Rare Dish Appearance Characteristic) is whether there are expensive dishes in the menu that have never appeared in the company's historical purchase records.

[0047] Fraud detection is then performed again using a fraud detection model, including inputting the feature vector composed of the aforementioned features into the trained fraud detection model for comprehensive judgment. The fraud detection model can employ supervised binary classification models (such as Gradient Boosting Decision Tree (GBDT) or Deep Neural Network (DNN)) or unsupervised anomaly detection models such as Isolation Forest. The fraud detection model can be trained using historical expense reimbursement data already marked as "normal" or "fraudulent" by auditors, along with their corresponding multi-source correlation data. The fraud detection model can output a fraud risk probability score (between 0 and 1). Furthermore, through model interpretability techniques (such as SHAP value analysis), it can be used to infer which features (such as "geographical inconsistency" or "unit price deviation") contribute most to the high fraud risk assessment, thereby categorizing the high fraud risk into specific types such as fraudulent merchant risk, time-based fraud risk, or abnormal menu price risk.

[0048] Finally, the fraud detection results are processed, including setting a risk threshold (e.g., risk probability score > 0.8). If the risk threshold is exceeded, the highest level of compliance anomaly is triggered, directly intercepting the case and transferring it to a special audit. The auditor's final ruling ("confirmed fraud" or "false alarm") will be used as a new label and fed back into the model training set for continuous iterative optimization of the model, forming an intelligent closed loop.

[0049] Optionally, the system also provides an intelligent assignment and review triggering mechanism based on confidence level grading. Specifically, during the AI ​​recognition process, the system not only returns the recognition result but also the recognition confidence level of each key piece of information (such as "total menu amount" and "payee"). Field mapping and assignment are processed according to confidence level grading: if the confidence level is higher than the first threshold (e.g., 95%), an automatic silent assignment is performed without user confirmation; if the confidence level is between the first and second thresholds (e.g., 80%-95%), the recognition result is highlighted in the corresponding field as a recommended value, which the user can accept or modify with one click; if the confidence level is lower than the second threshold, an information recognition anomaly is triggered. This improves automation while ensuring accuracy, achieving intelligent human-machine collaboration.

[0050] Optionally, the system also provides a self-learning and optimization function for verification rules based on historical data. Specifically, the system records the results of each compliance verification (such as dish screening) and the final ruling of the auditors on the warnings (such as "confirmed violation" or "special approval"). When the same dish name is "specially approved" multiple times by the auditors, the system can automatically recommend to the administrator that the dish name be removed from the list of prohibited reimbursement dishes or marked as an exception. This allows the verification rules to evolve with the company's actual business approval practices, solving the problem of rigid rules.

[0051] In some possible embodiments, this invention constructs a complete technical solution integrating "multi-scenario voucher collection, AI intelligent recognition, dynamic field mapping, automated verification, anomaly warning, and closed-loop handling." The core idea is to build an intelligent system based on an enterprise expense reimbursement platform, comprising four core modules: voucher collection, AI recognition, verification and control, and early warning and handling. By deploying multi-type voucher collection components, it enables multi-terminal uploading of attachments such as payment screenshots and hospitality menus. It calls AI large-scale model interfaces to automatically identify and extract key information. A dynamic field mapping mechanism is designed to flexibly match identified information with reimbursement detail fields. Multi-level verification rules are constructed to verify information consistency and compliance. An intelligent early warning and closed-loop handling mechanism is established to improve the efficiency of handling abnormal scenarios. Ultimately, it achieves automated, accurate, and compliant processing of expense reimbursement vouchers, reduces labor costs, and manages financial risks.

[0052] The core implementation process of this invention is as follows: 1) System configuration phase: The administrator completes the integration of AI recognition interface (connecting to the AI ​​large model interface provided by the customer), configuration of field mapping rules, configuration of verification rules (amount comparison, prohibited dishes list, etc.), configuration of early warning strategy and multi-terminal adaptation settings.

[0053] 2) Voucher collection stage: Users upload attachments (payment screenshots, hospitality menus) in the designated attachment fields of the reimbursement platform via PC / mobile. The system limits each field to one attachment at a time, supports mainstream image formats, and completes attachment legality verification and storage.

[0054] 3) AI recognition and information extraction stage: After the attachment is uploaded, the system automatically calls the AI ​​large model interface to recognize information such as the payee's name, payment amount, and payment time in the payment screenshot, as well as information such as the dish name and menu amount in the banquet menu; it receives the data returned by the interface and performs standardized processing and format conversion.

[0055] 4) Field mapping and assignment stage: According to the preset field mapping rules, the standardized identification information is automatically assigned to the corresponding fields of the reimbursement details (such as payment screenshot amount, payee name, menu amount, etc.), and intelligent adaptation of date type fields is supported (date type is assigned first, and text type is assigned if it cannot be adapted).

[0056] 5) Automated verification phase: The system performs dual verification according to the verification rules. The first is the consistency verification of the amount (comparing the menu amount with the invoice amount), and the second is the compliance verification (comparing the dish name with the list of prohibited reimbursements). If the verification fails, the corresponding abnormal scenario handling process is triggered.

[0057] 6) Anomaly Warning and Closed-Loop Handling Stage: In case of anomalies such as identification failure, amount mismatch, or the presence of prohibited reimbursement items, the system will automatically trigger an early warning (such as "menu amount is less than invoice amount" or "menu amount was manually entered"). Users supplement or correct information according to the early warning, and the reviewers complete the review, forming a closed-loop management of "identification-verification-early warning-handling-review".

[0058] The following reference Figure 2 The specific implementation scheme of the aforementioned AI-based expense reimbursement voucher information processing system is described in detail. For example... Figure 2 As shown, the system includes: 1) System Integration and Configuration Module 201.

[0059] This module is the foundation for the system to realize AI recognition, field mapping and verification control. Its core functions include AI interface integration, rule configuration, permission control and multi-terminal adaptation, ensuring smooth linkage and flexible adaptation of all aspects of the system.

[0060] 1.1 Interface Design and Integration: AI Recognition Interface: We have developed a standardized AI recognition interface that connects to the AI ​​large-scale model interface provided by our clients. It supports the recognition of information from two types of vouchers: payment screenshots and hospitality menus (printed / handwritten). The interface uses JSON format to transmit data, and the core transmission fields include voucher type, recognition result (payee name, payment amount, etc.), recognition status, and confidence level. It supports breakpoint resume and retry mechanisms to avoid data loss and has a recognition response time of ≤1 second.

[0061] Reimbursement Platform Linkage Interface: Develop a linkage interface with the expense reimbursement platform to realize functions such as synchronizing attachment upload status, assigning identification information values, and providing verification result feedback; support sending field assignment instructions to the reimbursement platform, synchronizing abnormal warning information, and ensuring smooth data flow between systems.

[0062] Warning push interface: Develop interfaces for integration with notification platforms such as WeChat Work and DingTalk to support multi-channel push of abnormal warning information; the push recipients can be configured according to the warning type (such as compliance warning, identification failure warning) to ensure that reviewers and users receive warning information in a timely manner.

[0063] 1.2 Rule Configuration Management: Field mapping configuration: Administrators can customize the mapping relationship between AI recognition fields and expense reimbursement detail fields. The preset mapping rules are shown in Table 1 below. Mapping fields can be flexibly modified or added through the backend database without adjusting the front-end interface.

[0064] Table 1

[0065] Verification rule configuration: Administrators can customize dual verification rules. The first is the amount comparison rule (such as "additional explanation is required when the menu amount is less than the invoice amount"), and the second is the compliance verification rule (preset list of prohibited reimbursement dishes: shark fin, bird's nest, sea cucumber, grouper, Buddha Jumps Over the Wall, wild game, cigarettes, alcohol, etc., and keywords can be added or deleted).

[0066] Attachment Configuration: Administrators can configure an "Whether to call AI recognition" switch for the attachment field. When checked, the field is limited to uploading only one attachment at a time, and the AI ​​recognition process is automatically triggered after the upload is completed. If not checked, the original upload logic is retained to adapt to different reimbursement scenarios.

[0067] 1.3 Multi-terminal adaptation configuration: Develop multi-terminal adaptation components to ensure stable operation of attachment upload and AI recognition functions on PC and mobile devices (phones, tablets); support compressed image upload on mobile devices to optimize recognition speed; adapt to different screen sizes to ensure the display effect of warning prompts and information entry interfaces, and improve the user experience.

[0068] 2) Voucher collection and AI recognition module 202 (i.e., the recognition module 101 mentioned above).

[0069] This module is responsible for collecting reimbursement vouchers from multiple terminals, verifying their legality, performing AI-powered intelligent recognition and information extraction, ensuring the integrity and accuracy of the identified information, and providing data support for subsequent field assignment and verification.

[0070] 2.1 Voucher Collection and Legality Verification: Multi-terminal data collection: Users can upload vouchers through designated attachment fields (payment screenshots, hospitality menus) on the reimbursement platform, supporting mainstream formats such as JPG, PNG, and PDF; mobile devices support photo uploads and album uploads, while PC devices support local file uploads and drag-and-drop uploads.

[0071] Legality verification: The system automatically verifies the format and size of attachments (single attachment ≤ 20MB). If the requirements are not met, a prompt will be triggered: "Please upload a JPG / PNG / PDF file within 20MB". At the same time, the system verifies the uniqueness of attachments to ensure that only one attachment is uploaded at a time for each field, avoiding duplicate uploads.

[0072] Voucher categorization and storage: After verification, the system categorizes and stores the vouchers according to their type (payment screenshot, hospitality menu) in a distributed file storage system, generates a unique attachment ID, and associates it with the reimbursement document information for easy traceability and querying later.

[0073] 2.2 AI Intelligent Recognition and Information Extraction: Recognition Trigger: When the attachment upload is completed and the "Call AI Recognition" switch is turned on, the system automatically triggers the AI recognition process, sends a recognition request to the AI large model interface, and carries parameters such as the attachment ID and voucher type.

[0074] Multi-scenario Recognition: 1. Payment Screenshot Recognition: It supports the recognition of screenshots from mainstream payment platforms, extracts key information such as the payee name, payment amount, payment time, and transaction number; automatically filters redundant information in the screenshot (such as advertisements, irrelevant button texts, etc.) to improve the recognition accuracy.

[0075] 2. Hospitality Menu Recognition: It supports the recognition of printed menus and handwritten menus, extracts key information such as the dish name, total menu amount, table information, etc.; for the problems of blurred handwriting and non-standard format in handwritten menus, it combines AI image enhancement technology to optimize the recognition effect. If the recognition confidence level < 80%, it is determined as "recognition failed".

[0076] Recognition Result Processing: Receive the recognition data returned by the AI interface and perform standardized processing, including data format verification (such as the payment amount needs to be a numerical type, and the payment time needs to conform to the format of "YYYY-MM-DD HH:mm:ss" or "YYYY年MM月DD日HH:mm:ss"), redundant data elimination, and key information extraction; store the processed standardized data in a relational database, associating the attachment ID with the reimbursement document ID.

[0077] 2.3 Recognition Exception Handling: When the AI recognition fails (such as the handwritten menu cannot be recognized, the screenshot is blurred), the system automatically triggers an exception prompt and pushes it to the user side: "The menu amount recognition failed, please fill in manually"; after the user fills in the information manually, the system marks the "manually filled" identifier and triggers an early warning "The menu amount is filled in manually" to remind the reviewer to conduct key review.

[0078] 3) Field Mapping and Assignment Module 203 (i.e., the above-mentioned Mapping Module 102).

[0079] This module realizes the dynamic mapping and automatic assignment of AI recognition information and reimbursement detail fields, supports field type adaptation, improves the efficiency of entering reimbursement information, and reduces the manual error rate.

[0080] 3.1 Dynamic Field Mapping: Based on the field mapping rules configured by the administrator, the system automatically matches the AI recognition fields and the reimbursement detail fields; supports the background dynamic adjustment of the mapping rules. The administrator can add and modify the mapping relationship through the database (such as adding the mapping of the "transaction number" field) without modifying the front-end code to adapt to the changes in the enterprise business requirements.

[0081] 3.2 Automatic Assignment and Type Adaptation: The standardized AI recognition information is automatically assigned to the corresponding fields in the reimbursement details, achieving intelligent adaptation of field types: the payment time field is assigned to the date type first, and if the recognition result format cannot be adapted to the date type (such as only containing "August 3, 2025" without a specific time), it is assigned to the text type; the amount field automatically verifies the validity of the value to ensure accurate assignment.

[0082] After the value is assigned, the system displays the information that identified the value (such as the payee's name and the amount shown in the payment screenshot) on the reimbursement details page. Users can view and modify the information, and the modified information will be marked with a "manually modified" label for easy traceability by auditors.

[0083] 4) Automated verification and compliance management module 204 (i.e., the verification module 103 mentioned above).

[0084] This module enables automated verification and compliance control of reimbursement information, including consistency verification of amounts and screening of prohibited dishes for reimbursement, thereby promptly identifying risks of violations and improving review efficiency.

[0085] 4.1 Amount consistency verification: The system automatically compares the menu amount with the invoice amount. If the menu amount is less than the invoice amount, an alert is triggered: "Menu amount is less than invoice amount, please provide supplementary information." The user is then forced to fill in supplementary information before submitting the reimbursement form. If the menu amount is the same as or greater than the invoice amount, the verification passes and the process proceeds to the next review stage.

[0086] 4.2 Compliance Verification (Screening of Dishes Prohibited from Being Reimbursed): The system will identify and extract the names of the dishes and match them with a preset list of prohibited dishes for reimbursement. If the dish is included in the list (such as shark fin or bird's nest), an alert will be triggered that "prohibited dishes are included, submission is not possible", and the reimbursement form will be blocked from being submitted. If no prohibited dishes are included, the verification will pass.

[0087] It supports flexible configuration of the list of dishes that are prohibited from being reimbursed. Administrators can add or delete keywords through the backend to adapt to changes in corporate compliance policies.

[0088] 5) Intelligent early warning and closed-loop handling module 205 (i.e., the early warning module 104 mentioned above).

[0089] This module enables intelligent early warning, multi-channel push notifications, and closed-loop handling of abnormal scenarios, ensuring timely processing of abnormal situations and improving the efficiency and compliance of reimbursement review.

[0090] 5.1 Intelligent Early Warning Classification and Alerts: The system categorizes abnormal scenarios into three types and triggers corresponding warning prompts: 1. Identify and issue alerts for anomalies: such as “Menu amount recognition failed, please fill in manually” or “Payment screenshot information recognition is incomplete”, and push them to the user’s end to guide the user to correct or supplement the information.

[0091] 2. Amount mismatch warning: such as "Menu amount is less than invoice amount, please provide additional information", a notification will be sent to the user, and the user can only submit the notification after providing additional information.

[0092] 3. Compliance warnings: such as "Includes prohibited reimbursement items, cannot submit" or "Menu amount is manually entered, please review carefully", will be pushed to the user's end and the reviewer's end respectively, to intercept non-compliant documents or remind them to conduct key reviews.

[0093] 5.2 Multi-channel early warning push: Warning information is pushed through multiple channels, including pop-up notifications on the reimbursement platform and push notifications via WeChat / DingTalk. For warnings concerning important compliance issues (such as prohibitions on reimbursing certain dishes), additional notifications are sent to the person in charge of financial auditing to ensure that the warning information is accurately delivered.

[0094] 5.3 Closed-loop handling process: 1. Warning Trigger: When the system detects an abnormal scenario, it automatically triggers the corresponding warning and records information such as the warning time, warning type, and associated expense report ID.

[0095] 2. Anomaly Handling: Users correct information based on the warning prompts (such as adding explanations or modifying menu information) or manually fill in missing information; reviewers focus on reviewing relevant information for warning scenarios such as manual filling and mismatched amounts.

[0096] 3. Status Synchronization: After the user completes the correction or supplement, the system automatically re-verifies. If the verification passes, the warning status is updated to "handled" and synchronized to the review process. If the verification still fails, the warning will continue to be triggered until the anomaly is resolved.

[0097] 4. Recording and Traceability: The system fully records the entire process of early warning triggering, handling, and review, including the operator, operation time, and handling content, forming a closed-loop traceability record to facilitate subsequent auditing and verification.

[0098] 6) Visual configuration and operation and maintenance module 206 (i.e. the visualization module mentioned above).

[0099] This module provides a visual configuration interface and operation and maintenance monitoring functions, enabling administrators to efficiently configure and maintain the system, and reduce maintenance costs.

[0100] 6.1 Visual configuration interface: It provides a visual rule configuration interface, allowing administrators to intuitively configure field mapping rules, validation rules, and lists of prohibited reimbursement items; it supports real-time application of configuration information, historical record query and rollback, facilitating configuration change management and problem troubleshooting.

[0101] 6.2 Operation and maintenance monitoring functions: Build an operation and maintenance monitoring dashboard to display key indicators such as AI recognition success rate, number of abnormal warnings, and warning handling time in real time; support statistical analysis by time range (day / week / month) to identify system operation bottlenecks (such as excessively high failure rate of a certain type of voucher recognition) and assist administrators in optimizing configuration and AI recognition models.

[0102] It supports multi-condition querying and exporting of identification logs, warning logs, and handling logs, with a log retention period of ≥3 months to meet business verification and auditing needs.

[0103] In summary, the key points of the embodiments of the present invention include: 1. Multi-scenario voucher AI recognition and integration mechanism: Enables intelligent recognition of multiple types of vouchers such as payment screenshots and hospitality menus (printed / handwritten), integrates AI large model interfaces, optimizes the recognition result processing flow, and improves recognition accuracy and efficiency.

[0104] 2. Dynamic field mapping and type adaptation technology: Supports dynamic configuration of AI-recognized fields and reimbursement detail fields in the background, realizing intelligent type adaptation of fields such as payment time, improving system adaptability and flexibility.

[0105] 3. Multi-level automated verification and compliance management system: Construct a dual verification rule for consistency of amount verification and screening of prohibited reimbursement items to achieve real-time control of reimbursement compliance and intercept non-compliant documents.

[0106] 4. Intelligent early warning and closed-loop handling mechanism for abnormal scenarios: For abnormal scenarios such as identification failure and amount mismatch, a classification early warning strategy and multi-channel push are designed to achieve closed-loop management of the entire process of "early warning-handling-review-traceability".

[0107] 5. Multi-terminal adaptation and visualized operation and maintenance architecture: It achieves full terminal adaptation for PC and mobile terminals, provides visualized configuration and operation and maintenance monitoring functions, reduces system maintenance costs, and improves operation and maintenance efficiency.

[0108] The embodiments of the present invention have the following beneficial effects: 1. Improve reimbursement efficiency and reduce labor costs: Automatically identify and assign values ​​to key information in payment screenshots and hospitality menus, replacing manual entry and reducing user input workload, improving reimbursement form filling efficiency by more than 70%; Automated verification and early warning mechanisms reduce the workload of manual verification by auditors, improving audit efficiency by more than 60%.

[0109] 2. Reduce data error rate and improve information accuracy: AI intelligent recognition replaces manual input, avoiding data errors caused by human negligence; standardized data processing and format verification further ensure the accuracy of recognized information, reducing the data error rate by more than 80%.

[0110] 3. Strengthen compliance management and reduce financial risks: By automatically screening for prohibited reimbursement items and verifying the consistency of amounts, real-time control of reimbursement compliance can be achieved, effectively intercepting irregular reimbursement documents and reducing corporate financial risks and compliance hazards.

[0111] 4. Improve system adaptability and reduce maintenance costs: The dynamic field mapping mechanism supports flexible backend configuration, which can adapt to changes in enterprise business needs without modifying the frontend interface; the multi-terminal adaptation function optimizes the operating experience in different scenarios and reduces system adaptation and maintenance costs.

[0112] 5. Achieve closed-loop management of abnormal scenarios and improve the standardization of audits: A comprehensive abnormal early warning and closed-loop handling process ensures timely handling of issues such as abnormality identification and compliance risks; full-process log traceability improves the standardization and auditability of reimbursement audits.

[0113] 6. Optimize user experience and improve employee satisfaction: Multi-terminal adaptation, automated identification and assignment simplify the process of filling out expense reimbursement forms; clear warning prompts and operation guides reduce the difficulty of user operation and improve the employee reimbursement experience and satisfaction.

[0114] 7. Achieve dynamic adaptive optimization of verification rules: By automatically downgrading verification rules that have not been triggered for a long time and actively strengthening high-frequency abnormal rules, the system automatically focuses on the most effective risk points at present, while avoiding interference from outdated rules, thereby continuously improving the accuracy and efficiency of reimbursement compliance management.

[0115] 8. Enhance the depth and effectiveness of enterprise anti-fraud and compliance risk control: By integrating multi-dimensional data such as payment, trajectory, and procurement, and using fraud identification models for correlation analysis and pattern mining, the system can upgrade from traditional single-point rule verification to intelligent multi-source investigation, thereby accurately identifying hidden fraudulent behaviors that are difficult to detect by traditional methods, such as "fake merchants," "time extraction," and "abnormal pricing."

[0116] This invention also provides an AI-based method for processing expense reimbursement voucher information. This method is applied to the aforementioned AI-based expense reimbursement voucher information processing system and can be executed by an electronic device with data processing capabilities. See also... Figure 3 The diagram shows a flowchart of an AI-based method for processing expense reimbursement voucher information. This method mainly includes the following steps S310 to S340: Step S310: When the expense voucher to be processed is obtained, the AI ​​recognition interface is called to recognize the information of the expense voucher and obtain the target recognition information; wherein, the expense voucher includes the image of the entertainment menu.

[0117] In some possible embodiments, the above-mentioned reimbursement voucher also includes a payment voucher image; step S310 may include: generating a voucher identifier and voucher type recognition request with the reimbursement voucher; wherein, the voucher type includes a payment voucher corresponding to the payment voucher image and a hospitality menu corresponding to the hospitality menu image; sending the recognition request to the AI ​​recognition interface to perform information recognition on the reimbursement voucher under the corresponding voucher type; receiving the original recognition information returned by the AI ​​recognition interface, and standardizing the original recognition information to obtain the target recognition information.

[0118] Step S320: Based on the dynamically configured field mapping rules, the target identification information is mapped to the fields corresponding to the reimbursement details to obtain the target reimbursement details information.

[0119] In some possible embodiments, step S320 above may include: obtaining a currently configured mapping relationship table, which defines the mapping relationship between fields identified by AI and fields corresponding to expense reimbursement details; assigning the target identification information to the fields corresponding to the expense reimbursement details according to the mapping relationship table to obtain the target expense reimbursement details information; wherein, during the assignment process, for the payment time field in the expense reimbursement details, the original time information obtained by identification is first converted into a standard date format before being assigned; if the conversion fails, the original time information is assigned as text.

[0120] Step S330: Automated verification of the target reimbursement details information. Automated verification includes consistency verification of menu amount and invoice amount, as well as compliance verification based on the list of prohibited reimbursement dishes.

[0121] In some possible embodiments, the consistency verification between the menu amount and the invoice amount may include: comparing the menu amount in the target reimbursement details with the invoice amount in the reimbursement document associated with the reimbursement voucher; if the menu amount is less than the invoice amount, an amount matching anomaly is triggered, and the submission operation of the reimbursement document is blocked until the consistency verification passes.

[0122] In some possible embodiments, the compliance verification based on the list of prohibited dishes may include: matching the names of dishes in the target reimbursement details with keywords from a preset list of prohibited dishes; wherein the list of prohibited dishes can be dynamically configured by the user; if a matching dish name exists in the target reimbursement details, a compliance exception is triggered, and the submission of the reimbursement voucher-related reimbursement document is blocked until the compliance verification passes.

[0123] Step S340: When an abnormal event is detected during the processing of reimbursement vouchers, the corresponding warning information is triggered so that the relevant user can handle and review the abnormality, and the processing status of the reimbursement voucher is updated according to the feedback results of the warning information; wherein, the abnormal event includes information identification abnormality, amount matching abnormality or compliance abnormality.

[0124] In some possible embodiments, the aforementioned trigger-corresponding warning information may include: generating warning information with differentiated handling guidelines based on the type of abnormal event, and pushing it to the user end, review end, or administrator end through at least one of the following channels: a pop-up window in the reimbursement system interface, a message from an enterprise instant messaging tool, or an email; wherein, the warning information corresponding to information identification abnormality is used to guide the user to manually fill in the missing information; the warning information corresponding to amount matching abnormality is used to guide the user to supplement the explanation of the reason; and the warning information corresponding to compliance abnormality is used to notify the reviewer or administrator to review.

[0125] The AI-based expense reimbursement voucher information processing method provided in this invention constructs an automated processing flow integrating "intelligent recognition, dynamic mapping, automatic verification, and closed-loop early warning." It effectively solves the technical problems of low efficiency, high error rate, high compliance risk, and system rigidity caused by the reliance on manual processing of unstructured vouchers in existing expense reimbursement systems. It realizes the automation, accuracy, compliance, and intelligence of expense reimbursement voucher processing, reduces labor costs, and manages financial risks.

[0126] Furthermore, the aforementioned AI-based expense reimbursement voucher information processing method also includes: performing statistical analysis on preset operation and maintenance indicators based on a preset statistical period; wherein, the operation and maintenance indicators include at least one of the following: information recognition success rate, trigger frequency and handling time of different types of abnormal events, and automatic verification interception rate; and displaying the analysis results in the form of visual charts on the administrator's operation and maintenance dashboard to assist in system optimization or risk strategy adjustment.

[0127] Optionally, the verification rules for the aforementioned automated verification can be executable rules automatically generated by parsing corporate financial system text using LLM. During automated verification, when performing keyword matching, a weighted calculation method is used: the trigger counts of each verification rule are weighted and summed based on its matching weight to obtain a total risk score. The risk level corresponding to the total risk score determines whether the verification passes and whether an early warning is triggered. Based on this, the aforementioned AI-based expense reimbursement voucher information processing method also includes: updating the verification rules for automated verification based on a pre-established rule effectiveness decay model; wherein, the rule effectiveness decay model is used to record the metadata of each verification rule, including creation time, most recent trigger time, historical trigger frequency, and current matching weight, and the matching weight of the verification rule is updated based on the metadata.

[0128] In some possible embodiments, updating the verification rules for automated verification based on a pre-established rule effectiveness decay model may include: periodically monitoring verification rules that have not been triggered for a long time; and when a first rule is detected as not triggered within a preset monitoring period, reducing its matching weight based on the most recent trigger time of the first rule. In one possible implementation, an exponential decay model can be used to calculate the updated matching weight: W i(t) =max(W min W i(t_last) ×exp(-α×Δt)), where W i(t) Let W be the matching weight of the i-th rule after the update at the current time t. min W is the preset minimum matching weight. i(t_last) Let be the matching weight of the i-th rule at the most recent trigger (i.e., the matching weight before the update), α be the preset decay coefficient, 0 < α < 1, and Δt be the interval between the current time t and the most recent trigger time t_last.

[0129] In other possible embodiments, updating the verification rules for automated verification based on a pre-established rule effectiveness decay model may include: updating the trigger record when the second rule is hit in automated verification, and increasing the matching weight of the second rule based on the updated trigger record. In one possible implementation, the updated matching weight can be determined by the following formula: W i(new) =min(W max W i(old) +β×(1+γ×F recent ), where W i(new) Let W be the matching weight updated for the i-th rule. max W is the preset maximum matching weight. i(old)Let F be the matching weight before the i-th rule is updated, β be the preset enhancement coefficient (β>0), γ be the preset frequency gain coefficient (γ>0), and F be the matching weight before the i-th rule is updated. recent This refers to the historical trigger frequency.

[0130] Optionally, the aforementioned automated verification further includes: acquiring multi-source associated data related to reimbursement vouchers, which includes at least: merchant names and transaction times identified from payment voucher images (such as payment screenshots), employee travel application and trajectory data, and historical food purchase price data from the enterprise's procurement system; constructing a fraud risk feature vector based on the multi-source associated data and inputting it into a pre-trained fraud identification model, which outputs fraud behavior identification results, including at least one of the following: fake merchant risk, time-based fraud risk, and abnormal food price risk; and triggering the highest level of compliance anomaly warning when the identified fraud risk exceeds a preset risk threshold.

[0131] The AI-based expense reimbursement voucher information processing method provided in this embodiment has the same implementation principle and technical effect as the aforementioned AI-based expense reimbursement voucher information processing system embodiment. For the sake of brevity, any parts not mentioned in the AI-based expense reimbursement voucher information processing method embodiment can be referred to the corresponding content in the aforementioned AI-based expense reimbursement voucher information processing system embodiment.

[0132] like Figure 4 As shown, an electronic device 400 provided in this embodiment of the invention includes: a processor 401, a memory 402 and a bus. The memory 402 stores a computer program that can run on the processor 401. When the electronic device 400 is running, the processor 401 and the memory 402 communicate through the bus. The processor 401 executes the computer program to implement the above-mentioned AI-based expense reimbursement voucher information processing method.

[0133] Specifically, the memory 402 and processor 401 mentioned above can be general-purpose memory and processor, without any specific limitations here.

[0134] This invention also provides a computer-readable storage medium storing a computer program. When a processor runs this computer program, it executes the AI-based expense reimbursement voucher information processing method described in the preceding method embodiments. The computer-readable storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), RAM, magnetic disks, or optical disks.

[0135] In this document, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0136] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limitations; therefore, other examples of exemplary embodiments may have different values.

[0137] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0138] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interface, and the indirect coupling or communication connection between modules may be electrical, mechanical, or other forms.

[0139] The modules described as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0140] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for processing expense reimbursement voucher information based on AI, characterized in that, include: When a reimbursement voucher to be processed is obtained, the AI ​​recognition interface is called to identify the information of the reimbursement voucher and obtain the target recognition information; wherein, the reimbursement voucher includes an image of a hospitality menu; Based on dynamically configured field mapping rules, the target identification information is mapped to the fields corresponding to the reimbursement details to obtain the target reimbursement details information; The target expense reimbursement details are automatically verified, including consistency verification between menu amount and invoice amount and compliance verification based on the list of prohibited reimbursement items. When an abnormal event is detected during the processing of the reimbursement voucher, a corresponding warning message is triggered so that the relevant user can handle and review the abnormality, and the processing status of the reimbursement voucher is updated based on the feedback results of the warning message; wherein, the abnormal event includes information identification abnormality, amount matching abnormality, or compliance abnormality.

2. The AI-based expense reimbursement voucher information processing method according to claim 1, characterized in that, The reimbursement voucher also includes a payment voucher image; the step of using an AI recognition interface to perform information recognition on the reimbursement voucher to obtain target recognition information includes: Generate a voucher identification request with the voucher identifier and voucher type; wherein, the voucher type includes the payment voucher corresponding to the payment voucher image and the hospitality menu corresponding to the hospitality menu image; The identification request is sent to the AI ​​identification interface to identify the information of the reimbursement voucher under the corresponding voucher type; The system receives the raw recognition information returned by the AI ​​recognition interface and performs standardization processing on the raw recognition information to obtain the target recognition information.

3. The AI-based expense reimbursement voucher information processing method according to claim 1, characterized in that, The dynamically configured field mapping rule maps the target identification information to the fields corresponding to the expense reimbursement details, thereby obtaining the target expense reimbursement detail information, including: Obtain the currently configured mapping table, which defines the mapping relationship between the fields identified by AI and the fields corresponding to the expense reimbursement details; According to the mapping table, the target identification information is assigned to the field corresponding to the reimbursement details to obtain the target reimbursement details information; wherein, during the assignment process, for the payment time field in the reimbursement details, the original time information obtained by identification is first converted into a standard date format before being assigned; if the conversion fails, the original time information is assigned as text.

4. The AI-based expense reimbursement voucher information processing method according to claim 1, characterized in that, The consistency verification between the menu amount and the invoice amount includes: Compare the menu amount in the target reimbursement details with the invoice amount in the reimbursement document associated with the reimbursement voucher; If the menu amount is less than the invoice amount, an amount matching error is triggered, and the submission of the reimbursement document is blocked until the consistency check passes.

5. The AI-based method for processing expense reimbursement voucher information according to claim 1, characterized in that, The compliance verification based on the list of prohibited reimbursement items includes: The names of dishes in the target reimbursement details are matched with keywords from a preset list of prohibited dishes; wherein, the list of prohibited dishes can be dynamically configured by the user. If a dish name is found to match in the target expense reimbursement details, a compliance exception is triggered, and the submission of the expense reimbursement document associated with the expense reimbursement voucher is blocked until the compliance verification is passed.

6. The AI-based method for processing expense reimbursement voucher information according to claim 1, characterized in that, The triggering of the corresponding warning information includes: Based on the type of the abnormal event, an early warning message with differentiated handling guidelines is generated and pushed to the user end, the review end, or the administrator end through at least one of the following channels: pop-up window of the reimbursement system interface, enterprise instant messaging tool message, or email. Specifically, the warning information corresponding to the information identification anomaly is used to guide the user to manually fill in the missing information; the warning information corresponding to the amount matching anomaly is used to guide the user to supplement the explanation of the reason; and the warning information corresponding to the compliance anomaly is used to notify the reviewer or administrator to review.

7. The AI-based method for processing expense reimbursement voucher information according to claim 1, characterized in that, The AI-based method for processing expense reimbursement voucher information also includes: Based on a preset statistical period, a statistical analysis is performed on preset operation and maintenance indicators; wherein, the operation and maintenance indicators include at least one of the following: information identification success rate, trigger frequency and processing time of different types of abnormal events, and automatic verification interception rate. The analysis results are displayed in the form of visual charts on the administrator's operation and maintenance dashboard to assist in system optimization or risk strategy adjustment.

8. An AI-based expense reimbursement voucher information processing system, characterized in that, include: The recognition module is used to identify the information of the reimbursement voucher by calling the AI ​​recognition interface when a reimbursement voucher to be processed is obtained, thereby obtaining target recognition information; wherein, the reimbursement voucher includes an image of a hospitality menu. The mapping module is used to map the target identification information to the fields corresponding to the reimbursement details based on dynamically configured field mapping rules, so as to obtain the target reimbursement details information. The verification module is used to automatically verify the target reimbursement details. The automatic verification includes consistency verification between the menu amount and the invoice amount, as well as compliance verification based on the list of prohibited reimbursement dishes. The early warning module is used to trigger corresponding early warning information when an abnormal event is detected during the processing of the reimbursement voucher, so that the relevant user can handle and review the abnormality, and update the processing status of the reimbursement voucher based on the feedback results of the early warning information; wherein, the abnormal event includes information identification abnormality, amount matching abnormality, or compliance abnormality.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the AI-based expense reimbursement voucher information processing method according to any one of claims 1-7.

10. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program, when run by the processor, executes the AI-based expense reimbursement voucher information processing method according to any one of claims 1-7.

Images