CN120047255A - Reimbursement data management system and method based on AI artificial intelligence - Google Patents

Abstract

The invention provides an AI artificial intelligence-based reimbursement data management system and method, wherein the system comprises a multi-mode data acquisition fusion module, an audit and compliance judging module, an approval process optimizing module, an analysis and prediction module, a data storage and blockchain module and a blockchain module, wherein the multi-mode data acquisition fusion module is used for providing an accessed reimbursement application entrance, the audit and compliance judging module is used for performing invoice authenticity identification, rational identification, reimbursement item classification and automatic audit, the approval process optimizing module is used for learning and optimizing reimbursement approval processes, automatically simplifying or increasing approval levels according to reimbursement application characteristics, the analysis and prediction module is used for performing multidimensional analysis on reimbursement data based on a big data processing frame and a data mining algorithm and generating a visual report and a visual result, and the data storage and blockchain module is used for constructing a distributed and non-tamperable data storage account book by adopting a blockchain technology and executing reimbursement process automation rules by using a blockchain intelligent contract. The invention comprehensively improves the efficiency, quality and safety of reimbursement data management, and promotes the enterprise financial management to develop towards intellectualization and refinement.

Description

Reimbursement data management system and method based on AI artificial intelligence

Technical Field

The invention relates to the technical field of reimbursement data management, in particular to an reimbursement data management system and method based on AI artificial intelligence.

Background

In the complex operating environment of modern enterprises, reimbursement services are a key component of financial management, and traditional management modes have difficulty in adapting to increasing data processing requirements and efficient management requirements.

The traditional reimbursement flow is highly dependent on manual operation, staff spends a great deal of time and effort to finish paper notes and fill in fussy reimbursement sheets, financial staff need to check the authenticity and compliance of massive reimbursement information and notes one by one, which not only results in long reimbursement period and low efficiency, but also causes the problems of data input errors, auditing deviation and the like due to factors such as fatigue, negligence and the like in manual operation, thereby seriously affecting the accuracy and reliability of financial data and possibly leading to adverse consequences such as financial statement distortion, decision errors and the like.

The traditional reimbursement data management has serious limitation in the aspect of data analysis, can only carry out simple classified summarization statistics, cannot deeply mine the hidden rich information behind the data, such as potential rules of cost expenditure, intelligent identification of abnormal consumption modes, accurate optimization strategies of cost control, internal association relations with other business data of enterprises and the like, and is difficult to meet urgent requirements of enterprise high-level decisions on comprehensive, deep and prospective data insight, and the strain capacity and innovation development potential of enterprises in market competition are greatly restricted.

In addition, as the enterprise scale is continuously enlarged, the business complexity is continuously improved, and the market environment is increasingly changeable, the traditional reimbursement management system is worry about coping with diversified reimbursement scenes, flexibly adapting to dynamic adjustment of internal management policies of the enterprise, effectively preventing various fraudulent reimbursement behaviors and the like, and an innovative reimbursement data management solution with higher intelligent degree is urgently needed so as to comprehensively improve the efficiency, quality and safety of reimbursement data management and promote the financial management of the enterprise to develop to an intelligent and refined direction.

Disclosure of Invention

The technical problem to be solved by the invention is to provide an AI artificial intelligence-based reimbursement data management system and method, so as to overcome the defects of the prior art in the background art, comprehensively improve the efficiency, quality and safety of reimbursement data management, and promote the enterprise financial management to develop to an intelligent and refined direction.

In order to solve the technical problems, the embodiment of the invention provides the following technical scheme:

in one aspect, an embodiment of the present invention provides an AI artificial intelligence based reimbursement data management system, including:

The multi-mode data acquisition fusion module is used for providing an accessed reimbursement application inlet which is used for selecting manual input reimbursement information, uploading invoice pictures and other evidence files and inputting reimbursement events by voice;

the verification and compliance judging module is used for constructing an invoice authenticity identification model based on a deep learning algorithm, interactively verifying invoice authenticity with a tax department database and a third party verification platform, constructing a reimbursement event analysis model by using a natural language understanding technology and a machine learning algorithm, and accurately classifying and automatically verifying reimbursement items by adopting a reimbursement item classification and verification model based on the combination of a rule engine and machine learning;

The approval process optimization module is used for determining an optimal approval path and an approval node sequence by using a reinforcement learning algorithm based on an enterprise organization structure, a business process, a risk control strategy and historical approval data;

The analysis and prediction module is used for carrying out multidimensional analysis on reimbursement data based on a big data processing frame and a data mining algorithm, mining potential association relations between reimbursement data and other business data by using association rule mining, cluster analysis and classification algorithms, predicting future reimbursement requirements, cost trends and cost changes by using historical data based on a machine learning prediction model, and generating a visual report and a visual result;

the data storage and blockchain module is used for constructing a distributed and non-tamperable data storage account book by adopting a blockchain technology, encrypting and storing reimbursement data and related auditing and approval information, executing reimbursement flow automation rules by utilizing a blockchain intelligent contract, and executing data verification logic.

Optionally, the auditing and compliance judging module comprises an invoice authenticity identifying sub-module, a reimbursement event analysis sub-module and a reimbursement item classifying sub-module, wherein:

the invoice authenticity identification sub-module is used for deep learning and training of massive invoice sample data, wherein the invoice sample data comprises an invoice printing format, an anti-counterfeiting mark and a code rule, and the invoice sample data is subjected to data interaction verification with an authoritative invoice database of a tax department and a third party invoice verification platform through verification data output by a deep learning algorithm;

The reimbursement event analysis sub-module is used for constructing an intelligent reimbursement event analysis model by using a natural language understanding technology and a machine learning algorithm, carrying out semantic understanding and logic analysis on reimbursement event description submitted by staff, and intelligently judging the rationality of reimbursement event by combining a detailed reimbursement policy knowledge base, an industry standard specification and a historical reimbursement data case base in an enterprise;

And the reimbursement item classification sub-module adopts a reimbursement item intelligent classification and auditing model based on the combination of a rule engine and machine learning, accurately classifies reimbursement items according to a reimbursement item classification function according to a preset reimbursement item classification rule and a classification mode automatically learned through machine learning, and automatically audits different types of reimbursement items according to corresponding auditing standards and threshold ranges.

Optionally, the approval process optimization module is configured to use a reinforcement learning algorithm to base on an enterprise organization structure, a business process, a risk control policy, and historical approval data, and specifically includes:

Setting a state space as S, wherein the state space comprises state information of the amount, the cost type, departments where the applicant is located and historical reimbursement credit records, the action space is A, the operation space comprises different operations in an approval process, a reward function is R (S, a), a reward value is determined according to an approval result and a business target, the state transition probability is P (S _t+1|s_t,a_t) and represents the probability of transition to a state S _t+1 after taking action a _t in the state S _t, the cost function is V (S), and an update formula of the cost function is based on a Belman equation:

where η is the learning rate and γ is the discount factor.

Optionally, the analyzing and predicting module uses association rule mining, cluster analysis mining and classification algorithm to mine potential association relation between reimbursement data and other service data, specifically:

(1) The potential association relation between the reimbursement data and other business data is mined through association rules specifically as follows:

integrating reimbursement data with other related business data, wherein the other related business data comprises sales data, purchasing data, inventory data and manpower resource data;

Cleaning the data, processing missing values, abnormal values and repeated data, coding and converting the data, and converting the classified variables into numerical values;

Performing association rule mining by selecting an Apriori algorithm or an FP-Growth algorithm, finding out frequent item sets and association rules between reimbursement data and other service data, analyzing the mined association rules, and understanding the service meaning of the association rules;

(2) The potential association relation between the reimbursement data and other business data is specifically:

Determining characteristics for cluster analysis, including key indexes in reimbursement data and related characteristics in other business data, and carrying out standardization processing on the selected characteristics so that different characteristics have the same dimension and data distribution;

Performing cluster analysis by adopting a K-Means clustering algorithm, a DBSCAN algorithm or a hierarchical clustering algorithm, calculating a cluster evaluation index, evaluating a clustering effect, and adjusting a clustering algorithm parameter or selecting a more suitable clustering algorithm according to an evaluation result;

Analyzing the clustering result, endowing business meaning to each cluster, and providing decision advice for enterprises according to the clustering result;

(3) The potential association relation between the reimbursement data and other service data is mined through a classification algorithm specifically as follows:

determining a classification task target, selecting and extracting characteristics related to the classification task, and selecting the characteristics with higher importance for classification model training by adopting a characteristic selection algorithm;

selecting a classification algorithm, training the selected classification algorithm by using a training data set, and adjusting algorithm parameters to optimize model performance;

And evaluating the trained classification model by using a test data set, calculating an evaluation index, applying the trained classification model with good evaluation index to actual reimbursement data management, and performing classification prediction on new reimbursement application or data.

Optionally, in the reimbursement event analysis submodule, a reimbursement event rationality judging function is set as follows:

G(R)=α×M_p(R)+β×M_s(R)+γ×M_c(R)

Wherein, M _p (R) represents a matching degree evaluation function of R based on a reimbursement policy knowledge base, M _s (R) represents an evaluation function of R based on an industry standard specification, M _c (R) represents a similarity evaluation function of R based on a historical reimbursement data case base, alpha, beta and gamma are corresponding weight coefficients, and alpha+beta+gamma=1. When G (R) is not less than T _G, the reimbursement event is reasonable, otherwise, the reimbursement event is unreasonable, wherein T _G is a preset rationality threshold, and R is an employee description event.

Optionally, in the reimbursement item classification submodule, the reimbursement item classification function is specifically:

Wherein Ln represents an nth class reimbursement item, K _n (X) represents a classification discriminant function of the nth class reimbursement item, and T _n is a corresponding classification threshold.

Optionally, the data storage and blockchain module adopts a blockchain technology to construct a distributed and non-tamperable data storage account book, which specifically comprises the following steps:

selecting blockchain architecture based on factors of enterprise size, data privacy requirements, and performance requirements

Setting all nodes and light nodes, wherein the all nodes store complete account book data and participate in transaction verification and block packing, establish safe P2P network connection and set inter-node communication protocols;

Encrypting the reimbursed data and the auditing and approving information by adopting an asymmetric encryption algorithm, and selecting a secure hash algorithm to hash the data;

constructing an intelligent contract to define a submission rule of a reimbursement application, embedding data verification logic in the intelligent contract, and deploying the compiled intelligent contract to nodes of a blockchain network;

defining a block structure, wherein each block comprises a block head and a block body, a hierarchical storage strategy is adopted, data which is accessed frequently in a near-term is stored in a high-speed storage medium, and historical data is stored in a high-capacity low-cost storage.

Optionally, the data storage and blockchain module executes reimbursement flow automation rules by using a blockchain intelligent contract, specifically:

Dividing the intelligent contract into a plurality of functional modules, including a reimbursement application module, an approval process module, a data verification module and a notification module;

Designing state transition logic of an intelligent contract based on a state machine model, wherein the state transition logic accords with predefined business rules and flows;

When the employee submits the reimbursement application, the intelligent contract automatically checks whether the data format meets the requirements, and verifies whether the reimbursement application meets the relevant business rules according to the business reimbursement policy;

the intelligent contract automatically determines an approval level and an approval sequence according to the amount of the reimbursement application, the type of the expense and the factors of departments where staff are located;

The intelligent contract automatically acquires information of corresponding approvers from an enterprise organization architecture database or a predefined approver list, and reminds the approvers to process reimbursement applications through system notification.

Optionally, the data verification logic of the data storage and blockchain module specifically includes:

checking reimbursement amount and budget, interacting with an enterprise budget management system or a database through an intelligent contract, acquiring budget amount information of departments, projects or individuals where staff is located, calculating accumulated reimbursement amount of the staff or the projects in a current period when reimbursement application is submitted, and comparing with residual budget;

And checking the compliance of reimbursement items, namely checking whether the reimbursement items submitted by staff are correctly classified according to the reimbursement item classification standard preset by the enterprise, and verifying whether the reimbursement items accord with the compliance policy and related laws and regulations of the enterprise.

On the other hand, the embodiment of the invention also provides a working method of the reimbursement data management system based on the AI artificial intelligence, which comprises the following steps:

providing an accessed reimbursement application inlet, wherein the reimbursement application inlet is used for selecting manual input reimbursement information, uploading invoice pictures and other evidence files, and inputting reimbursement events by voice;

Constructing an invoice authenticity identification model based on a deep learning algorithm, and interactively verifying invoice authenticity with a tax department database and a third party verification platform; the method comprises the steps of constructing a reimbursement event analysis model by using a natural language understanding technology and a machine learning algorithm, and accurately classifying and automatically auditing reimbursement items by adopting a reimbursement item classification and auditing model based on the combination of a rule engine and machine learning;

Utilizing a reinforcement learning algorithm to determine an optimal approval path and an approval node sequence based on enterprise organization structures, business processes, risk control strategies and historical approval data;

The method comprises the steps of carrying out multidimensional analysis on reimbursement data based on a big data processing frame and a data mining algorithm, mining potential association relations between reimbursement data and other business data by using association rule mining, cluster analysis and classification algorithms, predicting future reimbursement requirements, cost trends and cost changes by using historical data based on a machine learning prediction model, and generating a visual report and a visual result;

and constructing a distributed and non-tamperable data storage account book by adopting a blockchain technology, encrypting and storing reimbursement data and related auditing and approval information, executing reimbursement flow automation rules by utilizing a blockchain intelligent contract, and executing data verification logic.

The technical scheme of the invention has the following beneficial effects:

1. the multi-mode data acquisition fusion module supports multiple terminals and input modes, is in butt joint with an enterprise internal system to acquire multi-source data and automatically fuses and verifies, can feed back reimbursement states in real time, is convenient for staff to operate, and ensures accurate data. The approval process optimization module utilizes a reinforcement learning algorithm to automatically adjust the approval level according to the characteristics of the reimbursement application, simplifies the small-scale conventional reimbursement process, increases the approval links of the reimbursement of major risks, predicts the approval comments, and obviously improves the reimbursement efficiency.

2. The invoice authenticity identification sub-module ensures the authenticity and legality of the invoice through deep learning and multi-platform interactive verification. The reimbursement event analysis submodule uses natural language understanding and machine learning technology and combines a multi-aspect knowledge base and a case base to accurately judge the reasonability of the reimbursement event. The reimbursement item classification sub-module accurately classifies reimbursement items and automatically audits according to standards based on a rule engine and a machine learning model.

3. The analysis and prediction module of the invention utilizes various data mining algorithms to mine the association relation between reimbursement data and other business data, and provides cross-department business insight, power-assisted optimization resource allocation and business flow. Based on the machine learning prediction model, historical data is utilized to predict future reimbursement requirements and the like, and scientific basis and prospective guidance are provided for enterprise financial budget and the like.

4. The data storage and blockchain module adopts the blockchain technology to construct a distributed and non-tamperable storage account book, encrypts stored data, and ensures confidentiality, integrity and non-tamperable property of the data through reasonable architecture selection, node setting, encryption algorithm, intelligent contracts and the like. And executing reimbursement flow automation rules by using the blockchain intelligent contract, strictly executing data verification logic, including verification of invoice authenticity, money compliance, item classification accuracy and the like, managing user rights and ensuring flow compliance transparency.

Drawings

FIG. 1 is a flow and schematic block diagram of an AI artificial intelligence based reimbursement data management system of the present invention;

FIG. 2 is a schematic diagram of an AI-based artificial intelligence based reimbursement data management system in accordance with the present invention;

FIG. 3 is a schematic block diagram of an audit and compliance determination module of the AI artificial intelligence based reimbursement data management system of the invention.

Detailed Description

In order to make the technical problems, technical solutions and advantages to be solved more apparent, the following detailed description will be given with reference to the accompanying drawings and specific embodiments.

As shown in fig. 1 and 2, a reimbursement data management system based on AI artificial intelligence, the specific working method of the reimbursement data management system of the invention comprises the following steps:

The multimodal data acquisition fusion module 101 is configured to provide an access reimbursement application portal, where the reimbursement application portal is configured to select manual input reimbursement information, upload invoice pictures and other documents, and enter reimbursement event by voice. The system comprehensively supports access of various terminal devices (including but not limited to computers, smart phones, tablet computers and the like), staff can flexibly select various modes to input reimbursement information through the module, such as manually inputting detailed information of reimbursement matters, money amount, date and the like, uploading invoice pictures, travel slips, other related documents and the like, and simultaneously can describe the reimbursement matters by utilizing an advanced voice input function, and the system automatically converts voice into structured text data by means of voice recognition and natural language processing technology. The system can realize deep seamless connection with various business systems (such as a travel reservation system, an office supplies purchasing system, an enterprise resource planning system, a human resource system, a financial system and the like) in an enterprise, automatically acquire structural data related to reimbursement, such as journey information, order details, employee departments and job level information, budget information and the like, and conduct real-time intelligent fusion and preliminary verification on the multi-source data and data manually input by employees or converted by voice, so that the integrity, consistency and accuracy of the data are ensured. The system has a powerful information feedback mechanism, and shows the accurate state of the reimbursement application to staff and approvers in real time, such as to be submitted, in the process of auditing, passed, refused, not passed, and the like, and detailed processing results and targeted prompt information, so that users are ensured to know reimbursement progress at any time.

As shown in fig. 3, the audit and compliance determination module 102 includes an invoice authenticity identification sub-module 1021, a reimbursement event analysis sub-module 1022, and a reimbursement item classification sub-module 1023, wherein:

And the invoice authenticity identification submodule 1021 is used for deep learning and training of massive invoice sample data, wherein the invoice sample data comprises printing formats, anti-counterfeiting marks and code rules of invoices, and data interaction verification is carried out on verification data output through a deep learning algorithm, an authoritative invoice database of a tax department and a third party invoice verification platform, so that the authenticity and the legality of each reimbursement invoice are ensured. When verifying the authenticity of the invoice through the deep learning algorithm, mass invoice sample data including invoices of different types (such as value-added tax special invoices, common invoices and the like) and different sources (such as invoices issued in different areas and different industries) are collected, so that the diversity and representativeness of the sample are ensured. And extracting the characteristic information such as the overall layout, the text typesetting, the font style, the color distribution and the like of the invoice aiming at the printing format of the invoice, and converting the characteristic information into structured data which can be processed by a deep learning algorithm. For example, text areas, form areas, etc. on the invoice are divided and marked, and information such as the position, size, content, etc. is recorded. For the anti-counterfeiting mark, features such as patterns, textures, color changes and the like of the anti-counterfeiting mark, such as features of watermarks, fluorescent fibers, safety lines and the like on an invoice are extracted by utilizing an image recognition technology, and the features are quantized and encoded. Analyzing the code rule of the invoice, extracting key code information such as invoice codes, numbers, invoicing dates, check codes and the like, and carrying out standardization processing to ensure uniform data formats.

Then, the deep learning model is constructed and trained, for example, a Convolutional Neural Network (CNN) is selected as an infrastructure, and an invoice authenticity identification model is constructed. The neural network with a multi-branch structure can be designed, one branch is used for specially processing the related characteristic data of the printing format, the other branch is used for processing the anti-counterfeiting mark characteristic data, the other branch is used for processing the code rule data, and finally, the results of the three branches are fused to comprehensively judge the authenticity of the invoice.

The invoice sample data is divided into a training set, a verification set and a test set, the training set is used for training the constructed deep learning model, and the verification of the true and false is carried out through the test set or the invention needing verification after the training is finished. The verification data output by the deep learning model comprises the printing format related verification data, the anti-counterfeiting mark related verification data and the code rule related verification data, for example, the deep learning model is used for learning the overall invoice layout in the invoice sample data, and the matching degree of the invoice to be verified and the standard invoice layout template is output. For example, the difference value between the relative position and the size proportion of each element (such as invoice head-up, buyer information, seller information, invoice content area, price tax total area and the like) on the invoice and the standard template is calculated, the matching degree is expressed in percentage, and the matching degree of more than 95% can be preliminarily judged to be in accordance with the standard format.

The reimbursement event analysis submodule 1022 is used for constructing an intelligent reimbursement event analysis model by using a natural language understanding technology and a machine learning algorithm, carrying out semantic understanding and logic analysis on reimbursement event description submitted by staff, and intelligently judging the rationality of reimbursement event by combining a detailed reimbursement policy knowledge base, an industry standard specification and a historical reimbursement data case base in an enterprise, wherein the reimbursement event analysis submodule is provided with a reimbursement event rationality judging function as follows:

G(R)=α×M_p(R)+β×M_s(R)+γ×M_c(R)

Wherein, M _p (R) represents a matching degree evaluation function of R based on a reimbursement policy knowledge base, M _s (R) represents an evaluation function of R based on an industry standard specification, M _c (R) represents a similarity evaluation function of R based on a historical reimbursement data case base, alpha, beta and gamma are corresponding weight coefficients, and alpha+beta+gamma=1. When G (R) is not less than T _G, the reimbursement event is reasonable, otherwise, the reimbursement event is unreasonable, wherein T _G is a preset rationality threshold, and R is an employee description event.

(1) A match evaluation function for R based on a reimbursement policy knowledge base, comprising:

Keyword matching and semantic analysis, namely performing word segmentation processing on employee description matters by R, and extracting key nouns, verbs, adjectives and the like in the employee description matters, such as keywords of 'travel cost', 'conference cost', 'training', 'purchasing', 'urgent', and the like. These keywords are then searched in a reimbursement policy knowledge base, and the number and frequency of keyword matches are counted. For example, if the reimbursement policy specifies reimbursement scope and criteria for a particular project, the matching score is increased when keywords associated with those specifications are included in the employee description. And analyzing the semantic similarity of the employee description event R and the terms in the reimbursement policy knowledge base by using a semantic understanding technology. And converting employee descriptions into semantic vector representations, and performing cosine similarity calculation with semantic vectors of clauses in a knowledge base. For example, for a travel fare reimbursement, if an employee describes "traffic and accommodation fare generated by visiting a customer to a foreign place due to a business expansion need" as having a higher semantic similarity with respect to the travel fare "necessary traffic and accommodation fare generated by a travel business activity of a company" in the reimbursement policy, the matching degree is improved.

And (3) rule matching and logic judgment, namely judging whether the employee description event accords with basic conditions and flows of reimbursement according to rules in a reimbursement policy knowledge base. For example, some fee reimbursements may require prior application or specific approval procedures, and if such rules are not mentioned or are not met in the employee description, the matching score is reduced. It is checked whether the cost details in the employee descriptive matters meet the cost types and criteria specified in the reimbursement policy. Such as traffic cost standards (such as limit of the level of the airplane ticket, regulation of the train ticket seat, etc.), accommodation cost standards (upper limit of accommodation cost divided by region and level), etc., if the traffic cost standards exceed the standard range, the matching degree is correspondingly reduced.

(2) An evaluation function for R based on industry standard specifications, comprising:

Compliance checking, namely checking whether the cost items in the staff description matters R are compliance or not according to the general cost reimbursement standards and specifications in the industry. For example, in some industries, business hospitality fees may have strict people average consumption criteria and hospitality scope restrictions, and if the business hospitality fees described by employees exceed industry criteria (e.g., people average consumption is too high, hospitality objects do not meet business association scope), then the assessment score is reduced. For professional service fees (such as legal consultation fees, examination and billing, etc.), checking whether the service provider has corresponding qualification and industry approval, if the service provider does not qualify as industry specifications, recording violation conditions and reducing the evaluation value.

And (3) judging rationality, namely analyzing whether the cost expenditure in staff description matters is reasonable from the aspect of industry practice. For example, for office supplies purchase cost, the average industry purchase price level is compared, if office supplies price reimbursed by staff is significantly higher than the market normal price and there is no reasonable reason (such as special customization requirement, etc.), it is considered unreasonable, and the evaluation score is reduced. Considering the expense range of similar business activities in the industry, if the expense described by the staff is greatly different from the conventional level of the industry (such as too high or too low), the reasonability of the staff is further verified. For example, an average cost of similar training activities within an industry is 200 yuan per day, while a training cost of staff reimbursement is 300 yuan per day, and a sufficiently reasonable interpretation is not provided (e.g., a well-known expert is invited to give lessons but no relevant proof is provided), then the assessment is unreasonable.

(3) Similarity evaluation function for R based on historical reimbursement data case base

And calculating text similarity, namely calculating the similarity of the employee description event R and the cases in the historical reimbursement data case base by adopting a text similarity algorithm (such as an edit distance algorithm, a cosine similarity algorithm and the like). The employee descriptions and the historical cases are converted into text vector representations, and the similarity degree of the employee descriptions and the historical cases is measured by calculating the distance or the similarity degree between the vectors. For example, the edit distance algorithm calculates the minimum number of edit operations (insert, delete, replace characters, etc.) required to convert one text to another, the smaller the edit distance, the higher the similarity.

And (3) extracting key features in the employee description event R and the historical reimbursement cases, such as cost types, business activity contents, occurrence places, time and the like. Different weights are allocated to different features, and the setting is carried out according to the influence degree of the different features on the reimbursement rationality. For example, the cost type is weighted higher and the business activity content is weighted less frequently. And calculating the matching degree of the features, and accumulating scores according to the weights of the matched features. If the match degree between the travel expense reimbursement described by the staff and the travel expense reimbursement in the history case is higher in key characteristics such as expense type, destination, business time and the like, the similarity score is increased, otherwise, if more differences exist (such as different expense types and non-conforming to business logic of the destination), the similarity evaluation value is reduced.

Anomaly detection and adjustment if the employee description event R is less similar to all cases in the historical reimbursement data case base (less than a set threshold, e.g., 30%), then an anomaly detection mechanism is triggered. Further analysis of details in employee descriptions checks for anomalies (e.g., excessive cost and unreasonable interpretation, unusual business activities, etc.). If an anomaly is found, the similarity assessment score is reduced and may be marked as a high risk reimbursement requiring manual review. According to the business development and market change conditions of enterprises, the historical reimbursement data case base is updated regularly, so that the similarity evaluation function is ensured to be suitable for new business modes and cost expenditure conditions. For example, as business of enterprises expands to new fields, reimbursement cases related to the new fields are timely brought in, so that accuracy of similarity evaluation is improved.

The reimbursement item classification sub-module 1023 adopts a reimbursement item intelligent classification and auditing model based on the combination of a rule engine and machine learning, accurately classifies reimbursement items according to a reimbursement item classification function according to a preset reimbursement item classification rule and a classification mode automatically learned through machine learning, and automatically audits reimbursement items of different types according to corresponding auditing standards and threshold ranges, wherein the reimbursement item classification function specifically comprises:

Wherein Ln represents an nth class reimbursement item, K _n (X) represents a classification discriminant function of the nth class reimbursement item, and T _n is a corresponding classification threshold.

The approval process optimization module 103 is configured to determine an optimal approval path and an approval node sequence based on an enterprise organization structure, a business process, a risk control policy and historical approval data by using a reinforcement learning algorithm, automatically simplify or increase an approval level according to the characteristics of an approval application, and perform automatic circulation, approval information recording and overdue reminding of the approval process, where the reinforcement learning algorithm specifically includes:

The method comprises the steps of setting a state space as S, wherein the state space comprises state information of the amount, the cost type, departments where the applicant is located and historical reimbursement credit records, setting an action space as A, wherein the action space comprises different operations in an approval process, such as approval, rejection, switching to the next-stage approval and the like, setting a reward function as R (S, a), determining a reward value according to an approval result and a business target, setting a state transition probability as P (S _t+1|s_t,a_t), representing the probability of transition to a state S _t+1 after taking action a _t in the state S _t, setting a value function as V (S), and setting an update formula of the value function based on a Bellman equation:

where η is the learning rate and γ is the discount factor.

For small and conventional reimbursement applications, the system can automatically simplify the approval process, directly and quickly approve by a direct generic leader, intelligently predict the approval opinion of the leader according to the approval history data and the behavior mode of the leader, automatically skip part of approval links if the approval passes and the accuracy reaches a certain threshold, directly enter a financial audit or reimbursement payment stage, and remarkably improve reimbursement efficiency, and automatically increase approval levels and professional audit links for reimbursement applications involving major project investment, high cost expenditure or higher risk, such as cost benefit analysis of financial experts, compliance audit of legal departments, risk assessment of audit departments and the like, so as to ensure rigorous, detailed and controllable risk in the approval process.

The analysis and prediction module 104 performs multidimensional analysis on the reimbursement data based on a big data processing frame and a data mining algorithm, and uses association rule mining, cluster analysis and classification algorithm to mine potential association relations between the reimbursement data and other service data, specifically:

1. Association rule mining

First, data preparation and preprocessing, integrating reimbursement data with other related business data, such as sales data (sales amount, sales order number, etc.), purchase data (purchase amount, purchase item category, etc.), inventory data (inventory turnover, inventory level, etc.), human resource data (employee number, employee performance, etc.), etc. And ensuring that the data are in the same data warehouse or data lake, and performing data cleaning to process missing values, abnormal values and repeated data.

The data is then encoded and converted to a numerical type to convert the classification variables for processing by the association rule mining algorithm. For example, the reimbursement item categories (e.g., travel fees, office supplies fees, etc.) are thermally encoded alone.

And thirdly, carrying out algorithm selection and parameter setting, and carrying out association rule mining by selecting an Apriori algorithm or an FP-Growth algorithm. For the Apriori algorithm, the appropriate minimum support (min_support) and minimum confidence (min_confidence) parameters are set. The minimum support determines the lowest frequency threshold at which frequent item sets occur, and the minimum confidence measures the reliability of the association rule. For example, min_support=0.05 (indicating that the item set appears in the dataset at least 5%) and min_confidence=0.6 (indicating that the confidence of the association rule is at least 60%) may be initially set, and then the adjustment optimization is performed according to the mining result. For the FP-Growth algorithm, the memory use and performance optimization parameters are mainly focused on, so that the algorithm can efficiently process large-scale data.

And finally, carrying out association rule mining and analysis, and running an association rule mining algorithm to find out frequent item sets and association rules between the reimbursement data and other service data. For example, it may be found that there is a correlation between "the travel reimbursement amount is high" and "the sales expansion activity is frequent (the number of sales orders is increased)", with a support degree of 0.1 (i.e., both cases occur in 10% of the data at the same time), and a confidence degree of 0.7 (indicating that 70% of the probability sales expansion activities are frequent in the case of the travel reimbursement amount is high).

And analyzing the mined association rule to understand the business meaning. In addition to the above-described association of sales with travel fees, it is also possible to find that "increase in office supplies purchase amount" is associated with "new employee attendance (increase in employee count)" or "increase in equipment maintenance fee" is associated with "decrease in production yield (possibly suggesting equipment failure affects production)", or the like. These association rules may provide cross-department business insight to the enterprise that helps optimize resource allocation and business processes.

And visually displaying the association rule, such as using an association rule diagram or a matrix diagram, and visually presenting the relationship between the reimbursement data and other service data. For example, in the association rule diagram, a node represents a reimbursement item or a business index, an edge represents an association relationship, and the thickness or color of the edge may represent an association strength (support degree or confidence degree).

2. Cluster analysis

First, features for cluster analysis are determined, including key indicators in reimbursement data (e.g., reimbursement amount, reimbursement frequency, reimbursement item category, etc.) and related features in other business data (e.g., sales performance, inventory turnover, employee performance score, etc.). Features with representativeness and discrimination are selected, and dimension disasters caused by excessive features are avoided. And (3) carrying out standardization processing on the selected features to enable different features to have the same dimension and data distribution. For example, the Z-Score normalization method is used to convert the feature values into a standard normal distribution with a mean value of 0 and a standard deviation of 1, so that the clustering algorithm is ensured not to deviate due to the feature scale difference.

And secondly, adopting a K-Means clustering algorithm, a DBSCAN algorithm or a hierarchical clustering algorithm and the like to perform clustering analysis. And for the K-Means algorithm, primarily determining the clustering quantity K according to service understanding and data distribution. For example, if it is desired to divide an enterprise division or a staff into different groups of cost behavior patterns, different values such as k=3 or k=4 may be tried first, and then the optimal K value is selected by evaluating the index.

And calculating a clustering evaluation index, such as a contour coefficient (Silhouette Coefficient), a Calinski-Harabasz index and the like, and evaluating the clustering effect. The contour coefficient measures the average ratio of the distance from each data point to the cluster center to which the data point belongs to the distances from other cluster centers, the value range is between-1 and 1, and the closer to 1, the better the clustering effect is. The Calinski-Harabasz index evaluates cluster quality by calculating the ratio of the inter-class dispersion to the intra-class dispersion, the larger the ratio the better.

And adjusting the parameters of the clustering algorithm or selecting a more proper clustering algorithm according to the evaluation result so as to obtain the optimal clustering effect. For example, if the K-Means algorithm is found to be undesirable in some cases (e.g., there are more outliers or irregular cluster shapes), an attempt may be made to use the DBSCAN algorithm that automatically finds dense regions and noise points in the data, without much assumption about the data distribution shape, and is more suitable for handling clusters of complex shapes.

And analyzing the clustering result, and endowing business significance to each cluster. For example, by clustering analysis, enterprise personnel are divided into several groups, one group may be "high reimbursement high-performance staff" which reimburse for a higher amount but bring higher business value (such as sales performance prominence) to the enterprise, another group may be "low reimbursement low-performance staff" which needs to pay further attention to the working performance and resource utilization conditions, and an "abnormal reimbursement staff group" which reimburses for a significantly different behavior from other groups may have potential illegal reimbursement risks and need important audits.

And providing decision suggestions for enterprises according to the clustering result. For example, for a group of "high reimbursement high performance staff", it can be considered to provide more flexible reimbursement policies or incentive measures to encourage them to continue to create value for the enterprise, and for a group of "abnormal reimbursement staff", internal audit and cost management are enhanced, reimbursement flows are normalized, and enterprise resource waste and loss are prevented. Meanwhile, the clustering result can also help enterprises to find out the similarity and the difference between different business departments or business processes, and a reference basis is provided for organization architecture optimization and business process reconstruction.

3. Classification algorithm application

First, a classification task objective is determined, such as predicting whether there is a fraud risk (classified into fraud and normal) in the reimbursement application or determining whether the reimbursement fee exceeds a budget (classified into super-budget and non-super-budget), etc. And labeling the data according to the target, and constructing a training data set and a testing data set. The labeling process can combine enterprise internal audit results, financial rules, business expert experience, and the like.

And performing feature engineering, and selecting and extracting features related to the classification task. In addition to characteristics of the reimbursement data itself (e.g., reimbursement amount, item category, time, etc.), other business data introduction characteristics such as employee credit records (from human resource data), vendor reputation (from procurement data), market wave conditions (from external market data), etc. may also be considered as an aid. And screening and combining the features, removing redundant and irrelevant features, and improving the accuracy and efficiency of the classification model.

Feature selection algorithms, such as Chi-Square Test (Chi-Square Test), information Gain (Information Gain) and the like, can be used for evaluating the importance of the features, and features with higher importance are selected for classification model training. For example, the two features of finding a deviation of the reimbursement amount from the average reimbursement amount by chi-square inspection and whether the vendor is a new vendor are of high importance for predicting reimbursement fraud risk.

Suitable classification algorithms are selected, such as decision tree algorithms (C4.5, CART, etc.), support Vector Machines (SVMs), naive bayes classifiers, neural networks (e.g., multi-layer perceptron MLP), etc. The different algorithms have respective advantages and disadvantages and application scenes, such as decision tree algorithms are easy to understand and interpret and are suitable for processing data with a hierarchical structure, SVM (support vector machines) perform better when processing small-sample and high-dimensional data, and neural networks have strong nonlinear modeling capability and are suitable for complex classification tasks, but have higher computational complexity and relatively poor interpretation.

The selected classification algorithm is trained using the training dataset, and algorithm parameters are adjusted to optimize model performance. For example, the depth of the tree, split node selection criteria (e.g., information gain ratio, base index, etc.), the kernel function type (e.g., linear kernel, polynomial kernel, radial basis function kernel, etc.), and penalty parameter C may be adjusted for decision tree algorithms, and parameters such as number of hidden layers, learning rate, number of iterations, etc., may be adjusted for neural networks. And the optimal parameter combination is selected by methods such as cross verification and the like, so that the generalization capability of the classification model is improved.

The trained classification model is evaluated using the test dataset, and evaluation metrics such as Accuracy (Accuracy), precision (Precision), recall (Recall), F1 values, etc. are calculated. The accuracy rate represents the proportion of the number of samples of the model predicted to be correct to the total number of samples, the accuracy rate measures the proportion of the number of samples of the model predicted to be positive and the actual positive to the number of samples of the model predicted to be positive, the recall rate represents the proportion of the number of samples of the model predicted to be positive to the actual positive to the number of samples of the actual positive, and the F1 value is the harmonic mean value of the accuracy rate and the recall rate, and the influence of the accuracy rate and the recall rate is comprehensively considered. Judging whether the performance of the model meets the service requirement according to the evaluation index, and if the performance is not ideal, further optimizing the model or adjusting the characteristic engineering.

And applying the trained and well-performing classification model to actual reimbursement data management, and performing classification prediction on new reimbursement applications or data. For example, at the time of filing of a reimbursed application, the system automatically predicts whether the application is at risk of fraud or over-budget using the classification model, and takes corresponding measures based on the prediction. If the risk is predicted to be high, the system can automatically trigger a manual auditing process or remind related departments to pay important attention, so that the financial risk of enterprises is effectively prevented, and the intelligent level and decision support capability of reimbursement data management are improved. Meanwhile, new data are continuously collected, and classification models are updated and optimized regularly to adapt to the changes of enterprise business environments and reimbursement behavior patterns.

Then, based on a machine learning prediction model, predicting future reimbursement requirements, cost trends and cost changes by utilizing historical data, and generating a visual report and a visual result;

Based on a machine learning prediction model, such as a time sequence prediction model, a regression analysis model and the like, future reimbursement requirements, cost trends, cost changes and the like are accurately predicted by utilizing the time sequence characteristics of historical reimbursement data and influence factors of related business data. Let the historical reimbursement data sequence be Y1, Y2,..yt (t is the time point), the prediction model be M _F, the prediction function be Y _t+h (h is the prediction step), then:

Y _t+h＝M_F (Y1, Y2,., yt)

For a simple autoregressive moving average (ARIMA) model, it is in the form of

Wherein, B is a hysteresis operator,And θ (B) is a polynomial, Δ ^d is a difference operator, ε _t is a white noise sequence. Model parameters are determined through fitting of historical data, and further prediction of future reimbursement data is achieved. The prediction result can provide scientific basis and prospective guidance for enterprise financial budget planning, cost control strategy formulation, resource planning and allocation and the like, help enterprises to make preparation for coping in advance, and reduce management risk.

The data storage and blockchain module 105 is configured to construct a distributed, non-tamperable data storage account book by using a blockchain technology, and encrypt and store reimbursement data and related audit and approval information. The data storage and blockchain module adopts the blockchain technology to construct a distributed and non-tamperable data storage account book, which comprises the following specific steps:

1. Block chain architecture selection and network construction

Firstly, the block chain architecture is selected by comprehensively considering factors such as enterprise scale, data privacy requirements, performance requirements and the like. For large enterprises and reimbursement data management involving multi-department collaboration, a federated chain architecture is suitable. The method can ensure the decentralization to a certain extent and can meet the requirements of enterprises on data privacy and authority management. For example, financial departments, business departments, audit departments and the like within an enterprise can serve as nodes of a federation chain to jointly maintain the security and stability of an account book.

Then, the different federated chain platforms, such as HYPERLEDGER FABRIC, corda, etc., are compared. HYPERLEDGER FABRIC provides high modularity and expandability, supports various consensus algorithms (such as PBFT, raft and the like), and can be flexibly configured according to the actual situation of enterprises. The Channel (Channel) concept allows data isolation and sharing among different business scenes or departments, and is very suitable for reimbursement of data interaction requirements of different processes and departments in data management.

Next, node types and distributions are determined. And setting full nodes and light nodes, wherein the full nodes store complete account book data and participate in transaction verification and block packing, such as financial department nodes, and the light nodes can only store part of key data for query verification, such as department nodes where ordinary staff is located. Nodes are distributed reasonably, and reliability of the network and redundant backup of data are ensured.

Finally, network parameters are configured. And establishing a safe P2P network connection, and setting a communication protocol (such as a TLS protocol) between nodes to ensure the safety of data transmission. Meanwhile, a node joining and exiting mechanism is configured, and identity authentication is performed by issuing a digital certificate through CA, so that node authority is strictly managed, and illegal node access is prevented.

2. Data encryption scheme design and implementation

Firstly, adopting an asymmetric encryption algorithm (such as ECC or RSA) to encrypt the reimbursement data and the approval information. And for reimbursement data submitted by staff, the public key encryption of a receiver (such as a financial node) is used, so that only a financial department can decrypt and view the reimbursement data by using a private key, and the confidentiality of the data in the transmission and storage processes is ensured.

For sensitive fields (such as employee identity information, bank account numbers, etc.), additional encryption measures such as AES symmetric encryption algorithm are used in combination with the key management system. And generating a random symmetric key to encrypt the sensitive field, then encrypting the symmetric key by using an asymmetric encryption algorithm, and storing the encrypted symmetric key and encrypted data together to ensure the high security of sensitive information.

Then, a secure hash algorithm such as SHA-256 is selected to hash the data. Before data storage, a hash value is calculated for each data block (e.g., a reimbursement record) and stored in a blockchain ledger. Any data modification can cause the hash value to change, and whether the data is tampered or not can be quickly detected by comparing the hash values.

Finally, a merck tree (MERKLE TREE) structure is constructed. And constructing a merck tree by taking the hash values of the plurality of reimbursed data blocks as leaf nodes, wherein the hash value of the root node is stored in the block head. By verifying hash values on the merck tree path, the integrity of the data block can be efficiently verified while reducing the effort of data storage and verification.

3. Intelligent contract function development and deployment

First, the reimbursement flow is automated. The method comprises the steps of establishing an intelligent contract to define a submitting rule of a reimbursement application, such as data format verification, filling item inspection and the like, automatically triggering an approval process, determining an approval level and an approval person according to preset rules (such as the amount of money, the type of expense and the like), automatically recording and feeding back an approval result, and improving the efficiency and transparency of the reimbursement process.

Data authentication and rights management. Data verification logic is embedded in the smart contract, such as invoice authenticity verification (by interacting with an external invoice verification platform), reimbursement amount compliance checking (against budget data), reimbursement item classification accuracy verification, and the like. At the same time, user rights are managed, ensuring that only authorized personnel can perform specific operations, such as submitting reimbursements, approving reimbursements, etc.

The smart contract code is written using a suitable programming language, such as Solidity or the HYPERLEDGER FABRIC-based Chaincode language. And (3) following the security coding specification, carrying out strict code examination and testing, and preventing data security problems caused by contract loopholes.

And deploying the compiled intelligent contracts to nodes of the blockchain network. Prior to deployment, adequate testing, including unit testing, integration testing, and simulated environment testing, is performed to ensure proper performance in all but one situations. After deployment, a contract upgrading mechanism is established so that contracts can be updated in time when business requirements change or vulnerabilities are found.

4. Ledger data storage structure and index optimization

A block structure is defined. Each chunk contains a chunk header (containing a version number, timestamp, previous chunk hash value, merck root hash value, etc.) and a chunk body (containing a series of reimbursed transaction data). The reimbursement transaction data is used for recording reimbursement bill numbers, employee information, reimbursement project details, amounts, invoice information, audit opinions, approval results, time and the like in detail, so that the integrity and traceability of the data are ensured.

Optimizing the data storage mode. Data that is accessed frequently in recent times is stored in a high-speed storage medium (such as a memory or SSD) using a tiered storage strategy, and historical data is stored in a high-capacity low-cost storage (such as HDD). Meanwhile, the partition storage is performed according to the characteristics of the data, such as partition according to time or partition according to business type, so that the data query and management efficiency is improved.

A multi-dimensional index is established. Indexes such as B-tree indexes or hash indexes are established for common query conditions (such as reimbursement single numbers, employee numbers, reimbursement time ranges, approval states and the like), so that the data query speed is accelerated. Meanwhile, a composite index, such as an index of (employee number and reimbursement time), is established, so that the multi-condition query requirement is met.

And optimizing a query algorithm. The technology of paging inquiry, caching inquiry results and the like is adopted, so that performance degradation caused by loading a large amount of data at one time is avoided. For complex queries (such as related to multi-table association or statistical analysis), the intelligent contract function of the blockchain is utilized to perform pre-calculation and data aggregation, the results are stored in contracts, and the results are directly obtained during query, so that the response speed is improved.

The data storage and blockchain module 105 executes reimbursement flow automation rules using blockchain intelligence contracts, specifically:

1. intelligent contract architecture design

First, the intelligent contract is divided into a plurality of functional modules, such as a reimbursement application module, an approval process module, a data verification module, a notification module, and the like. Each module focuses on specific business logic, improving maintainability and extensibility of contracts. For example, the reimbursement application module is responsible for receiving and primarily verifying reimbursement application data submitted by staff, the approval process module determines an approval path and processes approval operation according to preset rules, the data verification module strictly verifies various aspects of reimbursement data (such as invoice authenticity, amount compliance and the like), and the notification module is used for sending reimbursement state update and important event reminding to related staff.

And secondly, communication and data interaction are carried out among the modules through standardized interfaces, so that the consistency of data and the smoothness of a flow are ensured. For example, the reimbursement application module transmits the data to the approval process module through the interface after receiving the complete and preliminary verified application data, and the data verification module acquires related data from other modules and returns the verification result to the calling module when the data needs to be verified.

Again, the state transition logic of the smart contract is designed based on the state machine model. Reimbursement flows typically include a number of states, such as "application submit", "approval in", "approval pass", "approval reject", "reimbursed", etc. The intelligent contract is triggered according to different events (such as submitting reimbursement application by staff, approving approval operation by an approver, and the like), so that legal conversion between states is realized. For example, when an employee successfully submits a reimbursement application and data verification passes, the contract state is converted from "initial" to "application submission", and after the approver approves the reimbursement application, the state is converted to "approval passing", and the subsequent flow continues to advance according to the state.

The intelligent contract can ensure that each state transition accords with the predefined business rule and flow requirement, and illegal or wrong state change is prevented. For example, the reimbursement application can only enter a reimbursed state if all necessary audit and approval links are completed and passed, and if a data problem or violation is found in the approval process, the contract state should be converted into an approval rejection state and the rejection reason is recorded.

2. Reimbursement flow automation rule execution

The intelligent method automatically checks whether the data format meets the requirements when the employee submits the reimbursement application, such as whether the date format is correct, whether the amount data is legal, whether the character filling is complete, and the like. For example, the reimbursement date must be in the format of "YYYY-MM-DD", the reimbursement amount must be positive and two decimal places remain, and the employee name, department, etc. must not be empty. If the data format is incorrect or incomplete, the contract refuses to accept the application and prompts the employee for modification.

And secondly, verifying whether the reimbursement application accords with the related business rules according to the business reimbursement policy. For example, checking whether the reimbursed project is within the limits of the business's allowed reimbursement (e.g., some businesses may limit reimbursement for entertainment consumption), determining whether the same project is subject to repeated reimbursement (by comparison with historical reimbursement data), verifying that the reimbursement amount exceeds the budget limits of the employee or project (by interacting with a budget management module or database to obtain budget information). If the business rule is violated, the contract automatically marks the application as abnormal and informs relevant personnel.

The intelligent contract automatically determines the approval level and the approval sequence according to the amount of the reimbursement application, the type of the expense, the department where the staff is located and other factors. For example, setting small reimbursement (for example, the amount is less than 1000 yuan) only needs to be directly under the upper-level approval, medium reimbursement (1000 yuan to 5000 yuan) needs to be approved by a department manager and a financial manager, and large reimbursement (more than 5000 yuan) needs to be approved by a company high-level lead. Also, depending on the type of fee (e.g., travel fee, office fee, business listing fee, etc.), different specialized approval personnel or departments may be involved.

The contract automatically acquires the information of the corresponding approver from the enterprise organization architecture database or the predefined approver list, and reminds the approver to timely process the reimbursement application through system notification (such as an e-mail, an enterprise internal message system and the like). After the approver logs in the system, the detailed information of the reimbursement application and the auxiliary auditing opinions (such as data verification results, historical reimbursement record references and the like) are displayed on the intelligent closing and closing interface, so that the approver can conveniently make an accurate decision. After the approval person finishes the approval operation (approving, rejecting or returning to modify), the contract automatically records the approval result and the opinion, and pushes the reimbursement application to the next approval link (if any) or ends the approval process according to the approval process rule.

In addition, the longest approval time of each approval link is set, such as 2 working days. If the approver does not process the reimbursement application within a specified time, the contract automatically triggers a reminding mechanism (such as sending an hastening notice), processes the reimbursement according to a preset rule after a certain period of time is exceeded (such as 1 working day is exceeded), and automatically transfers the application flow to a superior approver or marks the application flow as abnormal waiting for manual intervention.

3. Data verification logic implementation

The intelligent contract is interacted with an enterprise budget management system or a database to acquire budget amount information of departments, projects or individuals where staff are located, wherein the budget amount information comprises annual budget, monthly budget and budget allocation conditions of different expense types. At the time of filing the reimbursement application, the contract calculates the accumulated reimbursement amount of the employee or project in the current period and compares the accumulated reimbursement amount with the residual budget. If the reimbursement amount plus the accumulated reimbursement amount exceeds the budget amount, the contract marks the reimbursement application as "super budget" and operates according to the super budget process flow specified by the enterprise, such as prompting the employee to adjust the reimbursement amount, requesting to provide additional budget oversubscription, or starting a special approval flow (such as co-approving the super budget portion by a department manager and a financial manager).

According to the preset reimbursement item classification standard of the enterprise, the intelligent contract checks whether reimbursement items submitted by staff are correctly classified. For example, it is determined whether a fee should be classified as a travel fee or a business listing fee, ensuring accuracy of fee statistics and analysis. If the classification is inaccurate, the contract notifies the employee to modify the reimbursement item classification. Meanwhile, whether the reimbursement project accords with the compliance policy and related laws and regulations of enterprises is also verified. For example, checking if business hospitality fees exceed business specified people average criteria or hospitality scope, verifying if office supplies purchases are from business specified suppliers or within reasonable purchasing channels, ensuring travel schedules and cost details comply with business travel policies (e.g., traffic pattern selection, accommodation criteria, etc.). If the condition of the illegal is found, the contract marks the reimbursement application as 'illegal', and records the illegal item and reason in detail and informs relevant personnel to process.

In conclusion, the reimbursement data management system and method based on the AI artificial intelligence have various obvious beneficial effects, can effectively solve various problems of the traditional reimbursement management mode, improves the efficiency, quality and safety of the reimbursement data management of enterprises, and promotes the enterprise financial management to develop to an intelligent and fine direction. The method comprises the following steps:

1. improving efficiency and user experience

The multi-mode data acquisition fusion module supports multiple terminals and input modes, is in butt joint with an enterprise internal system to acquire multi-source data and automatically fuses and verifies, can feed back reimbursement states in real time, is convenient for staff to operate, and ensures accurate data. The approval process optimization module utilizes a reinforcement learning algorithm to automatically adjust the approval level according to the characteristics of the reimbursement application, simplifies the small-scale conventional reimbursement process, increases the approval links of the reimbursement of major risks, predicts the approval comments, and obviously improves the reimbursement efficiency.

2. Enhancing auditing accuracy and compliance

And the invoice authenticity identification submodule ensures the authenticity and legality of the invoice through deep learning and multi-platform interactive verification. The reimbursement event analysis submodule uses natural language understanding and machine learning technology and combines a multi-aspect knowledge base and a case base to accurately judge the reasonability of the reimbursement event. The reimbursement item classification sub-module accurately classifies reimbursement items and automatically audits according to standards based on a rule engine and a machine learning model.

3. Mining data value boost decisions

The analysis and prediction module utilizes various data mining algorithms to mine association relation between reimbursement data and other business data, and provides cross-department business insight, power-assisted optimization resource allocation and business processes. Based on the machine learning prediction model, historical data is utilized to predict future reimbursement requirements and the like, and scientific basis and prospective guidance are provided for enterprise financial budget and the like.

4. Ensuring data security and transparency

The data storage and blockchain module adopts the blockchain technology to construct a distributed and non-tamper-resistant storage account book, encrypts storage data, and ensures confidentiality, integrity and tamper-resistant property of the data through reasonable architecture selection, node setting, encryption algorithm, intelligent contract and the like. And executing reimbursement flow automation rules by using the blockchain intelligent contract, strictly executing data verification logic, including verification of invoice authenticity, money compliance, item classification accuracy and the like, managing user rights and ensuring flow compliance transparency.

While the foregoing is directed to the preferred embodiments of the present invention, it will be appreciated by those skilled in the art that various modifications and adaptations can be made without departing from the principles of the present invention, and such modifications and adaptations are intended to be comprehended within the scope of the present invention.

Claims (10)

1. An AI artificial intelligence based reimbursement data management system, comprising:

The multi-mode data acquisition fusion module is used for providing an accessed reimbursement application inlet which is used for selecting manual input reimbursement information, uploading invoice pictures and other evidence files and inputting reimbursement events by voice;

The verification and compliance judging module is used for constructing an invoice authenticity identification model based on a deep learning algorithm, interactively verifying invoice authenticity with a tax department database and a third party verification platform, constructing a reimbursement event analysis model by using a natural language understanding technology and a machine learning algorithm, and classifying and automatically verifying reimbursement items by adopting a reimbursement item classification and verification model based on the combination of a rule engine and machine learning;

the approval process optimization module is used for determining an optimal approval path and an approval node sequence by using a reinforcement learning algorithm based on an enterprise organization structure, a business process, a risk control strategy and historical approval data;

The analysis and prediction module is used for carrying out multidimensional analysis on reimbursement data based on a big data processing frame and a data mining algorithm, mining potential association relations between reimbursement data and other business data by using association rule mining, cluster analysis and classification algorithms, predicting future reimbursement requirements, cost trends and cost changes by using historical data based on a machine learning prediction model, and generating a visual report and a visual result;

the data storage and blockchain module is used for constructing a distributed and non-tamperable data storage account book by adopting a blockchain technology, encrypting and storing reimbursement data and related auditing and approval information, executing reimbursement flow automation rules by utilizing a blockchain intelligent contract, and executing data verification logic.

2. The AI artificial intelligence based reimbursement data management system of claim 1, wherein the audit and compliance determination module includes an invoice authenticity identification sub-module, a reimbursement event analysis sub-module, and a reimbursement item classification sub-module, wherein:

the invoice authenticity identification sub-module is used for deep learning and training of massive invoice sample data, wherein the invoice sample data comprises an invoice printing format, an anti-counterfeiting mark and a code rule, and the invoice sample data is subjected to data interaction verification with an authoritative invoice database of a tax department and a third party invoice verification platform through verification data output by a deep learning algorithm;

The reimbursement event analysis sub-module is used for constructing an intelligent reimbursement event analysis model by using a natural language understanding technology and a machine learning algorithm, carrying out semantic understanding and logic analysis on reimbursement event description submitted by staff, and intelligently judging the rationality of reimbursement event by combining a detailed reimbursement policy knowledge base, an industry standard specification and a historical reimbursement data case base in an enterprise;

And the reimbursement item classification sub-module adopts a reimbursement item intelligent classification and auditing model based on the combination of a rule engine and machine learning, accurately classifies reimbursement items according to a reimbursement item classification function according to a preset reimbursement item classification rule and a classification mode automatically learned through machine learning, and automatically audits different types of reimbursement items according to corresponding auditing standards and threshold ranges.

3. The AI-artificial-intelligence-based reimbursement data management system according to claim 1, wherein the approval process optimization module is configured to utilize a reinforcement learning algorithm, based on an enterprise organization structure, a business process, a risk control policy, and historical approval data, and specifically comprises:

Setting a state space as S, wherein the state space comprises state information of the amount, the cost type, departments where the applicant is located and historical reimbursement credit records, the action space is A, the operation space comprises different operations in an approval process, a reward function is R (S, a), a reward value is determined according to an approval result and a business target, the state transition probability is P (S _t+1|s_t,a_t) and represents the probability of transition to a state S _t+1 after taking action a _t in the state S _t, the cost function is V (S), and an update formula of the cost function is based on a Belman equation:

where η is the learning rate and γ is the discount factor.

4. The AI artificial intelligence based reimbursement data management system according to claim 1, wherein the analysis and prediction module utilizes association rule mining, cluster analysis mining, classification algorithm mining potential association relation between reimbursement data and other business data, specifically:

(1) The potential association relation between the reimbursement data and other business data is mined through association rules specifically as follows:

integrating reimbursement data with other related business data, wherein the other related business data comprises sales data, purchasing data, inventory data and manpower resource data;

Cleaning the data, processing missing values, abnormal values and repeated data, coding and converting the data, and converting the classified variables into numerical values;

Performing association rule mining by selecting an Apriori algorithm or an FP-Growth algorithm, finding out frequent item sets and association rules between reimbursement data and other service data, analyzing the mined association rules, and understanding the service meaning of the association rules;

(2) The potential association relation between the reimbursement data and other business data is specifically:

Determining characteristics for cluster analysis, including key indexes in reimbursement data and related characteristics in other business data, and carrying out standardization processing on the selected characteristics so that different characteristics have the same dimension and data distribution;

Performing cluster analysis by adopting a K-Means clustering algorithm, a DBSCAN algorithm or a hierarchical clustering algorithm, calculating a cluster evaluation index, evaluating a clustering effect, and adjusting a clustering algorithm parameter or selecting a more suitable clustering algorithm according to an evaluation result;

Analyzing the clustering result, endowing business meaning to each cluster, and providing decision advice for enterprises according to the clustering result;

(3) The potential association relation between the reimbursement data and other service data is mined through a classification algorithm specifically as follows:

determining a classification task target, selecting and extracting characteristics related to the classification task, and selecting the characteristics with higher importance for classification model training by adopting a characteristic selection algorithm;

selecting a classification algorithm, training the selected classification algorithm by using a training data set, and adjusting algorithm parameters to optimize model performance;

And evaluating the trained classification model by using a test data set, calculating an evaluation index, applying the trained classification model with good evaluation index to actual reimbursement data management, and performing classification prediction on new reimbursement application or data.

5. The AI artificial intelligence based reimbursement data management system of claim 2, wherein the reimbursement event analysis submodule sets a reimbursement event rationality judgment function to:

G(R)=α×M_p(R)+β×M_s(R)+γ×M_c(R)

Wherein, M _p (R) represents a matching degree evaluation function of R based on a reimbursement policy knowledge base, M _s (R) represents an evaluation function of R based on an industry standard specification, M _c (R) represents a similarity evaluation function of R based on a historical reimbursement data case base, alpha, beta and gamma are corresponding weight coefficients, and alpha+beta+gamma=1. When G (R) is not less than T _G, the reimbursement event is reasonable, otherwise, the reimbursement event is unreasonable, wherein T _G is a preset rationality threshold, and R is an employee description event.

6. The reimbursement data management system based on AI artificial intelligence of claim 2, wherein in the reimbursement item classification submodule, reimbursement item classification functions are specifically:

Wherein Ln represents an nth class reimbursement item, K _n (X) represents a classification discriminant function of the nth class reimbursement item, and T _n is a corresponding classification threshold.

7. The AI-artificial-intelligence-based reimbursement data management system of claim 1, wherein the data storage and blockchain module employs a blockchain technique to construct a distributed, non-tamperable data storage ledger, in particular:

selecting blockchain architecture based on factors of enterprise size, data privacy requirements, and performance requirements

Setting all nodes and light nodes, wherein the all nodes store complete account book data and participate in transaction verification and block packing, establish safe P2P network connection and set inter-node communication protocols;

Encrypting the reimbursed data and the auditing and approving information by adopting an asymmetric encryption algorithm, and selecting a secure hash algorithm to hash the data;

constructing an intelligent contract to define a submission rule of a reimbursement application, embedding data verification logic in the intelligent contract, and deploying the compiled intelligent contract to nodes of a blockchain network;

defining a block structure, wherein each block comprises a block head and a block body, a hierarchical storage strategy is adopted, data which is accessed frequently in a near-term is stored in a high-speed storage medium, and historical data is stored in a high-capacity low-cost storage.

8. The AI-artificial-intelligence-based reimbursement data management system of claim 1, wherein the data storage and blockchain module executes reimbursement flow automation rules using blockchain intelligence contracts, in particular:

Dividing the intelligent contract into a plurality of functional modules, including a reimbursement application module, an approval process module, a data verification module and a notification module;

Designing state transition logic of an intelligent contract based on a state machine model, wherein the state transition logic accords with predefined business rules and flows;

When the employee submits the reimbursement application, the intelligent contract automatically checks whether the data format meets the requirements, and verifies whether the reimbursement application meets the relevant business rules according to the business reimbursement policy;

the intelligent contract automatically determines an approval level and an approval sequence according to the amount of the reimbursement application, the type of the expense and the factors of departments where staff are located;

The intelligent contract automatically acquires information of corresponding approvers from an enterprise organization architecture database or a predefined approver list, and reminds the approvers to process reimbursement applications through system notification.

9. The AI artificial intelligence based reimbursement data management system of claim 1, wherein the data storage and blockchain module data validation logic specifically comprises:

checking reimbursement amount and budget, interacting with an enterprise budget management system or a database through an intelligent contract, acquiring budget amount information of departments, projects or individuals where staff is located, calculating accumulated reimbursement amount of the staff or the projects in a current period when reimbursement application is submitted, and comparing with residual budget;

And checking the compliance of reimbursement items, namely checking whether the reimbursement items submitted by staff are correctly classified according to the reimbursement item classification standard preset by the enterprise, and verifying whether the reimbursement items accord with the compliance policy and related laws and regulations of the enterprise.

10. A method of operating the AI-artificial-intelligence-based reimbursement data management system of any one of claims 1-9, comprising:

providing an accessed reimbursement application inlet, wherein the reimbursement application inlet is used for selecting manual input reimbursement information, uploading invoice pictures and other evidence files, and inputting reimbursement events by voice;

Constructing an invoice authenticity identification model based on a deep learning algorithm, and interactively verifying invoice authenticity with a tax department database and a third party verification platform; the method comprises the steps of constructing a reimbursement event analysis model by using a natural language understanding technology and a machine learning algorithm, and accurately classifying and automatically auditing reimbursement items by adopting a reimbursement item classification and auditing model based on the combination of a rule engine and machine learning;

Utilizing a reinforcement learning algorithm to determine an optimal approval path and an approval node sequence based on enterprise organization structures, business processes, risk control strategies and historical approval data;

The method comprises the steps of carrying out multidimensional analysis on reimbursement data based on a big data processing frame and a data mining algorithm, mining potential association relations between reimbursement data and other business data by using association rule mining, cluster analysis and classification algorithms, predicting future reimbursement requirements, cost trends and cost changes by using historical data based on a machine learning prediction model, and generating a visual report and a visual result;

and constructing a distributed and non-tamperable data storage account book by adopting a blockchain technology, encrypting and storing reimbursement data and related auditing and approval information, executing reimbursement flow automation rules by utilizing a blockchain intelligent contract, and executing data verification logic.