Contract text intelligent analysis-based collection and payment compliance automatic verification system

Abstract

The invention discloses a contract text intelligent analysis-based collection and payment compliance automatic verification system, and relates to the technical field of financial management, and the system comprises a data collection module which is in stable connection with an enterprise finance ERP system, a contract management SaaS platform and a business execution OA system through an API interface, synchronously acquiring multiple types of service data according to a preset frequency and performing format unification; core elements such as a collection and payment amount range, agreed account information, a payment time and performance progress association rule and the like are accurately extracted from a contract text, logic association among the elements is established, unstructured information is converted into standardized structured data, and in an anomaly identification link, four-dimensional features are used as input, so that an anomaly identification result is obtained. According to the method, key indexes such as the amount deviation degree, the time matching degree, the account inconsistency frequency and the clause change frequency are covered, the key indexes are input into a gradient boosting tree model after standardization processing, enterprise historical violation case data are used for training optimization, and the risk type and probability are accurately output.

Description

Technical Field

[0001] This invention relates to the field of financial management technology, specifically to an automatic verification system for payment and receipt compliance based on intelligent parsing of contract text. Background Technology

[0002] In the field of corporate financial management, compliance verification of receipts and payments is a crucial link in ensuring corporate fund security, standardizing business operations, and preventing financial risks. As corporate business scale continues to expand and become more complex, the data sources involved in receipts and payments are becoming increasingly diverse, covering not only the company's internal financial system and contract management system, but also closely related to business execution systems. At the same time, the diversity and dynamism of contract terms and the complexity of receipts and payments in the performance process pose challenges to compliance verification. Against this backdrop, using intelligent text processing technology to automate compliance verification of receipts and payments has become an urgent need to improve corporate financial management and enhance risk control capabilities.

[0003] Existing corporate payment compliance verification systems have several limitations in addressing the aforementioned challenges. First, these systems rely on static clauses of a single contract for verification, only comparing the numerical value of the payment amount with the amount stated in the contract and matching the characters of the payment account with the account specified in the contract. This verification method limits data processing to the current contract text and lacks technical integration with the company's historical payment database and real-time performance management system. It cannot access dynamic information such as historical payment records, historical contract performance completion data, and real-time performance progress data, failing to cover the dynamic data association across the entire payment process and making it difficult to comprehensively and accurately determine the compliance of payment activities. Second, in the anomaly identification stage, existing systems lack technical tracking mechanisms for high-frequency anomalies, failing to capture hidden risks such as multiple modifications to payment terms under the same contract and frequent deviations of the receiving account from the agreement. This makes it difficult to detect potential violations in a timely manner. Finally, the existing systems' early warning mechanisms only provide a single system prompt function, lacking a technical mapping relationship between risk levels and response processes. This prevents differentiated early warning and response measures based on different risk levels, resulting in low risk handling efficiency and an inability to effectively prevent and mitigate financial risks. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an automatic payment compliance verification system based on intelligent contract text parsing. This system constructs a multi-source data access architecture, establishing stable connections with the enterprise's financial ERP system, contract management SaaS platform, and business execution OA system. It synchronously acquires multi-source data at preset data collection frequencies, including historical payment records, historical contract performance data, electronic texts of contracts to be verified, real-time performance progress data, and pending payment application data. After format unification and redundancy processing, this data is stored in a distributed raw database, achieving the integration of multi-source business data and expanding the data source coverage for payment compliance verification. The contract text parsing module uses entity recognition and relation extraction algorithms to extract core payment agreement information from the contract and convert it into standardized structured data, storing it in a benchmark database. This provides an accurate data foundation for subsequent abnormal behavior identification. The abnormal behavior identification module uses four-dimensional features as input, trains a model using a gradient boosting tree algorithm, and outputs risk type and risk probability, improving the technical accuracy and coverage of risk identification. The graded early warning module presets risk probability-abnormal level mapping rules, triggering differentiated early warning processes and achieving technical differentiation in risk response. The results output module displays the verification results, generates analysis reports, and stores them synchronously in the log database, providing data support for the optimization of enterprise risk control rules and forming a technical risk control closed loop.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an automatic verification system for payment and receipt compliance based on intelligent parsing of contract text, the system comprising:

[0006] Data acquisition module: It establishes stable connections with the enterprise's financial ERP system, contract management SaaS platform and business execution OA system through API interface, and synchronously acquires multiple types of business data at preset frequency. After unifying the format and handling redundancy, the data is stored in a distributed raw database with a MySQL cluster architecture and data index.

[0007] Contract text parsing module: Extracts the range of payment and receipt amounts, agreed payment and receipt account information, rules for the association between payment time and performance progress, and approval specifications for changes to payment and receipt terms in the contract through entity recognition and relation extraction algorithms. It converts the data into standardized structured data and stores it into a PostgreSQL baseline database after verifying the integrity and correctness of the fields.

[0008] Abnormal behavior recognition module: Based on multi-source data, an abnormal behavior recognition model is built using a three-stage process of feature engineering, model training and parameter optimization. The four-dimensional feature parameters are standardized and then input into the gradient boosting tree algorithm for training, and the risk type and risk probability are output.

[0009] Tiered early warning module: It presets the mapping rules between risk probability ranges and anomaly levels, and uses the rule engine to match the anomaly level and trigger differentiated early warning technology processes including prompt sending, email notification, and process control;

[0010] Results output module: Displays verification results in a responsive visual interface on both web and mobile devices, generates anomaly behavior analysis reports in an exportable format according to preset templates, and synchronously stores them in a MongoDB log database with associated indexes, supporting multi-dimensional retrieval.

[0011] Furthermore, the data acquisition module synchronously acquires the enterprise's historical payment and receipt records, historical contract performance completion data, electronic text of contracts to be verified, real-time performance progress data, and pending payment and receipt application data at a preset data acquisition frequency. The historical payment and receipt records include amount, time, account, and business number; the historical contract performance completion data includes performance nodes, completion ratio, and acceptance results; and the pending payment and receipt application data includes application amount, target account, application department, and business association number. The acquired unstructured text data, semi-structured table data, and structured form data are uniformly processed, converted into JSON format, and null and duplicate values ​​are removed before being stored in the system's preset distributed raw database.

[0012] Furthermore, the contract text parsing module extracts the payment and receipt amount range, agreed payment and receipt account information, payment time nodes and performance progress association rules, and payment and receipt clause change approval specifications from the contract through entity recognition algorithms. The payment and receipt amount range includes upper limit, lower limit, and precise value; the agreed payment and receipt account information includes bank, account number, and account name; the payment time node and performance progress association rule is expressed as: X% performance completed, Y% payment is made; the payment and receipt clause change approval specifications include approval level and required materials. The module establishes logical relationships between the core information through relation extraction algorithms, converting the parsed unstructured information into standardized structured data. The structured data includes business number, amount range, account information, payment-performance rules, and change approval requirements fields. After data verification to check the completeness and format correctness of the fields, the data is stored in a PostgreSQL database.

[0013] Furthermore, the contract text parsing module extracts the range of payment and receipt amounts, agreed payment and receipt account information, association rules between payment time nodes and performance progress, and approval specifications for changes to payment and receipt terms in the contract through entity recognition algorithms. The specific steps are as follows: using the labeled payment and receipt related entity corpus as the training basis, a professional dictionary containing monetary units, account identifiers, performance terms, and approval vocabulary is pre-constructed. The candidate entity range is initially locked through dictionary matching and context feature fusion. Then, sequence labeling technology is used to capture the boundary information and type attributes of entities in the text. When extracting the range of payment and receipt amounts, the upper limit, lower limit, or precise value of the amount is determined by combining the numerical context. When extracting the agreed payment and receipt account information, the bank, account number, and account name association identifier are associated to form complete account data. When extracting the association rules between payment time nodes and performance progress, the corresponding logical relationship between performance completion of X% and payment of Y% is identified through semantic association analysis. When extracting the approval specifications for changes to payment and receipt terms, the specific content corresponding to the keywords of approval level, required materials, and authority nodes is focused on. Finally, the extracted entity information is cross-validated to form structured core agreed information data.

[0014] Furthermore, the contract text parsing module establishes logical connections between core information elements through a relation extraction algorithm, converting the parsed unstructured information into standardized structured data. The specific steps are as follows: Data preprocessing is performed on the text fragments containing the core entities of payment and receipt, output after entity recognition. Noise removal is achieved by eliminating irrelevant guiding words and redundant modifiers through regular expression matching. Then, part-of-speech tagging and dependency parsing are used to locate the predicate connection components between entities. Simultaneously, entity information is bound to text location indexes to form triplet data of entity type-entity content-location coordinates. Subsequently, the relation extraction process is initiated, employing a hybrid mode combining rule matching and model prediction. The rule layer constructs a relation template library for fixed sentence patterns, and the relation extraction is performed through syntactic matching... The model layer uses entity type, location distance, and predicate attributes as features to mine implicit relationships in complex sentence structures using a prediction model trained on labeled corpora, and outputs a triplet of relationship between entity A, relationship type, and entity B. Next, a relationship constraint system is established around the payment and receipt business logic, binding the amount range with payment time-performance progress rules, account information with change approval specifications, and terms and permissions nodes to form a closed-loop relationship chain. Low-reliability data is filtered through relationship frequency statistics and semantic consistency judgment. Finally, a structured field system including business number, amount boundary, full account information, payment trigger conditions, approval level, and material requirements is defined according to enterprise financial data specifications.

[0015] Furthermore, the abnormal behavior identification module constructs an abnormal behavior identification model based on historical business data stored in the data acquisition module, current business data, and benchmark data generated by the contract text parsing module. The model adopts a three-stage technical process of feature engineering, model training, and parameter optimization. In the feature engineering stage, the deviation between the payment amount and the contract agreement, the matching degree between the payment time and the performance progress, the frequency of inconsistencies between the payee's account and the contract agreement, and the frequency of changes to the payment terms of the same contract are set as core input feature parameters. In the model training stage, the gradient boosting tree algorithm is used, with historical violation case data of the enterprise as training samples. The historical violation case data is labeled with risk type and risk probability, and the training set and test set are divided in a 7:3 ratio. The model hyperparameters are adjusted through 5-fold cross-validation. The hyperparameters include learning rate, tree depth, and number of iterations. In the model output stage, the risk type and risk probability corresponding to the current payment behavior are output according to the preset classification logic. The risk types include fraudulent contracts to obtain funds, illegal transfer of funds, malicious default on payments, and illegal changes to terms.

[0016] Furthermore, the deviation between the payment amount and the contractual agreement is calculated by taking the absolute value of the difference between the application amount and the contractually agreed amount, and then comparing it with the contractually agreed amount. The matching degree between payment time and performance progress is calculated by taking the absolute value of the difference between the current performance ratio and the performance ratio corresponding to the agreed payment, and then comparing it with the performance ratio corresponding to the agreed payment. The frequency of inconsistencies between the payee's account and the contractual agreement is calculated on a monthly basis, accumulating the number of times the payee's account and the contractually agreed account do not match under the same business number. The frequency of changes to the same contract's payment and receipt terms is calculated on a contract validity period basis, accumulating the number of times the same contract's payment and receipt terms have been modified. All feature data are standardized so that the processed data values ​​are within the range of zero to one.

[0017] Furthermore, the graded early warning module presets technical mapping rules between risk probability ranges and anomaly levels, wherein a risk probability ≥80% corresponds to a level 1 anomaly, a risk probability of 50%-79% corresponds to a level 2 anomaly, and a risk probability <50% corresponds to a level 3 anomaly.

[0018] Furthermore, after receiving the risk type and risk probability output by the abnormal behavior identification module, the tiered early warning module matches the corresponding abnormality level through the rule engine and triggers a differentiated early warning technology process. Level 1 abnormalities send a mandatory pop-up notification via the system message push interface. This notification covers the payment approval page and requires manual confirmation to close. A real-time email is sent to the designated risk control department's email address via the SMTP protocol, and the interception function of the payment approval system is invoked via the interface to freeze the current approval process. Level 2 abnormalities send a pop-up notification via the system message push interface. This notification is displayed on the right side of the approval page and can be closed. An email is sent to the finance manager's email address via the SMTP protocol, and the pause function of the payment approval system is invoked via the interface to pause the current approval process. Level 3 abnormalities send a text notification through the system message center and automatically generate an abnormal log via the log recording interface. The log includes time, business number, risk type, and probability, and the log fields comply with enterprise log management standards.

[0019] Compared with existing technologies, this automatic payment compliance verification system based on intelligent contract text parsing has the following advantages:

[0020] I. This invention accurately extracts core elements from contract texts, such as the range of payment and receipt amounts, agreed account information, and rules relating payment time and performance progress. It also establishes logical connections between these elements, transforming unstructured information into standardized structured data. In the anomaly identification phase, four-dimensional features are used as input, covering key indicators such as amount deviation, time matching degree, frequency of account inconsistency, and frequency of clause changes. After standardization, these features are input into a gradient boosting tree model, which is trained and optimized using historical violation case data of enterprises. This model can accurately output risk types and probabilities, identifying not only explicit risks but also implicit risks, greatly improving the technical accuracy and coverage of risk identification.

[0021] Second, this invention uses a pre-defined mapping rule between risk probability ranges and anomaly levels to trigger differentiated early warning processes for different levels of anomalies. Level 1 anomalies are addressed with strict measures such as mandatory pop-up notifications, email notifications to the risk control department, and blocking of approval processes. Level 2 anomalies are addressed with pop-up notifications, email notifications to the finance manager, and suspension of approval processes. Level 3 anomalies are addressed primarily with system message notifications and anomaly logs, ensuring the timeliness and effectiveness of risk handling. Simultaneously, leveraging the visualization and data storage technology of the result output module, verification results are presented in real time, analysis reports are generated and exported, and results and reports are synchronized to the log database, providing data support for optimizing enterprise risk control rules. This forms a complete technical chain from data collection to result application, constructing a closed-loop risk control system at the technical level.

[0022] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from an examination of the following, or may be learned from the practice of the invention. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0024] Figure 1 This is a structural block diagram of an automatic payment compliance verification system based on intelligent parsing of contract text;

[0025] Figure 2 A flowchart of an automatic payment compliance verification system based on intelligent parsing of contract text;

[0026] Figure 3 This is a flowchart of the contract text parsing module of an automatic payment compliance verification system based on intelligent contract text parsing. Detailed Implementation

[0027] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0028] This invention provides an automatic payment compliance verification system based on intelligent contract text parsing, such as... Figure 1As shown, the system includes a data acquisition module, a contract text parsing module, an abnormal behavior identification module, a tiered early warning module, and a result output module. By constructing a multi-source data access technology architecture, it establishes stable connections with the enterprise's financial ERP system, contract management SaaS platform, and business execution OA system. It synchronously acquires multi-source data at preset data acquisition frequencies, including historical payment and receipt records, historical contract performance completion data, electronic texts of contracts to be verified, real-time performance progress data, and pending payment and receipt application data. After format unification and redundancy processing, the data is stored in a distributed raw database, achieving the integration of multi-source business data and expanding the data source coverage for payment and receipt compliance verification. The contract text parsing module uses entity recognition and relation extraction algorithms to extract core payment and receipt agreement information from contracts and convert it into standardized structured data, storing it in a benchmark database. This provides an accurate data foundation for subsequent abnormal behavior identification. The abnormal behavior identification module uses four-dimensional features as input, trains a model through a gradient boosting tree algorithm, and outputs risk type and risk probability, improving the technical accuracy and coverage of risk identification. The tiered early warning module presets risk probability-abnormal level mapping rules to trigger differentiated early warning processes, achieving technical differentiation in risk response. The results output module displays the verification results, generates analysis reports, and stores them synchronously in the log database, providing data support for the optimization of enterprise risk control rules and forming a technical risk control closed loop.

[0029] This embodiment first starts the data acquisition module, such as... Figure 2 As shown, stable connections are established with the enterprise's financial ERP system, contract management SaaS platform, and business execution OA system through preset API interfaces. Various types of data are synchronized every 15 minutes. The synchronized content includes the enterprise's historical payment and receipt records for the past five years, historical contract performance records, electronic texts of contracts currently awaiting payment and receipt verification, real-time performance progress data of the contracts, and corresponding pending payment and receipt application data. The historical payment and receipt records include specific amounts, payment or receipt times, corresponding account information, and business numbers. The historical contract performance records include each performance node, actual completion rate, and acceptance results. The pending payment and receipt application data includes the application amount, target account information, application department, and business association number. The data acquisition module performs unified format processing on the synchronized unstructured contract texts, semi-structured performance progress tables, and structured payment and receipt application forms, converting all data into the same text format. At the same time, preset filtering rules are used to remove null values, duplicate values, and invalid characters from the data. After data redundancy removal, the processed data is stored in the original database using a cluster architecture, and a data index is established with the business number as the core.

[0030] After the data acquisition module completes data storage, such as Figure 3As shown, the electronic text of the contract to be verified is transmitted to the contract text parsing module via an interface. The electronic text is preprocessed first, removing garbled characters through character encoding standardization and cleaning up redundant spaces and line breaks using a formatting tool. Then, the text is segmented into sentences and paragraphs based on punctuation marks such as periods and semicolons. Subsequently, a recognition model optimized for enterprise contract scenarios is loaded. This model, combined with a pre-built professional dictionary, narrows down the range of candidate information in the text. The professional dictionary includes various monetary units, account identifiers, performance-related terms, and approval-specific terms. Through dictionary matching and the fusion of contextual features, content related to payments and receipts is initially filtered out. Then, the boundaries and attributes of this content are captured using preset rules, extracting the payment and receipt amount range, agreed payment and receipt account information, the association rules between payment time points and performance progress, and the approval specifications for changes to payment and receipt terms. The payment and receipt amount range explicitly includes upper and lower limits or precise values; the agreed payment and receipt account information includes the bank, account number, and account name; the association rules between payment time points and performance progress include the corresponding logic between the two; and the approval specifications for changes to payment and receipt terms include the approval level and required materials. Next, the extracted core information is processed to establish relationships. First, the text fragments containing core information are denoised, and irrelevant introductory words and modifiers are removed. Then, part-of-speech tagging and syntactic analysis are used to locate the connecting components between various types of information. Each type of information is bound to a text location index to form combined data containing information type, specific content, and location coordinates. The relationships between information are locked through comparison with a pre-set template library and feature analysis. Combined data containing information subject, association type, and associated object is output. Then, a constraint system is established with payment and receipt business logic as the core. The amount range is bound to the payment time node, and the account information is bound to the approval specifications to form an association chain. Unreliable data is filtered through association frequency statistics and semantic consistency judgment. Finally, a structured field system is defined according to the enterprise financial data specifications. The fields include business number, amount boundary, full account information, payment trigger conditions, approval level, and material requirements. The associated core information is accurately filled into the corresponding fields. After field format validation, logical conflict detection, and data integrity verification, standardized structured data is generated and simultaneously stored in a dedicated benchmark database with a multi-field index.

[0031] After the contract text parsing module completes the structured data storage, the abnormal behavior identification module calls historical business data, current business data, and structured benchmark data from the benchmark database through interfaces. Based on this data, feature parameter calculations are initiated. When calculating the deviation between the payment amount and the contract stipulation, the module first obtains the application amount in the pending payment application and the agreed amount in the benchmark data, calculates the difference, takes the absolute value, and then performs a ratio calculation with the agreed amount. When calculating the matching degree between payment time and performance progress, the module obtains the actual performance ratio in the current performance progress data and the corresponding agreed payment in the benchmark data. The performance ratio is calculated, the difference between the two is taken as the absolute value, and then the absolute value is compared with the performance ratio corresponding to the agreed payment. When counting the frequency of inconsistencies between the payee's account and the contract agreement, the number of times the payee's account under the same business number does not match the agreed account in the benchmark data is accumulated, using the natural month as the time unit. When counting the frequency of changes to the payment terms of the same contract, the number of times the payment terms of the contract are modified is accumulated, using the contract validity period as the time range. After completing the calculation of the four types of characteristic parameters, the values ​​of all parameters are converted to the range of zero to one through preset standardized processing rules to form characteristic data that meets the requirements of subsequent processing.

[0032] After the abnormal behavior identification module completes the feature data processing, it inputs it into the pre-trained identification model. This model is trained based on the company's historical violation case data. These case data are all labeled with clear risk types and risk probabilities. Through the analysis and processing of feature data, the model outputs the risk type and risk probability corresponding to the current payment and receipt behavior. The risk types include fraudulent contracts to obtain funds, illegal transfer of funds, malicious default on payments, and illegal changes to terms. The accuracy of the risk probability is retained to two decimal places.

[0033] After the abnormal behavior identification module outputs risk information, the tiered early warning module immediately receives this information and invokes the preset mapping rules between risk probability and abnormality level. If the risk probability is ≥80%, it is judged as a Level 1 abnormality; if the risk probability is between 50% and 79%, it is judged as a Level 2 abnormality; if the risk probability is <50%, it is judged as a Level 3 abnormality. After determining the abnormality level, the tiered early warning module initiates the corresponding early warning process. For Level 1 abnormalities, a pop-up prompt is sent via the system message push interface. This prompt covers the payment approval page and requires manual confirmation to close. A real-time email is also sent to the risk control department's email address via a specified protocol. The email contains risk details, the business number, and related data links. For Level 2 exceptions, the system calls the interception function of the payment and receipt approval system via API to freeze the current approval process. For Level 2 exceptions, a pop-up notification is sent via the system message push interface. This notification is displayed on the right side of the approval page and can be manually closed. An email containing risk details and handling suggestions is sent to the finance manager's email address via a specified protocol. At the same time, the system calls the pause function of the payment and receipt approval system via API to suspend the current approval process. For Level 3 exceptions, a text notification is sent via the system message center. This notification is stored in the system's exception reminder section. At the same time, an exception log is automatically generated via the log recording interface. The log includes the occurrence time, business number, risk type, and risk probability. The log fields strictly follow the company's log management standards.

[0034] The tiered early warning module initiates the early warning process and simultaneously synchronizes risk information and early warning process information to the results output module. This module displays a visual interface on both the system's web and mobile platforms, employing a responsive design to adapt to different device sizes. It displays real-time payment compliance verification results, including verification conclusions, anomaly levels, risk types, the triggered early warning process name, and the current process status. Verification conclusions are categorized as pass or fail, anomaly levels as Level 1, Level 2, and Level 3, and the current process status as pending, processing, and completed. Simultaneously, it automatically generates an anomaly behavior analysis report based on a preset template. The template includes a title, basic business information, verification process data, risk analysis, and early warning records. The report supports exporting to common document and table formats. The results output module synchronizes the verification results and anomaly behavior analysis report to the system log database via a data synchronization interface. The verification results include fields such as business number, verification time, conclusion, level, and risk type. The anomaly behavior analysis report is stored as a binary stream. This database adopts an architecture supporting massive unstructured data storage and establishes data association indexes, supporting multi-dimensional data retrieval by business number, time range, and anomaly level.

[0035] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A system for automatic checking of payment and receipt compliance based on intelligent analysis of contract texts, characterized by, The system includes: Data acquisition module: It establishes stable connections with the enterprise's financial ERP system, contract management SaaS platform and business execution OA system through API interface, and synchronously acquires multiple types of business data at preset frequency. After unifying the format and handling redundancy, the data is stored in a distributed raw database with a MySQL cluster architecture and data index. Contract text parsing module: Extracts the range of payment and receipt amounts, agreed payment and receipt account information, rules for the association between payment time and performance progress, and approval specifications for changes to payment and receipt terms in the contract through entity recognition and relation extraction algorithms. It converts the data into standardized structured data and stores it into a PostgreSQL baseline database after verifying the integrity and correctness of the fields. Abnormal behavior recognition module: Based on multi-source data, an abnormal behavior recognition model is built using a three-stage process of feature engineering, model training and parameter optimization. The four-dimensional feature parameters are standardized and then input into the gradient boosting tree algorithm for training, and the risk type and risk probability are output. Tiered early warning module: It presets the mapping rules between risk probability ranges and anomaly levels, and uses the rule engine to match the anomaly level and trigger differentiated early warning technology processes including prompt sending, email notification, and process control; Results output module: Displays verification results in a responsive visual interface on both web and mobile devices, generates anomaly behavior analysis reports in an exportable format according to preset templates, and synchronously stores them in a MongoDB log database with associated indexes, supporting multi-dimensional retrieval.

2. The system for automatic checking of compliance of payment and receipt based on intelligent analysis of contract text according to claim 1, characterized in that, The data acquisition module synchronously acquires the enterprise's historical payment and receipt records, historical contract performance completion data, electronic text of contracts to be verified, real-time performance progress data, and pending payment and receipt application data at a preset data acquisition frequency. The historical payment and receipt records include amount, time, account, and business number. The historical contract performance completion data includes performance nodes, completion percentage, and acceptance results. The pending payment and receipt application data includes application amount, target account, application department, and business association number. The acquired unstructured text data, semi-structured table data, and structured form data are uniformly processed, converted into JSON format, and null and duplicate values ​​are removed before being stored in the system's preset distributed raw database.

3. The automatic payment compliance verification system based on intelligent contract text parsing according to claim 1, characterized in that, The contract text parsing module extracts the payment and receipt amount range, agreed payment and receipt account information, payment time nodes and performance progress association rules, and payment and receipt clause change approval specifications from the contract through entity recognition algorithms. The payment and receipt amount range includes upper limit, lower limit, and precise value. The agreed payment and receipt account information includes the opening bank, account number, and account name. The payment time node and performance progress association rule is expressed as: X% performance completed, Y% payment is made. The payment and receipt clause change approval specifications include approval level and required materials. The module establishes logical relationships between the core information through relation extraction algorithms, converting the parsed unstructured information into standardized structured data. The structured data includes business number, amount range, account information, payment-performance rules, and change approval requirements fields. After data verification to check the completeness and format correctness of the fields, the data is stored in a PostgreSQL database.

4. The automatic payment compliance verification system based on intelligent contract text parsing according to claim 3, characterized in that, The contract text parsing module extracts the range of payment and receipt amounts, agreed payment and receipt account information, association rules between payment time nodes and performance progress, and approval specifications for changes to payment and receipt terms from the contract through entity recognition algorithms. The specific steps are as follows: using annotated payment and receipt-related entity corpus as a training basis, a professional dictionary containing monetary units, account identifiers, performance terms, and approval vocabulary is pre-constructed. Candidate entity ranges are initially identified through dictionary matching and contextual feature fusion. Sequence labeling technology is then used to capture the boundary information and type attributes of entities in the text. When extracting the range of payment and receipt amounts, the upper limit, lower limit, or precise value of the amount is determined by combining numerical context. When extracting agreed payment and receipt account information, the bank, account number, and account name association identifiers are associated to form complete account data. When extracting the association rules between payment time nodes and performance progress, semantic association analysis is used to identify the corresponding logical relationship between performance completion of X% and payment of Y%. When extracting approval specifications for changes to payment and receipt terms, the specific content corresponding to keywords related to approval level, required materials, and authority nodes is considered. Finally, the extracted entity information is cross-validated to form structured core agreed information data.

5. The automatic payment compliance verification system based on intelligent contract text parsing according to claim 3, characterized in that, The contract text parsing module establishes logical connections between core information elements through a relation extraction algorithm, converting the parsed unstructured information into standardized structured data. The specific steps are as follows: Data preprocessing is performed on the text fragments containing the core entities of payment and receipt after entity recognition. Noise removal is achieved by eliminating irrelevant leading words and redundant modifiers through regular expression matching. Then, part-of-speech tagging and dependency parsing are used to locate the predicate connection components between entities. Simultaneously, entity information is bound to text location indexes to form triplet data of entity type-entity content-location coordinates. Subsequently, the relation extraction process is initiated, employing a hybrid mode combining rule matching and model prediction. The rule layer constructs a relation template library for fixed sentence patterns, and syntactic structure comparison is used to... The model layer identifies explicit relationships between entities, using entity type, location distance, and predicate attributes as features. Based on annotated corpora, a predictive model is trained to uncover implicit relationships in complex sentence structures, collectively outputting a relationship triplet: entity A - relationship type - entity B. Next, a relationship constraint system is established around the payment and receipt business logic. A closed-loop relationship chain is formed by binding the amount range with payment time-performance progress rules, account information with change approval specifications, and terms and permissions nodes. Low-reliability data is filtered through relationship frequency statistics and semantic consistency judgment. Finally, a structured field system including business number, amount boundaries, full account information, payment trigger conditions, approval levels, and material requirements is defined according to enterprise financial data specifications.

6. The automatic payment compliance verification system based on intelligent contract text parsing according to claim 1, characterized in that, The abnormal behavior identification module is based on historical business data stored in the data acquisition module, current business data, and benchmark data generated by the contract text parsing module. It constructs an abnormal behavior identification model using a three-stage process: feature engineering, model training, and parameter optimization. In the feature engineering stage, the deviation between payment and receipt amounts and contract stipulations, the matching degree between payment time and performance progress, the frequency of inconsistencies between the payee's account and contract stipulations, and the frequency of changes to payment and receipt terms within the same contract are set as core input feature parameters. In the model training stage, a gradient boosting tree algorithm is used, with historical violation case data as training samples. These historical violation case data are labeled with risk types and probabilities, and the training and test sets are divided in a 7:3 ratio. The model hyperparameters are adjusted using 5-fold cross-validation, including the learning rate, tree depth, and number of iterations. In the model output stage, the risk type and probability corresponding to the current payment and receipt behavior are output according to a preset classification logic. Risk types include obtaining funds through fraudulent contracts, illegally transferring funds, maliciously defaulting on payments, and illegally changing terms.

7. The automatic payment compliance verification system based on intelligent contract text parsing according to claim 6, characterized in that, The deviation between the payment amount and the contractual agreement is calculated by taking the absolute value of the difference between the application amount and the contractually agreed amount, and then comparing it with the contractually agreed amount. The matching degree between payment time and performance progress is calculated by taking the absolute value of the difference between the current performance ratio and the performance ratio corresponding to the agreed payment, and then comparing it with the performance ratio corresponding to the agreed payment. The frequency of inconsistencies between the payee's account and the contractually agreed account is calculated on a monthly basis, accumulating the number of times the payee's account and the contractually agreed account do not match under the same business number. The frequency of changes to the same contract's payment and receipt terms is calculated on a contract validity period basis, accumulating the number of times the same contract's payment and receipt terms have been modified. All feature data are standardized so that the processed data values ​​are within the range of zero to one.

8. The automatic payment compliance verification system based on intelligent contract text parsing according to claim 1, characterized in that, The graded early warning module presets technical mapping rules between risk probability ranges and anomaly levels, where a risk probability ≥ 80% corresponds to a Level 1 anomaly, a risk probability of 50%-79% corresponds to a Level 2 anomaly, and a risk probability < 50% corresponds to a Level 3 anomaly.

9. The automatic payment compliance verification system based on intelligent contract text parsing according to claim 1, characterized in that, After receiving the risk type and risk probability output by the abnormal behavior identification module, the tiered early warning module matches the corresponding abnormality level through the rule engine and triggers a differentiated early warning technology process. Level 1 abnormalities send a mandatory pop-up notification via the system message push interface. This notification covers the payment approval page and requires manual confirmation to close. A real-time email is sent to the designated risk control department's email address via the SMTP protocol, and the interception function of the payment approval system is invoked via the interface to freeze the current approval process. Level 2 abnormalities send a pop-up notification via the system message push interface. This notification is displayed on the right side of the approval page and can be closed. An email is sent to the finance manager's email address via the SMTP protocol, and the pause function of the payment approval system is invoked via the interface to pause the current approval process. Level 3 abnormalities send a text notification via the system message center and automatically generate an abnormal log through the log recording interface. The log includes time, business number, risk type, and probability, and the log fields comply with enterprise log management standards.

Images