Internet platform-based bill business online financing management system and method

Abstract

The application relates to the technical field of bill management, and particularly discloses a bill business online financing management system and method based on an Internet platform. Bill data from enterprise financial systems, bank core systems and third-party payment platforms are collected in real time through standardized interfaces, and a unified field mapping table is constructed by using a semantic embedding model and a K-means clustering algorithm. Semantic consistency characteristic values are calculated to identify field naming inconsistency problems. Meanwhile, the logical relationship between time fields is analyzed based on a Bayesian network modeling, and logical conflict rate characteristic values are calculated to identify time sequence errors. Further, the above characteristics are fused into a comprehensive credibility characteristic vector, which is input into a support vector machine model for training and prediction, so that automatic credibility scoring of bill data is realized. When it is judged that the data is not credible, a data verification compensation mechanism is automatically triggered, and intelligent error correction and manual review processes are started.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bill management, in particular to an online financing management system and method for bill business based on an Internet platform. BACKGROUND

[0002] With the development of financial technology, bill business gradually evolves towards online and platformization. In particular, in the process of enterprise financing, the collection and processing of electronic bill data become a key link. In practical applications, bill data usually comes from multiple heterogeneous systems, such as enterprise financial systems, bank core systems, and third-party payment platforms, etc. These systems have great differences in field naming, data format, and logical rules, which leads to problems such as semantic inconsistency and logical conflict in the process of data integration and analysis. In order to improve the efficiency and security of bill financing business, a technical means capable of automatically identifying field semantics, analyzing logical relationships, and evaluating data credibility is urgently needed to realize intelligent management and quality control of multi-source bill data.

[0003] The prior art has the following disadvantages:

[0004] Traditional methods mostly rely on manually set field mapping tables, which cannot adapt to the dynamic changes of multiple heterogeneous systems and lack semantic-level automatic recognition capabilities. Secondly, existing logical verification mechanisms mostly judge the time sequence relationship between fields based on fixed rules, which are difficult to meet the needs of logical conflict identification in complex scenarios and lack quantitative evaluation means. Thirdly, most systems have not established a unified data credibility evaluation model, which cannot comprehensively score bill data quality based on semantic consistency and logical rationality, thereby affecting the accuracy of subsequent risk control and intelligent decision-making. Therefore, it is necessary to propose an intelligent bill data management method combining semantic analysis, logical reasoning, and machine learning to solve the above problems and improve the automation level and data governance capability of bill financing business. SUMMARY

[0005] The present application aims to provide an online financing management system and method for bill business based on an Internet platform to solve the problems in the above background.

[0006] The object of the present application can be achieved by the following technical solutions:

[0007] The online financing management method for bill business based on an Internet platform comprises the following steps:

[0008] S1: In the process of bill financing application, real-time collection of electronic bill data from multiple heterogeneous systems;

[0009] S2: performing semantic analysis on the collected bill data, constructing a unified field mapping table, and calculating a semantic consistency feature value for evaluating the expression consistency of the same semantic field in different systems;

[0010] S3: analyzing the logical relationship between fields, identifying time sequence logical errors, and calculating a logical conflict rate feature value for evaluating the rationality of the internal logic of the bill data;

[0011] S4: constructing a comprehensive credibility feature vector from the semantic consistency feature value and the logical conflict rate feature value, analyzing the comprehensive credibility feature vector using a machine learning model, and evaluating the credibility of the current bill data according to the analysis result;

[0012] S5: If the current bill data is determined to be untrustworthy, automatically triggering a data verification compensation mechanism, and starting an artificial review and intelligent error correction process.

[0013] As a further scheme of the present application, the process of obtaining the semantic consistency feature value is:

[0014] Real-time collection of bill data, semantic analysis of the collected bill data, construction of a unified field mapping table, and calculation of a semantic consistency feature value, specifically including:

[0015] Text cleaning and standardization processing is performed on each field content in the collected bill data, and each field content is converted into a high-dimensional vector representation through a pre-trained semantic embedding model, thereby forming a bill field vector set;

[0016] The bill field vector set is input into a K-means clustering algorithm, and clustering analysis is performed on all field vectors according to the set number of clusters, obtaining a plurality of clustering clusters, wherein each clustering cluster contains a plurality of field vectors that are similar in semantic space;

[0017] For each clustering cluster, the cosine similarity between any two different field vectors in the corresponding cluster is calculated, and the semantic consistency feature value corresponding to the corresponding clustering cluster is obtained according to the ratio between the sum of all cosine similarity values and the product of the total number of fields in the corresponding cluster and one less than the total number of fields.

[0018] As a further scheme of the present application, the evaluation of the expression consistency of the same semantic field in different systems specifically includes:

[0019] Determine whether the semantic consistency feature value is greater than or equal to a preset threshold, if yes, the expression of the same semantic field in different systems is consistent, if not, the expression of the same semantic field in different systems is inconsistent.

[0020] As a further scheme of the present application, the process of obtaining the logical conflict rate feature value is:

[0021] Real-time collection of bill data, analyzing the logical relationship between the collected bill data fields, identifying time sequence logical errors, calculating logical conflict rate characteristic values, specifically including:

[0022] Extract the time-related fields in the collected bill data, and standardize each field to form a field set with a unified time format;

[0023] According to the preset time sequence logical rules between fields, a Bayesian network model is constructed, wherein the nodes represent the time fields in the bill, the edges represent the causal dependence relationship between the fields, and each node is configured with a corresponding conditional probability table;

[0024] For each bill record, the probability value of violating the time sequence logic is calculated using the Bayesian inference algorithm, specifically including: under the given parent node condition, the maximum conditional probability of the current field state is calculated, and the time logic error probability value of the corresponding bill record is obtained by subtracting the maximum conditional probability from one;

[0025] Set a logical error judgment threshold, when the logical error probability value of a bill record exceeds the logical error judgment threshold, it is determined that the corresponding bill has a time sequence logical conflict;

[0026] Statistical all bill records are judged to exist time sequence logical conflict, and the ratio of the number of logical conflict rate characteristic values to the total number of bills is taken as the ratio of the number of logical conflict rate characteristic values.

[0027] As a further scheme of the present application: the evaluation of the rationality of the internal logic of the bill data, specifically including:

[0028] Determine whether the logical conflict rate characteristic value is greater than or equal to the preset threshold, if yes, the internal logic of the bill data is not reasonable, if not, the internal logic of the bill data is reasonable.

[0029] As a further scheme of the present application: the use of a machine learning model to analyze the comprehensive credibility feature vector, specifically including:

[0030] Obtain the semantic consistency feature value and the logical conflict rate characteristic value of the bill data, construct the semantic consistency feature value and the logical conflict rate characteristic value into a comprehensive credibility feature vector, use the comprehensive credibility feature vector as the input of the machine learning model, take minimizing the error between the predicted credibility score and the actual credibility score as the training target, train the model, and output the credibility score of the bill data according to the trained model, the machine learning model is a support vector machine model.

[0031] As a further scheme of the present application: the training process of the machine learning model is:

[0032] The semantic consistency feature value and the logic conflict rate feature value of each bill are calculated respectively, and are combined into a comprehensive credibility feature vector as a feature dimension; the actual credibility score of each bill is labeled to form a labeled training sample set; the support vector machine model takes minimizing the error between the predicted credibility score and the actual credibility score as the objective function, uses the kernel function mapping strategy to map the feature vector to a high-dimensional space, and establishes a nonlinear regression model by solving the optimal classification hyperplane; a regularization term is introduced in the training process to control the model complexity and prevent overfitting, and finally a trained credibility prediction model is obtained for automatic credibility scoring output of newly collected bill data.

[0033] As a further scheme of the application: the credibility of the current bill data is evaluated according to the analysis result, and specifically includes:

[0034] It is judged whether the credibility score of the bill data is greater than or equal to a preset threshold, if yes, the corresponding bill data is credible, and if no, the corresponding bill data is not credible.

[0035] As a further scheme of the application: if the current bill data is judged to be not credible, an automatic data verification compensation mechanism is triggered, and specifically includes:

[0036] When the bill credibility score output by the support vector machine model is lower than the preset credibility threshold, the system automatically marks the corresponding bill as a to-be-reviewed state, and starts a multi-level data verification compensation process; the process includes automatic correction of fields with semantic inconsistency or logic conflict, and pushing to a manual review interface; at the same time, combined with the abnormal distribution of the semantic consistency feature value and the logic conflict rate feature value, the high-risk field area is located to assist the reviewer to quickly identify the problem source, so as to complete the closed-loop processing and quality improvement of the untrustworthy bill data.

[0037] The bill business online financing management system based on the Internet platform comprises:

[0038] A data acquisition module, which acquires electronic bill data from multiple heterogeneous systems in real time during a bill financing application process;

[0039] A bill semantic analysis module, which performs semantic analysis on the collected bill data, constructs a unified field mapping table, and calculates a semantic consistency feature value for evaluating the consistency of the same semantic field in different systems;

[0040] A bill rationality analysis module, which identifies time sequence logic errors by analyzing the logic relationship between fields, calculates a logic conflict rate feature value for evaluating the rationality of the internal logic of the bill data;

[0041] The bill credibility evaluation module constructs the semantic consistency feature value and the logical conflict rate feature value into a comprehensive credibility feature vector, analyzes the comprehensive credibility feature vector using a machine learning model, and evaluates the credibility of the current bill data according to the analysis result.

[0042] The bill correction management module automatically triggers a data verification compensation mechanism, starts an artificial review and intelligent error correction process if it is determined that the current bill data is not credible.

[0043] The beneficial effects of the present application are:

[0044] (1) The present application effectively solves the problem of inconsistent data semantics caused by different field naming rules in multi-source heterogeneous systems such as enterprise financial systems, bank core systems and third-party payment platforms, by introducing an intelligent field mapping mechanism based on semantic embedding and clustering analysis. Specifically, the system first performs text cleaning and standardization processing on the field content in the collected bill data, removes noise information and unifies the expression form; then uses a pre-trained semantic embedding model (such as BERT) to convert the field content into a high-dimensional vector representation, thereby realizing the mapping from text symbols to semantic space; On this basis, the K-means clustering algorithm is used to classify the semantic of all field vectors, so that fields with similar semantics are divided into the same cluster; further, by statistically analyzing the cosine similarity between field vectors in each cluster, the semantic consistency feature value is calculated, which is used to quantify the consistency degree of the expression of the same semantic field in different systems. If the semantic consistency feature value is lower than the preset threshold, it means that there is a problem of inconsistent field naming or semantic expression deviation, and the system can automatically identify and prompt the field combination that needs to be mapped and unified, and then construct a highly universal unified field mapping table. This mechanism not only significantly improves the accuracy and automation level of cross-system bill data integration, but also effectively reduces the cost and error rate caused by traditional manual experience-based field matching, laying a solid foundation for subsequent data quality evaluation and credibility modeling.

[0045] (2) The application realizes deep modeling of the time sequence relationship between the data fields of the bill and logical conflict identification by constructing a time logic reasoning mechanism based on a Bayesian network. Specifically, the system first extracts all time-related fields from the collected bill data and standardizes them to form a set of time fields in a unified format; then, a Bayesian network model describing the causal dependency relationship between the fields is established based on the preset business rules, wherein the nodes represent the time fields, the edges represent the logical sequence relationship between the fields, each node is configured with a corresponding conditional probability table for describing the probability distribution under different parent node states; on this basis, the system uses the Bayesian inference algorithm to analyze the logical consistency of each bill record, calculates the probability value of violating the time sequence logic, and determines whether there is a time sequence conflict by setting a logical error judgment threshold; finally, the ratio of the number of conflict bills to the total number of bills is taken as the logical conflict rate characteristic value for quantitatively evaluating the rationality of the overall bill data at the time logic level. Further, the system fuses the logical conflict rate characteristic value and the aforementioned semantic consistency characteristic value to construct a comprehensive credibility characteristic vector, and takes it as the input feature of the support vector machine model, introduces the kernel function mapping and regularization strategy to train the nonlinear regression model in the high-dimensional space to minimize the error between the predicted credibility score and the artificially annotated true score; after training, the model can automatically output the credibility score of the newly collected bill data, thereby realizing comprehensive and dynamic evaluation of the bill data quality. The application not only significantly improves the intelligentization and automation level of data checking in the bill financing business, but also provides high-quality data support for subsequent risk control, credit evaluation and intelligent approval and other key decision-making links, and has good engineering practicability and popularization prospect. BRIEF DESCRIPTION OF DRAWINGS

[0046] The application will be further described below in conjunction with the accompanying drawings.

[0047] Figure 1 is a flow chart of the bill business online financing management method based on the Internet platform of the application;

[0048] Figure 2 is a flow chart of the bill business online financing management system based on the Internet platform in the application. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the application will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0050] Please refer toFigure 1 The application is an Internet platform-based bill online financing management method, which comprises the following steps:

[0051] S1: In the bill financing application process, real-time collection of electronic bill data from multiple heterogeneous systems;

[0052] The heterogeneous systems include enterprise financial systems, bank core systems and third-party payment platforms, and the collected data covers different formats and field naming rules;

[0053] S2: Semantic analysis of the collected bill data, construction of a unified field mapping table, and calculation of semantic consistency feature values for evaluating the consistency of the same semantic fields in different systems;

[0054] S3: Analysis of the logical relationship between fields, identification of time sequence logical errors, and calculation of logical conflict rate feature values for evaluating the rationality of the internal logic of the bill data;

[0055] S4: The semantic consistency feature values and the logical conflict rate feature values are constructed into a comprehensive credibility feature vector, and a machine learning model is used to analyze the comprehensive credibility feature vector, and according to the analysis result, the credibility of the current bill data is evaluated;

[0056] S5: If the current bill data is determined to be untrustworthy, the data verification compensation mechanism is automatically triggered, and the manual review and intelligent error correction process is started.

[0057] In S1, in the bill financing application process, real-time collection of electronic bill data from multiple heterogeneous systems, the heterogeneous systems include enterprise financial systems, bank core systems and third-party payment platforms, and the collected data covers different formats and field naming rules, specifically including:

[0058] In the bill financing application process, the system is connected to multiple heterogeneous systems through standardized interfaces, including but not limited to enterprise financial systems, bank core systems and third-party payment platforms. Each system outputs electronic bill information in different data formats and transmission protocols according to its business characteristics, such as structured database tables, XML files, JSON messages or API interface calls. The system adopts a distributed data collection architecture to support concurrent access and real-time synchronization of multi-source heterogeneous data, ensuring that relevant bill data can be obtained in the first time after the bill financing application is initiated, and improving the overall business processing efficiency.

[0059] The collected electronic invoice data covers various formats and field naming conventions. For example, a corporate financial system may use "invoice date" as a field name, while a bank system names it "billing time", and a third-party payment platform uses English fields such as "InvoiceDate". In view of the above differences, the system has a built-in field recognition and semantic mapping module, which combines regular expression matching, keyword recognition, and field alias library to preliminarily normalize the field names of different systems, and uniformly converts the original data into a predefined data structure format, providing standardized input for subsequent semantic analysis and field consistency analysis.

[0060] In S2, the collected invoice data is subjected to semantic analysis, a unified field mapping table is constructed, and a semantic consistency feature value is calculated, which is used to evaluate the consistency of the expression of the same semantic field in different systems, specifically including:

[0061] Real-time collection of invoice data, semantic analysis of the collected invoice data, construction of a unified field mapping table, and calculation of a semantic consistency feature value, specifically including:

[0062] Text cleaning and standardization processing is performed on the contents of each field in the collected invoice data, and a pre-trained semantic embedding model (such as BERT) is used to convert each field content into a high-dimensional vector representation, thereby forming a set of invoice field vectors;

[0063] The set of invoice field vectors is input into a K-means clustering algorithm, and all field vectors are subjected to clustering analysis according to a set number of clusters, obtaining a plurality of clustering clusters, wherein each clustering cluster contains a plurality of field vectors that are similar in semantic space;

[0064] For each clustering cluster, the cosine similarity between any two different field vectors in the cluster is calculated, and according to the ratio between the sum of all cosine similarity values and the product of the total number of fields in the cluster and one less than the total number, the semantic consistency feature value corresponding to the corresponding clustering cluster is obtained, and the calculation expression is: ; In the formula, denotes the semantic consistency feature value of the th cluster, denotes the number of clustering clusters, denotes the th clustering cluster, denotes the th invoice field vector, denotes the th invoice field vector, and as an index, used to traverse all possible field vector pairs in the clustering cluster, so as to calculate the cosine similarity between these field vectors, denotes the clustering The number of field vectors contained therein;

[0065] The semantic consistency feature value is used to quantitatively evaluate the degree of consistency of semantic expression among fields in the same cluster. The closer the value is to 1, the more consistent the semantics of the fields in the corresponding cluster are, and the higher the consistency is; otherwise, it indicates that the semantic differences are large, and there may be misclassification or semantic inconsistency.

[0066] Determine whether the semantic consistency feature value is greater than or equal to a preset threshold. If it is, the same semantic field is expressed consistently in different systems; otherwise, the same semantic field is expressed inconsistently in different systems.

[0067] It should be noted that this method involves text cleaning and standardization of electronic invoice data from multiple heterogeneous systems. A pre-trained semantic embedding model is used to transform the field content from each system into high-dimensional vector representations. Then, the K-means clustering algorithm is employed to perform semantic space classification analysis of the fields, constructing a unified field mapping table. Based on this, the mean cosine similarity between field vectors within each cluster is calculated to obtain semantic consistency feature values, thereby achieving a quantitative assessment of the consistency of the same semantic field expressions across different systems. This method not only effectively identifies semantic differences caused by inconsistent field naming but also automatically judges the rationality of field mapping by setting a consistency threshold, significantly improving the accuracy and automation level of cross-system invoice data integration. It also exhibits good scalability and engineering practicality.

[0068] In S3, the logical relationships between fields are analyzed, chronological order logical errors are identified, and logical conflict rate characteristic values ​​are calculated to evaluate the rationality of the internal logic of the invoice data. Specifically, this includes:

[0069] Real-time acquisition of invoice data; analysis of logical relationships between fields in the acquired invoice data; identification of logical errors in time sequence; calculation of logical conflict rate characteristic values, specifically including:

[0070] The time-related fields in the collected invoice data are extracted and standardized to form a set of fields with a unified time format.

[0071] A Bayesian network model is constructed based on the preset time order logic rules between fields, where nodes represent the time field in the ticket, edges represent the causal dependencies between fields, and a corresponding conditional probability table is configured for each node.

[0072] For each ticket record, the probability of violating chronological order logic is calculated using a Bayesian inference algorithm. Specifically, this includes: given the parent node, calculating the maximum conditional probability of the current field state, and subtracting the maximum conditional probability from the result to obtain the chronological logic error probability of the corresponding ticket record. The calculation expression is as follows: In the formula, Represents a node The set of parent nodes, Represents a node The maximum conditional probability given the state of its parent node. Indicates the number of nodes. This represents the probability value of a time-based logic error. Indicates the first One bill record, Indicates the number of bill records. Indicates in Take the maximum value among the nodes. This indicates a logical error in the invoice record.

[0073] Set a logical error judgment threshold. When the logical error probability value of a certain ticket record exceeds the logical error judgment threshold, it is determined that the corresponding ticket has a time sequence logical conflict.

[0074] The number of tickets that are determined to have chronological logical conflicts is counted, and the ratio of this number to the total number of tickets is used as the logical conflict rate characteristic value.

[0075] Determine whether the logical conflict rate characteristic value is greater than or equal to a preset threshold. If it is, the internal logic of the invoice data is unreasonable; otherwise, the internal logic of the invoice data is reasonable.

[0076] It should be noted that this method involves extracting and standardizing time-related fields from invoice data to construct a dataset with a unified time format. A Bayesian network model describing the causal dependencies between these fields is then established based on pre-defined business rules. This Bayesian network model quantifies the logical relationships between time fields using conditional probability tables. During inference, it combines the actual values ​​of each invoice record to calculate the probability of violating time logic. Furthermore, a logical error judgment threshold is set to determine whether a single invoice has a time sequence conflict, and the proportion of conflicting invoices is used as a logical conflict rate feature to comprehensively assess the rationality of the internal logic of the invoice data. This method effectively identifies potential logical errors caused by misaligned or reversed field entries, exhibiting strong interpretability and generalization ability. It provides a quantitative analytical basis for invoice data quality control and significantly improves the intelligence level of data verification in invoice financing.

[0077] In S4, the semantic consistency feature value and the logical conflict rate feature value are constructed into a comprehensive credibility feature vector, a machine learning model is used to analyze the comprehensive credibility feature vector, and the credibility of the current bill data is evaluated according to the analysis result, specifically including:

[0078] The semantic consistency feature value and the logical conflict rate feature value of the bill data are obtained, the semantic consistency feature value and the logical conflict rate feature value are constructed into a comprehensive credibility feature vector, which is used as the input of the machine learning model, the error between the predicted credibility score and the actual credibility score is minimized as the training target, the model is trained, and the credibility score of the bill data is output according to the trained model. The machine learning model is a support vector machine model.

[0079] The training process of the machine learning model is:

[0080] The semantic consistency feature value and the logical conflict rate feature value of each bill are calculated respectively, and are combined into a comprehensive credibility feature vector as a feature dimension; the actual credibility score of each bill is labeled to form a labeled training sample set; the support vector machine model takes the error between the predicted credibility score and the actual credibility score as the objective function, uses the kernel function mapping strategy to map the feature vector to a high-dimensional space, and establishes a nonlinear regression model by solving the optimal classification hyperplane; a regularization term is introduced in the training process to control the model complexity and prevent overfitting, and finally a trained credibility prediction model is obtained for automatic credibility scoring output of newly collected bill data.

[0081] The credibility of the current bill data is evaluated according to the analysis result, specifically including:

[0082] It is judged whether the credibility score of the bill data is greater than or equal to a preset threshold, if yes, the corresponding bill data is credible, if not, the corresponding bill data is not credible.

[0083] In S5, if the current bill data is not credible, the data verification compensation mechanism is automatically triggered, and the manual review and intelligent error correction process is started, specifically including:

[0084] When the bill credibility score output by the trained support vector machine model is lower than the preset credibility threshold, the system will automatically determine that the bill data has a high risk of being not credible, and mark it as a to-be-reviewed state. The credibility threshold is dynamically adjusted according to historical data statistical analysis and actual business needs to adapt to different bill types, industry characteristics and risk control standards. Once the bill is marked as a to-be-reviewed state, the system starts a multi-level data verification compensation process and enters the automatic and manual collaborative processing stage.

[0085] In the first stage of the data verification compensation process, the system calls the built-in intelligent correction to automatically correct the identified semantic inconsistency fields or logical conflict fields. The intelligent correction combines the abnormal indicators of the semantic consistency feature value and the logical conflict rate feature value, analyzes the possible error patterns between fields, and provides several high-confidence correction candidates.

[0086] For example, for semantic mismatch problems caused by inconsistent field naming, the system can recommend unified field names; for time sequence errors, reasonable time adjustment suggestions can be made based on Bayesian network inference results. All correction suggestions will be pushed to the manual review interface together with the original ticket data for reference by the reviewer.

[0087] In the second stage, the system displays the distribution of the semantic consistency feature value and the logical conflict rate feature value in the ticket data in a visual manner to assist the reviewer in quickly locating high-risk field areas. Specifically, the system sorts and weightedly analyzes the semantic similarity scores of various fields and the logical consistency probabilities to form a risk heat map or a field anomaly index ranking, highlighting the key fields that are most likely to have problems. The reviewer can directly view the field details, correction suggestions, and related historical data comparison on the interface, thereby significantly improving the review efficiency and accuracy. Finally, the correction results confirmed by the human being are fed back to the system database and are simultaneously updated to the subsequent model training sample pool, realizing the continuous optimization and closed-loop control of the entire ticket data quality management system.

[0088] Please refer to Figure 2 The online financing management system for bill business based on an Internet platform includes:

[0089] A data acquisition module, which acquires electronic bill data from multiple heterogeneous systems in real time during a bill financing application process;

[0090] A bill semantic analysis module, which performs semantic analysis on the acquired bill data, constructs a unified field mapping table, and calculates a semantic consistency feature value for evaluating the consistency of the expression of the same semantic fields in different systems;

[0091] A bill rationality analysis module, which identifies time sequence logical errors by analyzing the logical relationship between fields, calculates a logical conflict rate feature value for evaluating the rationality of the internal logic of the bill data;

[0092] A bill credibility evaluation module, which constructs a comprehensive credibility feature vector from the semantic consistency feature value and the logical conflict rate feature value, analyzes the comprehensive credibility feature vector using a machine learning model, and evaluates the credibility of the current bill data according to the analysis result.

[0093] The bill correction management module automatically triggers a data verification compensation mechanism, starts an artificial review and intelligent error correction process if it is determined that the current bill data is not reliable.

[0094] The working principle of the present application is as follows: in the bill financing application process, electronic bill data from multiple heterogeneous systems such as enterprise financial systems, bank core systems and third-party payment platforms is collected in real time through a standardized interface, and the data has different formats and field naming rules; the system uses a distributed architecture to realize concurrent access and structure unification processing of data, and provides standardized input for subsequent analysis; the field content is converted into a high-dimensional vector using a pre-trained semantic embedding model, and a unified field mapping table is constructed by combining the K-means clustering algorithm; further, the semantic consistency feature value is obtained by calculating the cosine similarity within the cluster, so as to quantitatively evaluate the expression consistency of the same semantic field in different systems; the time sequence logical relationship between bill fields is analyzed based on a Bayesian network model, the time logic error probability of each bill record is calculated, and the logical conflict rate feature value is obtained by counting the conflict ratio, which is used to measure the rationality of the internal logic of the bill data; the above two feature values are constructed into a comprehensive credibility feature vector, which is used as the input of a support vector machine model; the credibility scoring model is trained by minimizing the prediction error, and the credibility score of the bill data is output; if it is determined that the bill data credibility is lower than the threshold, the data verification compensation mechanism is automatically triggered, the intelligent error correction module is called to generate a correction suggestion and push it to the artificial review interface, the visual risk prompt is combined to assist in quickly locating the problem field, and a closed-loop data quality management process is formed. The present application realizes intelligent management of the whole process of bill data collection, analysis and verification, significantly improves the data accuracy and processing efficiency of bill financing business, and has good application prospect and popularization value.

[0095] The above describes one embodiment of the present application in detail, but the content is only the preferred embodiment of the present application, and cannot be considered as limiting the scope of the present application. Any equivalent changes and improvements made according to the scope of the present application should still belong to the patent coverage range of the present application.

Claims

1. An internet platform based ticket business online financing management method, characterized in that, The method comprises the following steps: S1: In the bill financing application process, real-time collection of electronic bill data from multiple heterogeneous systems; S2: Semantic analysis of the collected bill data, construction of a unified field mapping table, and calculation of a semantic consistency feature value for evaluating the consistency of the expression of the same semantic field in different systems; The process of obtaining the semantic consistency feature value is: Real-time collection of bill data, semantic analysis of the collected bill data, construction of a unified field mapping table, and calculation of a semantic consistency feature value, specifically including: Text cleaning and standardization processing of each field content in the collected bill data, and conversion of each field content into a high-dimensional vector representation through a pre-trained semantic embedding model, thereby forming a bill field vector set; Input the bill field vector set into the K-means clustering algorithm, and perform clustering analysis on all field vectors according to the set number of clusters, to obtain multiple clustering clusters, wherein each clustering cluster contains several field vectors that are close in the semantic space; For each cluster, the cosine similarity between any two different field vectors in the cluster is calculated, and the semantic consistency feature value of the corresponding cluster is obtained according to the ratio between the sum of all cosine similarity values and the product of the total number of fields in the cluster and one less than the total number of fields in the cluster. The expression is: ; in the formula, the semantic consistency feature value of the i-th cluster is represented by , the number of clusters is represented by , the i-th cluster is represented by , the j-th field vector in the i-th cluster is represented by , the k-th field vector in the i-th cluster is represented by , and , and , and , and , and is used as an index to traverse all possible field vector pairs in the cluster so as to calculate the cosine similarity between the field vectors, , and the number of field vectors contained in the cluster is represented by . S3: Analyzing the logical relationship between fields, identifying time sequence logical errors, and calculating a logical conflict rate feature value for evaluating the rationality of the internal logic of the bill data; The process of obtaining the logical conflict rate feature value is: Real-time collection of bill data, analysis of the logical relationship between fields in the collected bill data, identification of time sequence logical errors, and calculation of a logical conflict rate feature value, specifically including: Extracting the fields related to time from the collected bill data, and standardizing each field to form a field set with a unified time format; Building a Bayesian network model according to the pre-set time sequence logical rules between fields, wherein the nodes represent the time fields in the bill, the edges represent the causal dependence relationship between the fields, and each node is configured with a corresponding conditional probability table; For each ticket record, the probability of violating chronological order logic is calculated using a Bayesian inference algorithm. Specifically, this includes: given the parent node, calculating the maximum conditional probability of the current field state, and subtracting the maximum conditional probability from the result to obtain the chronological logic error probability of the corresponding ticket record. The calculation expression is as follows: In the formula, Represents a node The set of parent nodes, Represents a node The maximum conditional probability given the state of its parent node. Indicates the number of nodes. This represents the probability value of a time-based logic error. Indicates the first One bill record, Indicates the number of bill records. Indicates in Take the maximum value among the nodes. This indicates a logical error in the bill record; Setting a logical error judgment threshold, when the logical error probability value of a bill record exceeds the logical error judgment threshold, it is determined that there is a time sequence logical conflict in the corresponding bill; Statistically counting the number of bill records that are determined to have a time sequence logical conflict, and taking the ratio of the number to the total number of bills as the logical conflict rate feature value; S4: Constructing a comprehensive credibility feature vector from the semantic consistency feature value and the logical conflict rate feature value, analyzing the comprehensive credibility feature vector using a machine learning model, and evaluating the credibility of the current bill data according to the analysis result; S5: If it is determined that the current bill data is not credible, automatically triggering a data verification compensation mechanism, and starting an artificial review and intelligent error correction process.

2. The internet platform based ticket business online financing management method according to claim 1, characterized in that, The evaluation of the consistency of the expression of the same semantic field in different systems specifically includes: Determining whether the semantic consistency feature value is greater than or equal to a pre-set threshold, if yes, the expression of the same semantic field in different systems is consistent, if not, the expression of the same semantic field in different systems is inconsistent. 3.The Internet platform-based ticket business online financing management method according to claim 1, characterized in that, The evaluation of the rationality of the internal logic of the bill data specifically includes: Determining whether the logical conflict rate feature value is greater than or equal to a pre-set threshold, if yes, the internal logic of the bill data is not reasonable, if not, the internal logic of the bill data is reasonable. 4.The Internet platform-based ticket business online financing management method according to claim 1, characterized in that, The analyzing the comprehensive credibility feature vector using the machine learning model specifically comprises: The semantic consistency feature value and the logical conflict rate feature value of the bill data are obtained, and the semantic consistency feature value and the logical conflict rate feature value are constructed into a comprehensive credibility feature vector as an input of the machine learning model, a support vector machine model, to minimize the error between the predicted credibility score and the actual credibility score as a training target to train the model, and the credibility score of the bill data is output according to the trained model. 5.The Internet platform-based ticket business online financing management method according to claim 4, characterized in that, The training process of the machine learning model is as follows: The semantic consistency feature value and the logical conflict rate feature value of each bill are calculated respectively and combined into a comprehensive credibility feature vector as a feature dimension; the actual credibility score of each bill is labeled to form a labeled training sample set; the support vector machine model takes minimizing the error between the predicted credibility score and the actual credibility score as an objective function, adopts a kernel function mapping strategy to map the feature vector to a high-dimensional space, and establishes a nonlinear regression model by solving an optimal classification hyperplane; a regularization term is introduced in the training process to control the model complexity and prevent overfitting, and finally a trained credibility prediction model is obtained for automatic credibility scoring and output of newly collected bill data. 6.The Internet platform-based ticket business online financing management method according to claim 1, characterized in that, The credibility of the current bill data is evaluated according to the analysis result, specifically comprising: It is judged whether the credibility score of the bill data is greater than or equal to a preset threshold, if yes, the corresponding bill data is credible, if not, the corresponding bill data is not credible. 7.The Internet platform-based online financing management method for a ticketing business according to claim 1, wherein, If it is judged that the current bill data is not credible, an automatic data verification compensation mechanism is triggered, specifically comprising: When the bill credibility score output by the support vector machine model is lower than the preset credibility threshold, the system automatically marks the corresponding bill as a to-be-reviewed state and starts a multi-level data verification compensation process; the process includes automatic correction of fields with semantic inconsistency or logical conflict and pushing to a manual review interface; at the same time, combined with the abnormal distribution of the semantic consistency feature value and the logical conflict rate feature value, the high-risk field area is located to assist the reviewer to quickly identify the problem source, thereby completing the closed-loop processing and quality improvement of the untrustworthy bill data.

8. The internet platform-based online financing management system for the internet platform-based online financing management method according to any one of claims 1-7, characterized in that, It comprises: A data acquisition module acquires electronic bill data from multiple heterogeneous systems in real time during the bill financing application process; A bill semantic analysis module performs semantic analysis on the collected bill data, constructs a unified field mapping table, and calculates a semantic consistency feature value for evaluating the consistency of the same semantic field in different systems; A bill rationality analysis module identifies time sequence logical errors by analyzing the logical relationship between fields, calculates a logical conflict rate feature value for evaluating the rationality of the internal logic of the bill data; The bill credibility evaluation module constructs a comprehensive credibility feature vector by combining the semantic consistency feature value and the logical conflict rate feature value, analyzes the comprehensive credibility feature vector using a machine learning model, and evaluates the credibility of the current bill data according to the analysis result. The bill correction management module automatically triggers a data verification compensation mechanism and starts an artificial review and intelligent error correction process if it is determined that the current bill data is not credible.

Images