A bank statement verification method and system based on OCR recognition technology

Abstract

The present application relates to the technical field of bank flow data verification, and in particular to a bank flow verification method and system based on OCR recognition technology. The method includes: obtaining a flow data set to be verified, analyzing the flow data set to be verified, determining a time-series structured flow data set and several target channel merchants; obtaining a channel merchant comprehensive document data set and a target channel merchant document data set based on several target channel merchants, analyzing the target channel merchant document data set based on the channel merchant comprehensive document data set, thereby determining a time-series verification rule set; obtaining reconciliation data, and performing a sliding time window reconciliation logic calculation on the time-series structured flow data set based on the time-series verification rule set and the reconciliation data to determine a reconciliation report; determining and outputting bank flow verification summary information based on the reconciliation report. The present application enables the automated verification process for bank flows to quickly respond to changes in channel merchant rules, reduce manual intervention, and improve the efficiency and accuracy of flow verification.

Description

Technical Field

[0001] The present application relates to the technical field of bank transaction data verification, and in particular to a bank transaction verification method and system based on OCR recognition technology. Background Art

[0002] With the popularity of electronic payments and online banking, bank statements have become important financial documents. In modern financial services, accurate verification of bank statements is an important part of ensuring compliance and preventing risks. A fast and accurate bank statement verification process is an important basis for ensuring the legality and authenticity of transactions.

[0003] Existing bank statement verification technology has limitations in its adaptability to the increasingly complex and dynamically changing channel merchant rules in the field involved. The automated processing flow in the existing bank statement verification cannot respond quickly to changes in channel merchant rules, resulting in frequent manual intervention and processing, which reduces the efficiency of bank statement verification, delays in user needs, and reduces user satisfaction and trust. Summary of the invention

[0004] The present application provides a bank statement verification method and system based on OCR recognition technology to solve the above technical problems.

[0005] In a first aspect, the present application provides a bank statement verification method based on OCR recognition technology, the method comprising:

[0006] Obtaining a pipeline data set to be verified, analyzing the pipeline data set to be verified, and determining a time-series structured pipeline data set and a number of target channel vendors;

[0007] According to the target channel merchants, a channel merchant comprehensive document dataset and a target channel merchant document dataset are obtained, and based on the channel merchant comprehensive document dataset, the target channel merchant document dataset is analyzed to determine a time series verification rule set;

[0008] Acquire reconciliation data, and perform sliding time window reconciliation logic calculation on the time series structured pipeline data set based on the time series verification rule set and the reconciliation data to determine a reconciliation report;

[0009] Based on the reconciliation report, determine and output the bank statement verification summary information.

[0010] Through this solution, the bank transaction data set to be verified is analyzed to determine the time-series structured bank transaction data set that is convenient for automated verification and the corresponding channel merchants of different payment channels involved in the bank transaction data. On this basis, combined with the comprehensive document data set of the channel merchants, the target channel merchant document data set is analyzed to determine the time-series verification rule set to ensure a high degree of matching between the bank transaction verification process and the channel merchant rules, so that the automated verification process for bank transactions can quickly respond to changes in channel merchant rules, reduce manual intervention, and improve the efficiency and accuracy of bank transaction verification. Based on the reconciliation data and the time-series verification rule set, the bank transaction data in the time-series structured bank transaction data set is verified by means of sliding time window reconciliation logic calculation, which further improves the verification efficiency while improving the flexibility and adaptability of the transaction verification process. Then, based on the reconciliation report, the bank transaction verification summary information is determined and output, so that users can intuitively understand the transaction verification results.

[0011] Optionally, analyzing the pipeline data set to be verified to determine a time-series structured pipeline data set and a number of target channel vendors includes:

[0012] Analyze the pipeline data set to be verified, determine the time period to be verified, and thereby determine the start time node and the end time node;

[0013] Based on the light character recognition algorithm, the transaction data set to be verified is analyzed, and the timestamp information, payment channel information and transaction change data set corresponding to each transaction data are extracted to construct a valid information set;

[0014] Determine a plurality of target channel merchants according to the plurality of payment channel information corresponding to the valid information set;

[0015] Based on the valid information set, the time-series structured pipeline data set is constructed according to the start time node and the end time node.

[0016] Through this solution, the start time node and the end time node are determined according to the time period to be verified corresponding to the transaction data set to be verified. At the same time, the timestamp information, payment channel information and transaction change data set in the transaction data set to be verified are identified and extracted to construct an effective information set, thereby determining several target channel merchants and constructing a time-series structured transaction data set to reduce redundant data and improve the verification efficiency and accuracy of the transaction verification process.

[0017] Optionally, based on the valid information set, the time-series structured pipeline data set is constructed according to the start time node and the end time node, specifically as follows:

[0018] ;

[0019] in, is the time-series structured pipeline dataset, is the valid information set, For the The timestamp information of the flow data, For the The payment channel information of the transaction data, For the The pipeline change data set of pipeline data, is the ray character recognition function, is the current flow data, is the pipeline data set to be verified, is the starting time node, is the end time node.

[0020] Through this solution, mathematical analysis methods are used, based on effective information sets, according to the start time node and the end time node, and through corresponding mathematical expressions, time-series structured pipeline data is automatically constructed to provide a scientific data basis for subsequent pipeline verification, improve data processing efficiency, and ensure the accuracy of pipeline verification.

[0021] Optionally, the step of analyzing the target channel vendor document dataset based on the channel vendor comprehensive document dataset to determine the time series verification rule set includes:

[0022] According to preset text preprocessing rules, the channel business comprehensive document data set is cleaned and segmented to determine a preprocessed comprehensive document data set;

[0023] Based on the preprocessed comprehensive document data set, perform word segmentation and vectorization processing on the keywords in the preprocessed comprehensive document data set to determine a document word vector data set corresponding to each target channel business;

[0024] Based on a preset channel business rule library, the document word vector data set corresponding to each target channel business is analyzed to determine a model training data set;

[0025] Constructing a channel business rule analysis model based on the model training data set;

[0026] Analyze the target channel business document data set according to the channel business rule analysis model and the preset channel business rule base to determine the rule set to be updated;

[0027] The timing verification rule set is determined according to the rule set to be updated and the preset channel provider rule base.

[0028] Through this solution, the comprehensive document data set of channel merchants is preprocessed to determine the preprocessed comprehensive document data set. On this basis, the word segmentation and vectorization technology is used to obtain the document word vector data set corresponding to each target channel merchant. Combined with the preset channel merchant rule library, a channel merchant rule analysis model is constructed, and the rule set to be updated is obtained through this analysis, and then the time series verification rule set is determined to achieve automated analysis of the target channel merchant document data set, thereby improving the accuracy and efficiency of the time series verification rule set analysis process.

[0029] Optionally, the timing verification rule set is determined according to the rule set to be updated and the preset channel provider rule base, specifically the following formula:

[0030] ;

[0031] in, is the temporal verification rule set, is the preset channel provider rule library, is the rule set to be updated.

[0032] Through this solution, mathematical analysis methods are used to merge the rule set to be updated and the preset channel merchant rule base according to the rule set to be updated. While retaining the channel merchant rules corresponding to different historical time nodes in the preset channel merchant rule base, the channel merchant rules that are not included in the preset channel merchant rule base or have inconsistent contents are included in the time series verification rule set to ensure the comprehensiveness and real-time nature of the time series verification rule set.

[0033] Optionally, based on the preset channel business rule library, the document word vector data set corresponding to each target channel business is analyzed to determine the model training data set, which is specifically the following formula:

[0034]

[0035] in, A training dataset for the model, For the The document word vector dataset corresponding to the target channel providers, For the The document comprehensive feature vector corresponding to the target channel business, is the vector fusion function, is the entity recognition function, is the entity vector set corresponding to the current document word vector dataset, is the relation extraction function, is the entity relationship vector set corresponding to the current document word vector dataset, For the A set of rule tags corresponding to each target channel partner in the preset channel partner rule library.

[0036] Through this solution, mathematical analysis methods are used to analyze the document word vector data set corresponding to each target channel merchant based on the preset channel merchant rule library. By analyzing the entities represented by each keyword vector in the document word vector data set and the relationship between each entity, an entity vector set and a corresponding entity relationship vector set are obtained. On this basis, a document comprehensive feature vector that can reflect the comprehensive characteristics of the corresponding target channel merchant document data set is formed, thereby improving the context perception ability of the subsequent channel merchant rule analysis model. Combined with the corresponding rule label set, a model training data set is jointly formed, reducing the dependence of the subsequent channel merchant rule analysis model on manually labeled data and improving model training efficiency.

[0037] Optionally, the channel business rule analysis model is constructed according to the model training data set, specifically the following formula:

[0038] ;

[0039] in, The target optimization parameters corresponding to the channel business rule analysis model, For the The comprehensive feature vector of the document corresponding to the target channel business, For the The rule tag set corresponding to each target channel partner in the preset channel partner rule library, A training dataset for the model, is the loss function, For the A set of rule labels corresponding to a target channel partner in the preset channel partner rule library, is the label prediction function.

[0040] Through this solution, mathematical analysis methods are used, according to the model training data set, and the channel merchant rule analysis model training process is described in mathematical language. The document comprehensive feature vector under the current target optimization parameters is processed using the label prediction function to obtain the rule labels corresponding to the current document comprehensive feature vector, and then the matching loss degree between each rule label and the rule label set is measured through the loss function. By processing the document comprehensive feature vectors corresponding to different target channel merchants, the matching loss degree is minimized, and finally the channel merchant rule analysis model is trained to enable the channel merchant rule analysis model to automatically parse the channel merchant rule documents, and extract the rule content corresponding to different rule labels in the channel merchant rule documents to form structured rule data, which provides a basis for channel merchant rules for subsequent pipeline verification.

[0041] Optionally, the performing a sliding time window reconciliation logic calculation on the time series structured pipeline data set based on the time series verification rule set and the reconciliation data to determine a reconciliation report includes:

[0042] Determine the time span of the sliding window according to the time period to be verified;

[0043] Analyze the time series structured pipeline data set according to the sliding window time span, and determine the window pipeline sum corresponding to the sliding time window span;

[0044] Based on the reconciliation data, the window transaction sum is analyzed, and the payment interval is determined according to the difference between the window transaction sum and the corresponding payment amount in the reconciliation data;

[0045] Based on the reconciliation data, the time-series structured transaction data set and the corresponding payment interval are checked according to the time-series verification rule set, and a reconciliation report is determined based on the checking result.

[0046] Through this solution, the corresponding sliding window time span is set according to the time period to be verified, so as to extract the total window flow corresponding to the sliding time window span in the time-series structured flow data set, and determine the payment interval based on the difference between the window flow total and the corresponding payment amount in the reconciliation data. On this basis, the verification results between the time-series verification rule set, the time-series structured flow data set and the corresponding payment interval are integrated to determine the reconciliation report.

[0047] Optionally, determining and outputting bank statement verification summary information according to the reconciliation report includes:

[0048] Analyze the reconciliation report, extract abnormal transaction data, and push the abnormal transaction data to the manual review process, and obtain manual review feedback provided by the manual review process;

[0049] The bank statement verification summary information is generated based on the reconciliation report and the manual review feedback, and the bank statement verification summary information is provided to the user.

[0050] Through this solution, the abnormal transaction data in the reconciliation report will be pushed to the manual review process, and the manual review feedback given by the manual review department on the abnormal transaction data will be received. Combined with the reconciliation report, the final bank transaction verification summary information will be generated to further ensure the comprehensiveness and accuracy of the transaction data verification.

[0051] In a second aspect, the present application provides a bank flow verification system based on OCR recognition technology, the system comprising:

[0052] A pipeline analysis module is used to obtain a pipeline data set to be verified, analyze the pipeline data set to be verified, and determine a time-series structured pipeline data set and a number of target channel vendors;

[0053] A channel analysis module, for obtaining a channel merchant comprehensive document data set and a target channel merchant document data set according to a number of the target channel merchants, and analyzing the target channel merchant document data set based on the channel merchant comprehensive document data set to determine a time series verification rule set;

[0054] A checking module, used to obtain reconciliation data, and based on the time series verification rule set and the reconciliation data, perform a sliding time window reconciliation logic calculation on the time series structured pipeline data set to determine a reconciliation report;

[0055] The output module is used to determine and output the bank statement verification summary information based on the reconciliation report.

[0056] Optionally, the pipeline analysis module is specifically used to:

[0057] Analyze the pipeline data set to be verified, determine the time period to be verified, and thereby determine the start time node and the end time node;

[0058] Based on the light character recognition algorithm, the transaction data set to be verified is analyzed, and the timestamp information, payment channel information and transaction change data set corresponding to each transaction data are extracted to construct a valid information set;

[0059] Determine a plurality of target channel merchants according to the plurality of payment channel information corresponding to the valid information set;

[0060] Based on the valid information set, the time-series structured pipeline data set is constructed according to the start time node and the end time node.

[0061] Optionally, when the pipeline analysis module constructs the time-series structured pipeline data set based on the valid information set, according to the start time node and the end time node, it is specifically the following formula:

[0062] ;

[0063] in, is the time-series structured pipeline dataset, is the valid information set, For the The timestamp information of the flow data, For the The payment channel information of the transaction data, For the The pipeline change data set of pipeline data, is the ray character recognition function, is the current flow data, is the pipeline data set to be verified, is the starting time node, is the end time node.

[0064] Optionally, the channel analysis module is specifically used to:

[0065] According to preset text preprocessing rules, the channel business comprehensive document data set is cleaned and segmented to determine a preprocessed comprehensive document data set;

[0066] Based on the preprocessed comprehensive document data set, perform word segmentation and vectorization processing on the keywords in the preprocessed comprehensive document data set to determine a document word vector data set corresponding to each target channel business;

[0067] Based on a preset channel business rule library, the document word vector data set corresponding to each target channel business is analyzed to determine a model training data set;

[0068] Constructing a channel business rule analysis model based on the model training data set;

[0069] Analyze the target channel business document data set according to the channel business rule analysis model and the preset channel business rule base to determine the rule set to be updated;

[0070] The timing verification rule set is determined according to the rule set to be updated and the preset channel provider rule base.

[0071] Optionally, when the channel analysis module determines the time series verification rule set according to the rule set to be updated and the preset channel business rule base, it is specifically the following formula:

[0072] ;

[0073] in, is the temporal verification rule set, is the preset channel provider rule library, is the rule set to be updated.

[0074] Optionally, the channel analysis module analyzes the document word vector data set corresponding to each target channel merchant based on a preset channel merchant rule base to determine the model training data set, specifically using the following formula:

[0075]

[0076] in, A training dataset for the model, For the The document word vector dataset corresponding to the target channel providers, For the The document comprehensive feature vector corresponding to the target channel business, is the vector fusion function, is the entity recognition function, is the entity vector set corresponding to the current document word vector dataset, is the relation extraction function, is the entity relationship vector set corresponding to the current document word vector dataset, For the A set of rule tags corresponding to each target channel partner in the preset channel partner rule library.

[0077] Optionally, the channel analysis module constructs a channel business rule analysis model according to the model training data set, which is specifically the following formula:

[0078] ;

[0079] in, The target optimization parameters corresponding to the channel business rule analysis model, For the The comprehensive feature vector of the document corresponding to the target channel business, For the The rule tag set corresponding to each target channel partner in the preset channel partner rule library, A training dataset for the model, is the loss function, For the A set of rule labels corresponding to a target channel partner in the preset channel partner rule library, is the label prediction function.

[0080] Optionally, the checking module is specifically used for:

[0081] Determine the time span of the sliding window according to the time period to be verified;

[0082] Analyze the time series structured pipeline data set according to the sliding window time span, and determine the window pipeline sum corresponding to the sliding time window span;

[0083] Based on the reconciliation data, the window transaction sum is analyzed, and the payment interval is determined according to the difference between the window transaction sum and the corresponding payment amount in the reconciliation data;

[0084] Based on the reconciliation data, the time-series structured transaction data set and the corresponding payment interval are checked according to the time-series verification rule set, and a reconciliation report is determined based on the checking result.

[0085] Optionally, the output module is specifically used to:

[0086] Analyze the reconciliation report, extract abnormal transaction data, and push the abnormal transaction data to the manual review process, and obtain manual review feedback provided by the manual review process;

[0087] The bank statement verification summary information is generated based on the reconciliation report and the manual review feedback, and the bank statement verification summary information is provided to the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0089] Figure 1 A schematic diagram of an application scenario provided for an embodiment of the present application.

[0090] Figure 2 A flowchart of a bank statement verification method based on OCR recognition technology provided in one embodiment of the present application.

[0091] Figure 3 A schematic diagram of the structure of a bank statement verification system based on OCR recognition technology provided in one embodiment of the present application. DETAILED DESCRIPTION

[0092] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0093] In addition, the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article, unless otherwise specified, generally means that the associated objects before and after are in an "or" relationship.

[0094] The embodiments of the present application are further described in detail below in conjunction with the drawings in the specification.

[0095] Existing bank statement verification technology has limitations in its adaptability to the increasingly complex and dynamically changing channel merchant rules in the field involved. The automated processing flow in the existing bank statement verification cannot respond quickly to changes in channel merchant rules, resulting in frequent manual intervention and processing, which reduces the efficiency of bank statement verification, delays in user needs, and reduces user satisfaction and trust.

[0096] Based on this, the present application provides a bank flow verification method and system based on OCR recognition technology. The flow data set to be verified is analyzed to determine the time-series structured flow data set that is convenient for automated verification processing and the corresponding channel merchants of different payment channels involved in the flow data. On this basis, combined with the comprehensive document data set of the channel merchant, the target channel merchant document data set is analyzed to determine the time-series verification rule set to ensure a high degree of matching between the flow verification process and the channel merchant rules, so that the automated verification process for bank flows can quickly respond to changes in channel merchant rules, reduce manual intervention, and improve the efficiency and accuracy of flow verification. Based on the reconciliation data and the time-series verification rule set, the flow data in the time-series structured flow data set is verified by means of sliding time window reconciliation logic calculation, further improving the verification efficiency while improving the flexibility and adaptability of the flow verification process, and then according to the reconciliation report, the bank flow verification summary information is determined and output, so that users can intuitively understand the flow verification results.

[0097] Figure 1 A schematic diagram of an application scenario provided by this application. In the bank statement verification process, the method provided by this application is applied to enable the automated verification process for bank statements to quickly respond to changes in channel merchant rules, reduce manual intervention, and improve the efficiency and accuracy of bank statement verification.

[0098] Specifically, the method of the present application is installed in any server, through which the data set of transaction flow to be verified provided by the user is obtained and analyzed, and the time-series structured transaction flow data set and the corresponding channel merchants of different payment channels involved in the transaction flow data that are convenient for automated verification processing are determined. On this basis, the target channel merchant document data set provided by the channel merchant is analyzed in combination with the channel merchant comprehensive document data set provided by the channel merchant, and the time-series verification rule set is determined to ensure a high degree of matching between the transaction flow verification process and the channel merchant rules, so that the automated verification process for bank transactions can quickly respond to changes in channel merchant rules, reduce manual intervention, and improve the efficiency and accuracy of transaction flow verification. Based on the reconciliation data and the time-series verification rule set provided by the bank, the transaction flow data in the time-series structured transaction flow data set is verified by means of a sliding time window reconciliation logic calculation, further improving the verification efficiency while improving the flexibility and adaptability of the transaction flow verification process, and then according to the reconciliation report, the bank transaction verification summary information is determined and provided to the user, so that the user can intuitively understand the transaction flow verification result. The specific implementation method can refer to the following embodiments.

[0099] Figure 2 This is a flow chart of a bank statement verification method based on OCR recognition technology provided in an embodiment of the present application. The method of this embodiment can be applied to the server in the above scenario. Figure 2 As shown, the method includes:

[0100] S201, obtaining a flow data set to be verified, analyzing the flow data set to be verified, and determining a time-series structured flow data set and a number of target channel vendors.

[0101] The transaction data set to be verified may be a transaction data set to be verified provided by a user, and may include image data, electronic document data, and the like.

[0102] The time-series structured pipeline data set can be a structured pipeline data set that is constructed based on the pipeline change time and is easy to analyze automatically according to the pipeline data set to be verified.

[0103] The target channel merchant may be the channel merchant corresponding to the payment channel in the transaction slip data provided by the current user.

[0104] Specifically, in the process of bank transaction verification, since the transaction information provided by users is of various types and usually involves multiple payment channels, and the corresponding fund flow cycles and additional fees in different payment channel transaction rules are different, if the bank transaction verification process only performs a single amount check, the accuracy and reliability of the transaction verification process cannot be guaranteed. Through OCR (Optical Character Recognition) technology, text recognition and analysis are performed on the transaction data set to be verified containing different types of transaction data, and the various amount types and corresponding amount data in different transaction slips, as well as their corresponding timestamps, are extracted to construct a time-series structured transaction data set, which provides a targeted data framework for subsequent automated transaction verification, and by identifying the payment channels corresponding to each transaction slip, several target channel merchants corresponding to the user's transaction data set to be verified are determined, providing guidance data for the subsequent transaction verification process based on channel merchant rules.

[0105] S202: According to a number of target channel merchants, a channel merchant comprehensive document dataset and a target channel merchant document dataset are obtained, and based on the channel merchant comprehensive document dataset, the target channel merchant document dataset is analyzed to determine a time series verification rule set.

[0106] The channel merchant comprehensive document data set may be a collection of document files of various historical periods corresponding to the transaction processing rules issued by the channel merchants corresponding to each payment channel and required to be agreed to when using its payment channel.

[0107] The target channel merchant document data set may be a set of transaction agreement documents at different time nodes corresponding to the payment channels involved in the current transaction data set to be verified.

[0108] The channel merchant rules may be the transaction processing rules of the channel merchant corresponding to the payment channel.

[0109] The time series verification rule set may be a verification rule for transaction data formulated based on the payment channel transaction rules at the corresponding time node of the transaction data.

[0110] Specifically, due to the differences in transaction rules formulated by channel merchants for different payment channels, the content and content description forms of the transaction rule documents issued by various channel merchants are different, and the content of the channel merchant transaction rule documents is updated irregularly. If the latest rules of the channel merchant are used as the benchmark to verify the transaction information corresponding to each historical time node in the transaction data set to be verified, it is easy to misjudge the accuracy of the transaction data. Therefore, it is necessary to automatically parse the rule documents issued by different channel merchants at different time nodes, extract the channel merchant rules corresponding to different time nodes, and use the channel merchant rules adopted by the corresponding channel merchant at the corresponding time node for the transaction data at different time nodes during the transaction data verification process to ensure the accuracy of the transaction data verification process. In order to realize the automatic and accurate analysis and rule extraction of the transaction rule documents corresponding to each channel merchant, NPL (Natural Language Processing (Natural Language Processing) technology, combined with mathematical analysis methods, is used to build and train a mathematical model for analyzing channel merchant rule documents based on the channel merchant comprehensive document dataset. Based on the mathematical model, the target channel merchant document dataset is analyzed to extract the transaction rules of the corresponding transaction channels of the channel merchant at different time nodes in the current transaction data set to be verified. This is used as a time-series verification rule set for subsequent verification of user transaction data, so that the verification of transaction data is highly consistent with the transaction rules of different transaction channels at different time nodes, significantly improving the accuracy and adaptability of the transaction verification process.

[0111] S203: Obtain reconciliation data, perform sliding time window reconciliation logic calculation on the time series structured flow data set based on the time series verification rule set and the reconciliation data, and determine a reconciliation report.

[0112] The reconciliation data may be various statement data provided by the bank.

[0113] The reconciliation report can be report information obtained by integrating the verification results of various amount data between the time-series structured flow data set and the reconciliation data based on the time-series verification rule set.

[0114] Specifically, in actual business, there may be delays or inconsistencies in the time of payment. To address this situation, a sliding time window reconciliation logic calculation process is designed. This process determines a sliding window with a fixed time span for the corresponding time period in the transaction data to be verified provided by the current user. Based on the sliding time window, the transaction data within the corresponding time span of the sliding time window is segmented and analyzed according to the time series verification rule set. By analyzing the difference between the transaction amount in the sliding time window and the payment amount of the corresponding period in the reconciliation data, the payment interval is located to serve as the reconciliation basis in the transaction verification process, thereby improving the analysis efficiency and enabling matching transaction records to be found quickly when the transaction data volume is large. The above sliding time window process allows flexible analysis under the boundary conditions of the data, and can discover some situations that may be overlooked, such as payment and transaction near certain specific dates. By performing a sliding time window reconciliation logic calculation on the time series structured transaction data set based on the time series verification rule set and reconciliation data, a reconciliation report is obtained, effectively improving the accuracy and verification efficiency of the transaction verification process.

[0115] S204. According to the reconciliation report, determine and output the bank statement verification summary information.

[0116] The bank statement verification summary information may be the bank statement verification summary information obtained by visually integrating the verification results of various transaction amounts based on the reconciliation report.

[0117] Specifically, after obtaining the reconciliation report, the verification results of each transaction amount in the reconciliation report are visualized and integrated through data visualization technology, and the bank transaction verification summary information that can be directly viewed and read by users is obtained. The bank transaction verification summary information is output through human-computer interaction equipment, such as high-definition display screens, so that users can intuitively understand the transaction verification results.

[0118] Through this solution, the bank transaction data set to be verified is analyzed to determine the time-series structured bank transaction data set that is convenient for automated verification and the corresponding channel merchants of different payment channels involved in the bank transaction data. On this basis, combined with the comprehensive document data set of the channel merchants, the target channel merchant document data set is analyzed to determine the time-series verification rule set to ensure a high degree of matching between the bank transaction verification process and the channel merchant rules, so that the automated verification process for bank transactions can quickly respond to changes in channel merchant rules, reduce manual intervention, and improve the efficiency and accuracy of bank transaction verification. Based on the reconciliation data and the time-series verification rule set, the bank transaction data in the time-series structured bank transaction data set is verified by means of sliding time window reconciliation logic calculation, which further improves the verification efficiency while improving the flexibility and adaptability of the transaction verification process. Then, based on the reconciliation report, the bank transaction verification summary information is determined and output, so that users can intuitively understand the transaction verification results.

[0119] In some embodiments, the transaction data set to be verified is analyzed to determine the time period to be verified, thereby determining the start time node and the end time node; based on the optical character recognition algorithm, the transaction data set to be verified is analyzed, and the timestamp information, payment channel information and transaction change data set corresponding to each transaction data are extracted to construct a valid information set; based on the corresponding payment channel information in the valid information set, a number of target channel merchants are determined; based on the valid information set, according to the start time node and the end time node, a time-series structured transaction data set is constructed.

[0120] The time period to be verified may be the overall time span of all the pipeline data in the pipeline data set to be verified.

[0121] The starting time node may be the starting time point of the time period to be verified corresponding to the flow data set to be verified.

[0122] The end time node may be the end time point of the time period to be verified corresponding to the flow data set to be verified.

[0123] The optical character recognition algorithm may be an OCR algorithm for recognizing transaction slip information.

[0124] The timestamp information may be a timestamp corresponding to the time when the flow data is generated.

[0125] The payment channel information may be corresponding information used to characterize the payment channels adopted by different transaction data.

[0126] The turnover change data set may be a data set used to characterize the turnover change amount.

[0127] Specifically, in the process of determining the time-series structured flow data set, by analyzing the generation time of each flow data in the flow data set to be verified, the overall time span of the flow data in the current flow data set to be verified, such as half a year or a year, is obtained, so as to extract the start time node and the end time node in the current flow data set to be verified, which are used as the overall time interval reference for subsequent flow data time series analysis and processing. The data format used to reflect the change of the user's flow amount in the flow data set to be verified is usually the image, scan or electronic document corresponding to the flow slip. In order to adapt to different data formats, the optical character recognition algorithm, namely the OCR algorithm, is used to identify the flow data content in different data formats. At the same time, due to There are some redundant data in the identified data that are not needed in the transaction data verification process, such as transaction records and placeholder information contained in the transaction slip. In order to improve the subsequent verification efficiency of the transaction data, when the transaction data content is identified and extracted through the OCR algorithm, the timestamp information, payment channel information and transaction change data set corresponding to the transaction data are extracted in a targeted manner to construct an effective information set to reduce redundant data. According to the payment channel information corresponding to each transaction data, the payment channels involved in each transaction data can be clarified, so as to determine a number of target channel merchants, and then through mathematical analysis, based on the effective information set, according to the start time node and the end time node, a time-series structured transaction data set is automatically constructed.

[0128] Through this solution, the start time node and the end time node are determined according to the time period to be verified corresponding to the transaction data set to be verified. At the same time, the timestamp information, payment channel information and transaction change data set in the transaction data set to be verified are identified and extracted to construct an effective information set, thereby determining several target channel merchants and constructing a time-series structured transaction data set to reduce redundant data and improve the verification efficiency and accuracy of the transaction verification process.

[0129] In some embodiments, based on the valid information set, a time-series structured pipeline data set is constructed according to the start time node and the end time node, specifically, the following formula (1):

[0130] (1)

[0131] in, It is a time-series structured pipeline dataset. is a valid information set, For the The timestamp information of the flow data, For the Payment channel information for transaction data, For the A dataset of running water changes of running water data, is the ray character recognition function, is the current flow data, is the flow data set to be verified, is the starting time node, The end time node.

[0132] Specifically, through formula (1) The OCR algorithm is used to identify and extract the timestamp information, payment channel information and transaction change data corresponding to each transaction data in the transaction data set to be verified, and a valid information set is constructed. Traverse each information in the valid information set, and then use the timestamp information corresponding to each information as the index and the corresponding flow change data as the content to construct a time-series structured flow data set. Constrain the time node range.

[0133] Through this solution, mathematical analysis methods are used, based on effective information sets, according to the start time node and the end time node, and through corresponding mathematical expressions, time-series structured pipeline data is automatically constructed to provide a scientific data basis for subsequent pipeline verification, improve data processing efficiency, and ensure the accuracy of pipeline verification.

[0134] In some embodiments, according to preset text preprocessing rules, the channel merchant comprehensive document data set is cleaned and segmented to determine a preprocessed comprehensive document data set; based on the preprocessed comprehensive document data set, the keywords in the preprocessed comprehensive document data set are segmented and vectorized to determine the document word vector data set corresponding to each target channel merchant; based on the preset channel merchant rule base, the document word vector data set corresponding to each target channel merchant is analyzed to determine the model training data set; based on the model training data set, a channel merchant rule analysis model is constructed; based on the channel merchant rule analysis model and the preset channel merchant rule base, the target channel merchant document data set is analyzed to determine the rule set to be updated; based on the rule set to be updated and the preset channel merchant rule base, a time series verification rule set is determined.

[0135] The preset text preprocessing rules may be preset preprocessing rules for text data in the channel business comprehensive document data set, and may include denoising and word segmentation rules.

[0136] The preprocessed comprehensive document data set may be a document data set obtained after preprocessing the channel provider comprehensive document data set.

[0137] Word segmentation vectorization processing can be the process of converting keywords obtained after word segmentation processing into corresponding vector information.

[0138] The document word vector dataset can be a data set containing the vector information corresponding to all keywords.

[0139] The preset channel business rule base may be a preset database for storing payment channel transaction rules and corresponding rule labels corresponding to different channel business.

[0140] The channel business rule analysis model may be a mathematical model used to analyze channel business document data and realize automatic analysis and extraction of channel business rules.

[0141] The model training data set may be a data set used to train and construct a channel business rule analysis model.

[0142] The rule set to be updated may be a target channel merchant corresponding rule set obtained by analyzing the target channel merchant document data set using a channel merchant rule analysis model.

[0143] Specifically, the channel merchant comprehensive document data set is cleaned and segmented by using the text denoising algorithm and word segmentation processing library specified in the preset text preprocessing rules, such as regular expressions and natural language processing libraries, to extract the preprocessed comprehensive document data set, and the keywords obtained by the word segmentation processing are converted into word vectors by using a word vector model, such as THULAC (THU Lexical Analyzer for Chinese, Tsinghua University Chinese word segmentation and part-of-speech tagging toolkit), so as to obtain the document word vector data set corresponding to each target channel merchant, and the channel merchant rule analysis model is trained and constructed by mathematical analysis means in combination with the preset channel merchant rule library, and the channel merchant rules in the target channel merchant document data set are analyzed and extracted according to the channel merchant rule analysis model, and the preset channel merchant rule library is retrieved according to the analysis results of the target channel merchant document data set, and if there are channel merchant rules in the analysis results of the target channel merchant document data set that are not included in the current preset channel merchant rule library or are included but have inconsistent contents, then this part of the channel merchant rules and their corresponding update time nodes are determined as the rule set to be updated, so as to realize the construction of a time-series structured pipeline data set based on the preset channel merchant rule library and the rule set to be updated.

[0144] Through this solution, the comprehensive document data set of channel merchants is preprocessed to determine the preprocessed comprehensive document data set. On this basis, the word segmentation and vectorization technology is used to obtain the document word vector data set corresponding to each target channel merchant. Combined with the preset channel merchant rule library, a channel merchant rule analysis model is constructed, and the rule set to be updated is obtained through this analysis, and then the time series verification rule set is determined to achieve automated analysis of the target channel merchant document data set, thereby improving the accuracy and efficiency of the time series verification rule set analysis process.

[0145] In some embodiments, the timing verification rule set is determined according to the rule set to be updated and the preset channel provider rule base, specifically as the following formula (2):

[0146] (2)

[0147] in, is the temporal verification rule set, is the preset channel provider rule library, is the rule set to be updated.

[0148] Specifically, the rule set to be updated is merged with the preset channel merchant rule base through formula (2). While retaining the channel merchant rules corresponding to different historical time nodes in the preset channel merchant rule base, the channel merchant rules that are not included in the preset channel merchant rule base or have inconsistent contents are included in the time series verification rule set to ensure the comprehensiveness and real-time performance of the time series verification rule set.

[0149] Through this solution, mathematical analysis methods are used to merge the rule set to be updated and the preset channel merchant rule base according to the rule set to be updated. While retaining the channel merchant rules corresponding to different historical time nodes in the preset channel merchant rule base, the channel merchant rules that are not included in the preset channel merchant rule base or have inconsistent contents are included in the time series verification rule set to ensure the comprehensiveness and real-time nature of the time series verification rule set.

[0150] In some embodiments, based on a preset channel business rule library, the document word vector dataset corresponding to each target channel business is analyzed to determine the model training dataset, which is specifically the following formula (3):

[0151] (3)

[0152] in, The dataset for model training, For the The document word vector dataset corresponding to the target channel providers, For the The document comprehensive feature vector corresponding to the target channel business, is the vector fusion function, is the entity recognition function, is the entity vector set corresponding to the current document word vector dataset, is the relation extraction function, is the entity relationship vector set corresponding to the current document word vector dataset, For the A set of rule labels corresponding to each target channel partner in the preset channel partner rule library.

[0153] The vector fusion function may be a mathematical function for fusing and splicing vectors, and the vector fusion function may adopt a vector splicing algorithm.

[0154] The entity recognition function may be a mathematical model used to identify all entities in a document word vector dataset, and the entity recognition function may adopt a deep learning model.

[0155] The relation extraction function may be a mathematical model for identifying the relationship between different entities in a document word vector data set, and the relation extraction function may adopt a convolutional neural network.

[0156] The document comprehensive feature vector may be vector information used to characterize the comprehensive features of the corresponding target channel business document data set.

[0157] The rule tag set may be an information set used to characterize the type of the current rules, such as a handling fee standard, a settlement cycle, and the like.

[0158] Specifically, the entity recognition function and relationship extraction function in formula (3) are used to extract the entities represented by each keyword vector in the document word vector data set and the relationship between each entity, respectively, to obtain an entity vector set and a corresponding entity relationship vector set, and the entity vector set and the corresponding entity relationship vector set are concatenated and fused through the vector fusion function to form a document comprehensive feature vector that can reflect the comprehensive characteristics of the corresponding target channel merchant document data set, thereby improving the context perception ability of the subsequent channel merchant rule analysis model, and retrieving the rule label set used to refer to each rule type in the current target channel merchant rule in the preset channel merchant rule library, which together constitute the model training data set, reduce the dependence of the subsequent channel merchant rule analysis model on manually labeled data, and improve the model training efficiency.

[0159] Through this solution, mathematical analysis methods are used to analyze the document word vector data set corresponding to each target channel merchant based on the preset channel merchant rule library. By analyzing the entities represented by each keyword vector in the document word vector data set and the relationship between each entity, an entity vector set and a corresponding entity relationship vector set are obtained. On this basis, a document comprehensive feature vector that can reflect the comprehensive characteristics of the corresponding target channel merchant document data set is formed, thereby improving the context perception ability of the subsequent channel merchant rule analysis model. Combined with the corresponding rule label set, a model training data set is jointly formed, reducing the dependence of the subsequent channel merchant rule analysis model on manually labeled data and improving model training efficiency.

[0160] In some embodiments, a channel business rule analysis model is constructed based on the model training data set, specifically the following formula (4):

[0161] (4)

[0162] in, Optimize the target parameters corresponding to the channel business rule analysis model. For the The document comprehensive feature vector corresponding to the target channel business, For the The rule tag set corresponding to each target channel partner in the preset channel partner rule library, The dataset for model training, is the loss function, For the The rule tag set corresponding to each target channel partner in the preset channel partner rule library, is the label prediction function.

[0163] The target optimization parameter may be a parameter used to characterize the training target of the channel business rule analysis model. When the parameter meets the model requirements, the model training is considered to be completed. The target optimization parameter may be set according to the model requirements.

[0164] The loss function may be a function used to measure the degree of matching loss between the comprehensive feature vector of the document and the rule label set, and the loss function may adopt a cross entropy loss algorithm.

[0165] The label prediction function may be a mathematical model for processing a comprehensive feature vector of a document and predicting a corresponding rule label. The label prediction function may adopt a random forest model.

[0166] Specifically, through formula (4) , use the label prediction function to call the corresponding random forest model to perform multi-decision tree integration processing on the comprehensive feature vector of the document under the current target optimization parameters, so as to obtain the rule labels corresponding to the current comprehensive feature vector of the document, and then use the loss function to measure the matching loss degree between each rule label and the rule label set. By inputting the comprehensive feature vectors of the documents corresponding to different target channel merchants into formula (4), the matching loss degree is minimized and optimized, and finally the channel merchant rule analysis model is trained, so that the channel merchant rule analysis model can automatically parse the channel merchant rule documents, and extract the rule content corresponding to different rule labels in the channel merchant rule documents, forming structured rule data, and providing the channel merchant rule basis for subsequent flow verification.

[0167] Through this solution, mathematical analysis methods are used, according to the model training data set, and the channel merchant rule analysis model training process is described in mathematical language. The document comprehensive feature vector under the current target optimization parameters is processed using the label prediction function to obtain the rule labels corresponding to the current document comprehensive feature vector, and then the matching loss degree between each rule label and the rule label set is measured through the loss function. By processing the document comprehensive feature vectors corresponding to different target channel merchants, the matching loss degree is minimized, and finally the channel merchant rule analysis model is trained to enable the channel merchant rule analysis model to automatically parse the channel merchant rule documents, and extract the rule content corresponding to different rule labels in the channel merchant rule documents to form structured rule data, which provides a basis for channel merchant rules for subsequent pipeline verification.

[0168] In some embodiments, the sliding window time span is determined based on the time period to be verified; based on the sliding window time span, the time series structured flow data set is analyzed to determine the window flow sum corresponding to the sliding time window span; based on the reconciliation data, the window flow sum is analyzed, and the payment interval is determined based on the difference between the window flow sum and the corresponding payment amount in the reconciliation data; based on the reconciliation data, the time series structured flow data set and the corresponding payment interval are checked according to the time series verification rule set, and a reconciliation report is determined based on the checking result.

[0169] The sliding window time span may be the time span between the start time and the end time of the sliding window used to perform time segmentation processing on a time series structured pipeline data set.

[0170] The window transaction sum can be the sum of the transaction amount data corresponding to the time period of the sliding window in the time series structured transaction data set.

[0171] The payment range may be an amount range used to measure whether the payment is effective.

[0172] Specifically, the time period to be verified reflects the overall time span of all the flow data in the current flow data set to be verified. The sliding window time span needs to be adjusted according to the time period to be verified. The macro rule is: the longer the time period to be verified, the longer the corresponding sliding window time span should be. The specific rules can be: divide the sliding window time span into three categories, short-term span (limited to one week, span of 7 days), medium-term span (limited to one month, span of 28 days) and long-term span (limited to one quarter, span of 84 days). If the time period to be verified is less than one month, the sliding window time span adopts a short-term span. If the time period to be verified exceeds one month and is less than one quarter, the sliding window time span adopts a medium-term span. If the time period to be verified exceeds one quarter, the sliding window time span adopts a long-term span, so that the sliding window time span can flexibly adapt to the flow data corresponding to different time periods to be verified.

[0173] After determining the time span of the sliding window, based on the time span, extract the flow data within the corresponding time period of the time series structured flow data set, count the total window flow within the corresponding time period, and measure the difference between the total window flow and the corresponding payment amount in the reconciliation data through a preset threshold. Determine the payment interval based on the amount span of the difference between the total window flow and the corresponding payment amount within the preset threshold. Then, using the various transaction rules in the time series verification rule set as the basis for reconciliation, check and match the flow data, payment interval and amount change data in the time series structured flow data set one by one, and mark the verification result of each flow data (correct verification or abnormal verification), and generate a corresponding reconciliation report after integrating the verification results.

[0174] Through this solution, the corresponding sliding window time span is set according to the time period to be verified, so as to extract the total window flow corresponding to the sliding time window span in the time-series structured flow data set, and determine the payment interval based on the difference between the window flow total and the corresponding payment amount in the reconciliation data. On this basis, the verification results between the time-series verification rule set, the time-series structured flow data set and the corresponding payment interval are integrated to determine the reconciliation report.

[0175] In some embodiments, the reconciliation report is analyzed, abnormal transaction data is extracted, and the abnormal transaction data is pushed to the manual review process, and the manual review feedback provided by the manual review process is obtained; based on the reconciliation report and the manual review feedback, bank transaction verification summary information is generated, and the bank transaction verification summary information is provided to the user.

[0176] The abnormal transaction data may be transaction data marked as verification abnormality.

[0177] Specifically, when there are abnormal verification data in the reconciliation report, the corresponding abnormal transaction data will be pushed to the manual review department, and the relevant reviewers will manually review the abnormal transaction data according to the pre-established manual review process, and receive manual review feedback from the manual review department on the abnormal transaction data. Combined with the reconciliation report, the final bank transaction verification summary information is generated and provided to users through human-computer interaction devices, such as high-definition display screens.

[0178] Through this solution, the abnormal transaction data in the reconciliation report will be pushed to the manual review process, and the manual review feedback given by the manual review department on the abnormal transaction data will be received. Combined with the reconciliation report, the final bank transaction verification summary information will be generated to further ensure the comprehensiveness and accuracy of the transaction data verification.

[0179] Figure 3 A schematic diagram of a bank statement verification system based on OCR recognition technology is provided in one embodiment of the present application. Figure 3 As shown, a bank transaction verification system 300 based on OCR recognition technology in this embodiment includes: a transaction analysis module 301, a channel analysis module 302, a verification module 303 and an output module 304.

[0180] The pipeline analysis module 301 is used to obtain a pipeline data set to be verified, analyze the pipeline data set to be verified, and determine a time-series structured pipeline data set and a number of target channel merchants;

[0181] The channel analysis module 302 is used to obtain a channel merchant comprehensive document data set and a target channel merchant document data set according to the target channel merchants, and analyze the target channel merchant document data set based on the channel merchant comprehensive document data set to determine a time series verification rule set;

[0182] A verification module 303 is used to obtain reconciliation data, and based on the time series verification rule set and the reconciliation data, perform a sliding time window reconciliation logic calculation on the time series structured pipeline data set to determine a reconciliation report;

[0183] The output module 304 is used to determine and output the bank statement verification summary information according to the reconciliation report.

[0184] Optionally, the pipeline analysis module 301 is specifically used for:

[0185] Analyze the pipeline data set to be verified, determine the time period to be verified, and thereby determine the start time node and the end time node;

[0186] Based on the light character recognition algorithm, the transaction data set to be verified is analyzed, and the timestamp information, payment channel information and transaction change data set corresponding to each transaction data are extracted to construct a valid information set;

[0187] Determine a plurality of target channel merchants according to the plurality of payment channel information corresponding to the valid information set;

[0188] Based on the valid information set, the time-series structured pipeline data set is constructed according to the start time node and the end time node.

[0189] Optionally, when the pipeline analysis module 301 constructs the time-series structured pipeline data set based on the valid information set, according to the start time node and the end time node, the specific formula is as follows:

[0190] ;

[0191] in, is the time-series structured pipeline dataset, is the valid information set, For the The timestamp information of the flow data, For the The payment channel information of the transaction data, For the The pipeline change data set of pipeline data, is the ray character recognition function, is the current flow data, is the pipeline data set to be verified, is the starting time node, is the end time node.

[0192] Optionally, the channel analysis module 302 is specifically used to:

[0193] According to preset text preprocessing rules, the channel business comprehensive document data set is cleaned and segmented to determine a preprocessed comprehensive document data set;

[0194] Based on the preprocessed comprehensive document data set, perform word segmentation and vectorization processing on the keywords in the preprocessed comprehensive document data set to determine a document word vector data set corresponding to each target channel business;

[0195] Based on a preset channel business rule library, the document word vector data set corresponding to each target channel business is analyzed to determine a model training data set;

[0196] Constructing a channel business rule analysis model based on the model training data set;

[0197] Analyze the target channel business document data set according to the channel business rule analysis model and the preset channel business rule base to determine the rule set to be updated;

[0198] The timing verification rule set is determined according to the rule set to be updated and the preset channel provider rule base.

[0199] Optionally, when the channel analysis module 302 determines the time series verification rule set according to the rule set to be updated and the preset channel business rule base, it is specifically the following formula:

[0200] ;

[0201] in, is the temporal verification rule set, is the preset channel provider rule library, is the rule set to be updated.

[0202] Optionally, the channel analysis module 302 analyzes the document word vector data set corresponding to each target channel business based on a preset channel business rule base to determine the model training data set, specifically using the following formula:

[0203]

[0204] in, A training dataset for the model, For the The document word vector dataset corresponding to the target channel providers, For the The document comprehensive feature vector corresponding to the target channel business, is the vector fusion function, is the entity recognition function, is the entity vector set corresponding to the current document word vector dataset, is the relation extraction function, is the entity relationship vector set corresponding to the current document word vector dataset, For the A set of rule tags corresponding to each target channel partner in the preset channel partner rule library.

[0205] Optionally, the channel analysis module 302 constructs a channel business rule analysis model according to the model training data set, which is specifically the following formula:

[0206] ;

[0207] in, The target optimization parameters corresponding to the channel business rule analysis model, For the The comprehensive feature vector of the document corresponding to the target channel business, For the The rule tag set corresponding to each target channel partner in the preset channel partner rule library, A training dataset for the model, is the loss function, For the A set of rule labels corresponding to a target channel partner in the preset channel partner rule library, is the label prediction function.

[0208] Optionally, the checking module 303 is specifically used for:

[0209] Determine the time span of the sliding window according to the time period to be verified;

[0210] Analyze the time series structured pipeline data set according to the sliding window time span, and determine the window pipeline sum corresponding to the sliding time window span;

[0211] Based on the reconciliation data, the window transaction sum is analyzed, and the payment interval is determined according to the difference between the window transaction sum and the corresponding payment amount in the reconciliation data;

[0212] Based on the reconciliation data, the time-series structured transaction data set and the corresponding payment interval are checked according to the time-series verification rule set, and a reconciliation report is determined based on the checking result.

[0213] Optionally, the output module 304 is specifically used to:

[0214] Analyze the reconciliation report, extract abnormal transaction data, and push the abnormal transaction data to the manual review process, and obtain manual review feedback provided by the manual review process;

[0215] The bank statement verification summary information is generated based on the reconciliation report and the manual review feedback, and the bank statement verification summary information is provided to the user.

[0216] The system of this embodiment can be used to execute the method of any of the above embodiments. The implementation principles and technical effects are similar and will not be described in detail here.

Claims

1. A bank statement verification method based on OCR recognition technology, characterized in that: include: Obtaining a pipeline data set to be verified, analyzing the pipeline data set to be verified, and determining a time-series structured pipeline data set and a number of target channel vendors; According to the target channel merchants, a channel merchant comprehensive document dataset and a target channel merchant document dataset are obtained, and based on the channel merchant comprehensive document dataset, the target channel merchant document dataset is analyzed to determine a time series verification rule set; Acquire reconciliation data, and perform a sliding time window reconciliation logic calculation on the time series structured pipeline data set based on the time series verification rule set and the reconciliation data to determine a reconciliation report; Determine and output bank statement verification summary information based on the reconciliation report; The step of analyzing the target channel vendor document dataset based on the channel vendor comprehensive document dataset to determine the time series verification rule set includes: According to preset text preprocessing rules, the channel business comprehensive document data set is cleaned and segmented to determine a preprocessed comprehensive document data set; Based on the preprocessed comprehensive document data set, perform word segmentation and vectorization processing on the keywords in the preprocessed comprehensive document data set to determine a document word vector data set corresponding to each target channel business; Based on a preset channel business rule library, the document word vector data set corresponding to each target channel business is analyzed to determine a model training data set; Constructing a channel business rule analysis model based on the model training data set; Analyze the target channel business document data set according to the channel business rule analysis model and the preset channel business rule base to determine the rule set to be updated; The timing verification rule set is determined according to the rule set to be updated and the preset channel provider rule base.

2. The method according to claim 1, characterized in that The analyzing the pipeline data set to be verified to determine the time-series structured pipeline data set and several target channel vendors includes: Analyze the pipeline data set to be verified, determine the time period to be verified, and thereby determine the start time node and the end time node; Based on the light character recognition algorithm, the transaction data set to be verified is analyzed, and the timestamp information, payment channel information and transaction change data set corresponding to each transaction data are extracted to construct a valid information set; Determine a plurality of target channel merchants according to the plurality of payment channel information corresponding to the valid information set; Based on the valid information set, the time-series structured pipeline data set is constructed according to the start time node and the end time node.

3. The method according to claim 2, characterized in that Based on the valid information set, the time series structured pipeline data set is constructed according to the start time node and the end time node, specifically as follows: ； in, is the time series structured pipeline dataset, is the valid information set, For the The timestamp information of the flow data, For the The payment channel information of the transaction data, For the The pipeline change data set of pipeline data, is the ray character recognition function, is the current flow data, is the pipeline data set to be verified, is the starting time node, is the end time node.

4. The method according to claim 1, characterized in that The timing verification rule set is determined according to the rule set to be updated and the preset channel business rule base, specifically the following formula: ； in, is the temporal verification rule set, is the preset channel provider rule library, is the rule set to be updated.

5. The method according to claim 2, characterized in that: Based on the preset channel business rule base, the document word vector data set corresponding to each target channel business is analyzed to determine the model training data set, which is specifically the following formula: in, A training dataset for the model, For the The document word vector dataset corresponding to the target channel business, For the The document comprehensive feature vector corresponding to the target channel business, is the vector fusion function, is the entity recognition function, is the entity vector set corresponding to the current document word vector dataset, is the relation extraction function, is the entity relationship vector set corresponding to the current document word vector dataset, For the A set of rule tags corresponding to each target channel partner in the preset channel partner rule library.

6. The method according to claim 5, characterized in that The channel business rule analysis model is constructed according to the model training data set, which is specifically the following formula: ； in, The target optimization parameters corresponding to the channel business rule analysis model, For the The comprehensive feature vector of the document corresponding to the target channel business, For the The rule tag set corresponding to each target channel partner in the preset channel partner rule library, A training dataset for the model, is the loss function, For the A set of rule labels corresponding to a target channel partner in the preset channel partner rule library, is the label prediction function.

7. The method according to claim 6, characterized in that The step of performing a sliding time window reconciliation logic calculation on the time series structured flow data set based on the time series verification rule set and the reconciliation data to determine a reconciliation report includes: Determine the time span of the sliding window according to the time period to be verified; Analyze the time series structured pipeline data set according to the sliding window time span, and determine the window pipeline sum corresponding to the sliding time window span; Based on the reconciliation data, the window transaction sum is analyzed, and the payment interval is determined according to the difference between the window transaction sum and the corresponding payment amount in the reconciliation data; Based on the reconciliation data, the time-series structured transaction data set and the corresponding payment interval are checked according to the time-series verification rule set, and a reconciliation report is determined based on the checking result.

8. The method according to claim 7, characterized in that Determining and outputting bank statement verification summary information based on the reconciliation report includes: Analyze the reconciliation report, extract abnormal transaction data, push the abnormal transaction data to the manual review process, and obtain manual review feedback provided by the manual review process; The bank statement verification summary information is generated based on the reconciliation report and the manual review feedback, and the bank statement verification summary information is provided to the user.

9. A bank statement verification system based on OCR recognition technology, characterized in that: The method as claimed in any one of claims 1 to 8 comprises: A pipeline analysis module is used to obtain a pipeline data set to be verified, analyze the pipeline data set to be verified, and determine a time-series structured pipeline data set and a number of target channel vendors; A channel analysis module, for obtaining a channel merchant comprehensive document data set and a target channel merchant document data set according to a number of the target channel merchants, and analyzing the target channel merchant document data set based on the channel merchant comprehensive document data set to determine a time series verification rule set; A checking module, used to obtain reconciliation data, and based on the time series verification rule set and the reconciliation data, perform a sliding time window reconciliation logic calculation on the time series structured pipeline data set to determine a reconciliation report; The output module is used to determine and output the bank statement verification summary information based on the reconciliation report.

Images