Account checking method and device based on deep learning model, equipment and medium

Abstract

The invention relates to the technical field of computers, and discloses an account checking method and device based on a deep learning model, equipment and a medium, and the method comprises the steps: extracting original data from a financial system to a big data platform, and constructing a data basic set in the big data platform; performing structure mapping on the data basic set according to financial dimensions to obtain a data basic set after structure mapping; extracting a field combination from the data basic set after the structure mapping to construct a feature vector, and inputting the feature vector into a pre-trained large model to obtain a deep learning model; identifying target financial data through the deep learning model to obtain an identification result; wherein the identification result is used for representing a difference type, a deviation degree and an abnormal label of the target financial data and the original data. The account checking method can be applied to a financial science and technology business management program system, and the account checking method which is more intelligent, real-time and more accurate can be achieved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a reconciliation method, apparatus, device, and medium based on a deep learning model. Background Art

[0002] In the current context of highly digitalized financial industry development, the volume of financial data is growing exponentially, with monthly transaction data often increasing by billions. With the increasing complexity of financial business types and the frequent cross-system data interaction, reconciliation, a core component of ensuring financial data consistency and compliance, faces multiple challenges, including heterogeneous data sources, inconsistent structures, and lengthy processing links.

[0003] Reconciliation methods in related technologies often rely on manual verification or static comparison based on rule engines, which makes it difficult to effectively handle the problems of abnormal pattern identification and matching high-dimensional data. In particular, in high-frequency trading, cross-currency settlement, and large-value payment scenarios, problems such as account discrepancies and difficulty in locating differences often arise, seriously affecting data accuracy and risk control efficiency. Therefore, it is necessary to propose a reconciliation method that can automatically identify complex difference characteristics and potential abnormal patterns in large-scale data, improve the intelligence level of reconciliation, and adapt to the real-world needs of large-scale financial data, strong real-time performance, and complex business logic. Summary of the Invention

[0004] The present invention provides a reconciliation method, device, computer equipment and medium based on a deep learning model to solve the technical problems in the related art that the reconciliation method is not intelligent enough, has poor real-time performance and poor accuracy.

[0005] First, a reconciliation method based on a deep learning model is provided, including:

[0006] Extracting raw data from the financial system to the big data platform, and building a data foundation set in the big data platform based on the raw data;

[0007] Performing structural mapping on the data foundation set according to financial dimensions to obtain a structurally mapped data foundation set; wherein the financial dimensions include an account field, a currency field, and a business date field;

[0008] Extracting field combinations from the data base after the structure mapping to construct feature vectors, and inputting the feature vectors into a pre-trained large model to obtain a deep learning model;

[0009] The target financial data is identified through the deep learning model to obtain an identification result; wherein the identification result is used to characterize the difference type, deviation degree and abnormal label between the target financial data and the original data.

[0010] In a second aspect, a reconciliation device based on a deep learning model is provided, comprising:

[0011] A construction module is used to extract raw data from the financial system to the big data platform, and to construct a data foundation set in the big data platform based on the raw data;

[0012] a structure mapping module, configured to perform structure mapping on the data foundation set according to financial dimensions to obtain a structure-mapped data foundation set; wherein the financial dimensions include an account field, a currency field, and a business date field;

[0013] An input module is used to extract field combinations from the data base set after the structure mapping to construct a feature vector, and input the feature vector into the pre-trained large model to obtain a deep learning model;

[0014] A reconciliation module is used to identify the target financial data through the deep learning model to obtain an identification result; wherein the identification result is used to characterize the difference type, degree of deviation and abnormal label between the target financial data and the original data.

[0015] In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned reconciliation method based on the deep learning model are implemented.

[0016] In a fourth aspect, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the deep learning model-based reconciliation method. In the solution implemented by the deep learning model-based reconciliation method, apparatus, computer device, and storage medium, raw data can be extracted from a financial system and stored in a big data platform. A data foundation set is constructed in the big data platform based on the raw data, and the data foundation set is structured according to financial dimensions to obtain a structured data foundation set. The financial dimensions include account fields, currency fields, and transaction date fields. Furthermore, a feature vector can be constructed by extracting field combinations from the structured data foundation set. The feature vector is then input into a pre-trained big model to obtain a deep learning model. The deep learning model can then be used to identify target financial data and obtain a recognition result. The recognition result is used to characterize the type of difference, degree of deviation, and anomaly label between the target financial data and the raw data. In this invention, automatic identification and intelligent reconciliation of financial data can be achieved. By extracting raw data from a financial system and constructing a structured data foundation set in the big data platform, the standardization and consistency of data processing are improved. Further combining this with a pre-trained large model to build a deep learning model enables efficient recognition and anomaly detection of large-scale, multi-dimensional financial data, significantly improving reconciliation efficiency and accuracy while reducing manual verification costs and human error. Furthermore, recognition results include the type of discrepancy, degree of deviation, and anomaly labels, facilitating rapid identification of the source of issues and enhancing risk early warning capabilities and internal control. This approach is applicable to financial institutions such as banks, insurance companies, and securities companies, and is particularly well-suited for handling high-frequency trading and multi-currency settlement reconciliation tasks in complex business scenarios, demonstrating excellent generalizability and scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0018] Figure 1 2 is a schematic diagram of an application environment of a reconciliation method based on a deep learning model in one embodiment of the present invention;

[0019] Figure 2 1 is a flow chart of a reconciliation method based on a deep learning model in one embodiment of the present invention;

[0020] Figure 3 yes Figure 1 A schematic flow chart of a specific implementation of step S10;

[0021] Figure 4 2 is a schematic structural diagram of a reconciliation device based on a deep learning model in one embodiment of the present invention;

[0022] Figure 5 is a structural diagram of a computer device in one embodiment of the present invention;

[0023] Figure 6 FIG. 2 is another structural diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0025] The reconciliation method based on deep learning model provided by the embodiment of the present invention can be applied in Figure 1In an application environment, wherein the client communicates with the server through a network. The server can extract raw data from the financial system to the big data platform through the client, and construct a data foundation set in the big data platform based on the raw data; structurally map the data foundation set according to the financial dimension to obtain a data foundation set after structural mapping; wherein the financial dimension includes an account field, a currency field, and a business date field; extract a field combination from the data foundation set after structural mapping to construct a feature vector, and input the feature vector into a pre-trained big model to obtain a deep learning model; identify the target financial data through the deep learning model to obtain a recognition result; wherein the recognition result is used to characterize the difference type, degree of deviation, and abnormal label between the target financial data and the raw data, and finally feed the recognition result back to the client. In the present invention, automatic recognition and intelligent reconciliation of financial data can be realized, and the standardization and consistency of data processing are improved by extracting raw data from the financial system and constructing a structured data foundation set in the big data platform. Further combining the pre-trained large model to build a deep learning model can achieve efficient recognition and anomaly detection of large-scale, multi-dimensional financial data, thereby greatly improving the efficiency and accuracy of reconciliation, and reducing manual verification costs and human errors. In addition, the recognition results include the type of difference, degree of deviation and anomaly label, which facilitates the rapid location of the source of the problem and improves risk warning capabilities and internal control levels. This method is suitable for financial institutions such as banks, insurance, and securities, and is particularly suitable for handling high-frequency transactions and multi-currency settlement reconciliation tasks in complex business scenarios, with good promotion and scalability. Among them, the client can be but is not limited to various personal computers, laptops, smart phones, tablets and portable wearable devices. The server can be implemented with an independent server or a server cluster consisting of multiple servers. The present invention is described in detail below through specific embodiments.

[0026] See also Figure 2 As shown, Figure 2 A flowchart of a reconciliation method based on a deep learning model provided in an embodiment of the present invention includes the following steps:

[0027] S10: Extracting original data from the financial system to the big data platform, and constructing a data foundation set in the big data platform based on the original data.

[0028] For example, raw data can be extracted from the financial system. It should be noted that the financial system stores a large amount of structured financial data generated in the daily operation of the enterprise, such as accounting vouchers, account details, report data, etc. These data contain multiple fields such as subjects, currencies, amounts, business dates, etc. In order to conduct subsequent data analysis and modeling, these raw data can be efficiently and securely migrated to the big data platform. For example, data can be extracted from multiple financial subsystems (such as general ledgers, accounts receivable and payable, and expense reimbursement systems), and then format cleaning, field mapping, and unified coding processing are performed. Finally, the standardized raw data is batch loaded into the big data platform to ensure data consistency, integrity, and traceability.

[0029] Furthermore, after the original data is successfully imported into the big data platform, a data foundation set can be further constructed based on this data, thereby providing a unified data input source for subsequent data modeling and intelligent reconciliation tasks. The construction process includes filtering, aggregating, and reorganizing data according to financial processing logic; classifying data according to business dimensions (such as accounting subjects, transaction types, time periods, etc.); and standardizing key fields (such as unified currency units, time formats), etc. This application does not limit this. In addition, auxiliary fields can be added according to business needs for the construction of subsequent model input features.

[0030] By building a data foundation set, we not only improve the efficiency and accuracy of data analysis, but also lay a solid data foundation for subsequent deep learning model training, ensuring that the model can learn and reason on high-quality, structured data.

[0031] Among them, such as Figure 3 As shown, step S10, that is, extracting the original data from the financial system to the big data platform and constructing a data foundation set in the big data platform based on the original data, includes the following steps:

[0032] S11: Acquire original data in the financial system.

[0033] The original data includes voucher details and original transaction flow records.

[0034] S12: The original data is transferred to the original data area of ​​the big data platform through a data extraction tool, and a cleaning operation is performed on the original data in the original data area to obtain cleaned original data.

[0035] S13: constructing the data basic set based on the cleaned original data.

[0036] For steps S11-S13, the original data can first be obtained from the financial system. This data includes two types of voucher details, namely the accounting voucher data of the company's daily economic operations, such as the voucher number, accounting subject, debit and credit direction, amount, business date, department and employee information, etc. It can also include original transaction flow records, such as bank receipts, POS transaction records, e-commerce platform settlement records, etc.

[0037] Due to the complex structure of these data, the dispersed data sources, the inconsistent formats, and the varying quality, data extraction tools (such as ETL tools or distributed data acquisition frameworks) can be used to transfer raw data from the financial system to the raw data area of ​​the big data platform. In this area, the transferred raw data can be cleansed, including removing redundant fields, repairing missing data, unifying formats (such as currency and time formats), identifying and removing duplicate records, and marking or repairing outliers or dirty data, thereby improving data consistency, integrity, and availability.

[0038] Furthermore, after data cleaning is completed, the original financial information can be converted into a structured, standardized analytical data set as the basis for subsequent model training and reconciliation analysis. The construction process can include classifying and organizing the cleaned data according to business rules, such as aggregating and reorganizing data by dimensions such as accounting subjects, transaction types, counterparties, and time periods; the original fields can also be converted and expanded in combination with external reference data (such as exchange rate tables and subject mapping rules). In addition, a unified data format and encoding standard can be set for the foundation set to facilitate subsequent feature extraction and model input. After construction is completed, the data foundation set has good business readability and algorithm adaptability, and can directly support subsequent structure mapping, feature vector construction, and deep learning modeling, thereby realizing an efficient and automated intelligent reconciliation process.

[0039] S20: Structural mapping is performed on the data foundation set according to the financial dimension to obtain a structurally mapped data foundation set.

[0040] The financial dimension includes an account field, a currency field, and a business date field.

[0041] For example, the data foundation set may be structurally mapped based on financial dimensions, that is, the original data foundation set may be reorganized and structured according to specific business perspectives and analysis requirements.

[0042] Specifically, financial dimensions include account fields (representing specific accounting items such as revenue, costs, and expenses), currency fields (identifying the currency in which transactions occurred, facilitating aggregation and conversion), and transaction date fields (reflecting the actual time of transactions, supporting time-based reconciliation analysis). By mapping the data structure through these dimensions, complex, multi-source financial data can be organized into a structured data model with unified logical relationships. This provides a clear and organized data foundation for subsequent feature extraction and model analysis, helping to improve data alignment accuracy and model recognition effectiveness.

[0043] In some embodiments, the structural mapping of the data basic set according to the financial dimension to obtain the structurally mapped data basic set includes: converting the original fields in the basic data set with the fields in the financial dimension according to a preset field mapping relationship to obtain the converted data basic set; and sorting the converted data basic set to obtain the structurally mapped data basic set.

[0044] For example, the original fields in the data foundation set can be converted to corresponding financial dimension fields based on preset field mapping relationships. In other words, field mapping rules can be predefined to map non-standard fields in the original data (such as internal system names, interface return names, etc.) to standard financial dimension fields (such as accounting subjects, currencies, and transaction dates). This conversion process not only unifies the data naming conventions and logical caliber, but also ensures that data from different sources can be compared and integrated on the same analytical dimensions, forming a converted data foundation set with business semantics.

[0045] Furthermore, the converted data foundation set can be sorted according to specific sorting rules to form the final data foundation set after structural mapping. Sorting is typically performed based on fields relevant to business logic, such as prioritizing ascending order by transaction date, followed by currency and accounting subject fields. This enhances data readability and comparability, facilitating subsequent feature extraction, difference identification, and model input. Furthermore, the ordered data structure helps improve the computational efficiency and accuracy of deep learning models.

[0046] S30: extracting field combinations from the data base set after the structure mapping to construct feature vectors, and inputting the feature vectors into the pre-trained large model to obtain a deep learning model.

[0047] In some embodiments, the extracting of field combinations from the data base set after the structure mapping to construct a feature vector, and inputting the feature vector into the pre-trained large model to obtain a deep learning model, includes: extracting a target field combination from the data after the structure mapping, the target field combination including at least two combinations of the subject field, amount field, business type, and transaction time; performing a data cleaning operation on the target field combination to obtain a cleaned field combination, wherein the data cleaning operation includes a numerical normalization operation, a category number, and a missing value filling operation; converting the cleaned field combination into the feature vector, and inputting the feature vector into the pre-trained large model to obtain the deep learning model.

[0048] For example, the first paragraph: Extraction of target field combinations and their significance

[0049] In step S30, a target field combination can be extracted from the data base set after structural mapping to build a deep learning model. The target field combination is selected based on the financial business characteristics and model recognition requirements, including but not limited to at least two of the accounting subject field, amount field, business type field and transaction time field. These fields represent the financial nature of the transaction (such as income, expenditure), numerical characteristics (such as amount scale), business background (such as procurement, reimbursement, sales) and time dimension (such as day / month / quarter). By combining these fields together, the multi-dimensional characteristics of a financial record can be fully reflected, which helps the model capture complex business patterns and potential abnormal behaviors.

[0050] After extracting the target field combination, data cleaning operations can be performed on it to ensure the quality and consistency of the model input data. The first is the numerical normalization operation, which uniformly maps numerical fields with orders of magnitude differences, such as amounts, to a standard range (such as 0 to 1) to prevent the model from being biased towards high-value features during training. The second is the category number (encoding) operation, which converts text fields such as business types or accounting subjects into numerical labels or one-hot encodings so that they can be processed by the neural network. Finally, the missing value filling operation is the reasonable filling of missing fields (such as mean filling, forward filling, or using special marker bits) to prevent model training interruptions or misjudgments. These cleaning steps are the core of feature engineering, ensuring that the constructed data is unified in form, complete in semantics, and meets the conditions for input into the model.

[0051] After cleaning, the processed field combinations can be converted into one-dimensional numerical feature vectors, which serve as the model input format. Feature vectors encode multi-field, multi-type financial information into numerical matrices that a neural network can understand, enabling the deep learning model to learn underlying correlation patterns between fields. This feature vector can then be input into a pre-trained large model, typically a deep neural network pre-trained on massive amounts of financial data, with strong sequence learning and anomaly recognition capabilities. During further training, the model gradually adjusts its weight parameters to fit the current enterprise's financial data characteristics, ultimately forming a customized deep learning model capable of accurately identifying data anomalies, discrepancy types, and deviation levels, providing intelligent judgment support for subsequent reconciliation.

[0052] S40: Identify the target financial data through the deep learning model to obtain an identification result.

[0053] The recognition result is used to characterize the difference type, deviation degree and abnormal label between the target financial data and the original data.

[0054] For example, a trained deep learning model can be used to automatically identify and analyze target financial data, generating corresponding recognition results. This recognition process, based on the model's understanding of feature vectors, can determine whether the target financial data differs from the original data and further refine the recognition results, including: the type of difference (such as inconsistent amounts, incorrect currencies, mismatched business types, etc.), the degree of deviation (such as deviation values ​​or percentages to quantify the degree of error), and whether there are abnormal labels (such as those marked as high risk and requiring manual review, etc.).

[0055] Through the above identification results, not only can errors or anomalies in accounting data be automatically discovered, but the specific reasons for the differences and risk levels can also be provided, greatly improving the intelligence, refinement and efficiency of financial reconciliation.

[0056] In some embodiments, identifying the target financial data through the deep learning model to obtain the identification result includes: constructing a corresponding target feature vector based on the target financial data; inputting the target feature vector into the deep learning model to obtain the identification result.

[0057] For example, a target feature vector can be constructed based on the fields contained in the target financial data (such as account, amount, business type, transaction time, etc.), ensuring that the vector is consistent with the feature structure used during training. Subsequently, the target feature vector is input into the trained deep learning model. The model judges and classifies the input data based on its learned patterns and parameters, and ultimately outputs a recognition result. This recognition result is used to accurately characterize whether the target data has anomalies, the type of differences, and the degree of deviation, thereby enabling intelligent judgment and auxiliary decision support for specific data during the reconciliation process.

[0058] In some embodiments, the method further includes: obtaining labeled historical reconciliation data as a training sample data set; constructing a corresponding feature vector for each training sample set as a model input, inputting the model input into the pre-trained large model, and using supervised learning to perform model fitting to obtain the deep learning model.

[0059] For example, to obtain a trained deep learning model, a training set is first constructed from historical reconciliation data. The data contained in the training set is used to learn and fit the model parameters. Subsequently, the training set is input into the deep learning model, and the model is trained using supervised learning. Specifically, given known labels, the goal is to minimize the error between the predicted output and the true value. After multiple rounds of training and validation, a deep learning model is obtained.

[0060] Based on the above embodiment, after obtaining the deep learning model, it also includes: iteratively training the deep learning model based on the training sample data set to extract data features, and calculating the loss function; using a preset method to iteratively train the loss function for the purpose of reducing the loss function value until the loss function value is less than a preset threshold; based on the loss function after iterative training, obtaining the iterative deep learning model.

[0061] It is understandable that in order to train a deep learning model with higher accuracy, the deep learning model can be repeatedly trained iteratively to continuously reduce the loss function until the loss function meets the expected threshold requirement.

[0062] It should be noted that this application does not limit the above-mentioned preset method and expected threshold. For example, the preset method can be a gradient descent algorithm, a batch gradient descent algorithm, a stochastic gradient descent algorithm, etc. This application takes the gradient descent algorithm as an example for explanation.

[0063] The purpose of the gradient descent algorithm is to find the minimum value of the loss function or converge to the minimum value through iteration. In a geometric sense, the gradient descent algorithm is that the gradient decreases fastest in the direction opposite to the vector where the function changes the fastest, making it easier to find the minimum value of the function. Based on this, in an embodiment of the present application, the gradient descent algorithm can be used to repeatedly iteratively train the deep learning model so that the loss function is continuously reduced, thereby reducing the error of the calculation result.

[0064] In an embodiment of the present application, the gradient descent algorithm is used to repeatedly iteratively train the deep learning model so that the loss function is continuously reduced to obtain the iterative deep learning model, and then a more accurate analysis result can be obtained based on the iterative deep learning model.

[0065] It can be seen that the above solution can realize automatic recognition and intelligent reconciliation of financial data. By extracting raw data from the financial system and building a structured data foundation set in the big data platform, the standardization and consistency of data processing are improved. Further combining the pre-trained large model to build a deep learning model can achieve efficient recognition and anomaly detection of large-scale, multi-dimensional financial data, thereby greatly improving reconciliation efficiency and accuracy, and reducing manual verification costs and human errors. In addition, the recognition results include the type of difference, degree of deviation and anomaly labels, which facilitates the rapid location of the source of the problem and improves risk warning capabilities and internal control levels. This method is suitable for financial institutions such as banks, insurance companies, and securities companies. It is particularly suitable for handling high-frequency transactions and multi-currency settlement reconciliation tasks in complex business scenarios, and has good generalizability and scalability.

[0066] It should be understood that the order of execution of the steps in the above embodiments does not necessarily mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0067] In one embodiment, a reconciliation device based on a deep learning model is provided, and the reconciliation device based on a deep learning model corresponds one-to-one to the reconciliation method based on a deep learning model in the above embodiment. Figure 4 As shown, the reconciliation device based on the deep learning model includes a construction module 101, a structure mapping module 102, an input module 103, and a reconciliation module 104. The functional modules are described in detail as follows:

[0068] Construction module 101 is used to extract original data from the financial system to the big data platform, and construct a data foundation set in the big data platform based on the original data;

[0069] A structure mapping module 102 is configured to perform structure mapping on the data foundation set according to financial dimensions to obtain a structure-mapped data foundation set; wherein the financial dimensions include an account field, a currency field, and a transaction date field;

[0070] An input module 103 is configured to extract field combinations from the structure-mapped data base set to construct feature vectors, and input the feature vectors into a pre-trained large model to obtain a deep learning model;

[0071] The reconciliation module 104 is used to identify the target financial data through the deep learning model to obtain an identification result; wherein the identification result is used to characterize the difference type, deviation degree and abnormal label between the target financial data and the original data.

[0072] In one embodiment, the construction module 101 is specifically configured to:

[0073] Acquiring original data from the financial system, wherein the original data includes voucher details and original transaction flow records;

[0074] The raw data is transferred to the raw data area of ​​the big data platform by a data extraction tool, and the raw data is cleaned in the raw data area to obtain cleaned raw data;

[0075] The data basic set is constructed based on the cleaned original data.

[0076] In one embodiment, the structure mapping module 102 is specifically configured to:

[0077] Convert the original fields in the basic data set to the fields in the financial dimension according to the preset field mapping relationship to obtain a converted data basic set;

[0078] The converted data basic set is sorted to obtain the structure-mapped data basic set.

[0079] In one embodiment, the input module 103 is further configured to:

[0080] Extracting a target field combination from the structure-mapped data, the target field combination including at least two combinations of the subject field, amount field, business type, and transaction time;

[0081] Performing a data cleaning operation on the target field combination to obtain a cleaned field combination, wherein the data cleaning operation includes a value normalization operation, a category numbering operation, and a missing value filling operation;

[0082] The cleaned field combination is converted into the feature vector, and the feature vector is input into the pre-trained large model to obtain the deep learning model.

[0083] In one embodiment, the construction module 101 is specifically configured to:

[0084] Constructing a corresponding target feature vector based on the target financial data;

[0085] The target feature vector is input into the deep learning model to obtain the recognition result.

[0086] In one embodiment, the construction module 101 is specifically configured to:

[0087] Obtain labeled historical reconciliation data as a training sample dataset;

[0088] A corresponding feature vector is constructed for each training sample set as a model input, the model input is input into the pre-trained large model, and the model is fitted using a supervised learning method to obtain the deep learning model.

[0089] In one embodiment, the construction module 101 is specifically configured to:

[0090] Iteratively training the deep learning model based on the training sample data set to extract data features and calculate the loss function;

[0091] Iteratively training the loss function using a preset method for the purpose of reducing the loss function value until the loss function value is less than a preset threshold;

[0092] Based on the loss function after iterative training, an iterative deep learning model is obtained.

[0093] The present invention provides a reconciliation device based on a deep learning model, which can realize automatic recognition and intelligent reconciliation of financial data. By extracting raw data from the financial system and constructing a structured data base set in the big data platform, the standardization and consistency of data processing are improved. Further combining the pre-trained large model to build a deep learning model can achieve efficient recognition and anomaly detection of large-scale, multi-dimensional financial data, thereby greatly improving reconciliation efficiency and accuracy, and reducing manual verification costs and human errors. In addition, the recognition results include the type of difference, degree of deviation and anomaly labels, which facilitates rapid location of the source of the problem and improves risk warning capabilities and internal control levels. This method is applicable to financial institutions such as banks, insurance companies, and securities companies, and is particularly suitable for handling high-frequency transactions and multi-currency settlement reconciliation tasks in complex business scenarios. It has good generalizability and scalability.

[0094] For the specific limitations of the deep learning model-based reconciliation device, please refer to the limitations of the deep learning model-based reconciliation method above and will not be repeated here. The various modules in the above-mentioned deep learning model-based reconciliation device can be implemented in whole or in part through software, hardware, or a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of the above modules.

[0095] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, memory, network interface and database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it implements the functions or steps on the service side of a reconciliation method based on a deep learning model.

[0096] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 6 As shown. The computer device includes a processor, memory, network interface, display screen and input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps on the client side of a reconciliation method based on a deep learning model.

[0097] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are performed:

[0098] Extracting raw data from the financial system to the big data platform, and building a data foundation set in the big data platform based on the raw data;

[0099] Performing structural mapping on the data foundation set according to financial dimensions to obtain a structurally mapped data foundation set; wherein the financial dimensions include an account field, a currency field, and a business date field;

[0100] Extracting field combinations from the data base after the structure mapping to construct feature vectors, and inputting the feature vectors into a pre-trained large model to obtain a deep learning model;

[0101] The target financial data is identified through the deep learning model to obtain an identification result; wherein the identification result is used to characterize the difference type, deviation degree and abnormal label between the target financial data and the original data.

[0102] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0103] Extracting raw data from the financial system to the big data platform, and building a data foundation set in the big data platform based on the raw data;

[0104] Performing structural mapping on the data foundation set according to financial dimensions to obtain a structurally mapped data foundation set; wherein the financial dimensions include an account field, a currency field, and a business date field;

[0105] Extracting field combinations from the data base after the structure mapping to construct feature vectors, and inputting the feature vectors into a pre-trained large model to obtain a deep learning model;

[0106] The target financial data is identified through the deep learning model to obtain an identification result; wherein the identification result is used to characterize the difference type, deviation degree and abnormal label between the target financial data and the original data.

[0107] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can be found in the relevant descriptions of the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0108] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchl ink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0109] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0110] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A reconciliation method based on a deep learning model, characterized in that: The method comprises: Extracting raw data from the financial system to the big data platform, and building a data foundation set in the big data platform based on the raw data; Performing structural mapping on the data foundation set according to financial dimensions to obtain a structurally mapped data foundation set; wherein the financial dimensions include an account field, a currency field, and a business date field; Extracting field combinations from the data base after the structure mapping to construct feature vectors, and inputting the feature vectors into a pre-trained large model to obtain a deep learning model; The target financial data is identified through the deep learning model to obtain an identification result; wherein the identification result is used to characterize the difference type, deviation degree and abnormal label between the target financial data and the original data.

2. The method according to claim 1, characterized in that Extracting raw data from the financial system to the big data platform and constructing a data foundation set in the big data platform based on the raw data includes: Acquiring original data from the financial system, wherein the original data includes voucher details and original transaction flow records; The raw data is transferred to the raw data area of ​​the big data platform by a data extraction tool, and the raw data is cleaned in the raw data area to obtain cleaned raw data; The data basic set is constructed based on the cleaned original data.

3. The method according to claim 1, characterized in that The step of performing structural mapping on the data foundation set according to the financial dimension to obtain the structurally mapped data foundation set includes: Convert the original fields in the basic data set to the fields in the financial dimension according to the preset field mapping relationship to obtain a converted data basic set; The converted data basic set is sorted to obtain the structure-mapped data basic set.

4. The method according to claim 1, wherein The extracting of fields from the data base set after the structure mapping and combining them to construct a feature vector, and inputting the feature vector into a pre-trained large model to obtain a deep learning model, includes: Extracting a target field combination from the structure-mapped data, the target field combination including at least two combinations of the subject field, amount field, business type, and transaction time; Performing a data cleaning operation on the target field combination to obtain a cleaned field combination, wherein the data cleaning operation includes a value normalization operation, a category numbering operation, and a missing value filling operation; The cleaned field combination is converted into the feature vector, and the feature vector is input into the pre-trained large model to obtain the deep learning model.

5. The method according to claim 1, characterized in that The method further comprises: Obtain labeled historical reconciliation data as a training sample dataset; A corresponding feature vector is constructed for each training sample set as a model input, the model input is input into the pre-trained large model, and the model is fitted using a supervised learning method to obtain the deep learning model.

6. The method according to claim 5, characterized in that After obtaining the deep learning model, the method further includes: Iteratively training the deep learning model based on the training sample data set to extract data features and calculate the loss function; Iteratively training the loss function using a preset method for the purpose of reducing the loss function value until the loss function value is less than a preset threshold; Based on the loss function after iterative training, an iterative deep learning model is obtained.

7. The method according to claim 1, characterized in that The target financial data is identified by the deep learning model to obtain an identification result, including: Constructing a corresponding target feature vector based on the target financial data; The target feature vector is input into the deep learning model to obtain the recognition result.

8. A reconciliation device based on a deep learning model, characterized in that: include: A construction module is used to extract raw data from the financial system to the big data platform, and to construct a data foundation set in the big data platform based on the raw data; a structure mapping module, configured to perform structure mapping on the data foundation set according to financial dimensions to obtain a structure-mapped data foundation set; wherein the financial dimensions include an account field, a currency field, and a business date field; An input module is used to extract field combinations from the data base set after the structure mapping to construct a feature vector, and input the feature vector into the pre-trained large model to obtain a deep learning model; A reconciliation module is used to identify the target financial data through the deep learning model to obtain an identification result; wherein the identification result is used to characterize the difference type, degree of deviation and abnormal label between the target financial data and the original data.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the reconciliation method based on the deep learning model as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the reconciliation method based on the deep learning model as described in any one of claims 1 to 7 are implemented.

Images