Accounting data management method, system, device and medium based on edge computing

Abstract

The application provides an accounting data management method, system, device and medium based on edge computing, which extracts a structured feature vector of newly added accounting voucher data; extracts a similar voucher set similar to the newly added accounting voucher data from a high-frequency subject area and a sparse subject area in historical accounting voucher data according to a data partition granularity; determines the distance between the structured feature vector of the newly added accounting voucher data and the feature vector of each accounting voucher in the similar voucher set, and then generates a risk confidence value of the newly added accounting voucher data based on the occurrence frequency of similar accounting vouchers and the feature vector distance; and when the risk confidence value exceeds a set threshold, updates the compliance verification result and the structured feature vector corresponding to the newly added accounting voucher data to an accounting voucher database. The technical scheme provided by the application can overcome the interference caused by the natural long-tail distribution characteristics of accounting subjects on similar voucher retrieval in accounting data management.

Description

Technical Field

[0001] This application relates to the field of data management technology, and more specifically, to an accounting data management method, system, device, and medium based on edge computing. Background Technology

[0002] With the sweeping wave of digitalization, data has become a core production factor. Its explosive growth has driven the continuous evolution of data management technology. Early file system and database technologies solved the problem of storing and querying structured data, but traditional architectures have gradually become inadequate in the face of the processing needs of massive amounts of unstructured data (such as videos and logs) and real-time streaming data (such as IoT sensor data). The popularization of cloud computing has given rise to distributed storage and computing frameworks, which have achieved elastic scaling and parallel processing. Edge computing pushes data processing to the terminal, reducing latency and saving bandwidth.

[0003] In existing data management, the core principle is full lifecycle control, which runs through the entire process of data generation, storage, processing, transmission, application, and destruction. Through classification and grading mechanisms, differentiated strategies are formulated based on data sensitivity and business value. At the same time, by establishing a data governance framework, clarifying the division of responsibilities, and combining automation tools to achieve process standardization, it is ensured that data serves business decisions under the premise of compliance. However, in the process of accounting data management, traditional edge-side accounting risk identification methods are difficult to overcome the interference caused by the natural long-tail distribution characteristics of accounting subjects to similar voucher retrieval. This leads to a decrease in the accuracy of risk assessment due to the uneven frequency of use of subjects in accounting edge nodes. Therefore, how to overcome the interference caused by the natural long-tail distribution characteristics of accounting subjects to similar voucher retrieval in accounting data management has become a difficult problem for the industry. Summary of the Invention

[0004] This application provides an accounting data management method, system, device, and medium based on edge computing, which can overcome the interference caused by the inherent long-tail distribution characteristics of accounting subjects to the retrieval of similar vouchers in accounting data management.

[0005] In a first aspect, this application provides an accounting data management method based on edge computing, comprising the following steps:

[0006] A lightweight accounting voucher database is built at the edge computing node to store key-value pairs formed by accounting voucher data and their corresponding compliance verification results;

[0007] When new accounting voucher data is received, the structured feature vector of the new accounting voucher data is extracted;

[0008] The data partitioning granularity is determined based on the distribution density of historical accounting voucher data, and a set of similar vouchers similar to the newly added accounting voucher data is extracted from the high-frequency account area and sparse account area of ​​the historical accounting voucher data according to the data partitioning granularity.

[0009] The distance between the structured feature vector of the newly added accounting voucher data and the feature vector of each similar voucher in the similar voucher set is determined, and then the risk confidence value of the newly added accounting voucher data is generated based on the frequency of occurrence of similar accounting vouchers and the feature vector distance.

[0010] When the risk confidence value exceeds the set threshold, the compliance verification result corresponding to the newly added accounting voucher data is identified, and the compliance verification result and structured feature vector corresponding to the newly added accounting voucher data are updated to the accounting voucher database at the edge computing node.

[0011] In some embodiments, when newly added accounting voucher data is received, extracting the structured feature vector of the newly added accounting voucher data specifically includes:

[0012] The newly added accounting voucher data is parsed to extract a standardized set of data fields;

[0013] Based on the attributes of accounting elements, the standardized data field set is divided into a qualitative feature subset and a quantitative feature subset;

[0014] The qualitative feature subset and the quantitative feature subset are respectively converted into qualitative feature vectors and quantitative feature vectors;

[0015] The initial structured feature vector is determined based on the qualitative feature vector and the quantitative feature vector;

[0016] The initial structured feature vector is reduced in dimensionality to obtain the structured feature vector of the newly added accounting voucher data.

[0017] In some embodiments, determining the data partitioning granularity based on the distribution density of historical accounting voucher data specifically includes:

[0018] Density space estimation is performed on the structured feature vector of the historical accounting voucher data to obtain the local density value of each sample point in the density space;

[0019] Density cluster centers of the historical accounting voucher data are extracted based on all local density values;

[0020] Calculate the distribution entropy of the neighborhood samples of the density cluster centers;

[0021] The data partitioning granularity is determined based on the comparison between the distribution entropy of the neighborhood samples and the preset entropy threshold.

[0022] In some embodiments, extracting a set of similar vouchers similar to the newly added accounting vouchers from the high-frequency account area and sparse account area of ​​the historical accounting voucher data according to the data partitioning granularity specifically includes:

[0023] Based on the data partitioning granularity, the high-frequency account area and sparse account area in the historical accounting voucher data are partitioned to obtain the high-frequency account partitioned dataset and the sparse account partitioned dataset.

[0024] Determine the target high-frequency subject partition and the target sparse subject partition based on the high-frequency subject partition dataset and the sparse subject partition dataset;

[0025] Within the target high-frequency account partition and the target sparse account partition, calculate the cosine similarity between the structured feature vector of each historical accounting voucher data and the structured feature vector of the newly added accounting voucher data.

[0026] A set of similar vouchers that are similar to the newly added accounting voucher data is identified based on all cosine similarity scores.

[0027] In some embodiments, determining the distance between the structured feature vector of the newly added accounting voucher data and the feature vector of each similar voucher in the similar voucher set specifically includes:

[0028] The structured feature vectors of newly added accounting voucher data are matched and standardized with the structured feature vectors of each accounting voucher in the similar voucher set to obtain vector pairs to be calculated;

[0029] For each vector pair to be calculated, the distance is solved to obtain the distance between the structured feature vector of the newly added accounting voucher data and the feature vector of each similar voucher in the similar voucher set.

[0030] In some embodiments, when the risk confidence value exceeds a set threshold, identifying the compliance verification result corresponding to the newly added accounting voucher data specifically includes:

[0031] When the risk confidence value exceeds the set threshold, the target verification rule set related to the newly added accounting voucher data is obtained from the preset accounting compliance verification rule base;

[0032] The core elements of the newly added accounting voucher data are matched and verified item by item with the rule entries in the target verification rule set to obtain the compliance verification results corresponding to the newly added accounting voucher data.

[0033] In some embodiments, the key-value pair refers to a data storage format that uses the structured feature vector of the accounting voucher data as the key and the compliance verification result corresponding to the accounting voucher data as the value.

[0034] Secondly, this application provides an accounting data management system based on edge computing, comprising:

[0035] The building module is used to build a lightweight accounting voucher database at the edge computing node, which stores the key-value pairs formed by accounting voucher data and their corresponding compliance verification results;

[0036] The processing module is used to extract the structured feature vector of the newly added accounting voucher data when it receives the newly added accounting voucher data;

[0037] The processing module is further configured to determine the data partitioning granularity based on the distribution density of historical accounting voucher data, and extract a set of similar vouchers that are similar to the newly added accounting voucher data from the high-frequency account area and sparse account area in the historical accounting voucher data according to the data partitioning granularity.

[0038] The processing module is further configured to determine the distance between the structured feature vector of the newly added accounting voucher data and the feature vector of each similar voucher in the similar voucher set, and then generate a risk confidence value of the newly added accounting voucher data based on the frequency of occurrence of similar accounting vouchers and the feature vector distance;

[0039] The execution module is used to identify the compliance verification result corresponding to the newly added accounting voucher data when the risk confidence value exceeds a set threshold, and update the compliance verification result and structured feature vector corresponding to the newly added accounting voucher data to the accounting voucher database at the edge computing node.

[0040] Thirdly, this application provides a computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described edge computing-based accounting data management method.

[0041] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned edge computing-based accounting data management method.

[0042] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0043] The edge computing-based accounting data management method, system, device, and medium provided in this application firstly construct a lightweight accounting voucher database at the edge computing node to store key-value pairs formed by accounting voucher data and their corresponding compliance verification results; secondly, when new accounting voucher data is received, the structured feature vector of the new accounting voucher data is extracted; further, the data partitioning granularity is determined based on the distribution density of historical accounting voucher data, and a set of similar vouchers similar to the new accounting voucher data is extracted from the high-frequency account area and sparse account area in the historical accounting voucher data according to the data partitioning granularity; then, the distance between the structured feature vector of the new accounting voucher data and the feature vector of each similar voucher in the set of similar vouchers is determined, and then a risk confidence value of the new accounting voucher data is generated based on the occurrence frequency of similar accounting vouchers and the feature vector distance; finally, when the risk confidence value exceeds a set threshold, the compliance verification result corresponding to the new accounting voucher data is identified, and the compliance verification result and structured feature vector corresponding to the new accounting voucher data are updated to the accounting voucher database at the edge computing node.

[0044] Therefore, this application can overcome the interference caused by the inherent long-tail distribution of accounting subjects in similar document retrieval in accounting data management. First, by constructing a lightweight key-value pair accounting document database on edge computing nodes, it can fully adapt to the limited storage and computing resources of edge nodes, avoiding the excessive occupation of edge resources by traditional heavy databases. At the same time, the key-value pair structure can realize the associated storage of accounting document data and compliance verification results, reducing cross-source calls during subsequent data queries. Second, by extracting the structured feature vector of newly added accounting document data, the semi-structured / unstructured information in accounting documents can be transformed into standardized quantitative data forms, ensuring the operability and consistency of subsequent similarity matching and risk assessment processes. Furthermore, by determining the data partitioning granularity based on the distribution density of historical accounting document data, and extracting a set of similar documents similar to the newly added accounting document data from the high-frequency subject area and sparse subject area of ​​historical accounting document data based on the data partitioning granularity, it can effectively avoid the computational redundancy caused by traditional global traversal retrieval. The problem is to ensure both the efficiency of high-frequency account data processing to avoid resource waste and the reduction of feature information loss in sparse account areas to ensure matching accuracy, thereby avoiding the interference caused by the inherent long-tail distribution of accounting subjects to similar voucher retrieval. Then, by determining the feature vector distance and combining it with the frequency of occurrence of similar accounting documents to generate a risk confidence value, the risk assessment of accounting documents can be transformed from qualitative judgment relying on human experience to quantitative assessment integrating business logic (frequency of similar documents) and data differences (feature vector distance), improving the objectivity and accuracy of risk assessment and outputting results that meet the real-time processing needs of accounting business. Finally, when the risk confidence value exceeds a set threshold, the compliance verification result corresponding to the newly added accounting document data is identified, and the compliance verification result and structured feature vector corresponding to the newly added accounting document data are updated to the accounting document database at the edge computing node. In summary, the technical solution provided in this application can overcome the interference caused by the inherent long-tail distribution of accounting subjects to similar voucher retrieval in accounting data management. Attached Figure Description

[0045] Figure 1 This is an exemplary flowchart of an edge computing-based accounting data management method according to some embodiments of this application;

[0046] Figure 2 This is an exemplary flowchart illustrating the determination of structured feature vectors according to some embodiments of this application;

[0047] Figure 3 This is a schematic diagram of the structure of an edge computing-based accounting data management system according to some embodiments of this application;

[0048] Figure 4This is a schematic diagram of the structure of a computer device implementing an edge computing-based accounting data management method according to some embodiments of this application. Detailed Implementation

[0049] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] refer to Figure 1 The figure is an exemplary flowchart of an edge computing-based accounting data management method according to some embodiments of this application. The edge computing-based accounting data management method mainly includes the following steps:

[0051] In step 101, a lightweight accounting voucher database is built at the edge computing node to store the structured feature vectors of accounting voucher data and key-value pairs formed by compliance verification results.

[0052] In practical implementation, based on the characteristics of accounting voucher data and the storage and computing power limitations of edge computing nodes, a key-value pair storage structure is designed with the structured feature vector of accounting voucher data as the key and the corresponding compliance verification result of accounting voucher data as the value, thereby obtaining a lightweight accounting voucher database. The accounting voucher database is used to store key-value pairs formed by accounting voucher data and its corresponding compliance verification results. The structured feature vector refers to the multi-dimensional data points formed by digitizing and vectorizing the key business attributes of an accounting voucher. Specifically, the structured feature vector is a feature vector generated by combining voucher type, accounting subject code, amount range, and transaction object classification.

[0053] It should be noted that, in this application, the accounting voucher database refers to a structured data set of key-value pairs formed by using the structured feature vector of accounting voucher data as the key and the corresponding compliance verification result as the value. By establishing the accounting voucher database, efficient management and fast access to historical accounting voucher data can be effectively achieved, while taking into account the storage and computing power limitations of edge devices, providing data support for similarity comparison, risk assessment, and compliance verification of newly added accounting voucher data. In this application, the key-value pair refers to a data storage format formed by using the structured feature vector of accounting voucher data as the key and the corresponding compliance verification result of accounting voucher data as the value, which is used to achieve efficient storage, query, and management of accounting voucher-related information in a lightweight accounting database.

[0054] In step 102, when new accounting voucher data is received, the structured feature vector of the new accounting voucher data is extracted.

[0055] In some embodiments, reference Figure 2As shown in the figure, this is an exemplary flowchart of determining a structured feature vector according to some embodiments of this application. In this embodiment, when new accounting voucher data is received, the extraction of the structured feature vector of the new accounting voucher data can be achieved by the following steps:

[0056] In step 1021, the newly added accounting voucher data is parsed to extract a standardized set of data fields;

[0057] In step 1022, the standardized data field set is divided into a qualitative feature subset and a quantitative feature subset based on the accounting element attributes;

[0058] In step 1023, the qualitative feature subset and the quantitative feature subset are converted into qualitative feature vectors and quantitative feature vectors, respectively;

[0059] In step 1024, an initial structured feature vector is determined based on the qualitative feature vector and the quantitative feature vector;

[0060] In step 1025, the initial structured feature vector is subjected to dimensionality reduction processing to obtain the structured feature vector of the newly added accounting voucher data.

[0061] In specific implementation, firstly, existing structured data parsing tools (such as Extensible Markup Language parser) are used to parse the format of the newly added accounting voucher data through field mapping rules, extracting a standardized data field set including, but not limited to, accounting entity, voucher date, account code, and transaction amount. This standardized data field set refers to a set of accounting data items with standardized and uniform formats and names. Secondly, based on accounting element attributes, the standardized data field set is divided into qualitative and quantitative feature subsets. Specifically, according to the data type attributes in the accounting element attributes, non-numerical information (such as summary text and account name) in the standardized data field set is classified as the qualitative feature subset, and numerical information (such as transaction amount and quantity) in the standardized data field set is classified as the quantitative feature subset. The qualitative feature subset refers to a set of text features composed of descriptive, non-quantitative data, and the quantitative feature subset refers to a set of numerical features composed of quantifiable, numerical data. Further, a word embedding encoding algorithm (such as Word2Vec) is used to extract the text features from the qualitative feature subset. The qualitative feature vector is converted into a low-dimensional, dense qualitative feature vector. The numerical features in the quantitative feature subset are then combined according to the order of the textual features in the qualitative feature vector to generate a quantitative feature vector. The qualitative feature vector refers to a numerical vector reflecting the textual semantics of the accounting voucher data, and the quantitative feature vector refers to a specific numerical vector reflecting the accounting voucher data. Then, the standardized qualitative and quantitative feature vectors are combined to form an initial structured feature vector containing all feature information. This initial structured feature vector is a feature vector that comprehensively represents the accounting voucher data. Finally, principal component analysis is used to calculate eigenvalues ​​and eigenvectors of the initial structured feature vector. The eigenvalues ​​are arranged in descending order, and the ratio of the cumulative sum of the first n eigenvalues ​​to the sum of all eigenvalues ​​is calculated to obtain the cumulative contribution rate. The first n eigenvalues ​​corresponding to the first time the cumulative contribution rate reaches a preset threshold are selected, and their corresponding eigenvectors are extracted to form a projection matrix. The initial structured feature vector and the projection matrix are multiplied to obtain the structured feature vector of the newly added accounting voucher data.

[0062] It should be noted that, in this application, the structured feature vector refers to a low-dimensional feature vector that retains the main feature information in the newly added accounting voucher data. The determination of the structured feature vector transforms the accounting voucher data from the original unstructured or semi-structured form into a unified numerical vector form. This transformation firstly achieves the standardized representation of accounting data, eliminates the heterogeneity between different formats and types of data, and enables accounting information that was originally difficult to compare directly to be quantitatively analyzed in the same feature space. Secondly, through feature extraction and dimensionality reduction, the core feature information of the accounting voucher is retained, and redundant noise is eliminated. This not only reduces the computational complexity of subsequent data processing, but also ensures that the analysis process focuses on the key elements that have a substantial impact on accounting business.

[0063] In step 103, the data partitioning granularity is determined based on the distribution density of historical accounting voucher data, and a set of similar vouchers similar to the newly added accounting voucher data is extracted from the high-frequency account area and sparse account area in the historical accounting voucher data according to the data partitioning granularity.

[0064] In some embodiments, determining the data partitioning granularity based on the distribution density of historical accounting voucher data can be achieved through the following steps:

[0065] Density space estimation is performed on the structured feature vector of the historical accounting voucher data to obtain the local density value of each sample point in the density space;

[0066] Density cluster centers of the historical accounting voucher data are extracted based on all local density values;

[0067] Calculate the distribution entropy of the neighborhood samples of the density cluster centers;

[0068] The data partitioning granularity is determined based on the comparison between the distribution entropy of the neighborhood samples and the preset entropy threshold.

[0069] In specific implementation, firstly, the k-nearest neighbor density estimation method is used, with each structured feature vector in the historical accounting voucher data as a sample point. The number of neighboring sample points within a preset k-value range is counted. Then, by calculating the reciprocal of the average Euclidean distance between neighboring sample points, the local density value of each sample point in the density space is obtained. The local density value refers to a quantitative indicator of the density of the surrounding sample distribution of a single sample point in the feature space. Secondly, based on all local density values, the density clustering centers of the historical accounting voucher data are extracted, i.e., the mean of the local density values ​​of all sample points is calculated. Based on the standard deviation, sample points with a local density value ≥ mean + 1.5 times the standard deviation are marked as high-density sample points. For each high-density sample point, all other high-density sample points with a density greater than the high-density sample point are traversed, and the Euclidean distance between them is calculated. The minimum Euclidean distance is taken as the relative distance of the high-density sample point. The relative distances of all high-density sample points are counted, and their mean and standard deviation are calculated. Then, sample points with a relative distance ≥ mean + 1.5 times the standard deviation are selected as density cluster centers. The density cluster center refers to the core sample point with a local density significantly higher than the surrounding samples in the feature space. Next, a neighborhood range with a fixed radius (which can be set according to actual needs) is defined centered on the density cluster center. The distribution frequency of all sample points within the neighborhood is counted, and the distribution frequency is used as an input variable to calculate the degree of disorder in the feature distribution of samples within the neighborhood using the Shannon entropy function. This yields the neighborhood sample distribution entropy of the density cluster center, which is a quantitative indicator measuring the disorder of the distribution of samples in the feature dimension within the neighborhood of the cluster center. Finally, the data partitioning granularity is determined based on the comparison result of the neighborhood sample distribution entropy and the preset entropy threshold value, i.e., a baseline data partitioning granularity is set. If the entropy of the neighborhood sample distribution is greater than a preset entropy threshold, the value obtained by multiplying the baseline partition granularity by an adjustment factor of 0.5 is used as the candidate granularity of the data partition granularity; if the entropy of the neighborhood sample distribution is less than or equal to the preset entropy threshold, the value obtained by multiplying the baseline partition granularity by an adjustment factor of 2 is used as the candidate granularity of the data partition granularity. The specific baseline data partition granularity can be set according to actual needs or according to experimental simulation. For example, the average entropy of the historical neighborhood sample distribution can be used as the baseline data partition granularity. The specific entropy threshold can also be set according to actual needs, which will not be elaborated here.

[0070] It should be noted that, in this application, the data partitioning granularity refers to the standard for the number of samples contained in each partition when dividing historical accounting voucher data into different data partitions according to the characteristic distribution. Determining the data partitioning granularity is the core link to achieve efficient partitioning and accurate retrieval of historical accounting voucher data. Its role is to avoid excessive heterogeneity of data within partitions due to overly coarse partitioning (such as the mixing of samples in high-frequency account areas and sparse account areas), or the problem of a surge in the number of partitions and waste of computing resources caused by overly fine partitioning, thereby laying an efficient data organization foundation for the subsequent extraction of similar voucher sets from high-frequency account areas and sparse account areas.

[0071] In some embodiments, extracting a set of similar vouchers similar to the newly added accounting vouchers from the high-frequency account area and sparse account area of ​​the historical accounting voucher data according to the data partitioning granularity can be achieved by the following steps:

[0072] Based on the data partitioning granularity, the high-frequency account area and sparse account area in the historical accounting voucher data are partitioned respectively to obtain the high-frequency account partitioned dataset and the sparse account partitioned dataset.

[0073] The target high-frequency subject partition and the target sparse subject partition are determined based on the high-frequency subject partition dataset and the sparse subject partition dataset.

[0074] Within the target high-frequency account partition and the target sparse account partition, calculate the cosine similarity between the structured feature vector of each historical accounting voucher data and the structured feature vector of the newly added accounting voucher data.

[0075] A set of similar vouchers that are similar to the newly added accounting voucher data is identified based on all cosine similarity scores.

[0076] In specific implementation, firstly, the high-frequency account area (i.e., the set of accounting vouchers in which the account code appears more frequently than a preset frequency threshold) and the sparse account area (i.e., the set of accounting vouchers in which the account code appears less frequently than a preset frequency threshold) in historical accounting voucher data are obtained. Then, the existing grid partitioning algorithm is used to map the structured feature vectors of the high-frequency account area and the sparse account area to the feature space grid, respectively. The number of samples in each grid cell is controlled according to the data partitioning granularity, thereby obtaining the high-frequency account partition dataset and the sparse account partition dataset. The high-frequency account partition dataset represents multiple subsets of the high-frequency account area, and each subset corresponds to a grid cell in the feature space. The sparse account partition dataset represents multiple subsets of the sparse account area. Secondly, the Euclidean distance formula is used to calculate the distance between the structured feature vector of the newly added accounting voucher data and the center vector distance of each partition in the high-frequency account partition dataset and the sparse account partition dataset. The high-frequency account partition with the smallest center vector distance is marked as the target high-frequency account partition. The sparse account partition with the smallest vector distance is marked as the target sparse account partition. The target high-frequency account partition represents the high-frequency account subset with the closest features to the newly added voucher, and the target sparse account partition represents the sparse account subset with the closest features to the newly added voucher. Then, within the target high-frequency account partition and the target sparse account partition, the cosine similarity between the structured feature vector of each historical accounting voucher data and the structured feature vector of the newly added accounting voucher data is calculated using the cosine similarity formula. The cosine similarity represents a quantitative index that measures the consistency between the two feature vectors. Finally, historical accounting vouchers with a cosine similarity greater than the similarity threshold within the target high-frequency account partition and the target sparse account partition are selected, and duplicates are removed using the unique identifier of the accounting voucher (such as the voucher number). (If the same accounting voucher exists in both target partitions, only one record is retained). Finally, a set of similar vouchers similar to the newly added accounting voucher data is obtained. The similarity threshold is a critical value determined based on the criteria for judging similar vouchers in accounting business. The specific value can be set according to actual needs, and it usually ranges from 0.7 to 0.9.

[0077] It should be noted that, in this application, the similar voucher set refers to a dataset composed of historical accounting vouchers similar to the newly added accounting vouchers. The similar voucher set contains multiple similar vouchers, which are historical accounting vouchers similar to the newly added accounting vouchers. The determination of the similar voucher set provides a core reference for subsequent accounting data risk assessment and compliance verification. Its function is to screen out samples that are highly correlated with the characteristics of the newly added vouchers from massive historical accounting voucher data, avoiding indiscriminate analysis of all historical data, significantly reducing the computing power consumption and data processing latency of edge computing nodes, and providing a comparable historical benchmark for the risk attribute determination of the newly added vouchers. Secondly, the similar vouchers in the high-frequency and sparse account areas included in this set can respectively reflect the feature matching degree of the newly added vouchers in regular business scenarios (high-frequency account areas) and special business scenarios (sparse account areas), helping to balance the stability of regular business and the particularity of special business when generating risk confidence values ​​based on the frequency of occurrence of similar vouchers and the distance of feature vectors, and avoiding risk misjudgment caused by a single dimension reference.

[0078] In step 104, the distance between the structured feature vector of the newly added accounting voucher data and the feature vector of each similar voucher in the similar voucher set is determined, and then the risk confidence value of the newly added accounting voucher data is generated based on the frequency of occurrence of similar accounting vouchers and the feature vector distance.

[0079] In some embodiments, determining the distance between the structured feature vector of the newly added accounting voucher data and the feature vector of each similar voucher in the similar voucher set can be achieved by the following steps:

[0080] The structured feature vectors of newly added accounting voucher data are matched and standardized with the structured feature vectors of each similar voucher in the similar voucher set to obtain vector pairs to be calculated;

[0081] For each pair of vectors to be calculated, the distance is obtained by solving the distance between the structured feature vector of the newly added accounting voucher data and each similar feature vector in the set of similar vouchers.

[0082] In specific implementation, firstly, a preset feature dimension mapping table (i.e., a reference table recording the definition of each dimension of the structured feature vector of accounting vouchers and its corresponding business meaning) is used to align the dimension names and order of the structured feature vector of the newly added accounting voucher data with the structured feature vector of each similar voucher in the similar voucher set. If a dimension of a certain vector is missing, the mean of that dimension in the historical accounting voucher data is used to fill in the missing dimension. Then, the Z-score standardization algorithm is used to eliminate the difference in dimensions between different dimensions, resulting in vector pairs to be calculated that are consistent in dimensions and comparable in value. The vector pairs to be calculated represent vector combinations composed of the structured feature vector of the newly added accounting voucher data and the structured feature vector of a single similar voucher, which can be directly used for distance calculation. Then, when solving the distance for each vector pair to be calculated, Euclidean distance is used to calculate the vector distance between the vector pairs to be calculated, and the vector distance is used as the distance between the structured feature vector of the newly added accounting voucher data and the feature vector of the similar voucher in the similar voucher set. This will not be elaborated further here.

[0083] It should be noted that in this application, the feature vector distance represents a quantitative indicator reflecting the degree of difference in the geometric positions of two types of feature vectors in the feature space. The determination of the feature vector distance is the core quantitative bridge connecting the newly added accounting voucher data with the set of similar vouchers and supporting the subsequent assessment of accounting business risks and compliance verification. On the one hand, by integrating accounting business logic (such as giving higher weight to key dimensions such as account codes and transaction amounts), this distance value accurately quantifies the degree of difference between the new vouchers and similar vouchers in core features, avoiding the limitation that relying solely on a single similarity indicator (such as cosine similarity) cannot distinguish the importance of feature dimensions, and providing a more suitable basis for judging the "substantial similarity" between the new vouchers and similar vouchers in accounting business scenarios.

[0084] In some embodiments, generating the risk confidence value of the newly added accounting document data based on the frequency of occurrence of similar accounting documents and the feature vector distance can be achieved by the following steps:

[0085] Based on the comparison results of the feature vector distance and the preset distance threshold, a subset of similar vouchers that are similar to the newly added accounting voucher data is divided from the similar voucher set;

[0086] The frequency of occurrence of each type of voucher in the aforementioned subset of vouchers is calculated in historical accounting voucher data to obtain the frequency of occurrence of each type of voucher.

[0087] The feature vector distances of each similar voucher in the subset of similar vouchers are normalized to obtain the normalized distance values ​​corresponding to each similar voucher.

[0088] The weighted distance value is obtained by weighting the normalized distance values ​​by using the frequency of occurrence of each similar voucher as the weight.

[0089] The weighted distance value is mapped to the risk confidence value of the newly added accounting voucher data through a preset risk transformation function.

[0090] In specific implementation, firstly, accounting vouchers in the similar voucher set whose feature vector distance is less than or equal to a preset distance threshold are filtered out to form a subset of similar vouchers of the same type as the newly added accounting voucher data. This subset of similar vouchers refers to a set of historical vouchers of the same transaction type as the newly added accounting vouchers. Secondly, by traversing the historical accounting voucher database, the frequency of occurrence of similar vouchers in the subset of similar vouchers in the historical accounting voucher data is statistically analyzed to obtain the occurrence frequency of each similar voucher. Further, the min-max normalization algorithm is used to map the feature vector distance of each similar voucher in the subset of similar vouchers to the [0,1] interval to obtain the normalized distance value corresponding to each similar voucher. Then, using the occurrence frequency of each similar voucher as the weight, each... The normalized distance value of each type of voucher is multiplied by its corresponding frequency of occurrence (the frequency of occurrence is used as a weight to reflect the impact of the regularity of this type of transaction on risk assessment). All the product results are then summed to obtain a weighted distance value, which is a quantitative indicator used to reflect the degree of deviation between the new voucher and the regular voucher of the same type. Finally, the historical sigmoid function is selected as the preset risk transformation function, and the weighted distance value is used as the input variable to the risk transformation function. The output result of the risk transformation function is used as the risk confidence value of the new accounting voucher data. The sigmoid function is a non-linear mathematical function whose core principle is to map any real number input to the (0,1) interval through exponential operation.

[0091] It should be noted that the risk confidence value in this application is an indicator that represents the compliance anomaly of newly added accounting vouchers. By determining the risk confidence value, the risk characteristics of newly added accounting vouchers can be transformed into standardized quantitative indicators in the range of [0,1]. This provides a threshold-based decision-making basis for risk management of accounting data at edge nodes. On the one hand, it realizes the transformation of risk judgment from subjective experience to objective quantification, eliminating the bias of manual judgment. On the other hand, it supports real-time hierarchical management. By comparing with preset thresholds, it automatically triggers in-depth verification of high-risk vouchers or rapid release of low-risk vouchers, optimizing the allocation of edge computing resources. At the same time, it provides traceable and verifiable quantitative records for subsequent risk tracing and auditing, ensuring the standardization and repeatability of accounting data compliance management.

[0092] In step 105, when the risk confidence value exceeds a set threshold, the compliance verification result corresponding to the newly added accounting voucher data is identified, and the compliance verification result and structured feature vector corresponding to the newly added accounting voucher data are updated to the accounting voucher database at the edge computing node.

[0093] In some embodiments, when the risk confidence value exceeds a set threshold, identifying the compliance verification result corresponding to the newly added accounting voucher data can be achieved through the following steps:

[0094] When the risk confidence value exceeds the set threshold, the target verification rule set related to the newly added accounting voucher data is obtained from the preset accounting compliance verification rule base;

[0095] The core elements of the newly added accounting voucher data are matched and verified item by item with the rule entries in the target verification rule set to obtain the compliance verification results corresponding to the newly added accounting voucher data.

[0096] In practice, when the risk confidence value exceeds a set threshold, rule retrieval keywords are constructed based on the core elements of the newly added accounting voucher data (such as account code, transaction type, and amount range). An existing rule matching algorithm based on structured query language is used to filter rule entries related to the newly added accounting voucher data from a pre-set accounting compliance verification rule library, resulting in a target verification rule set. This target verification rule set refers to a standardized set of verification rules highly adapted to the business scenario of the newly added accounting voucher, selected from the pre-set accounting compliance verification rule library. Then, the core elements of the newly added accounting voucher data (such as account code, transaction amount, and transaction date) are extracted and converted into a rule-recognizable structured data format (such as key-value pairs). A pattern matching mechanism from a rule engine (such as the Drools rule engine) is then used to logically compare each core element with the rule entries in the target verification rule set in sequence. If a core element meets all the conditions of a rule entry, it is determined to comply with the rule; if it does not meet any condition, it is determined to violate the rule. The judgment results of all rules are then integrated to obtain the compliance verification result corresponding to the newly added accounting voucher data.

[0097] It should be noted that the compliance verification result in this application is a conclusive report characterizing whether the newly added accounting vouchers comply with accounting compliance requirements. The determination of the compliance verification result is a key link in transforming the high-risk signal of the newly added accounting vouchers into a specific basis for compliance judgment, supporting the closed-loop management of accounting business risks, and serving as a core component of the accounting voucher compliance file. Subsequent audits and risk tracing can directly trace the rule basis and verification process for the violation judgment, ensuring the verifiability and reproducibility of the compliance judgment.

[0098] In some embodiments, updating the compliance verification result and structured feature vector corresponding to the newly added accounting voucher data to the accounting voucher database at the edge computing node can be achieved by the following steps:

[0099] Obtain the compliance verification results and structured feature vectors corresponding to the newly added accounting voucher data;

[0100] The compliance verification results and structured feature vectors of the newly added accounting voucher data are written to the preset storage partition according to the key-value pair storage format of the accounting voucher database, and the index update mechanism of the accounting voucher database at the edge computing node is triggered to synchronously update the storage information of the storage partition.

[0101] In practice, firstly, the compliance verification result and structured feature vector corresponding to the newly added accounting voucher data are obtained; then, the compliance verification result and structured feature vector of the newly added accounting voucher data are written as the value and key of the key-value pair, respectively, into a preset storage partition, and the index update mechanism of the accounting voucher database at the edge computing node is triggered to synchronously update the storage information of the storage partition.

[0102] Furthermore, in another aspect of this application, in some embodiments, this application provides an accounting data management system based on edge computing, see reference. Figure 3 The figure is a schematic diagram of the structure of an edge computing-based accounting data management system according to some embodiments of this application. The edge computing-based accounting data management system includes: a construction module 201, a processing module 202, and an execution module 203, which are described below:

[0103] Construction module 201, in this application, is mainly used to build a lightweight accounting voucher database at the edge computing node, for storing accounting voucher data and key-value pairs formed by the corresponding compliance verification results;

[0104] Processing module 202, in this application, is mainly used to extract the structured feature vector of the newly added accounting voucher data when the newly added accounting voucher data is received;

[0105] The processing module 202 is further configured to determine the data partitioning granularity based on the distribution density of historical accounting voucher data, and extract a set of similar vouchers that are similar to the newly added accounting voucher data from the high-frequency account area and sparse account area in the historical accounting voucher data according to the data partitioning granularity.

[0106] In addition, the processing module 202 is also used to determine the distance between the structured feature vector of the newly added accounting voucher data and the feature vector of each similar voucher in the similar voucher set, and then generate the risk confidence value of the newly added accounting voucher data based on the frequency of occurrence of similar accounting vouchers and the feature vector distance;

[0107] The execution module 203 in this application is mainly used to identify the compliance verification result corresponding to the newly added accounting voucher data when the risk confidence value exceeds a set threshold, and update the compliance verification result and structured feature vector corresponding to the newly added accounting voucher data to the accounting voucher database at the edge computing node.

[0108] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described edge computing-based accounting data management method.

[0109] In some embodiments, reference Figure 4 The figure is a schematic diagram of the structure of a computer device implementing an edge computing-based accounting data management method according to some embodiments of this application. The edge computing-based accounting data management method in the above embodiments can... Figure 4 The computer device shown is used to implement this, and the computer device includes at least one processor 301, a communication bus 302, a memory 303, and at least one communication interface 304.

[0110] The processor 301 can be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more devices used to control the execution of the edge computing-based accounting data management method described in this application.

[0111] The communication bus 302 can be used to transmit information between the aforementioned components.

[0112] The memory 303 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 303 may exist independently and be connected to the processor 301 via the communication bus 302. The memory 303 may also be integrated with the processor 301.

[0113] The memory 303 stores program code for executing the scheme of this application, and its execution is controlled by the processor 301. The processor 301 executes the program code stored in the memory 303. The program code may include one or more software modules. In the above embodiments, the determination of the accounting data management method based on edge computing can be implemented by the processor 301 and one or more software modules in the program code in the memory 303.

[0114] Communication interface 304 uses any transceiver-like device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0115] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0116] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0117] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described edge computing-based accounting data management method.

[0118] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0119] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. An accounting data management method based on edge computing, characterized in that, Includes the following steps: A lightweight accounting voucher database is built at the edge computing node to store key-value pairs formed by accounting voucher data and their corresponding compliance verification results; When new accounting voucher data is received, the structured feature vector of the new accounting voucher data is extracted; The data partitioning granularity is determined based on the distribution density of historical accounting voucher data, and a set of similar vouchers similar to the newly added accounting voucher data is extracted from the high-frequency account area and sparse account area of ​​the historical accounting voucher data according to the data partitioning granularity. The structured feature vector of the newly added accounting voucher data is determined to be the distance between the feature vector of each similar voucher in the similar voucher set. Then, a risk confidence value of the newly added accounting voucher data is generated based on the frequency of occurrence of similar accounting vouchers and the feature vector distance. The similar accounting vouchers refer to accounting vouchers selected from the similar voucher set whose feature vector distance is less than or equal to a preset distance threshold and are similar to the newly added accounting voucher. When the risk confidence value exceeds the set threshold, the compliance verification result corresponding to the newly added accounting voucher data is identified, and the compliance verification result and structured feature vector corresponding to the newly added accounting voucher data are updated to the accounting voucher database at the edge computing node.

2. The method as described in claim 1, characterized in that, When new accounting voucher data is received, extracting the structured feature vector of the new accounting voucher data specifically includes: The newly added accounting voucher data is parsed to extract a standardized set of data fields; Based on the attributes of accounting elements, the standardized data field set is divided into a qualitative feature subset and a quantitative feature subset; The qualitative feature subset and the quantitative feature subset are respectively converted into qualitative feature vectors and quantitative feature vectors; The initial structured feature vector is determined based on the qualitative feature vector and the quantitative feature vector; The initial structured feature vector is reduced in dimensionality to obtain the structured feature vector of the newly added accounting voucher data.

3. The method as described in claim 1, characterized in that, Determining the data partitioning granularity based on the distribution density of historical accounting voucher data specifically includes: Density space estimation is performed on the structured feature vector of the historical accounting voucher data to obtain the local density value of each sample point in the density space; Density cluster centers of the historical accounting voucher data are extracted based on all local density values; Calculate the distribution entropy of the neighborhood samples of the density cluster centers; The data partitioning granularity is determined based on the comparison between the distribution entropy of the neighborhood samples and the preset entropy threshold.

4. The method as described in claim 1, characterized in that, Based on the data partitioning granularity, a set of similar vouchers similar to the newly added accounting vouchers is extracted from the high-frequency account area and sparse account area of ​​the historical accounting voucher data. Specifically, this includes: Based on the data partitioning granularity, the high-frequency account area and sparse account area in the historical accounting voucher data are partitioned respectively to obtain the high-frequency account partitioned dataset and the sparse account partitioned dataset. The target high-frequency subject partition and the target sparse subject partition are determined based on the high-frequency subject partition dataset and the sparse subject partition dataset. Within the target high-frequency account partition and the target sparse account partition, calculate the cosine similarity between the structured feature vector of each historical accounting voucher data and the structured feature vector of the newly added accounting voucher data. A set of similar vouchers that are similar to the newly added accounting voucher data is identified based on all cosine similarity scores.

5. The method as described in claim 1, characterized in that, Determining the distance between the structured feature vector of the newly added accounting voucher data and the feature vector of each similar voucher in the similar voucher set specifically includes: The structured feature vectors of newly added accounting voucher data are matched and standardized with the structured feature vectors of each accounting voucher in the similar voucher set to obtain vector pairs to be calculated; For each vector pair to be calculated, the distance is obtained by solving the distance between the structured feature vector of the newly added accounting voucher data and the feature vector of each similar voucher in the similar voucher set.

6. The method as described in claim 1, characterized in that, When the risk confidence value exceeds a set threshold, the specific steps for identifying the compliance verification result corresponding to the newly added accounting voucher data include: When the risk confidence value exceeds the set threshold, the target verification rule set related to the newly added accounting voucher data is obtained from the preset accounting compliance verification rule base; The core elements of the newly added accounting voucher data are matched and verified item by item with the rule entries in the target verification rule set to obtain the compliance verification results corresponding to the newly added accounting voucher data.

7. The method as described in claim 1, characterized in that, The key-value pair refers to a data storage format that uses the structured feature vector of accounting voucher data as the key and the corresponding compliance verification result of the accounting voucher data as the value.

8. An accounting data management system based on edge computing, which manages accounting data using the method described in any one of claims 1 to 7, characterized in that, The system includes: The building module is used to build a lightweight accounting voucher database at the edge computing node, which stores the key-value pairs formed by accounting voucher data and their corresponding compliance verification results; The processing module is used to extract the structured feature vector of the newly added accounting voucher data when it receives the newly added accounting voucher data; The processing module is further configured to determine the data partitioning granularity based on the distribution density of historical accounting voucher data, and extract a set of similar vouchers that are similar to the newly added accounting voucher data from the high-frequency account area and sparse account area in the historical accounting voucher data according to the data partitioning granularity. The processing module is further configured to determine the distance between the structured feature vector of the newly added accounting voucher data and the feature vector of each similar voucher in the similar voucher set, and then generate a risk confidence value of the newly added accounting voucher data based on the frequency of occurrence of similar accounting vouchers and the feature vector distance; The execution module is used to identify the compliance verification result corresponding to the newly added accounting voucher data when the risk confidence value exceeds a set threshold, and update the compliance verification result and structured feature vector corresponding to the newly added accounting voucher data to the accounting voucher database at the edge computing node.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing code, and the processor being configured to retrieve the code and execute the edge computing-based accounting data management method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the edge computing-based accounting data management method as described in any one of claims 1 to 7.

Images