Image upstream and downstream association map generation method, equipment and medium

Abstract

The invention discloses an image upstream and downstream association map generation method and device and a medium, and the method comprises the steps: carrying out the feature extraction of an image data set of a financial system, and obtaining the business entity feature information of the financial system; according to the service entity feature information and a pre-constructed service exchange probability prediction network model, obtaining service exchange probability information with an upstream and downstream relationship between different service entities; determining an association relationship between different service entity nodes according to the service exchange probability information; the association relationship comprises whether an edge and an edge attribute exist or not; and generating an image upstream and downstream association map of the financial system according to the business entity nodes and the association relationship among the business entity nodes. Potential association and business logic behind image data can be mined.

Description

Technical Field

[0001] This application relates to the technical field of financial data processing, and particularly to a method, device, and medium for generating an upstream and downstream association graph of images. Background Art

[0002] In recent years, with the rapid development of big data and artificial intelligence technologies, the application of machine learning in the financial field has gradually increased. Currently, many enterprises have begun to try to use machine learning algorithms for financial statement analysis, fraud detection, budget preparation, etc. However, most of these applications focus on the preliminary analysis and prediction of data, and there are still significant technical challenges in the processing and correlation analysis of complex financial image data (such as invoices, contracts, etc.).

[0003] Currently, the sources of financial image data are diverse and the formats are inconsistent. For complex financial image data, manual participation in analysis is usually required. Analyzing the correlation between image data manually is time-consuming and the accuracy is limited by the experience of personnel. Especially when dealing with a large amount of data, it is difficult to ensure timeliness and correctness, and most are based on simple rules or statistical methods, making it difficult to discover the deep correlations between data, and the correlation analysis is not deep enough.

[0004] Therefore, it usually stays at the surface data integration and cannot deeply explore the potential correlations and business logics behind the image data, making it difficult to automatically establish complex upstream and downstream association relationships. Summary of the Invention

[0005] Embodiments of this application provide a method, device, and medium for generating an upstream and downstream association graph of images, which are used to solve the problem of being unable to deeply explore the potential correlations and business logics behind the image data.

[0006] Embodiments of this application adopt the following technical solutions: On the one hand, embodiments of this application provide a method for generating an upstream and downstream association graph of images. The method includes: extracting features from the image data set of the financial system to obtain the business entity feature information of the financial system; according to the business entity feature information and a pre-constructed business transaction probability prediction network model, obtaining the business transaction probability information indicating the existence of upstream and downstream relationships between different business entities; determining the association relationships between different business entity nodes according to the business transaction probability information; the association relationships include whether there are edges and edge attributes; generating the upstream and downstream association graph of images of the financial system according to the business entity nodes and the association relationships between the business entity nodes.

[0007] In one example, determining the association relationship between different business entity nodes according to the business transaction probability information specifically includes: when the business transaction probability between business entity nodes is higher than a preset probability threshold, determining that there is an edge between the business entity nodes; the edge is used to represent the upstream and downstream business relationship between business entity nodes; determining the business transaction probability between business entity nodes as the weight of the edge; when the business transaction probability between business entity nodes is lower than or equal to the preset probability threshold, determining that there is no edge between the business entity nodes.

[0008] In one example, after generating the image upstream and downstream association graph of the financial system according to the business entity nodes and the association relationship between the business entity nodes, the method further includes: obtaining the new business entity feature information of the financial system; inputting the new business entity feature information and the business entity feature information into the business transaction probability prediction network model to obtain the business transaction probability information of the upstream and downstream relationship between the business entity and the new business entity, so as to determine the association relationship between the new business entity node and the business entity node; adding the new business entity node to the image upstream and downstream association graph according to the association relationship between the new business entity node and the business entity node.

[0009] In one example, after generating the image upstream and downstream association graph of the financial system according to the business entity nodes and the association relationship between the business entity nodes, the method further includes: performing field matching on the feature information of the corresponding business entity nodes with an association relationship; comparing the field contents between the matching fields, and when they are inconsistent, sending the error association notification information of the corresponding business entity node to the client.

[0010] In one example, before obtaining the business transaction probability information of the upstream and downstream relationship between different business entities according to the business entity feature information and the pre-constructed business transaction probability prediction network model, the method further includes: extracting features from the sample image data of the financial system to obtain sample image data features; marking the upstream and downstream business relationships between the sample image data features; using the sample image data features as samples and the upstream and downstream business relationships between the sample image data features as sample labels to train the graph convolutional network model to obtain the business transaction probability prediction model.

[0011] In one example, marking the upstream and downstream business relationships between the sample image data features specifically includes: matching the sample image data features according to the business rule library; preliminarily marking the sample image data features with a matching relationship as sample image data features with an upstream and downstream business relationship; sending the sample image data features without a matching relationship and the sample image data features with a preliminary mark to the client to obtain the finally marked sample image data features.

[0012] In one example, before extracting the feature information of the business entities of the financial system from the image data set of the financial system, the method further includes: obtaining image files and image file metadata from the financial system, where the image files include invoices and contracts; performing standard formatting processing on the image files to obtain standard image files; and associating and integrating the image file metadata and the standard image files to obtain an image data set.

[0013] In one example, extracting the feature information of the business entities of the financial system from the image data set of the financial system specifically includes: performing image recognition on the image data set to obtain initial business entity feature information; performing spelling check on the initial business entity feature information to obtain initial corrected entity feature information; performing format structuring processing on the initial corrected entity feature information to distinguish the feature information of the business entity in each part; and performing feature cleaning and encoding on the feature information of the business entity in each part to obtain the feature information of the business entities of the financial system.

[0014] On the other hand, an embodiment of the present application provides an image upstream and downstream association graph generation device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute an image upstream and downstream association graph generation method as described in any one of the above.

[0015] On the other hand, an embodiment of the present application provides a non-volatile computer storage medium for generating an image upstream and downstream association graph, storing computer-executable instructions, and the computer-executable instructions can execute an image upstream and downstream association graph generation method as described in any one of the above.

[0016] The above at least one technical solution adopted in the embodiment of the present application can achieve the following beneficial effects: Deeply mining business relationships: By extracting the feature information of the business entities from the image data set of the financial system, the feature information of the business entities can be mined from a large amount of image data. These feature information are the key attributes of the business entities and provide a basis for subsequent analysis.

[0017] On this basis, using the business transaction probability prediction network model to calculate the probability of the existence of upstream and downstream relationships between different business entities realizes the deep mining of the potential relationships between business entities, breaking through the limitation of the traditional method that can only obtain surface relationships.

[0018] Even if the sources of financial image data are diverse and the formats are inconsistent, effective integration can still be carried out. In the existing technology, it is difficult to effectively integrate financial image data due to its diverse sources and inconsistent formats.

[0019] Precisely determine the association relationship: Based on the probability information of business transactions, it is not only possible to determine whether there is an edge (i.e., an association relationship) between different business entity nodes, but also to clarify the edge attributes (such as the degree of closeness of the association, etc.). This precise way of determining the association relationship can more accurately reflect the actual business transactions between business entities compared to simple qualitative judgments, providing a more reliable basis for the analysis and decision-making of the financial system.

[0020] Mine the precise association graph: The upstream and downstream association graph of images generated based on business entity nodes and their association relationships visually displays the upstream and downstream relationships between various business entities in the financial system in a graphical manner. This visual presentation enables financial personnel and decision-makers to quickly and clearly understand the business architecture and processes of the entire financial system, facilitating the discovery of potential risk points and business opportunities.

[0021] Risk warning and prevention and control: By analyzing the relationships between business entities in the association graph, some abnormal associations or potential risk factors can be discovered in a timely manner. For example, if a certain business entity has a close upstream and downstream relationship with a high-risk business entity, the system can issue a warning in a timely manner to remind relevant personnel to conduct further review and risk prevention and control, reducing financial risks.

[0022] Adapt to complex business environments: It can handle the complex and changeable business relationships in the financial system. Whether it is the multi-level business transactions within a large enterprise group or cross-industry and cross-regional business cooperation, it can accurately construct an association graph through feature extraction and probability prediction, adapting to financial systems of different scales and complexities. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the present application, the following will detail some embodiments of the present application in conjunction with the drawings, where: Figure 1 is a schematic flowchart of a method for generating an upstream and downstream association graph of images provided by an embodiment of the present application; Figure 2 is a schematic structural diagram of a device for generating an upstream and downstream association graph of images provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application without creative efforts fall within the scope of protection of this application.

[0025] The following will refer to the drawings to elaborate on some embodiments of this application in detail.

[0026] Figure 1 It is a schematic flowchart of a method for generating an upstream and downstream association map of images provided by an embodiment of this application. Some input parameters or intermediate results in this process allow manual intervention and adjustment to help improve accuracy.

[0027] The implementation of the analysis method involved in the embodiments of this application can be a terminal device or a server, and this application does not make special restrictions on this. For the convenience of understanding and description, the following embodiments will be described in detail using a server as an example.

[0028] It should be noted that this server can be a single device or a system composed of multiple devices, that is, a distributed server, and this application does not make specific limitations on this.

[0029] Figure 1 The process in S101: Extract features from the image data set of the financial system to obtain the business entity feature information of the financial system.

[0030] In some embodiments of this application, the process of obtaining the image data set of the financial system is as follows: First, obtain image files and image file metadata from the financial system. The image files include invoices and contracts. Then, perform standard formatting processing on the image files to obtain standard image files. Then, associate and integrate the image file metadata and the standard image files to obtain an image data set.

[0031] It should be noted that the image files can also include table files.

[0032] For example, to obtain image files of function modules such as online reporting, finance, taxation, contracts, and supply chain from an enterprise ERP system, API functions can be used to obtain relevant data at a predetermined time interval or business trigger event (such as new invoice generation, contract signing, etc.). This method can achieve real-time or near-real-time data collection and ensure the accuracy and integrity of the data.

[0033] Collect the image files related to employees' reimbursement from the online reporting module, including expense reimbursement application forms and invoices. The invoices include travel expense invoices (such as air tickets, hotel invoices, etc.), office supplies purchase invoices, etc. At the same time, also collect the metadata of the expense reimbursement application forms and invoices, such as reimbursement date, reimburser's department, reimbursement amount and other information.

[0034] Collect the procurement order information files from the supply chain module, which include supplier names, specifications, quantities, procurement prices of the procured goods or services, etc. At the same time, obtain the contract images related to the procurement orders, and these contracts clarify important information such as the rights and obligations of both parties to the procurement, delivery terms, payment conditions, etc.

[0035] Collect relevant image data such as sales orders and delivery orders, as well as the corresponding sales contract images from the sales business. These data reflect the business transactions between the enterprise and its customers.

[0036] Collect the accounting voucher image files from the financial module, including balance sheet, income statement files, etc. These data reflect the overall financial situation of the enterprise.

[0037] Mainly collect the image data related to tax declarations from the tax module, including image files such as value-added tax return forms and enterprise income tax return forms.

[0038] Store the collected image data in a data warehouse according to a unified data structure, format the image files in different formats, and convert all image files into electronic scans or image formats with a unified resolution; at the same time, associate and integrate the relevant metadata with the image files to form a unified image data set for subsequent processing.

[0039] For example, the invoice feature information includes the following: The invoice number is the unique identifier of an invoice, which is unique and deterministic, and can be directly used as a feature to distinguish different invoice operations. The invoice date reflects the chronological order of the operations corresponding to the invoice, which is very important for analyzing the timeliness and periodicity of operations. The amount is one of the core elements of an invoice, which is related to the cost and revenue calculation of an enterprise; in addition to the numerical value of the amount itself, features such as the range of the amount and the fluctuation of the amount can also be extracted. The information of the purchaser and the seller includes the name, taxpayer identification number, etc. of the enterprise. This information can be used to identify business association relationships, such as judging whether there is a long-term cooperation relationship with a specific supplier or customer; business logic features, the type of goods or services on the invoice reflects the scope of business activities of the enterprise. By classifying and counting the types of goods or services, the main business areas of the enterprise and the proportion of different business types can be understood; the applicable tax rate of the invoice is also an important feature. Goods or services with different tax rates reflect the nature of the enterprise's business (such as whether it involves tax-exempt operations, special industry operations with high tax rates, etc.), and are of great significance for tax calculation and compliance inspection.

[0040] On the other hand, the contract feature information includes the following: The contract number, like the invoice number, is the unique identifier of the contract and is used to accurately identify different contract operations; the information of the signing parties includes the names, addresses, contact information, etc. of both parties. This information helps to determine the business entities involved in the contract and their association relationships; the contract amount is the core economic indicator of the contract and has an association relationship with the invoice amount. By analyzing the size, change trend of the contract amount and its comparison with the actual executed amount (which can be reflected by invoices), the deviation and potential risks in contract execution can be found; the contract term clarifies the effective time range of the contract, which is of great significance for analyzing the sustainability and stability of operations. For example, long-term contracts may involve more complex financial arrangements and risk response measures. Business content features, the types of goods or services involved in the contract are similar to those of invoices, but may differ in the degree of detail of the description. By analyzing the types of goods or services in the contract, the business cooperation content and strategic direction of the enterprise can be deeply understood; the payment terms in the contract (such as prepayment, installment payment, final payment, etc.) are an important feature. The payment terms affect the cash flow and fund arrangement of the enterprise and have a logical association with the invoice issuance time and amount.

[0041] In some embodiments of the present application, the process of feature extraction from the image data set of the financial system is as follows: First, perform image recognition on the image data set to obtain the initial business entity feature information.

[0042] For example, for invoices and contract images in the form of scanned documents, OCR technology is used for text recognition. OCR technology can convert the text in the images into editable text information, thereby improving the accuracy and efficiency of feature extraction.

[0043] Then, perform a spelling check on the initial business entity feature information to obtain the initial corrected entity feature information.

[0044] For example, during the text recognition process, a rule-based method needs to be adopted for the recognition results. For example, according to the format specifications of invoices or contracts, check whether the recognized content is reasonable, and at the same time use the spelling check algorithm in machine learning to perform grammar and semantic checks on the recognized text.

[0045] Then, perform format structuring on the initial corrected entity feature information to distinguish the feature information of the business entity in each part.

[0046] For example, divide the invoice according to a fixed format (such as the header, item details, total column, etc.) to determine the corresponding feature content for each part. For the contract, classify and organize the contract terms according to different categories (such as rights and obligations terms, payment terms, liability for breach of contract terms, etc.) to better extract and analyze features.

[0047] Finally, perform feature cleaning and encoding on the feature information of the business entity in each part to obtain the business entity feature information of the financial system.

[0048] For example, remove duplicate feature values: For example, during the data collection process, there may be duplicate invoice or contract records due to data entry errors or system failures. It is necessary to identify and delete the duplicate data by comparing key features (such as invoice numbers or contract numbers).

[0049] Clear invalid feature values: For example, some data that is clearly illogical, such as the invoicing date in the future or the contract amount being negative, needs to be corrected or deleted.

[0050] Handle missing values: For cases where certain features are missing, according to the business logic, methods such as filling (such as filling numerical features with the mean or median) or deletion (if the proportion of missing values is too high and cannot be reasonably filled) can be used for processing.

[0051] For the feature encoding method, for text-type features such as product or service type, contract type, etc., one-hot encoding is used to convert them into numerical features. For example, "purchase contract" is encoded as [1, 0, 0], "sales contract" is encoded as [0, 1, 0], "service contract" is encoded as [0, 0, 1], etc. This can convert text features into a numerical form that machine learning algorithms can handle, while avoiding the complexity and errors brought by the algorithm's direct processing of the text itself.

[0052] S102: According to the business entity feature information and the pre-constructed business transaction probability prediction network model, obtain the business transaction probability information indicating the upstream and downstream relationship between different business entities.

[0053] Among them, the trained prediction model analyzes and processes the entire financial image dataset. For each pair of possible business entities (such as an invoice and a contract), the model calculates the probability of the existence of an upstream and downstream business transaction between them based on the learned features and business relationship patterns. This probability reflects the confidence of the model in the relationship between the two based on data features. For example, the model may calculate that the upstream and downstream association probability between a certain invoice and a certain contract is 0.8, which means that according to the model's analysis of the data, there is an 80% possibility that the invoice and the contract have an upstream and downstream business relationship.

[0054] It should be noted that the model adopts a loss function optimization method and can also convert the probability value into a more semantic confidence representation. For example, the probability value of 0.8 - 1.0 is defined as a high-confidence association, 0.5 - 0.8 as a medium-confidence association, and 0.0 - 0.5 as a low-confidence association.

[0055] In some embodiments of the present application, it is necessary to pre-construct a business transaction probability prediction network model, and the construction process is as follows: First, extract features from the sample image data of the financial system to obtain sample image data features.

[0056] It should be noted that the feature extraction process is as described in S101.

[0057] Then, mark the upstream and downstream business relationships between the sample image data features.

[0058] Among them, the marking process is as follows: First, match the sample image data features according to the business rule library. Then, preliminarily mark the sample image data features with a matching relationship as sample image data features with an upstream and downstream business relationship. Then, send the sample image data features without a matching relationship and the sample image data features with a preliminary mark to the client to obtain the finally marked sample image data features.

[0059] It should be noted that marking sample data is crucial for model training. Due to the complexity of financial business relationships, it is difficult to obtain completely accurate marked data. Therefore, a combination of manual marking and partial automatic marking is adopted. Financial experts or personnel familiar with business processes mark some typical invoice - contract association relationships. For example, for some clear procurement businesses, mark that a certain invoice is issued for the goods or services purchased for a specific contract. Such marking has relatively high accuracy. At the same time, some known business rules are used for automatic marking. For instance, if the invoice issuer is the same as the supplier name in the contract and the invoice date is within the contract validity period, it can be initially marked as associated. However, there are certain errors in this automatic marking and further manual review is required.

[0060] After the marking is completed, the sample image data features are used as sample inputs, and the upstream and downstream business relationships between the sample image data features are used as sample labels to train the graph convolutional network model to obtain a business transaction probability prediction model.

[0061] It should be noted that a graph neural network can be used to construct a prediction model because there are naturally complex graph structure relationships among business entities (such as invoices, contracts, etc.) in the financial system. GNN can directly learn on graph structure data and can effectively capture the mutual relationship information between nodes (business entities). Different from traditional neural networks, GNN can spread information between nodes in the graph through a message passing mechanism. For example, when dealing with the association between invoices and contracts, it can pass the feature information of invoices to the relevant contract nodes and can also receive information from contract nodes, thereby better learning the association patterns between them.

[0062] Use multiple layers of Graph Convolutional Network (GCN) layers as the network - side structure. In each layer, the features of nodes are updated according to the features of their neighbor nodes. The number of neurons in the hidden layer can be set according to the scale and complexity of the data. If the volume of financial business data to be processed is large and the relationships are complex, more neurons are required to fully learn the relationships between features. At the same time, add the ReLU activation function to introduce non - linear factors and improve the expressive ability of the model.

[0063] For example, the labeled sample data is divided into a training set, a validation set, and a test set. The training set accounts for 70 - 80% of the total samples, the validation set accounts for 10 - 15%, and the test set accounts for 10 - 15%. During the training process, the model learns based on the input feature vectors (entity features such as invoices and contracts after feature engineering) and the labeled upstream and downstream association relationships (e.g., 1 indicates an association, 0 indicates no association), and uses the cross-entropy loss function to measure the difference between the model's prediction results and the true labels. Hyperparameter tuning is used to optimize the model. Hyperparameters such as the learning rate, the number of layers of the graph neural network, the number of neurons in each layer, and the number of message passing steps can be adjusted. The performance of the model under different hyperparameter combinations is evaluated using the cross-validation method. K-fold cross-validation (e.g., K = 5) is used, where the training set is divided into 5 subsets. Each time, 4 subsets are used as training data and 1 subset is used as validation data. This is repeated 5 times, and the average performance metrics (such as accuracy, recall, etc.) are calculated. The hyperparameter combination with the best performance is selected to finally determine the structure of the model.

[0064] S103: Determine the association relationship between different business entity nodes according to the business transaction probability information; the association relationship includes whether there is an edge and edge attributes.

[0065] In some embodiments of the present application, when the business transaction probability between business entity nodes is higher than a preset probability threshold, it is determined that there is an edge between the business entity nodes.

[0066] Among them, the edge is used to represent the upstream and downstream business relationship between business entity nodes.

[0067] Determine the business transaction probability between business entity nodes as the weight of the edge.

[0068] When the business transaction probability between business entity nodes is lower than or equal to the preset probability threshold, it is determined that there is no edge between the business entity nodes.

[0069] That is to say, in the association graph, the nodes represent different financial business entities, which include invoices, contracts, online reporting business records, supply chain orders, financial accounting vouchers, tax declaration records, etc. Each node contains relevant feature information after feature engineering. For example, the invoice node contains information such as invoice number, invoicing date, amount, purchaser, and seller as part of its feature vector. This feature information not only helps the model calculate the relationship between business entities but also provides more information about the nodes when analyzing and visualizing the graph subsequently.

[0070] Edges represent the upstream and downstream relationships between business entities. The existence and attributes of the edges are determined based on the association probability or confidence level output by the model. If the association probability or confidence level is higher than a pre-set threshold (e.g., 0.7), an edge is established between the corresponding two business entity nodes. The weight of the edge can be determined according to the confidence level of the association. For example, if the confidence level is 0.8, the weight of the edge can be set to 0.8. Edges with higher weights indicate closer upstream and downstream relationships between business entities, while edges with lower weights may indicate relatively weaker or less certain relationships.

[0071] S104: Generate an upstream and downstream association graph of the financial system's images based on business entity nodes and the association relationships between them.

[0072] That is to say, in this scenario of upstream and downstream associations of financial images, nodes represent different financial business entities; edges represent the upstream and downstream association relationships between business entities. For example, if an invoice is issued for a contract procurement business, then there is an edge between the invoice node and the contract node; the attribute of the edge can be set as the confidence level of the association (initially, a default value can be set according to manual marking or business logic), and the attribute of the node is the feature vector after feature engineering.

[0073] Among them, all business entity nodes and the edges between them are integrated according to the calculated relationships to form a complete upstream and downstream association graph of financial images. This graph is a complex network structure that contains various types of business entities and the intricate upstream and downstream relationships between them; then it is visually presented. For the convenience of understanding and analysis, the association graph needs to be visually presented, and D3.js is used to display the graph.

[0074] In the visual graph, different types of business entities can be represented by nodes of different colors or shapes, and the thickness of the edges can be determined according to the weights of the edges (i.e., the confidence levels of the associations). In this way, financial personnel and relevant management personnel can intuitively see the upstream and downstream relationship network of business transactions within the entire financial system, quickly identify key business nodes, core business processes, and areas of association relationships that may pose risks.

[0075] It should be noted that for business entity nodes with an association relationship, some characteristic information should be corresponding. For example, the amounts of the contract and the invoice. Therefore, when there is no corresponding relationship, it may be that the association relationship is misidentified, or the invoice or the contract has a false risk.

[0076] Therefore, perform field matching on the characteristic information of the corresponding business entity nodes with an association relationship.

[0077] Compare the field content between the matching fields. When there is an inconsistency, send the error association notification information of the corresponding business entity node to the client to identify the reason for the inconsistency.

[0078] It should be noted that during the construction process, there may be some isolated nodes. These nodes may be generated due to data loss, virtual data, or the model's failure to accurately identify their relationships. These isolated nodes can be analyzed separately to determine whether further data needs to be supplemented or the model needs to be adjusted.

[0079] In some implementations of the present application, when new image data (such as invoices, contracts, etc.) is generated, it will trigger an event for updating the knowledge graph. For example, when an enterprise newly signs a procurement contract and at the same time generates a corresponding invoice in the financial system. These new data need to be incorporated into the existing associated knowledge graph system. The new data first goes through the same processing process as the original data, including steps such as data collection and feature engineering. In the data collection stage, new data and its related metadata are obtained from the corresponding financial modules (such as collecting contract information from the contract management module and invoice information from the invoice management system). Then, feature engineering is carried out. Features such as invoice number, invoicing date, and amount are extracted from the new invoice, and features such as contract number, contracting party, and contract amount are extracted from the new contract.

[0080] Based on this, obtain the new business entity feature information of the financial system. Then, input the new business entity feature information and the business entity feature information into the business transaction probability prediction network model to obtain the business transaction probability information indicating an upstream and downstream relationship between the business entity and the new business entity, so as to determine the association relationship between the new business entity node and the business entity node. According to the association relationship between the new business entity node and the business entity node, add the new business entity node to the image upstream and downstream association knowledge graph.

[0081] Among them, if the new data has an upstream and downstream relationship with the existing nodes, an edge is established or updated between the corresponding nodes. For example, if the new invoice has an upstream and downstream association with an existing contract node and the association confidence level is higher than the threshold, an edge is established between the invoice node and the contract node or the weight of the existing edge is updated (if there was an association before but the confidence level has changed).

[0082] It should be noted that in addition to the generation of new data, when the original image data changes, the atlas also needs to be updated. For example, if the amount of an invoice is adjusted for some reason, or the payment terms in a contract are modified, for data change situations, feature engineering also needs to be re-performed, and then the association relationship is recalculated through the model. If the change results in a change in the association relationship with other business entities (such as no association originally becomes associated, or the association confidence level drops below the threshold), then the relationships of the nodes and edges in the association atlas are updated accordingly.

[0083] In some embodiments of the present application, the atlas needs to be verified regularly, and the verification process is as follows: Randomly select some business association relationships from the association atlas for manual inspection regularly. For example, select several invoice - contract association relationships from the atlas, and have financial experts or personnel familiar with the business process check according to the actual financial documents (original invoices, contract texts, etc.). The content of the manual random inspection includes whether the association relationship is correct (such as whether the invoice was actually issued for the business corresponding to the contract), whether the association confidence level is reasonable (for example, whether a high-confidence association conforms to the actual business logic), etc. If it is found that there are differences between the results of the manual random inspection and the association relationships in the atlas, the reasons need to be further analyzed.

[0084] Conduct a review of some upstream and downstream business associations based on the actual business logic. For example, review whether the association between the purchase contract and the corresponding goods receipt note and invoice in the supply chain business conforms to the enterprise's purchase process. If the purchase contract stipulates a certain delivery quantity and time, then the quantity and time of the goods receipt note and invoice should match it. If the association relationship in the atlas shows a mismatch, there may be an error.

[0085] According to the results of the manual random inspection and the business logic review, if it is found that there are errors in the atlas, the model needs to be adjusted. If the model misjudges due to new data types or special situations, the model can be retrained, new training data can be supplemented, or the structure and parameters of the model can be adjusted to improve the accuracy of the atlas.

[0086] It should be noted that although the embodiments of the present application are described with reference to Figure 1 to introduce and explain steps S101 to S104 in sequence, this does not mean that steps S101 to S104 must be executed in a strict order. The reason why the embodiments of the present application introduce and explain steps S101 to S104 in the order shown in Figure 1 is to facilitate those skilled in the art to understand the technical solution of the embodiments of the present application. In other words, in the embodiments of the present application, the order between steps S101 to S104 can be appropriately adjusted according to actual needs.

[0087] By Figure 1 the method of, deeply mining business relationships: By extracting features from the image data set of the financial system, the feature information of business entities can be mined from a large amount of image data. These feature information are the key attributes of business entities and provide a basis for subsequent analysis.

[0088] On this basis, using the business transaction probability prediction network model to calculate the probability of the existence of upstream and downstream relationships between different business entities, the in-depth mining of potential relationships between business entities is realized, breaking through the limitation that traditional methods can only obtain surface relationships.

[0089] Even if the sources of financial image data are diverse and the formats are inconsistent, effective integration can still be carried out. In the existing technology, it is difficult to carry out effective integration for the diverse sources and inconsistent formats of financial image data.

[0090] Precisely determining the association relationship: According to the business transaction probability information, not only can it be determined whether there is an edge (i.e., whether there is an association relationship) between different business entity nodes, but also the edge attributes (such as the degree of closeness of the association, etc.) can be clarified. This precise way of determining the association relationship can more accurately reflect the actual business transactions between business entities compared with simple qualitative judgments, providing a more reliable basis for the analysis and decision-making of the financial system.

[0091] Mining the precise association graph: The image upstream and downstream association graph generated based on business entity nodes and their association relationships visually shows the upstream and downstream relationships between business entities in the financial system in a graphical way. This visual presentation method enables financial personnel and decision-makers to quickly and clearly understand the business architecture and processes of the entire financial system, facilitating the discovery of potential risk points and business opportunities.

[0092] Risk warning and prevention: By analyzing the relationships between business entities in the association graph, some abnormal associations or potential risk factors can be discovered in a timely manner. For example, if a business entity has a close upstream and downstream relationship with a high-risk business entity, the system can issue a warning in a timely manner to remind relevant personnel to conduct further reviews and risk prevention and control to reduce financial risks.

[0093] Adapting to complex business environments: It can handle the complex and changeable business relationships in the financial system. Whether it is the multi-level business transactions within a large enterprise group or cross-industry and cross-regional business cooperation, it can accurately construct an association graph through feature extraction and probability prediction to adapt to financial systems of different scales and complexities.

[0094] That is to say, a high degree of automation in image data processing is achieved. Machine learning is used to automatically identify and process the characteristics of financial image data, greatly improving the processing speed and accuracy and reducing the dependence on manual operations. Deep correlation mining is performed on the existing financial image data of the enterprise. Through machine learning technology, the system can discover the deep correlations and hidden patterns among the image data and obtain the probability information of business transactions with upstream and downstream relationships.

[0095] Then, machine learning is combined with the association graph technology. Deep learning technology is used to enhance the processing and analysis capabilities of image data, and then data visualization and association analysis are performed through the association graph technology. Finally, the constructed association graph is used to realize real-time monitoring and early warning of financial risks. The deep mining and analysis of financial image data are realized, the risk control ability of the financial system is enhanced, potential financial risks are discovered and warned in time, and the enterprise is helped to respond quickly and take measures.

[0096] In summary, this application is divided into five modules as follows: (1) Data acquisition module: Collect relevant data containing image data (such as invoices, contracts, etc.) from various modules of the financial system (online reporting, supply chain, finance, taxation, contracts, accounts receivable and payable, etc.). In addition to the information of the image file itself, these data also include relevant metadata, such as the creation time of the image, the identifier of the associated business, etc. The collected data is formatted according to a unified standard for subsequent analysis and processing.

[0097] (2) Feature engineering module: Extract features from the image data. Based on image recognition technology, text recognition and structured processing are performed on documents such as invoices and contracts to further ensure the accuracy of feature extraction. The extracted features are cleaned and encoded, and some text features are converted into numerical features for the processing of machine learning algorithms. For example, for invoices, features such as invoice number, invoicing date, amount, purchaser and seller information are extracted; for contracts, features such as contract number, signing parties, contract amount, contract term, and types of goods or services involved are extracted.

[0098] (3) Model construction and training module: Select appropriate machine learning models, such as deep neural networks (e.g., multi-layer perceptrons), graph neural networks, etc. Graph neural networks can better handle the graph structure relationships in the data and are very suitable for constructing association graphs.

[0099] The model is trained using labeled sample data (which can be some manually labeled business transaction relationships as the initial training set). During the training process, the model learns the mapping relationship between the image data features and the upstream and downstream business relationships. The performance of the model is optimized by means of cross-validation, adjusting hyperparameters, etc., to improve the model's ability to accurately associate different financial business upstream and downstream relationships.

[0100] (4) Association graph construction module: Use the trained model to perform association analysis on the entire financial image data. The model outputs the probability or confidence level of upstream and downstream business transactions between different image data.

[0101] According to the association results output by the model, construct an upstream and downstream association graph of financial images. The nodes in the graph represent different financial business entities (such as the business corresponding to invoices, the business corresponding to contracts, etc.), the edges represent the upstream and downstream relationships between the businesses, and the weights of the edges can be determined according to the confidence level of the association.

[0102] (5) Graph update and verification module: When new image data is generated or the original image data changes, update the association graph in a timely manner. Re-perform steps such as feature extraction and model prediction to ensure that the graph reflects the latest business association relationships. Verify the association graph through methods such as manual sampling inspection and actual business logic review of some upstream and downstream businesses, and adjust the model in a timely manner if errors are found.

[0103] Based on the same idea, some embodiments of the present application also provide devices and non-volatile computer storage media corresponding to the above methods.

[0104] Figure 2 FIG. is a schematic structural diagram of an apparatus for generating an upstream and downstream association graph of images provided by an embodiment of the present application, including: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a method for generating an upstream and downstream association graph of images according to any one of the above.

[0105] A non-volatile computer storage medium for generating an upstream and downstream association graph of images provided by some embodiments of the present application stores computer-executable instructions, and the computer-executable instructions can execute a method for generating an upstream and downstream association graph of images according to any one of the above.

[0106] The various embodiments in the present application are all described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other, and the key points of each embodiment are the differences from other embodiments. In particular, for the device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0107] The devices, media, and methods provided by the embodiments of this application correspond one-to-one. Therefore, the devices and media also have beneficial technical effects similar to those of their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be elaborated here.

[0108] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.

[0109] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the specified functions in Figure 1 one or more of the flows Figure 1 or blocks or combinations of blocks.

[0110] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the specified functions in Figure 1 one or more of the flows Figure 1 or blocks or combinations of blocks.

[0111] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are performed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the specified functions in Figure 1 one or more of the flows Figure 1 or blocks or combinations of blocks.

[0112] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0113] The memory may include non - permanent memory in the form of computer - readable media, random access memory (RAM) and / or non - volatile memory such as read - only memory (ROM) or flash RAM. The memory is an example of computer - readable media.

[0114] Computer - readable media includes permanent and non - permanent, removable and non - removable media which can store information by any method or technology. The information can be computer - readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase - change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read - only memory (ROM), electrically erasable programmable read - only memory (EEPROM), flash memory or other memory technologies, compact disc read - only memory (CD - ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices or any other non - transitory media that can be used to store information accessible by a computing device. As defined herein, computer - readable media does not include transitory computer - readable media such as modulated data signals and carrier waves.

[0115] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non - exclusive inclusion, such that a process, method, article or apparatus comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or apparatus. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising the element.

[0116] The above - mentioned are only embodiments of the present application and are not used to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the technical principle of the present application shall fall within the protection scope of the present application.

Claims

1. A method for generating an upstream and downstream association map of images, characterized in that, The method includes: Performing feature extraction on the image data set of the financial system to obtain the business entity feature information of the financial system; According to the business entity feature information and the pre-constructed business transaction probability prediction network model, obtaining the business transaction probability information of the upstream and downstream relationships between different business entities; Determining the association relationship between different business entity nodes according to the business transaction probability information; the association relationship includes whether there is an edge and edge attributes; Generating an upstream and downstream association map of the financial system according to the business entity nodes and the association relationship between the business entity nodes.

2. The method according to claim 1, characterized in that, The determining the association relationship between different business entity nodes according to the business transaction probability information specifically includes: When the business transaction probability between business entity nodes is higher than a preset probability threshold, determining that there is an edge between the business entity nodes; the edge is used to represent the upstream and downstream business relationship between the business entity nodes; Determining the business transaction probability between business entity nodes as the weight of the edge; When the business transaction probability between business entity nodes is lower than or equal to the preset probability threshold, determining that there is no edge between the business entity nodes.

3. The method according to claim 1, characterized in that After generating the upstream and downstream association map of the financial system according to the business entity nodes and the association relationship between the business entity nodes, the method further includes: Obtaining the new business entity feature information of the financial system; Inputting the new business entity feature information and the business entity feature information into the business transaction probability prediction network model to obtain the business transaction probability information of the upstream and downstream relationships between the business entity and the new business entity, so as to determine the association relationship between the new business entity node and the business entity node; Adding the new business entity node to the upstream and downstream association map according to the association relationship between the new business entity node and the business entity node.

4. The method according to claim 1, wherein After generating the upstream and downstream association map of the financial system according to the business entity nodes and the association relationship between the business entity nodes, the method further includes: Performing field matching on the feature information of the corresponding business entity nodes with an association relationship; Comparing the field contents between the matching fields, and when they are inconsistent, sending the error association notification information of the corresponding business entity node to the client.

5. The method according to claim 1, characterized in that, Before obtaining the business transaction probability information of the upstream and downstream relationships between different business entities according to the business entity feature information and the pre-constructed business transaction probability prediction network model, the method further includes: Performing feature extraction on the sample image data of the financial system to obtain sample image data features; Marking the upstream and downstream business relationships between the sample image data features; Taking the sample image data features as samples and the upstream and downstream business relationships between the sample image data features as sample labels to train a graph convolutional network model to obtain a business transaction probability prediction model.

6. The method according to claim 5, characterized in that The marking the upstream and downstream business relationships between the sample image data features specifically includes: Matching the sample image data features according to the business rule library; Preliminarily marking the sample image data features with a matching relationship as sample image data features with an upstream and downstream business relationship. Send the sample image data features without matching relationships and the sample image data features with preliminary markings to the client to obtain the sample image data features with final markings.

7. The method according to claim 1, wherein Before extracting the feature information of the business entities of the financial system from the image data set of the financial system, the method further includes: Obtain image files and image file metadata from the financial system, where the image files include invoices and contracts; Perform standard formatting processing on the image files to obtain standard image files; Associate and integrate the image file metadata and the standard image files to obtain an image data set.

8. The method according to claim 1, wherein The extracting the feature information of the business entities of the financial system from the image data set of the financial system specifically includes: Perform image recognition on the image data set to obtain initial business entity feature information; Perform spelling checks on the initial business entity feature information to obtain initial corrected entity feature information; Perform format structuring processing on the initial corrected entity feature information to distinguish the feature information of business entities in each part; Perform feature cleaning and encoding on the feature information of business entities in each part to obtain the feature information of business entities of the financial system.

9. An upstream and downstream correlation map generation device for images, characterized in that, Includes: At least one processor; And, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute an image upstream and downstream association graph generation method according to any one of claims 1-8 above.

10. A non-volatile computer storage medium for generating an upstream and downstream associated image map, storing computer-executable instructions, characterized in that, The computer-executable instructions can execute an image upstream and downstream association graph generation method according to any one of claims 1-8 above.

Images