Invoice entry classification recognition method and recognition system

Abstract

The present invention relates to the field of invoice data management, and discloses an invoice entry classification identification method and identification system, including: a data initial processing module piles up the invoice data to be classified, obtains multiple invoice data piles, and classifies them according to classification patterns, the classification patterns include invoice classification identification based on invoice content and invoice classification identification based on invoice amount, identifies invoice classification identification features, and completes invoice classification identification. The present invention significantly improves the processing speed of invoice classification through data pile-up and multi-container parallel processing mechanism, and is particularly suitable for processing large-scale invoice data. The dynamic classification feature extraction and merging strategy is adopted to effectively cope with the diversity and complexity of invoice data and improve the accuracy of classification.

Description

Technical Field

[0001] The present invention relates to the field of invoice management, and in particular to an invoice entry classification recognition method and recognition system. Background Art

[0002] With the continuous expansion of corporate business and the popularization of e-commerce, the number of invoices generated daily has increased dramatically, bringing huge challenges to corporate financial management. Traditional invoice classification work mainly relies on manual work, which is not only time-consuming and labor-intensive, but also prone to errors, and it is difficult to meet the needs of modern enterprises for efficient and accurate financial management. In recent years, although some invoice recognition software based on OCR (optical character recognition) technology has appeared on the market, most of these software can only achieve basic extraction of invoice information. In-depth classification and archiving of invoices still require manual intervention, especially when faced with complex and diverse invoice types, the classification efficiency and accuracy still need to be improved.

[0003] The existing invoice classification technology mainly has the following problems: Inconsistent classification standards: Different companies may have different classification standards for invoices, such as by product category, service item, tax amount, etc. This requires the classification system to be highly flexible and configurable.

[0004] Low processing efficiency: Faced with massive invoice data, single thread or simple parallel processing can hardly meet the needs of efficient processing, especially when the classification rules are complex, the processing speed drops significantly.

[0005] Insufficient classification accuracy: Due to the diversity of invoice formats, incomplete information and OCR recognition errors, the existing system cannot guarantee high-accuracy automatic classification.

[0006] Poor scalability: As the business of an enterprise changes, the invoice classification rules may need to be adjusted. Existing systems often find it difficult to quickly adapt to such changes, requiring a long development cycle and cost. Summary of the invention

[0007] The purpose of the present invention is to overcome the shortcomings of the prior art and provide an invoice entry classification identification method, comprising the following steps: Step 1: The data initial processing module divides the invoice data to be classified into piles to obtain multiple invoice data piles, and classifies them according to the classification mode. If the classification mode is invoice classification recognition based on invoice content, then go to step 2; if the classification mode is invoice classification recognition based on invoice amount, then go to step 5; Step 2: the invoice classification recognition module generates a first recognition container, randomly selects an invoice data pile and sends it to the first recognition container, the first recognition container obtains the classification data of the selected invoice data pile, and obtains the first invoice classification feature; the invoice classification recognition module generates the same number of recognition containers according to the number of the remaining invoice data piles, sends each invoice data pile to each recognition container, and each recognition container is connected to the first recognition container respectively; Step 3: According to the first invoice classification feature, each identification container classifies the invoices respectively to obtain classification results. According to the classification results of each identification container, the invoice classification completeness is obtained respectively. If the invoice classification is complete, the identification container classification is completed, and the first invoice classification identification feature is the invoice classification identification feature, and the process proceeds to step 6. If there is an identification container with incomplete invoice classification, the process proceeds to step 4. Step 4: the first identification container classifies the remaining invoice data after classification in the identification containers with incomplete invoice classification, and obtains a second classification feature; the first classification feature and the second classification feature constitute the invoice classification identification feature, and then proceeds to step 6; Step 5: The invoice classification recognition module obtains the first invoice classification amount range, generates the same number of recognition containers according to the number of invoice data piles, and classifies each recognition container according to the invoice classification amount range to obtain a classification result. According to the classification results of each recognition container, the invoice classification completeness is obtained respectively. If the invoice classification is complete, the classification of the recognition container is completed, and the process proceeds to step 6; if there are recognition containers with incomplete invoice classification, the second invoice amount range of the remaining invoice data after classification in each invoice classification recognition container is obtained. The second invoice amount range and the first invoice classification amount range constitute the invoice classification recognition feature, and the process proceeds to step 6; Step six, obtain the invoice classification identification features and complete the invoice classification identification.

[0008] Furthermore, the data initial processing module divides the invoice data to be classified into piles to obtain multiple invoice data piles, including: The data initialization module piles the invoice data to be classified according to the set data size; the data size is the number of invoices or the invoicing cycle.

[0009] Furthermore, the first identification container obtains the classification data of the selected invoice data pile to obtain the first invoice classification feature, including: The first identification container obtains the invoice types of all invoices in the selected invoice data pile, and all invoice types constitute a first invoice classification feature.

[0010] Furthermore, the classification completeness of the invoices is obtained according to the classification results of each identification container, including: Each identification container classifies the invoices in the container according to the first invoice classification feature. After the classification is completed, if there is no unclassified invoice data in the container, the invoice classification is complete; if there is unclassified invoice data in the container, the invoice classification is incomplete.

[0011] Furthermore, the first identification container classifies the remaining invoice data after classification in the identification containers with incomplete invoice classification, and obtains the second classification features, including: The first identification container classifies and identifies the remaining invoice data to obtain classification information of the remaining invoice data, and the classification information of the remaining invoice data is the second classification feature.

[0012] Furthermore, if there is an identification container with incomplete invoice classification, the second invoice amount range of the remaining invoice data after classification in each identification container with incomplete invoice classification is obtained, including: A second invoice amount range is obtained based on the invoice amount of the remaining invoice data after classification in the identification container for each incomplete invoice classification.

[0013] An invoice entry classification identification system, characterized in that it applies the invoice entry classification identification method, comprises a cloud data server, a data initial processing module, an invoice classification identification module, a data processing module, a communication module, a data acquisition module and a data storage module; The data initial processing module, invoice classification and identification module, data acquisition module, communication module and data storage module are respectively connected to the data processing module; the cloud data server is communicatively connected to the communication module.

[0014] The beneficial effects of the present invention are: through data stacking and multi-container parallel processing mechanism, the processing speed of invoice classification is significantly improved, which is particularly suitable for processing large-scale invoice data. The dynamic classification feature extraction and merging strategy is adopted to effectively deal with the diversity and complexity of invoice data and improve the accuracy of classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The figure is a flowchart of a method for identifying the classification of invoices; Figure 2 The figure is a schematic diagram of the principle of an invoice entry classification and identification system. DETAILED DESCRIPTION

[0016] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings, but the protection scope of the present invention is not limited to the following.

[0017] The features and performance of the present invention are further described in detail below in conjunction with the embodiments.

[0018] like Figure 1 As shown, a method for identifying invoice entry classification includes the following steps: Step 1: The data initial processing module divides the invoice data to be classified into piles to obtain multiple invoice data piles, and classifies them according to the classification mode. If the classification mode is invoice classification recognition based on invoice content, then go to step 2; if the classification mode is invoice classification recognition based on invoice amount, then go to step 5; Step 2: the invoice classification recognition module generates a first recognition container, randomly selects an invoice data pile and sends it to the first recognition container, the first recognition container obtains the classification data of the selected invoice data pile, and obtains the first invoice classification feature; the invoice classification recognition module generates the same number of recognition containers according to the number of the remaining invoice data piles, sends each invoice data pile to each recognition container, and each recognition container is connected to the first recognition container respectively; Step 3: According to the first invoice classification feature, each identification container classifies the invoices respectively to obtain classification results. According to the classification results of each identification container, the invoice classification completeness is obtained respectively. If the invoice classification is complete, the identification container classification is completed, and the first invoice classification identification feature is the invoice classification identification feature, and the process proceeds to step 6. If there is an identification container with incomplete invoice classification, the process proceeds to step 4. Step 4: the first identification container classifies the remaining invoice data after classification in the identification containers with incomplete invoice classification, and obtains a second classification feature; the first classification feature and the second classification feature constitute the invoice classification identification feature, and then proceeds to step 6; Step 5: The invoice classification recognition module obtains the first invoice classification amount range, generates the same number of recognition containers according to the number of invoice data piles, and classifies each recognition container according to the invoice classification amount range to obtain a classification result. According to the classification results of each recognition container, the invoice classification completeness is obtained respectively. If the invoice classification is complete, the classification of the recognition container is completed, and the process proceeds to step 6; if there are recognition containers with incomplete invoice classification, the second invoice amount range of the remaining invoice data after classification in each invoice classification recognition container is obtained. The second invoice amount range and the first invoice classification amount range constitute the invoice classification recognition feature, and the process proceeds to step 6; Step six, obtain the invoice classification identification features and complete the invoice classification identification.

[0019] The data initial processing module divides the invoice data to be classified into piles to obtain multiple invoice data piles, including: The data initialization module piles the invoice data to be classified according to the set data size; the data size is the number of invoices or the invoicing cycle.

[0020] The first identification container obtains the classification data of the selected invoice data pile to obtain the first invoice classification feature, including: The first identification container obtains the invoice types of all invoices in the selected invoice data pile, and all invoice types constitute a first invoice classification feature.

[0021] The invoice classification uses the following algorithm: This model defines a new class "BERT_AVG" based on BERT-BASE-CHINESE, which extends the BERT-BASE-CHINESE model. This class implements a modified version of the original BERT base model, including the following customizations or improvements: Average Pooling: Implements an average pooling operation on the last hidden state of the BERT model. This is an alternative to the [CLS] token representation commonly used in sequence classification. Average pooling can provide a more robust representation by considering the average of all token embeddings instead of relying on the representation of a single token.

[0022] Custom output layer: The "Linear" layer is a custom output layer where we can adjust the hidden size and activation function when debugging the model. This provides flexibility in designing the model architecture for specific tasks.

[0023] Embedding Freezing Option: Freezing embedding layers can reduce training time and prevent overfitting when training data is limited.

[0024] Loss function: The BCEWithLogitsLoss loss used by the model can be customized for multi-label classification problems.

[0025] Configurability and flexibility: The algorithm introduces additional customization options, allowing users to specify various model parameters. At the same time, the algorithm also provides the option of directly returning logits without calculating the loss, which is useful during inference.

[0026] Custom prediction processing: The model converts logits into binary predictions and organizes them into a list of labels for each instance. This is a custom implementation that can be used for the task at hand, such as multi-label classification.

[0027] In addition, this algorithm also post-processes the output data to reduce the number of easily confused labels. Entropy is used to calculate the disorder in a system, that is, to calculate the degree of confusion in the system. For POC, the more average the probability distribution of the result, the easier it is to confuse the category of the invoice. When the entropy value of the probability distribution of the invoice data obtained by the BERT_AVG algorithm exceeds the threshold we set (40% of the maximum entropy, i.e. 0.6437), it can be considered that the category of the invoice is more easily confused. Therefore, the top five categories of probability are used as alternative sets for this type of invoice data. After such processing, invoices whose categories are not easily confused can be accurately identified, and invoices whose categories are easily confused can use the alternative set method to improve the hit rate of the category.

[0028] The invoice classification completeness is obtained according to the classification results of each identification container, including: Each identification container classifies the invoices in the container according to the first invoice classification feature. After the classification is completed, if there is no unclassified invoice data in the container, the invoice classification is complete; if there is unclassified invoice data in the container, the invoice classification is incomplete.

[0029] The first identification container classifies the remaining invoice data after classification in each incompletely classified identification container to obtain a second classification feature, including: The first identification container classifies and identifies the remaining invoice data to obtain classification information of the remaining invoice data, and the classification information of the remaining invoice data is the second classification feature.

[0030] If there is an identification container with incomplete invoice classification, the second invoice amount range of the remaining invoice data after classification in each identification container with incomplete invoice classification is obtained, including: A second invoice amount range is obtained based on the invoice amount of the remaining invoice data after classification in the identification container for each incomplete invoice classification.

[0031] like Figure 2 As shown, an invoice entry classification recognition system, applying the invoice entry classification recognition method, includes a cloud data server, a data initial processing module, an invoice classification recognition module, a data processing module, a communication module, a data collection module and a data storage module; The data initial processing module, invoice classification and identification module, data acquisition module, communication module and data storage module are respectively connected to the data processing module; the cloud data server is communicatively connected to the communication module.

[0032] Specifically, the present invention provides a method and system for identifying invoice entry classification, which effectively solves the problems existing in the prior art and improves the efficiency and accuracy of invoice classification through intelligent data stacking, multi-container parallel processing, dynamic classification feature extraction and merging and other strategies.

[0033] A method for identifying invoice entry classification includes the following steps: Step 1: Initial data processing The data initial processing module divides the invoice data to be classified into multiple piles according to the set data size (such as the number of invoices or the billing cycle). This step helps to carry out subsequent parallel processing and improve processing efficiency. According to the classification mode set by the user, choose to enter the classification recognition path based on the invoice content (step 2) or the classification recognition path based on the invoice amount (step 5).

[0034] Step 2: Preliminary classification based on invoice content The invoice classification recognition module generates a first recognition container and randomly selects an invoice data pile to send to the container. The first recognition container analyzes the selected invoice data pile, extracts the type information of all invoices, and forms a first invoice classification feature. According to the number of remaining invoice data piles, a corresponding number of recognition containers are generated, and each data pile is assigned to each container. All containers are connected to the first recognition container and share the first invoice classification feature.

[0035] Step 3: Parallel classification and completeness check Each identification container classifies the invoices in its own container according to the first invoice classification feature. After the classification is completed, it is checked whether there is unclassified invoice data in each container. If all invoices are correctly classified, it is considered that the classification task of the container is completed; otherwise, it goes to step 4.

[0036] Step 4: Supplementary classification and feature merging The first identification container reclassifies the remaining invoice data in the incompletely classified container and extracts the second classification feature. The first classification feature is combined with the second classification feature to form a final invoice classification identification feature for subsequent processing or storage.

[0037] Step 5: Classification based on invoice amount The invoice classification recognition module first determines the first invoice classification amount range, generates a corresponding number of recognition containers according to the number of invoice data piles, and each container is classified according to the set amount range.

[0038] If the invoice classification in a container is incomplete, the amount range of the remaining invoices in the container is extracted as the second invoice amount range, which is combined with the first invoice classification amount range for subsequent analysis.

[0039] Step 6: Output results Classification feature output: Regardless of whether it is content-based classification or amount-based classification, the final invoice classification recognition feature is obtained, marking the completion of the invoice classification recognition task.

[0040] The stacking strategy of the data initial processing module can be flexibly set according to the actual needs of the enterprise, such as stacking by daily, weekly, monthly and other invoicing cycles, or stacking by a fixed number of invoices (such as 1,000 invoices per stack) to balance processing efficiency and resource allocation.

[0041] First, random selection of identification containers: Randomness helps to avoid the influence of the particularity of a specific data pile on the initial classification feature extraction and improve the universal applicability of the classification features.

[0042] Parallel processing mechanism: Utilize multi-threading or distributed computing resources to enable multiple recognition containers to work simultaneously, significantly improving processing speed.

[0043] Classification integrity check: Ensure the accuracy and completeness of the classification process by comparing the total amount of invoice data before and after classification, or setting a specific classification success flag.

[0044] Dynamic feature merging: allows dynamic adjustment and optimization of classification features during the classification process, improving the system's adaptability to complex and changeable invoice data.

[0045] The present invention also provides an invoice entry classification identification system, which applies the above method and includes: Cloud data server: used to store and process large amounts of invoice data and provide a data access interface. Data initial processing module: responsible for the stacking and preprocessing of invoice data. Invoice classification and identification module: executes the core logic of invoice classification, including feature extraction, classification rule application, etc. Data processing module: coordinates the data flow between modules to ensure efficient data transmission and processing. Communication module: realizes data communication between the system and the cloud data server, and supports remote data access and update. Data collection module: responsible for collecting invoice data from various sources (such as electronic invoice platforms, scanned copies, etc.). Data storage module: stores the classified invoice data and its classification features for subsequent query and analysis.

[0046] Example 1: Identification of invoice entry categories based on invoice content A large retail enterprise needs to process thousands of invoices from different suppliers and customers every day, and the invoice content covers various types such as commodity sales, service fees, returns and refunds, etc. In order to improve the efficiency of financial management, the enterprise decided to adopt the invoice entry classification identification method and system provided by the present invention to automatically classify invoices according to the invoice content.

[0047] Specific implementation steps: Initial data processing: Data stacking: The data initial processing module sets the number of invoices in each stack to 1,000 based on the actual needs of the enterprise, and stacks all the invoice data received that day to form multiple invoice data stacks.

[0048] Classification mode selection: The user selects the classification recognition mode based on the invoice content.

[0049] Preliminary classification based on invoice content: Initialization of the first identification container: The invoice classification identification module generates a first identification container and randomly selects an invoice data pile to send to the container.

[0050] First invoice classification feature extraction: The first recognition container analyzes the selected invoice data pile and extracts the type information of all invoices (such as product sales, service fees, returns and refunds, etc.) to form the first invoice classification feature.

[0051] Multi-container parallel processing: Generate a corresponding number of identification containers according to the number of remaining invoice data piles, and assign each data pile to each container. All containers are connected to the first identification container and share the first invoice classification feature.

[0052] Parallel classification and integrity verification: Parallel classification: Each identification container classifies the invoices in its own container according to the first invoice classification feature. For example, merchandise sales invoices are classified into one category, service fee invoices are classified into another category, and return and refund invoices are classified into a third category.

[0053] Completeness check: After the classification is completed, the system checks whether there is any unclassified invoice data in each container. After verification, all invoices are correctly classified and the classification task of each container is considered completed.

[0054] Result output: Classification feature output: The system outputs classification features based on the invoice content, including the number of various invoices, total amount and other information, marking the completion of the invoice classification and identification task. This information is stored in the data storage module for subsequent query and analysis.

[0055] Example 2: Identification of invoice entry categories based on invoice amount: A manufacturing enterprise needs to process a large number of raw material purchase invoices from suppliers every month, and the invoice amounts vary greatly. In order to better manage financial processes and control costs, the enterprise decides to adopt the invoice entry classification identification method and system provided by the present invention to automatically classify invoices according to the invoice amount.

[0056] Specific implementation steps: Initial data processing: Data stacking: The data initial processing module stacks the invoice data according to the actual needs of the enterprise and the invoicing cycle (such as monthly) to form multiple invoice data stacks.

[0057] Classification mode selection: The user selects the classification recognition mode based on the invoice amount, and the system enters step five.

[0058] Classification based on invoice amount: Amount range determination: The invoice classification identification module first determines the first invoice classification amount range, such as 0-1000 yuan, 1001-5000 yuan, 5001 yuan and above, etc.

[0059] Multi-container parallel processing: Generate a corresponding number of identification containers according to the number of invoice data piles, and each container is classified according to the set amount range. For example, the first container processes invoices of 0-1000 yuan, the second container processes invoices of 1001-5000 yuan, and the third container processes invoices above 5001 yuan.

[0060] Completeness check and scope adjustment: Completeness check: After the classification is completed, the system checks whether there is unclassified invoice data in each container. After inspection, it is found that some invoices in the third container are still unclassified.

[0061] Range adjustment: The system extracts the amount range of the remaining invoices in the third container and finds that the amounts of these invoices are mainly concentrated between 5001 and 10000 yuan. Therefore, the system takes 5001-10000 yuan as the second invoice amount range and merges it with the original amount range to form a new classification standard.

[0062] Result output: Classification feature output: The system outputs classification features based on invoice amount, including the number of invoices within each amount range, total amount and other information, marking the completion of the invoice classification recognition task. This information is stored in the data storage module for subsequent query and analysis.

Claims

1. A method for identifying invoice entry classification, characterized in that: The steps include: Step 1: The data initial processing module divides the invoice data to be classified into piles to obtain multiple invoice data piles, and classifies them according to the classification mode. If the classification mode is invoice classification recognition based on invoice content, then go to step 2; if the classification mode is invoice classification recognition based on invoice amount, then go to step 5; Step 2: the invoice classification recognition module generates a first recognition container, randomly selects an invoice data pile and sends it to the first recognition container, the first recognition container obtains the classification data of the selected invoice data pile, and obtains the first invoice classification feature; the invoice classification recognition module generates the same number of recognition containers according to the number of the remaining invoice data piles, sends each invoice data pile to each recognition container, and each recognition container is connected to the first recognition container respectively; Step 3: According to the first invoice classification feature, each identification container classifies the invoices respectively to obtain classification results. According to the classification results of each identification container, the invoice classification completeness is obtained respectively. If the invoice classification is complete, the identification container classification is completed, and the first invoice classification identification feature is the invoice classification identification feature, and the process proceeds to step 6. If there is an identification container with incomplete invoice classification, the process proceeds to step 4. Step 4: the first identification container classifies the remaining invoice data after classification in the identification containers with incomplete invoice classification, and obtains a second classification feature; the first classification feature and the second classification feature constitute the invoice classification identification feature, and then proceeds to step 6; Step 5: The invoice classification recognition module obtains the first invoice classification amount range, generates the same number of recognition containers according to the number of invoice data piles, and classifies each recognition container according to the invoice classification amount range to obtain a classification result. According to the classification results of each recognition container, the invoice classification completeness is obtained respectively. If the invoice classification is complete, the classification of the recognition container is completed, and the process proceeds to step 6; if there are recognition containers with incomplete invoice classification, the second invoice amount range of the remaining invoice data after classification in each invoice classification recognition container is obtained. The second invoice amount range and the first invoice classification amount range constitute the invoice classification recognition feature, and the process proceeds to step 6; Step six, obtain the invoice classification identification features and complete the invoice classification identification.

2. The invoice entry classification identification method according to claim 1 is characterized in that: The data initial processing module divides the invoice data to be classified into piles to obtain multiple invoice data piles, including: The data initialization module piles the invoice data to be classified according to the set data size; the data size is the number of invoices or the invoicing cycle.

3. The invoice entry classification identification method according to claim 2 is characterized in that: The first identification container obtains the classification data of the selected invoice data pile to obtain the first invoice classification feature, including: The first identification container obtains the invoice types of all invoices in the selected invoice data pile, and all invoice types constitute a first invoice classification feature.

4. The invoice entry classification identification method according to claim 3 is characterized in that: The invoice classification completeness is obtained according to the classification results of each identification container, including: Each identification container classifies the invoices in the container according to the first invoice classification feature. After the classification is completed, if there is no unclassified invoice data in the container, the invoice classification is complete; if there is unclassified invoice data in the container, the invoice classification is incomplete.

5. The invoice entry classification identification method according to claim 4 is characterized in that: The first identification container classifies the remaining invoice data after classification in each incompletely classified identification container to obtain a second classification feature, including: The first identification container classifies and identifies the remaining invoice data to obtain classification information of the remaining invoice data, and the classification information of the remaining invoice data is the second classification feature.

6. The invoice entry classification identification method according to claim 5 is characterized in that: If there is an identification container with incomplete invoice classification, the second invoice amount range of the remaining invoice data after classification in each identification container with incomplete invoice classification is obtained, including: A second invoice amount range is obtained based on the invoice amount of the remaining invoice data after classification in the identification container for each incomplete invoice classification.

7. An invoice entry classification and identification system, characterized in that: An invoice entry classification identification method according to any one of claims 1 to 6 is applied, comprising a cloud data server, a data initial processing module, an invoice classification identification module, a data processing module, a communication module, a data acquisition module and a data storage module; The data initial processing module, invoice classification and identification module, data acquisition module, communication module and data storage module are respectively connected to the data processing module; the cloud data server is communicatively connected to the communication module.

Images