Document review method, device, computer equipment and storage medium
By predicting the incremental document data and dynamic update number of unaudited documents, the problem of low resource utilization in traditional document audit systems is solved, and a more efficient document audit process is achieved.
Patent Information
- Application Number
- CN202210378179.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-12
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2042-04-12
AI Technical Summary
In traditional document review systems, the server resource utilization rate is low, and the number of containers cannot be dynamically adjusted according to the number of documents at the moment, resulting in insufficient resources during peak periods or excessive resources during trough periods, resulting in inefficiency.
By obtaining the document increment data of the current unit time, enter the document increment prediction model completed by training, predict the document increment data of the next unit time, update the container number based on the number of documents that have not been reviewed, and ensure the container resources Reasonable configuration.
This improves the utilization rate of equipment resources, avoids the problems of insufficient containers during peak periods and excessive containers during trough periods, and achieves a more efficient document review process.
Smart Images

Figure CN114638543B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and in particular to a document review method, apparatus, computer equipment, storage medium and computer program product. Background Art
[0002] Document review refers to the review of document elements in documents submitted by customers (such as international documents) based on the principles of "document consistency" and "document consistency" to determine the authenticity of the documents.
[0003] In traditional technology, the number of servers performing document review tasks is pre-configured and cannot be adjusted; however, the number of documents at each unit time is different, and there are peak and trough periods, resulting in the inability to reasonably utilize server resources, which in turn leads to low utilization of equipment resources. Summary of the invention
[0004] Based on this, it is necessary to provide a document review method, device, computer equipment, computer-readable storage medium and computer program product that can improve resource utilization in response to the above technical problems.
[0005] In a first aspect, the present application provides a document review method. The method comprises:
[0006] Get the incremental data of the document at the current unit moment;
[0007] Inputting the document increment data of the current unit time into the trained document increment prediction model to obtain the document increment data of the next unit time;
[0008] According to the number of unaudited completed documents at the current unit time and the document increment data at the next unit time, the number of unaudited completed documents at the next unit time is obtained;
[0009] According to the number of unaudited completed documents at the next unit time, the number of containers at the current unit time is updated to obtain an updated number of containers;
[0010] The container corresponding to the updated number of containers is called, and the unaudited completed documents corresponding to the number of unaudited completed documents at the next unit time are audited to obtain an audit result.
[0011] In one embodiment, the trained document increment prediction model is obtained by training using the following method:
[0012] Acquire sample data; the sample data includes document increment data at each unit time and actual document increment data at the next unit time of each unit time;
[0013] Inputting the document increment data of each unit time into the document increment prediction model to be trained to obtain the predicted document increment data of the next unit time of each unit time;
[0014] According to the difference between the predicted document increment data and the actual document increment data at the next unit moment of each unit moment, the document increment prediction model to be trained is trained to obtain the trained document increment prediction model.
[0015] In one embodiment, the updating of the number of containers at the current unit time according to the number of unaudited completed documents at the next unit time to obtain the updated number of containers includes:
[0016] Dividing the number of unaudited completed documents at the next unit time by the throughput of a single container, to obtain the predicted number of containers at the next unit time;
[0017] Comparing the predicted number of containers at the next unit time with the number of containers at the current unit time to obtain a comparison result;
[0018] According to the comparison result, the number of containers at the current unit time is updated to obtain an updated number of containers.
[0019] In one embodiment, the auditing of the unaudited completed documents corresponding to the number of unaudited completed documents at the next unit time to obtain the audit result includes:
[0020] Extracting the document elements of the unaudited document;
[0021] Convert the document element into an object instance of a corresponding document element type;
[0022] The rule in the rule base that matches the object instance of the document element type is called, and the object instance of the document element type is audited to obtain the audit result.
[0023] In one embodiment, the method further comprises:
[0024] In response to the rule editing operation, a rule file is created; the rule file includes document review rule information;
[0025] Compiling the rule file to obtain corresponding rules;
[0026] The rules are loaded into the rule base.
[0027] In one embodiment, extracting the document elements of the unaudited completed document includes:
[0028] Obtaining the document image of the unreviewed document;
[0029] Extracting text content from the document image;
[0030] According to a document element recognition model corresponding to the document type of the unaudited document, document elements in the text content are extracted as document elements of the unaudited document.
[0031] In a second aspect, the present application also provides a document review device. The device comprises:
[0032] The acquisition module is used to obtain the incremental data of the document at the current unit moment;
[0033] A prediction module, used for inputting the document increment data of the current unit time into the trained document increment prediction model to obtain the document increment data of the next unit time;
[0034] A determination module, configured to obtain the number of unaudited completed documents at the next unit time according to the number of unaudited completed documents at the current unit time and the document increment data at the next unit time;
[0035] An updating module, configured to update the number of containers at the current unit time according to the number of unaudited completed documents at the next unit time, to obtain an updated number of containers;
[0036] The audit module is used to call the container corresponding to the updated container quantity, audit the unaudited completed documents corresponding to the number of unaudited completed documents at the next unit time, and obtain the audit result.
[0037] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:
[0038] Get the incremental data of the document at the current unit moment;
[0039] Inputting the document increment data of the current unit time into the trained document increment prediction model to obtain the document increment data of the next unit time;
[0040] According to the number of unaudited completed documents at the current unit time and the document increment data at the next unit time, the number of unaudited completed documents at the next unit time is obtained;
[0041] According to the number of unaudited completed documents at the next unit time, the number of containers at the current unit time is updated to obtain an updated number of containers;
[0042] The container corresponding to the updated number of containers is called, and the unaudited completed documents corresponding to the number of unaudited completed documents at the next unit time are audited to obtain an audit result.
[0043] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:
[0044] Get the incremental data of the document at the current unit moment;
[0045] Inputting the document increment data of the current unit time into the trained document increment prediction model to obtain the document increment data of the next unit time;
[0046] According to the number of unaudited completed documents at the current unit time and the document increment data at the next unit time, the number of unaudited completed documents at the next unit time is obtained;
[0047] According to the number of unaudited completed documents at the next unit time, the number of containers at the current unit time is updated to obtain an updated number of containers;
[0048] The container corresponding to the updated number of containers is called, and the unaudited completed documents corresponding to the number of unaudited completed documents at the next unit time are audited to obtain an audit result.
[0049] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:
[0050] Get the incremental data of the document at the current unit moment;
[0051] Inputting the document increment data of the current unit time into the trained document increment prediction model to obtain the document increment data of the next unit time;
[0052] According to the number of unaudited completed documents at the current unit time and the document increment data at the next unit time, the number of unaudited completed documents at the next unit time is obtained;
[0053] According to the number of unaudited completed documents at the next unit time, the number of containers at the current unit time is updated to obtain an updated number of containers;
[0054] The container corresponding to the updated number of containers is called, and the unaudited completed documents corresponding to the number of unaudited completed documents at the next unit time are audited to obtain an audit result.
[0055] The document review method, device, computer equipment, storage medium and computer program product described above obtain the document increment data of the current unit moment by inputting the trained document increment prediction model to obtain the document increment data of the next unit moment; then obtain the number of unaudited completed documents of the next unit moment according to the number of unaudited completed documents of the current unit moment and the document increment data of the next unit moment; then update the number of containers of the current unit moment according to the number of unaudited completed documents of the next unit moment to obtain the updated number of containers; finally, call the container corresponding to the updated number of containers to review the unaudited completed documents corresponding to the number of unaudited completed documents of the next unit moment to obtain the review result. In this way, by predicting the document increment data of the next unit moment, the number of unaudited completed documents of the next unit moment is determined, and then the number of containers of the current unit moment is updated, so as to achieve the purpose of automatically updating the number of containers of the current unit moment according to the number of unaudited completed documents of the next unit moment, avoid the defects of insufficient number of containers in peak period and excessive number of containers in valley period, make the equipment resources reasonably used, and improve the utilization rate of equipment resources. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 A schematic diagram of a process of a document review method in one embodiment;
[0057] Figure 2 A schematic diagram of a flow chart of steps for obtaining an audit result in one embodiment;
[0058] Figure 3 A schematic diagram of a process flow of a document review method in another embodiment;
[0059] Figure 4 It is a structural block diagram of an adaptive intelligent order review decision-making device based on rule engine and machine learning in one embodiment;
[0060] Figure 5 The following is a processing flow chart of each functional module in an intelligent document examination and decision-making device in one embodiment;
[0061] Figure 6 A structural block diagram of a document verification device in an embodiment;
[0062] Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION
[0063] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0064] In one embodiment, Figure 1 As shown, a document review method is provided, which is described by taking the method applied to a server as an example, and includes the following steps:
[0065] Step S101, obtaining the incremental data of the document at the current unit time.
[0066] The unit moment refers to the time step, such as half an hour, and the current unit moment refers to the current time step, such as the current half an hour.
[0067] Among them, documents refer to paper documents that need to be reviewed, such as international documents. International documents refer to the documents, certificates and files used in international settlements, which can be divided into contract agreements, applications, letters of authorization, documents, certificates, etc.; documents can be divided into ocean bills of lading, air waybills, packing lists, commercial invoices, insurance policies, etc.
[0068] Among them, the incremental data of documents refers to the number of documents added; the incremental data of documents at the current unit moment refers to the number of documents added within the current time step, such as the number of documents added within the current half hour.
[0069] Specifically, the server analyzes the document record data in the database to obtain the incremental data of the document at the current unit time. For example, the business personnel scan the paper documents submitted by the customer for review into electronic document images, and separately package the document images of different document types, and then submit a review request in the document review system to upload document images of different document types; the document review system automatically generates different image batch numbers according to the document images of different document types, and stores them in the review request information table of the relational database; the review request information table of the relational database records the attribute information such as the review request number, image batch number, document major category, document minor category, etc. The server analyzes the review request information table in the relational database to obtain the incremental data of the document at the current unit time.
[0070] Step S102, inputting the incremental data of the document at the current unit time into the trained incremental prediction model of the document to obtain the incremental data of the document at the next unit time.
[0071] Among them, the document increment prediction model is a model used to predict the document increment data of the next unit moment, such as the XGBoost (eXtreme Gradient Boosting) prediction model, which is obtained through pre-training.
[0072] Among them, the incremental data of documents at the next unit time refers to the number of documents added in the next time step, such as the number of documents added in the next half hour.
[0073] Specifically, the server obtains a trained document incremental prediction model, and then inputs the document incremental data of the current unit moment into the trained document incremental prediction model. The server analyzes and processes the document incremental data of the current unit moment through the trained document incremental prediction model to obtain the document incremental data of the next unit moment.
[0074] For example, the server inputs the incremental document data A of the current half hour (such as 9:00 am) into the trained incremental document prediction model to obtain the incremental document data B of the next half hour (such as 9:30 am).
[0075] Step S103, according to the number of unaudited completed documents at the current unit time and the document increment data at the next unit time, the number of unaudited completed documents at the next unit time is obtained.
[0076] Among them, documents that have not been reviewed include documents awaiting review and documents under review; documents awaiting review refer to documents that have not been processed, and documents under review refer to documents that are being reviewed.
[0077] Specifically, the server obtains the number of documents to be reviewed and the number of documents under review at the current unit moment, adds the number of documents to be reviewed and the number of documents under review at the current unit moment, and obtains the number of unreviewed documents at the current unit moment; adds the number of unreviewed documents at the current unit moment and the document increment data of the next unit moment, and obtains the number of unreviewed documents at the next unit moment.
[0078] For example, assuming that the number of documents to be reviewed and the number of documents under review at the current unit time are A1 and A2 respectively, and the incremental data of documents at the next unit time is A3, then the number of unreviewed documents at the next unit time is A1+A2+A3.
[0079] Step S104, updating the number of containers at the current unit time according to the number of unaudited documents at the next unit time to obtain an updated number of containers.
[0080] Among them, container refers to the equipment used to perform document review tasks. It should be noted that the intelligent document review system is deployed through containerization, and containers are lighter than traditional virtual machines and more conducive to rapid replication and deployment, which also provides a basis for the rapid expansion and contraction of intelligent document review.
[0081] The number of containers at the current unit time refers to the number of containers currently deployed. The updated number of containers refers to the number of containers at the next unit time, specifically the number of containers required for the next time step, such as the number of containers required in the next half hour.
[0082] Specifically, the server divides the number of unaudited documents at the next unit time by the single container throughput to obtain the number of containers at the next unit time; based on the number of containers at the next unit time, the server updates the number of containers at the current unit time to obtain the updated number of containers.
[0083] For example, assuming that the number of containers at the next unit moment is 8 and the number of containers at the current unit moment is 5, the number of containers at the current unit moment is expanded so that the number of containers at the current unit moment increases to 8. For another example, assuming that the number of containers at the next unit moment is 8 and the number of containers at the current unit moment is 12, the number of containers at the current unit moment is reduced. At the same time, in order to ensure that the system has a certain redundant processing capability, the number of containers at the current unit moment can be reduced to 11.
[0084] Step S105, calling the container corresponding to the updated container quantity, and auditing the unaudited completed documents corresponding to the number of unaudited completed documents at the next unit time to obtain the audit result.
[0085] Specifically, the server distributes the unaudited completed documents corresponding to the number of unaudited completed documents at the next unit time to each container in the containers corresponding to the updated number of containers; and audits the unaudited completed documents through each container to obtain an audit result.
[0086] In the above document review method, the incremental data of the document at the current unit moment is input into the trained incremental prediction model to obtain the incremental data of the document at the next unit moment; then, the number of unaudited completed documents at the current unit moment and the incremental data of the document at the next unit moment are used to obtain the number of unaudited completed documents at the next unit moment; then, the number of containers at the current unit moment is updated according to the number of unaudited completed documents at the next unit moment to obtain the updated number of containers; finally, the container corresponding to the updated number of containers is called to review the unaudited completed documents corresponding to the number of unaudited completed documents at the next unit moment to obtain the review result. In this way, the number of unaudited completed documents at the next unit moment is determined by predicting the incremental data of the document at the next unit moment, and then the number of containers at the current unit moment is updated, so as to achieve the purpose of automatically updating the number of containers at the current unit moment according to the number of unaudited completed documents at the next unit moment, avoid the defects of insufficient number of containers at peak times and excessive number of containers at low times, make reasonable use of equipment resources, and thus improve the utilization rate of equipment resources.
[0087] In one embodiment, a trained document increment prediction model is obtained by training through the following method: obtaining sample data; the sample data includes the document increment data at each unit moment and the actual document increment data at the next unit moment of each unit moment; inputting the document increment data at each unit moment into the document increment prediction model to be trained to obtain the predicted document increment data at the next unit moment of each unit moment; training the document increment prediction model to be trained based on the difference between the predicted document increment data at the next unit moment of each unit moment and the actual document increment data to obtain a trained document increment prediction model.
[0088] Among them, the document incremental data at each unit moment refers to the document incremental data at each historical unit moment; the actual document incremental data at the next unit moment of each unit moment refers to the actual document incremental data at the next historical unit moment of each historical unit moment.
[0089] Specifically, the server obtains the document increment data at each unit moment and the actual document increment data at the next unit moment of each unit moment as sample data; divides the sample data into a training set and a test set; then inputs the document increment data at each unit moment in the training set into the document increment prediction model to be trained, and analyzes and processes the document increment data at each unit moment through the document increment prediction model to be trained to obtain the predicted document increment data at the next unit moment of each unit moment; obtains a loss value based on the difference between the predicted document increment data at the next unit moment of each unit moment and the actual document increment data; adjusts the model parameters of the document increment prediction model to be trained based on the loss value to obtain an adjusted document increment prediction model; retrains the adjusted document increment prediction model based on the sample data until the trained document increment prediction model converges, then stops the training, and uses the trained document increment prediction model as the trained document increment prediction model. Furthermore, the server can also use the sample data of the validation set to judge the prediction accuracy of the trained document increment prediction model; if the prediction accuracy of the document increment prediction model does not meet the requirements, the document increment prediction model is iteratively tuned again.
[0090] For example, assuming that the document increment prediction model is an XGBoost prediction model, its training steps are as follows:
[0091] Step 1): Install the XGBoost library; select the open source XGBoost python library and install it directly through the pip command. The python library refers to a collection of codes in Python that completes certain functions and is a combination of codes for users to use; pip is a tool for installing and managing Python packages.
[0092] Step 2): Dataset preparation. The data from 2019-01-01 to 2021-12-31 are used as the XGBoost dataset. The dataset will be preprocessed and converted into time series data with a step length of half an hour. The corresponding transaction volume is the incremental value in the following format.
[0093] 2019-01-01 00:00,0
[0094] 2019-01-01 00:30,0
[0095] 2019-01-01 01:00,0 ...
[0097] 2019-01-01 08:30,100
[0098] 2019-01-01 09:00,150 ...
[0100] Time series data can be formulated for supervised learning by using the previous time steps as input variables and the next time step as the output variable; this approach is called a sliding window because the input and expected values are moved forward in time, creating new samples for the supervised learning model.
[0101] Step 3): Fit and evaluate the model. The model needs to be trained with historical data and predict the future, so we need to use a technique called forward validation. In forward validation, we first need to select a cut point to divide the training data set and the test data set; the data from 2019-01-01 to 2020-12-31 is used as the training set, and the data from 2021-01-01 to 2021-12-31 is used as the test set. After the training set and the test set are divided, a single-step prediction is performed to train the training set and predict the first step of the test set to evaluate the model. Then, the true value of the test set is added to the training set, the model is refitted, and the model is allowed to predict the second step of the test set. Repeating this process for the entire test data set allows the error metric to be calculated from it to evaluate the model.
[0102] Step 4): Determine the model configuration. Evaluate the model through single-step prediction and determine to use the first three time steps as the input of the model parameters for training. After the training is completed, new data can be predicted, that is, the value corresponding to the next time step is predicted using the most recent three time steps as input.
[0103] Step 5): Iterative model fitting. Because the time series data is always moving forward, the model will continuously add the most recent data to the training set for fitting and retraining to continuously optimize the prediction model.
[0104] In this embodiment, the document increment prediction model to be trained is iteratively trained based on sample data to obtain a trained document increment prediction model, so that the document increment data for the next unit moment output by the document increment prediction model is more accurate, further improving the prediction accuracy of the document increment data for the next unit moment, thereby improving the accuracy of subsequent container quantity adjustments.
[0105] In one embodiment, the above step S104, based on the number of unaudited completed documents at the next unit moment, updates the number of containers at the current unit moment to obtain an updated number of containers, specifically including: dividing the number of unaudited completed documents at the next unit moment by the single container throughput to obtain a predicted number of containers at the next unit moment; comparing the predicted number of containers at the next unit moment with the number of containers at the current unit moment to obtain a comparison result; and updating the number of containers at the current unit moment according to the comparison result to obtain an updated number of containers.
[0106] The throughput of a single container refers to the maximum number of documents that a single container can handle. It should be noted that when the number of unchecked documents at the next unit time is divided by the throughput of a single container, if the result is not an integer, it will be processed uniformly.
[0107] Specifically, the server divides the number of unaudited completed documents at the next unit moment by the throughput of a single container to obtain the result of the division; if the result is an integer, the result is used as the predicted number of containers at the next unit moment; if the result is not an integer, the result is rounded to an integer and then added with 1 to obtain the predicted number of containers at the next unit moment; the predicted number of containers at the next unit moment is compared with the number of containers at the current unit moment to obtain a comparison result; if the comparison result is greater than 0, the number of containers at the current unit moment is amplified so that the amplified number of containers is equal to the predicted number of containers at the next unit moment, that is, the updated number of containers is equal to the predicted number of containers at the next unit moment. If the comparison result is equal to 0, the number of containers at the current unit moment is not updated, that is, the predicted number of containers at the next unit moment is still the same as the number of containers at the current unit moment. If the comparison result is greater than or equal to -3 and less than 0, the number of containers at the current unit moment is not updated, that is, the predicted number of containers at the next unit moment is still the same as the number of containers at the current unit moment. If the comparison result is less than -3, the number of containers at the current unit time is reduced so that the reduced number of containers is equal to the predicted number of containers at the next unit time + 3, that is, the updated number of containers is equal to the predicted number of containers at the next unit time + 3.
[0108] For example, assuming that the predicted number of containers at the next unit moment is 8 and the number of containers at the current unit moment is 5, the number of containers at the current unit moment is expanded so that the number of containers at the current unit moment increases to 8; for another example, assuming that the predicted number of containers at the next unit moment is 8 and the number of containers at the current unit moment is any value between 8 and 11, the number of containers at the current unit moment is not updated; for another example, assuming that the predicted number of containers at the next unit moment is 8 and the number of containers at the current unit moment is greater than 11, the number of containers at the current unit moment is reduced so that the number of containers at the current unit moment is reduced to be equal to the predicted number of containers at the next unit moment + 3.
[0109] In this embodiment, the number of containers at the current unit moment is updated according to the number of unaudited completed documents at the next unit moment to obtain an updated number of containers; the purpose of automatically updating the number of containers at the current unit moment according to the number of unaudited completed documents at the next unit moment is achieved, avoiding the defects of insufficient number of containers during peak periods and excessive number of containers during trough periods, so that equipment resources are reasonably utilized, thereby improving the utilization rate of equipment resources.
[0110] In one embodiment, Figure 2 As shown, the above step S105 is to review the unreviewed completed documents corresponding to the number of unreviewed completed documents at the next unit time to obtain the review result, which specifically includes the following steps:
[0111] Step S201, extracting the document elements of the unaudited documents.
[0112] Among them, document elements refer to the information that needs to be identified in the unaudited documents; the document elements that need to be extracted are different for different document types. For example, for sea waybills, the document elements that need to be extracted are type, bill of lading number, loading port, unloading port, ship name, container number, carrier, shipping company, loading date, etc.; for invoices, the document elements that need to be extracted are type, invoice number, invoicing date, company name, issuer, invoice payee, and total price.
[0113] Step S202: convert the document element into an object instance of the corresponding document element type.
[0114] Among them, the document element type refers to the element type to which the document element belongs, such as invoice number, invoice date, company name, issuer, invoice payee, total price, etc.
[0115] The object instance refers to the specific content of the document element type. For example, if the document element type is the invoice number, the corresponding object instance is the specific invoice number. It should be noted that the interaction between the rule engine and the business data is realized through the instance of the data object.
[0116] Step S203, calling the rule in the rule base that matches the object instance of the document element type, auditing the object instance of the document element type, and obtaining the audit result.
[0117] The rule base includes rules that are customized by users.
[0118] Specifically, the server obtains the document type and text content of the unaudited document; according to the document element recognition model corresponding to the document type, the document elements in the text content are extracted as the document elements of the unaudited document; the document elements of the unaudited document are converted to obtain object instances of the corresponding document element types; the rule library is queried to obtain rules that match the object instances of the document element types, and the object instances of the document element types are audited using the rules that match the object instances of the document element types to obtain audit results of each document element of the unaudited document; based on the audit results of each document element of the unaudited document, the audit results of the unaudited document are obtained; if among the audit results corresponding to all the document elements, one document element has a false audit result, then it is confirmed that the unaudited document has failed the audit; if all the audit results corresponding to the document elements are true, then it is confirmed that the unaudited document has passed the audit. For example, for a sea waybill, if the review results corresponding to the document elements such as type, bill of lading number, loading port, unloading port, ship name, container number, carrier, shipping company, and loading date are all true, then the sea waybill is confirmed to have passed the review; otherwise, it is confirmed that the sea waybill has failed the review.
[0119] For example, the server obtains the corresponding image compression package according to the document review request number, extracts the text content of the image one by one through OCR (Optical Character Recognition) technology after decompression, obtains the elements of the corresponding document, and finally stores the elements in the relational database; then, the server calls the interface for activating the document review rule, and the input parameter is the document review request number; the Drools rule engine queries the relational database according to the document review request number, obtains the elements of each document type, and converts them into object instances of the corresponding type as the input of the Drools rule engine, makes rule matching decisions, obtains the document review conclusion, and stores the document review conclusion in the relational database. Among them, the Drools rule engine is an open source rule engine written in Java language, which uses the Rete algorithm to evaluate the rules written.
[0120] Furthermore, the rule engine can match instances with rules in the rule base, make inference decisions based on the matched rules, and obtain conclusions. The specific implementation steps are as follows: Step 1): Create a rule base, use the API (Application Porgramming Interface) provided by Drools to obtain the latest deployed rules, and load them into the rule base; Step 2): Insert object instances, and obtain the extracted elements of each document type from the relational database according to the document review request number. The elements correspond one-to-one with the corresponding data object attributes, so the corresponding data objects can be instantiated according to the document type and the extracted elements, and these object instances are loaded into the Working Memory. Step 3): Call the API provided by Drools, activate the rules, review the document according to the defined rules, and store the review conclusions in the relational database. Step 4): Release resources, call the API provided by Drools to release resources.
[0121] In this embodiment, the container corresponding to the updated container quantity is called to automatically audit the unaudited completed documents corresponding to the number of unaudited completed documents at the next unit time to obtain the audit result, thereby avoiding the defects of reduced audit efficiency due to server resource bottlenecks during peak business volume and waste of server resources during trough business volume, thereby improving the audit efficiency and the utilization rate of equipment resources.
[0122] In one embodiment, before step S203, the following contents are also included: in response to the rule editing operation, a rule file is created; the rule file includes document review rule information; the rule file is compiled to obtain corresponding rules; and the rules are loaded into the rule library.
[0123] The rule file refers to the DRL file, which is used to edit the document review rules. The document review rule information refers to the rule details, which can be obtained through visual configuration without adjusting the program logic.
[0124] Specifically, the server responds to the user's rule editing operation on the rule editing page, obtains a rule file created by the user including document review rule information, compiles the rule file, obtains corresponding rules, and loads the rules into the rule library.
[0125] For example, the visual configuration of the order review rules can be achieved through the following steps:
[0126] Step 1): Create a project. Create a "Smart Order Review Rules" project on the workbench. You only need to create the project once, and all subsequent order review rules will be attached to this project.
[0127] Step 2): Create a data object. The data object defines the attributes involved in the corresponding rules, and the interaction between the rule engine and business data is realized through the instance of the data object. One document type corresponds to one rule and one data object type. Taking the invoice type as an example, the data object type defines attributes such as "document type", "invoice number", "invoice date", "company name", "issuer", "invoice payee", and "total price". The attributes in the data object also correspond to the elements to be extracted from the document of the corresponding type, and the data object information is stored in a relational database.
[0128] Step 3): Create a DRL file. The DRL file is used to edit rules. The format of the DRL file is as follows. One document type corresponds to one document review rule, which is a different rule name in the DRL file.
[0129]
[0130] Step 4): Compile and deploy. Compile and deploy the DRL file. After the deployment is complete, the Drools rules can obtain the latest order review rules.
[0131] In this embodiment, in response to the rule editing operation, the corresponding rules are created and loaded into the rule library, thereby achieving the purpose of visually configuring the order review rules. This avoids the defect in the prior art that the order review rules are communicated offline and then implemented by program developers through code, resulting in a high degree of coupling between the order review rules and the program, and business personnel are unable to adjust the order review rules in real time as needed.
[0132] In one embodiment, the above step S201, extracting the document elements of the unaudited document, specifically includes: obtaining a document image of the unaudited document; extracting the text content of the document image; and extracting the document elements in the text content as the document elements of the unaudited document based on a document element recognition model corresponding to the document type of the unaudited document.
[0133] Among them, the document image refers to the electronic image corresponding to the unaudited document, which is obtained by scanning the unaudited document in paper form.
[0134] Among them, text content refers to the text content presented on the document image.
[0135] Among them, the document element recognition model is a model used to identify and extract corresponding document elements from the text content of the document image, mainly using natural language processing technology.
[0136] Specifically, the server obtains the document image and document type of the unaudited document; extracts the document image line by line through OCR technology to obtain the text content corresponding to the document image; determines the document elements to be extracted corresponding to the document type from the correspondence between the document type and the document elements to be extracted; determines the document element recognition model corresponding to the document elements to be extracted from the correspondence between the document elements and the document element recognition model, as the document element recognition model corresponding to the document type; extracts the corresponding document elements from the text content of the document image through the document element recognition model corresponding to the document type, as the document elements of the unaudited document.
[0137] For example, assuming that the unaudited document is a sea waybill, the document elements extracted from the sea waybill are type, bill of lading number, loading port, unloading port, ship name, container number, carrier, shipping company, loading date, etc.
[0138] In this embodiment, based on the document element recognition model corresponding to the document type of the unaudited document, the document elements in the text content of the document image of the unaudited document are automatically extracted as the document elements of the unaudited document; the purpose of automatically extracting the document elements is achieved, and the entire process does not require human participation, thereby improving the efficiency of document element extraction, and further improving the efficiency of subsequent document audits.
[0139] In one embodiment, Figure 3 As shown, another document review method is provided, which is described by taking the method applied to a server as an example, and includes the following steps:
[0140] Step S301, obtaining the document increment data at the current unit time.
[0141] Step S302, inputting the incremental data of the document at the current unit time into the trained incremental prediction model of the document to obtain the incremental data of the document at the next unit time.
[0142] Step S303, according to the number of unaudited completed documents at the current unit time and the document increment data at the next unit time, the number of unaudited completed documents at the next unit time is obtained.
[0143] Step S304: Divide the number of unaudited documents at the next unit time by the throughput of a single container to obtain the predicted number of containers at the next unit time.
[0144] Step S305, comparing the predicted container quantity at the next unit time with the container quantity at the current unit time to obtain a comparison result.
[0145] Step S306: updating the number of containers at the current unit time according to the comparison result to obtain an updated number of containers.
[0146] Step S307, calling the container corresponding to the updated container quantity, and obtaining the document image of the unaudited document corresponding to the number of unaudited documents at the next unit time.
[0147] Step S308, extracting the text content of the document image; extracting the document elements in the text content as the document elements of the unaudited document according to the document element recognition model corresponding to the document type of the unaudited document.
[0148] Step S309, converting the document element into an object instance of the corresponding document element type; calling the rule in the rule library that matches the object instance of the document element type, reviewing the object instance of the document element type, and obtaining the review result.
[0149] The above-mentioned document review method determines the number of unreviewed documents at the next unit moment by predicting the incremental document data at the next unit moment, and then updates the number of containers at the current unit moment, thereby achieving the purpose of automatically updating the number of containers at the current unit moment according to the number of unreviewed documents at the next unit moment, avoiding the defects of insufficient number of containers during peak periods and excessive number of containers during trough periods, making rational use of equipment resources and thus improving the utilization rate of equipment resources.
[0150] In one embodiment, in order to more clearly illustrate the document verification method provided by the embodiment of the present application, the document verification method is specifically described below with a specific embodiment. In one embodiment, the present application also provides an adaptive intelligent document review decision method based on rule engine and machine learning. With the help of the Drools rule engine, the document review rules are decoupled from the program, and the document review rules are maintained by business personnel and take effect in real time; with the help of the XGBoost model, the transaction volume of the document review business is predicted in advance, and the autonomous decision-making of system expansion and contraction is realized, which can not only avoid the waste of server resources, but also ensure that the document review efficiency will not be reduced due to server resources. It solves the problem that the document review rules are coupled with the program, which makes the document review rules difficult to expand and cannot quickly respond to the needs of intelligent document review business, and the server cannot be automatically expanded and contracted according to the business volume, resulting in server resource bottlenecks during peak business volume, server resource waste during business volume troughs, and the defect of not being able to reasonably use server resources. Specifically including the following contents:
[0151] Figure 4 An adaptive intelligent document review decision-making device based on rule engine and machine learning is provided. Figure 4The device includes a review request submission module, a review rule editing module, a review result feedback module, a review task allocation module, an intelligent review module, a Drools rule engine, an XGBoost prediction module, and a scaling decision module. Among them, the review request submission module, the review rule editing module, and the review result feedback module are front-end operation modules; the review task allocation module and the intelligent review module are back-end master-slave processing modules; the Drools rule engine, the XGBoost prediction module, and the scaling decision module are common components.
[0152] For example, the document business involves a variety of document and certificate types, and the document review rules used for different types of documents and certificates are different. For example, in the sea waybill, "bill of lading number", "loading port", "unloading port", "container number", "carrier" and other elements must be extracted, and the "carrier" must be submitted to the blacklist for inspection; for invoices, "invoice number", "invoice date" and "invoice amount" are elements that must be extracted, and the "invoice amount" must be checked with the amount in the contract. Intelligent document review will first identify each image type and determine which document type it belongs to. The document type is used as the input of the Drools rule engine. The rule engine determines the subsequent document review rules based on the defined "document type processing rules" and verifies the documents of the corresponding document type. At present, the intelligent document review system is deployed through containerization, and containers are lighter than traditional virtual machines and are more conducive to rapid replication and deployment, which also provides a basis for the rapid expansion and contraction of intelligent document review. With the help of the XGBoost prediction model, the transaction volume of order review in the next half hour is predicted in advance. The expansion and contraction ratio of the container is determined based on the predicted value, the current order review transaction volume to be processed, the current order review transaction volume being processed, the single container throughput, and the number of containers currently executing the order review task.
[0153] Figure 5 A processing flow chart of each functional module in an intelligent document review and decision-making device is provided. Figure 5 ,The specific processing process of each functional module is as follows:
[0154] 1. Document review request submission module: This module is mainly used by business personnel to convert the documents and bills submitted by customers into electronic images, forming a document review request, which is then transferred to the system. The main processing steps are as follows:
[0155] Step 1): The customer submits relevant paper documents and the business staff scans the paper documents into electronic images.
[0156] Step 2): Submit a request for document review. The business personnel package the scanned electronic image into a compressed package, upload it to the system, submit the request for document review, and the system generates a document review request number. The status of this document review request is "pending review". The image file is finally stored in the image document system, and the image document system returns an impact batch number, which is subsequently downloaded based on the image batch. The document review request record is stored in a table in a relational database. The document review request record contains fields such as "document review request number", "impact batch number", and "status". The dictionary value of the "status" field is "pending review", "under review", and "review completed".
[0157] 2. Order review task allocation module: The background processing module is a master-slave architecture. The master node is responsible for managing the slave nodes and allocating tasks. The slave nodes complete the order review tasks and send heartbeat packets to the master node at regular intervals. The main information in the heartbeat packet includes "timestamp", "number of tasks being processed by the current node", "node IP", etc. The master node determines whether the slave node is available based on the heartbeat packet sent by the slave node. If no heartbeat packet is received for three consecutive cycles, the slave node will be deleted from the available list. The removed node can only be re-added to the available list by the master node after sending three available heartbeat packets to the master node in succession. At the same time, the master node will summarize the "number of tasks being processed by the current node" information in the heartbeat packet sent by the slave node, and count the number of order review requests currently being processed by the system. The processing flow of self-assessment task allocation is as follows:
[0158] Step 1): Get the queue for pending review. The master control node scans the order review request information table in the relational database in real time to obtain the order review requests with a status of "pending review";
[0159] Step 2): The slave node sends a heartbeat packet to the master node at regular intervals. The main information in the heartbeat packet includes "timestamp", "number of tasks currently being processed by the current node", "node IP", etc. If the master node does not receive a heartbeat packet for three consecutive cycles, the slave node will be deleted from the available list. The removed node can only be re-added to the available list by the master node after sending three available heartbeat packets to the master node in succession. At the same time, the master node will register the task processing status of each point based on the "node IP" and "number of tasks currently being processed by the current node" in the heartbeat packet sent by the slave node, and summarize and count the number of order review requests currently being processed by the system. This information will be updated to the relational database.
[0160] Step 3): Assign the order review task. The master control node selects the slave node with the least current tasks to perform the order review task based on the task processing status of each registered slave node, and adjusts the status of the request to "under review".
[0161] 3. Review rule editing module: It mainly uses Drools workbench to realize the visual configuration of review rules, and packages and publishes the rules through the workbench.
[0162] 4. Intelligent document review module: It mainly extracts the corresponding element content according to different document types, and stores the element content in a relational database. It then uses the document review request number as an input parameter and calls the document review rule activation interface. The corresponding interface obtains all types of elements according to the document review request number and converts them into object instances as the input of the Drools rule engine to make rule matching decisions and obtain the document review conclusion.
[0163] 5. Drools rule engine: Drools is a rule engine based on the Rete algorithm and is developed in Java. The working mode of Drools is to first obtain the required rules from the deployed rules and load them into the Production Memory (rule base), and the data object instances to be matched are loaded into the Working Memory (working memory). The rule engine matches the instances with the rules in the rule base, makes inference decisions based on the matched rules, and obtains conclusions.
[0164] 6. Review result feedback module: For applications that fail the review, business personnel can also conduct manual review, mainly to avoid approval failure caused by inaccurate OCR recognition due to the low quality of the uploaded electronic image. For such cases, business personnel can re-scan and upload the electronic image, and directly adjust the element content and review results on the page.
[0165] 7. Scaling decision module: It mainly determines the scaling ratio of containers based on the predicted value of future order review transaction volume, the current order review transaction volume to be processed, the current order review transaction volume being processed, the single container throughput, and the number of containers currently executing order review tasks. The predicted value of future order review transaction volume is predicted with the help of the XGBoost prediction model. The main processing steps of this module are as follows:
[0166] Step 1): query the relational database to obtain the number of review requests with the status of "pending review" and "under review". The number of review requests "pending review" is the number of review transactions to be processed by the current system, and the number of review requests "under review" is the number of review transactions currently being processed;
[0167] Step 2): Get the single container throughput, which is preset in the system after stress testing. The container specifications used in the stress testing environment are consistent with those in production, which are 4C4G. Only one container is deployed to execute the order review task; the relational database configuration is also consistent with production, 16C32G. By gradually increasing the number of 100, 200, and 300 transactions, the final single container throughput for the order review task is 12TPS.
[0168] Step 3): Get the predicted value of the order review transaction volume in the next half hour, and predict the transaction volume in the next half hour through the XGBoost prediction model.
[0169] Step 4): Determine the expansion / contraction ratio. The transaction volume to be processed plus the transaction volume in the next half hour plus the transaction volume currently being processed is the total transaction volume that the system needs to process in the next half hour. Divide this value by 1800 seconds, which is the total processing capacity that the system needs to achieve in the next half hour. Divide this value by the single container throughput, which is the number of containers roughly required by the system. The difference is processed as is. The number of containers required by the system in the next half hour minus the number of containers in the current system. If this value is positive, it is the number of containers that need to be expanded. If this value is zero, no processing is performed. If this value is negative, in order to ensure that the system has a certain redundant processing capacity, it is processed according to the following Table 1.
[0170] Table 1
[0171] Greater than or equal to -3 less than 0 No treatment Less than -3 The actual number of shrinking containers = the absolute value of the required number of expansion and shrinking containers minus 3
[0172] Step 5): Call the interface to scale the container according to the scaling ratio.
[0173] 8. XGBoost prediction module: XGBoost is an effective implementation of the stochastic gradient boosting machine learning algorithm, which is used for classification and regression of tabular data sets and can be used for time series prediction. Currently, there is a mature XGBoost open source algorithm library that can be used directly.
[0174] In the implementation example, the pain point of coupling business rules with applications in intelligent order review is solved. The Drools rule engine is used to implement business-defined order review rules and make them effective in real time to meet the requirements of rapid response to business needs. At the same time, the XGBoost model is used to predict future order review transaction volumes to achieve automatic expansion and contraction of containers, greatly improving the utilization of equipment resources and reducing operating costs.
[0175] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.
[0176] Based on the same inventive concept, the embodiment of the present application also provides a document review device for implementing the document review method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more document review device embodiments provided below can refer to the limitations of the document review method above, and will not be repeated here.
[0177] In one embodiment, Figure 6 As shown, a document review device is provided, including: an acquisition module 610, a prediction module 620, a determination module 630, an update module 640 and a review module 650, wherein:
[0178] The acquisition module 610 is used to obtain the document increment data at the current unit time.
[0179] The prediction module 620 is used to input the document increment data of the current unit time into the trained document increment prediction model to obtain the document increment data of the next unit time.
[0180] The determination module 630 is used to obtain the number of unaudited completed documents at the next unit time according to the number of unaudited completed documents at the current unit time and the document increment data at the next unit time.
[0181] The updating module 640 is used to update the number of containers at the current unit time according to the number of unaudited completed documents at the next unit time to obtain an updated number of containers.
[0182] The review module 650 is used to call the container corresponding to the updated container quantity, review the unreviewed completed documents corresponding to the number of unreviewed completed documents at the next unit time, and obtain the review result.
[0183] In one embodiment, the document review device also includes a training module for acquiring sample data; the sample data includes document incremental data at each unit moment and actual document incremental data at the next unit moment of each unit moment; the document incremental data at each unit moment is input into the document incremental prediction model to be trained to obtain the predicted document incremental data at the next unit moment of each unit moment; the document incremental prediction model to be trained is trained based on the difference between the predicted document incremental data at the next unit moment of each unit moment and the actual document incremental data to obtain a trained document incremental prediction model.
[0184] In one embodiment, the update module 640 is further used to divide the number of unaudited completed documents at the next unit moment by the single container throughput to obtain the predicted number of containers at the next unit moment; compare the predicted number of containers at the next unit moment with the number of containers at the current unit moment to obtain a comparison result; and update the number of containers at the current unit moment according to the comparison result to obtain an updated number of containers.
[0185] In one embodiment, the audit module 650 is also used to extract the document elements of the documents that have not been audited; convert the document elements into object instances of the corresponding document element types; call the rules in the rule library that match the object instances of the document element types, audit the object instances of the document element types, and obtain the audit results.
[0186] In one embodiment, the document review device also includes an editing module for creating a rule file in response to a rule editing operation; the rule file includes document review rule information; the rule file is compiled to obtain corresponding rules; and the rules are loaded into a rule library.
[0187] In one embodiment, the audit module 650 is also used to obtain a document image of an unaudited document; extract text content from the document image; and extract document elements from the text content as document elements of the unaudited document based on a document element recognition model corresponding to the document type of the unaudited document.
[0188] Each module in the above-mentioned document verification device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module.
[0189] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as the incremental document data of the current unit moment, the incremental document data of the next unit moment, the number of unaudited documents at the next unit moment, the number of updated containers, and the audit results. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a document audit method is implemented.
[0190] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0191] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.
[0192] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.
[0193] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.
[0194] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.
[0195] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.
[0196] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0197] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. A document review method, characterized in that: The method comprises: Get the incremental data of the document at the current unit moment; Inputting the document increment data of the current unit time into the trained document increment prediction model to obtain the document increment data of the next unit time; According to the number of unaudited completed documents at the current unit time and the document increment data at the next unit time, the number of unaudited completed documents at the next unit time is obtained; According to the number of unaudited completed documents at the next unit time, the number of containers at the current unit time is updated to obtain an updated number of containers; The container corresponding to the updated number of containers is called, and the unaudited completed documents corresponding to the number of unaudited completed documents at the next unit time are audited to obtain an audit result.
2. The method according to claim 1, characterized in that The trained document increment prediction model is obtained by training using the following method: Acquire sample data; the sample data includes document increment data at each unit time and actual document increment data at the next unit time of each unit time; Inputting the document increment data of each unit time into the document increment prediction model to be trained to obtain the predicted document increment data of the next unit time of each unit time; According to the difference between the predicted document increment data and the actual document increment data at the next unit moment of each unit moment, the document increment prediction model to be trained is trained to obtain the trained document increment prediction model.
3. The method according to claim 1, characterized in that The updating of the number of containers at the current unit time according to the number of unaudited completed documents at the next unit time to obtain the updated number of containers includes: Dividing the number of unaudited completed documents at the next unit time by the throughput of a single container, to obtain the predicted number of containers at the next unit time; Comparing the predicted number of containers at the next unit time with the number of containers at the current unit time to obtain a comparison result; According to the comparison result, the number of containers at the current unit time is updated to obtain an updated number of containers.
4. The method according to claim 1, characterized in that: The auditing of the unaudited completed documents corresponding to the number of unaudited completed documents at the next unit time to obtain the audit result includes: Extracting the document elements of the unaudited document; Convert the document element into an object instance of a corresponding document element type; The rule in the rule base that matches the object instance of the document element type is called, and the object instance of the document element type is audited to obtain the audit result.
5. The method according to claim 4, characterized in that The method further comprises: In response to the rule editing operation, a rule file is created; the rule file includes document review rule information; Compiling the rule file to obtain corresponding rules; The rules are loaded into the rule base.
6. The method according to claim 4, characterized in that The extracting of the document elements of the unaudited completed document includes: Obtaining the document image of the unreviewed document; Extracting text content from the document image; According to a document element recognition model corresponding to the document type of the unaudited document, document elements in the text content are extracted as document elements of the unaudited document.
7. A document review device, characterized in that: The device comprises: The acquisition module is used to obtain the incremental data of the document at the current unit moment; A prediction module, used for inputting the document increment data of the current unit time into the trained document increment prediction model to obtain the document increment data of the next unit time; A determination module, configured to obtain the number of unaudited completed documents at the next unit time according to the number of unaudited completed documents at the current unit time and the document increment data at the next unit time; An updating module, configured to update the number of containers at the current unit time according to the number of unaudited completed documents at the next unit time, to obtain an updated number of containers; The audit module is used to call the container corresponding to the updated container quantity, audit the unaudited completed documents corresponding to the number of unaudited completed documents at the next unit time, and obtain the audit result.
8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.
10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.
