A documentary credit review method and system

Abstract

The present invention provides a method and system for reviewing documentary letters of credit, the method comprising: obtaining a letter of credit message corresponding to a letter of credit to be reviewed, and obtaining document data corresponding to a document to be reviewed; performing a first natural language processing on the letter of credit message to obtain letter of credit instantiation data; performing a second natural language processing on the document data to obtain document instantiation data; inputting the letter of credit instantiation data and the document instantiation data into a document review rule engine, so that the document review rule engine reviews the letter of credit instantiation data and the document instantiation data based on the review rules of a preset document review knowledge graph, and obtains and outputs the review results of the letter of credit to be reviewed and the document to be reviewed. There is no need for manual review of the letter of credit and the document, which reduces the review cost and improves the review efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and system for reviewing documentary letters of credit. Background Art

[0002] In the letter of credit and document review business, it is still dependent on document reviewers to manually review letters of credit and documents. However, due to the wide variety of documents involved in the letter of credit business and the complex business scenarios, and the document review needs to follow international practices such as UCP600 and ISBP745, it takes a lot of time (usually three years) to train a qualified document reviewer. Therefore, the manual review of letters of credit and documents has problems such as high review costs and low review efficiency. Summary of the invention

[0003] In view of this, an embodiment of the present invention provides a method and system for reviewing a documentary letter of credit to solve the problems of high review cost and low review efficiency in the manual review of letters of credit and documents.

[0004] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0005] A first aspect of an embodiment of the present invention discloses a method for reviewing a documentary letter of credit, the method comprising:

[0006] Obtain the letter of credit message corresponding to the letter of credit to be reviewed, and obtain the document data corresponding to the document to be reviewed;

[0007] Performing a first natural language processing on the letter of credit message to obtain letter of credit instantiation data;

[0008] Performing a second natural language processing on the document data to obtain document instantiation data;

[0009] The instantiation data of the letter of credit and the instantiation data of the document are input into a document review rule engine, so that the document review rule engine reviews the instantiation data of the letter of credit and the instantiation data of the document based on the review rules of a preset document review knowledge graph, and obtains and outputs the review results of the letter of credit to be reviewed and the document to be reviewed.

[0010] Preferably, the first natural language processing includes at least: message splicing, extraction processing, processing of semi-structured fields, structuring processing of order review-related terms, structuring processing of product information, classification and understanding processing of order review-irrelevant terms, standardization processing and knowledge mapping processing.

[0011] Preferably, performing a second natural language processing on the document data to obtain document instantiation data includes:

[0012] Extracting the elements to be reviewed and the values ​​of the elements to be reviewed from the document data;

[0013] The elements to be reviewed and their values ​​are sorted, standardized and knowledge mapped to obtain document instantiation data.

[0014] Preferably, extracting the elements to be reviewed and the values ​​of the elements to be reviewed from the document data includes:

[0015] Performing preliminary element recognition processing on the document data to extract the elements to be reviewed;

[0016] The elements to be reviewed are processed using preset summary rules, product information or a PICK model to obtain the values ​​of the elements to be reviewed. The PICK model is based on a graph neural network and a conditional random field model.

[0017] Preferably, the process of outputting the audit results of the pending letter of credit and the pending document includes:

[0018] The audit results of the pending letter of credit and the pending document are displayed on the audit result display page.

[0019] Preferably, after the audit result display page displays the audit results of the pending letter of credit and the pending document, it also includes:

[0020] When an abandonment instruction to abandon the review result is detected, switch to a page for executing a traditional document review business operation process.

[0021] Preferably, the method further comprises:

[0022] In response to the viewing instruction, a letter of credit display page or a bill display page is displayed, wherein the letter of credit display page displays the segmentation result and classification result of the letter of credit to be reviewed, and the bill display page displays the bill classification result and the elements to be reviewed of the document to be reviewed.

[0023] A second aspect of an embodiment of the present invention discloses a documentary credit review system, the system comprising:

[0024] An acquisition unit, used to acquire the letter of credit message corresponding to the letter of credit to be reviewed, and to acquire the document data corresponding to the document to be reviewed;

[0025] A first processing unit, configured to perform a first natural language processing on the letter of credit message to obtain instantiation data of the letter of credit;

[0026] A second processing unit, configured to perform a second natural language processing on the document data to obtain document instantiation data;

[0027] The review unit is used to input the instantiation data of the letter of credit and the instantiation data of the document into the review rule engine, so that the review rule engine reviews the instantiation data of the letter of credit and the instantiation data of the document based on the review rules of the preset review knowledge graph, and obtains and outputs the review results of the letter of credit to be reviewed and the document to be reviewed.

[0028] Preferably, the first natural language processing includes at least: message splicing, extraction processing, processing of semi-structured fields, structuring processing of order review-related terms, structuring processing of product information, classification and understanding processing of order review-irrelevant terms, standardization processing and knowledge mapping processing.

[0029] Preferably, the second processing unit includes:

[0030] An extraction module, used for extracting the elements to be reviewed and the values ​​of the elements to be reviewed from the document data;

[0031] The processing module is used to perform element sorting, standardization processing and knowledge mapping processing on the elements to be reviewed and the values ​​of the elements to be reviewed, so as to obtain document instantiation data.

[0032] Based on the above embodiment of the present invention, a method and system for auditing a documentary credit is provided. The method comprises: obtaining a credit message corresponding to the credit to be audited, and obtaining document data corresponding to the document to be audited; performing a first natural language processing on the credit message to obtain credit instantiation data; performing a second natural language processing on the document data to obtain document instantiation data; inputting the credit instantiation data and the document instantiation data into a document auditing rule engine, so that the document auditing rule engine audits the credit instantiation data and the document instantiation data based on the auditing rules of the preset document auditing knowledge graph, and obtains and outputs the audit results of the credit to be audited and the document to be audited. In this scheme, the credit instantiation data and the document instantiation data corresponding to the audited credit and the document to be audited are respectively obtained. The credit instantiation data and the document instantiation data are input into the document auditing rule engine configured with the document auditing rules for auditing, and the audit results of the credit to be audited and the document to be audited are obtained and output. There is no need to manually audit the credit and the document, which reduces the auditing cost and improves the auditing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0034] Figure 1 A flowchart of a method for reviewing a documentary letter of credit provided by an embodiment of the present invention;

[0035] Figure 2 A flowchart of extracting elements to be reviewed provided by an embodiment of the present invention;

[0036] Figure 3 A schematic diagram of the workflow of the document review rule engine provided by an embodiment of the present invention;

[0037] Figure 4 A schematic diagram of the architecture of a big data cloud provided by an embodiment of the present invention;

[0038] Figure 5 A schematic diagram of the processing logic of the audit service provided by an embodiment of the present invention;

[0039] Figure 6 A schematic diagram of the processing logic architecture of a method for reviewing a documentary letter of credit provided by an embodiment of the present invention;

[0040] Figure 7 A schematic diagram of the business architecture of a method for reviewing a documentary letter of credit provided by an embodiment of the present invention;

[0041] Figure 8 A structural block diagram of a documentary credit review system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0044] As can be seen from the background technology, the current review of letters of credit and documents still relies on manual review. However, due to the wide variety of documents involved in the letter of credit business and the complex business scenarios, and the need to follow international practices such as UCP600 and ISBP745 in document review, it takes a lot of time to train reviewers, resulting in the high review cost and low review efficiency of manual review of letters of credit and documents.

[0045] Therefore, the embodiment of the present invention provides a method and system for reviewing documentary letters of credit, which respectively obtains letter of credit instantiation data and document instantiation data corresponding to the review letter of credit and the document to be reviewed. The letter of credit instantiation data and the document instantiation data are input into a document review rule engine configured with document review rules for review, and the review results of the review letter of credit and the document to be reviewed are obtained and output. There is no need for manual review of the letter of credit and the document, so as to reduce the review cost and improve the review efficiency.

[0046] See also Figure 1 , which shows a flow chart of a method for reviewing a documentary credit provided by an embodiment of the present invention, the review method comprising:

[0047] Step S101: obtaining the letter of credit message corresponding to the letter of credit to be reviewed, and obtaining the document data corresponding to the document to be reviewed.

[0048] It should be noted that letters of credit have two formats: image and text. When obtaining letter of credit messages, the corresponding acquisition method should be determined according to the format of the letter of credit. Documents usually only have the format of images, so when obtaining document data, it is necessary to identify the image of the document to extract the document data.

[0049] In the specific implementation of step S101, when the format of the letter of credit to be reviewed is an image, the image of the letter of credit to be reviewed is subjected to intelligent image recognition processing to extract the letter of credit message from the image of the letter of credit to be reviewed; when the format of the letter of credit to be reviewed is a text, the letter of credit message corresponding to the letter of credit to be reviewed can be directly obtained (without the need for intelligent image recognition processing).

[0050] The image of the document to be reviewed is subjected to intelligent image recognition processing, and the document data corresponding to the document to be reviewed is extracted from the image of the document to be reviewed. The extracted document data can also be referred to as the content after OCR (Optical Character Recognition) recognition of the document image.

[0051] In some embodiments, intelligent image recognition includes but is not limited to: image correction, text detection and recognition, table detection, watermark detection, hand signature detection, seal detection and recognition and other processing methods.

[0052] Step S102: Perform a first natural language processing on the letter of credit message to obtain letter of credit instantiation data.

[0053] In the process of specifically implementing step S102, a first natural language processing is performed on the letter of credit message to obtain corresponding letter of credit instantiation data, wherein, in the process of performing the first natural language processing on the letter of credit message, the structured data of the letter of credit message is first determined, and then the letter of credit instantiation data is determined based on the structured data of the letter of credit message.

[0054] In some embodiments, the first natural language processing includes at least: message splicing, extraction processing, processing of semi-structured fields, structuring processing of order review-related terms, structuring processing of product information, classification and understanding processing of order review-irrelevant terms, standardization processing and knowledge mapping processing.

[0055] The processing of semi-structured fields includes: name and address analysis, sentence / rule analysis, etc. The processing of structured clauses related to document review includes: clause splitting, semantic template matching analysis, and other clause processing.

[0056] In some embodiments, in the process of performing the first natural language on the letter of credit message, it is also necessary to pre-process and segment the letter of credit message, and determine the relevance of the letter of credit message to the document review. If there is relevance to the document review, the letter of credit message can be structured to process the clauses related to the document review; if there is no relevance to the document review, the letter of credit message can be processed to classify and understand the clauses irrelevant to the document review.

[0057] It should be noted that knowledge mapping specifically refers to: mapping language symbols to formal ontology representation (that is, the review knowledge graph) through a series of standardized processing.

[0058] Step S103: Perform a second natural language processing on the document data to obtain document instantiation data.

[0059] In the specific implementation of step S103, the document data is corrected, the document name is identified and classified, and the elements to be reviewed and the values ​​of the elements to be reviewed are extracted from the document data that has completed the error correction, document name identification and classification, and the elements to be reviewed and the values ​​of the elements to be reviewed are sorted, standardized and mapped to obtain document instantiation data. Among them, the elements to be reviewed can be understood as the data that needs to be reviewed in the document to be reviewed, and the element sorting includes the splitting and merging of SPAN.

[0060] In some embodiments, in the process of extracting the elements to be reviewed from the document data, the document data is subjected to document form processing, document element model extraction and other document processing, so as to extract the elements to be reviewed and the values ​​of the elements to be reviewed.

[0061] In some embodiments, the specific form of the model factory is predefined to constitute a collection of model factories, and the model factory is used to extract the elements to be reviewed and the values ​​of the elements to be reviewed from the document data; specifically, the document data is subjected to preliminary element identification processing to extract the elements to be reviewed; the elements to be reviewed are processed using preset summary rules, product information or a PICK model to obtain the values ​​of the elements to be reviewed, and the PICK model is based on a graph neural network and a conditional random field model.

[0062] It should be noted that for sparse elements or elements to be reviewed with a relatively simple expression form, summary rules can be used for processing (equivalent to rule-based content parsing) to obtain the value of the element to be reviewed; for elements to be reviewed that appear many times and have flexible expressions, they can be processed using multimodal PICK models based on image information and HTML to obtain the value of the element to be reviewed; if the commodity information analysis of the document to be reviewed is similar to 45A in the letter of credit to be reviewed, the commodity information can be used for processing (equivalent to content parsing of the commodity information) to obtain the value of the element to be reviewed.

[0063] To better explain how to use the model factory to extract the elements to be reviewed and their values ​​from the document data, Figure 2 The flowchart of extracting the elements to be reviewed is shown as an example for explanation; the overall process of extracting the elements to be reviewed and the values ​​of the elements to be reviewed by the model factory can be specifically divided into four processes: preliminary identification of elements, identification of elements, arrangement of elements, standardization of elements and mapping; Figure 2 The following steps are involved:

[0064] Step S201: Input HTML file.

[0065] It should be noted that the input HTML file is an HTML file of the document to be reviewed with complete location information and text information.

[0066] Step S202: Analyze using preliminary rules to identify elements to be reviewed.

[0067] Step S203: using the preset summary rules, product information or PICK model, the elements to be reviewed are processed to obtain the values ​​of the elements to be reviewed.

[0068] Step S204: sorting the elements to be reviewed and the values ​​of the elements to be reviewed.

[0069] It should be noted that element arrangement includes the splitting and merging of SPANs. Multiple SPANs represent the same element value or the same element value requires secondary analysis.

[0070] Step S205: The elements to be reviewed and their values ​​to be reviewed that have been sorted are subjected to standardization processing and knowledge mapping processing to obtain document instantiation data.

[0071] It should be noted that the knowledge mapping process is specifically the knowledge representation mapping with the document review knowledge graph. The element preliminary identification process corresponds to step S202, the element identification process corresponds to step S203, the element sorting process corresponds to step S204, and the element standardization and mapping process corresponds to step S205.

[0072] Step S104: Input the instantiation data of the letter of credit and the instantiation data of the document into the document review rule engine, so that the document review rule engine reviews the instantiation data of the letter of credit and the instantiation data of the document based on the review rules of the preset document review knowledge graph, and obtains and outputs the review results of the letter of credit and the document to be reviewed.

[0073] In the specific implementation of step S104, the instantiation data of the letter of credit and the instantiation data of the document are input into the document review rule engine. The document review rule engine reviews the instantiation data of the letter of credit and the instantiation data of the document based on the review rules of the preset document review knowledge graph, obtains and outputs the review results of the letter of credit and the document to be reviewed, and the review results include: all matching comparison data items and non-matching comparison data items.

[0074] In some embodiments, after the audit results are output, if the audit results do not meet the requirements, the information entered by the business personnel on the knowledge maintenance page can be used to complete the knowledge of the letter of credit instantiation data and the document instantiation data. When the audit instruction is detected, the completed letter of credit instantiation data and the document instantiation data are reviewed again through the document review rule engine to obtain an updated audit result.

[0075] In some embodiments, a document review knowledge graph is constructed based on the document entity. The document review knowledge graph is divided into a model layer and a data layer in terms of logical architecture. In the model layer, entity and relationship specifications are agreed upon based on ontological knowledge to form a graph structure definition. In the data layer, an attribute graph is used to store business and related document instances. In the application of the document review knowledge graph, the document review rule engine is used to implement compliance review of business documents based on the document review business scenario and deductive reasoning method.

[0076] For example Figure 3The workflow diagram of the document review rule engine is provided. The review data structure is standardized in the business ontology knowledge definition, which is the reference and interaction basis for the implementation of modules such as the entity data protocol converter, the document review knowledge graph construction, and the document review rule engine. The document review rule engine is divided into two parts: rule mapping and binding and rule logic execution. Rule mapping and binding are mainly used to map the review rule representation of the model layer to the business bill instance of the data layer, and establish the binding relationship between the rule instance and the bill instance attribute in the form of an edge. Among them, the rule instance defines the calculation logic of the review, and the bill instance stores the actual data. The document review rule engine fills the bill instance into the calculation logic of the rule instance for review. The rule logic execution function uses conditional triggering to assemble factual data and then executes rule reasoning, and finally outputs the review result (or audit conclusion).

[0077] In some embodiments, for the verified and published ontology representation (that is, the order review knowledge graph) and rule representation files, they can first be converted into node and edge files recognizable by the graph system through a knowledge storage tool, and then loaded into the graph system for storage through a graph system loading tool. The ontology SCHEMA knowledge and basic knowledge base entities are stored in the TigerGraph graph system.

[0078] It should be noted that all modules of the order review rule engine are deployed and run on the big data cloud, such as Figure 4 The architecture diagram of the big data cloud provided is divided into the collection layer, the stream computing layer and the data layer. The purpose of the streaming collection module of the collection layer is: the rule stream and the data stream enter Kafka through the streaming collection module; the purpose of Kafka of the stream computing layer is: the data stream enters the dimension table through Kafka; the purpose of Flink of the stream computing layer is: the ETL operator execution module, and the main program of the order review rule engine is executed in Flink; Mysql of the data layer is used to store the rule base, and TigerGraph of the data layer is used to store the ontology and basic knowledge base.

[0079] In some embodiments, the audit service of the audit rule engine is an independent service that can be called by the AI ​​automatic task scheduling module or when an audit instruction triggered by a business person is detected; the processing logic of the audit service is as follows: Figure 5 A processing logic diagram of the audit service provided. Figure 5 The following steps are involved:

[0080] Step S501: Determine whether the audit rule is excluded. If the audit rule is excluded, skip the audit rule and proceed to the next audit rule; if the audit rule is not excluded, execute step S502.

[0081] Step S502: According to the filtering conditions of the audit rule (such as applicable business type and document type, etc.), determine whether the audit rule is skipped. If the audit rule is skipped, skip the audit rule and enter the next audit rule; if the audit rule is not skipped, execute step S503.

[0082] Step S503: After calculating the antecedent conditions of the audit rule, determine whether the audit rule is triggered. If the audit rule is not triggered, or if the audit rule is skipped, generate the audit result; if the audit rule is triggered, execute step S504.

[0083] Step S504: after calculating the consequent condition of the audit rule, determine whether the audit rule is triggered. If the audit rule is not triggered, or if the audit rule is triggered, or if the audit rule is skipped, generate an audit result.

[0084] For example, suppose a certain audit rule is "exist(InvoiceClause.show_creditNumber)->exist(Invoice.creditNumber)&&equal(Invoice.creditNumber,DocumenteryCredit.creditNumber)"; the antecedent of this audit rule is "exist(InvoiceClause.show_creditNumber) = when the letter of credit requires the invoice to show the letter of credit number", and the consequent of this audit rule is "exist(Invoice.creditNumber)&&equal(Invoice.creditNumber,DocumenteryCredit.creditNumber) = the invoice must have the letter of credit number and it must be consistent with the letter of credit number". When the antecedent of this audit rule is true, the consequent of this audit rule will be triggered for audit.

[0085] In the embodiment of the present invention, the instantiation data of the letter of credit and the instantiation data of the document corresponding to the audited letter of credit and the document to be audited are obtained respectively. The instantiation data of the letter of credit and the instantiation data of the document are input into the audit rule engine configured with the audit rule for auditing, and the audit results of the audited letter of credit and the document to be audited are obtained and output. There is no need to manually audit the letter of credit and the document, which reduces the audit cost and improves the audit efficiency.

[0086] It can be seen from the contents of the above embodiments that after the relevant processing of the letter of credit and the documents to be reviewed, the review results of the letter of credit and the documents to be reviewed can be obtained by using the review rule engine. In actual application, multiple display pages can be set to display corresponding information to business personnel.

[0087] In some embodiments, the audit results of the pending letters of credit and pending documents are displayed on the audit result display page.

[0088] In other embodiments, in response to a viewing instruction, a letter of credit display page or a bill display page is displayed, the letter of credit display page displays the segmentation results and classification results of the letter of credit to be reviewed, and the bill display page displays the bill classification results and elements to be reviewed of the documents to be reviewed.

[0089] That is to say, set up the audit result display page, letter of credit display page and bill display page; business personnel can view information such as letter of credit terms segmentation and classification results through the letter of credit display page.

[0090] After maintenance, business personnel can view the analysis display page of the entire letter of credit, such as viewing the simple column display page of the letter of credit, or viewing the column display page of the letter of credit terms.

[0091] When the bill display page is displayed, first enter the bill classification page. The folders on the bill classification page are set with different color labels. Specifically, the current folder is set to purple, the folder with images is set to dark blue, and the folder without images is set to light blue. Business personnel can drag the image of the bill to the specified folder on the bill classification page for classification and maintenance, and process it according to the updated bill classification results when parsing the bill. In the bill element display page, the entity extraction results of the bill can be displayed according to the bill type. In each bill page, business personnel can edit or confirm the extraction results while comparing the sliced ​​image of the bill on one side of the screen. Some fields prompt error correction information and entity link results in the form of rich text. Business personnel can also use the online regional OCR function to slice and identify bills.

[0092] It should be noted that in order to meet the needs of business personnel, business personnel can choose AI processing (using the method of this solution) or traditional document review business operation process processing. Business personnel can also choose to use AI processing results after confirming all bill entities. The AI ​​system will output the review results according to the bill entities and review rules confirmed by the business personnel, and display the review results on the review result display page. The review result display page displays the discrepancy output statement (business description when the review rules do not meet the requirements), review type, review data item, review data source, and review rule source; prompts are given for "consistency" or "inconsistency" in the review results, and a single review rule can also be traced back to the location of the original document.

[0093] After the audit result is displayed on the audit result display page, the business personnel can abandon the audit result and enter the traditional audit business operation process. In some embodiments, when an abandonment instruction to abandon the audit result is detected, the page for executing the traditional audit business operation process is switched.

[0094] Similarly, business personnel can choose to use the audit results displayed in the audit result display page. Specifically, in response to a submit instruction, they enter the product operation page with the field information automatically filled in.

[0095] The above content is about the relevant description of the review method of documentary credit, please refer to Figure 6 , Figure 6 A schematic diagram of the processing logic architecture of the documentary credit review method provided by an embodiment of the present invention is provided. Figure 6 It includes five parts: intelligent image recognition, audit data preparation and processing, knowledge completion, document review knowledge graph and AI pre-review results. Among them, the intelligent image recognition part is used to extract the letter of credit message and document data from the images of the letter of credit to be reviewed and the documents to be reviewed. It should be noted that if the format of the letter of credit to be reviewed is text, there is no need to perform intelligent image recognition on the letter of credit to be reviewed, and the letter of credit message can be directly obtained from the letter of credit to be reviewed; the audit data preparation processing part is used to obtain the letter of credit instantiation data and document instantiation data, the knowledge completion part is used to complete the letter of credit instantiation data and document instantiation data (non-essential step), the document review knowledge graph part is used to review the letter of credit instantiation data and document instantiation data, and the AI ​​pre-review result part is used to display the audit results.

[0096] Applying this solution in actual business can provide business personnel with a variety of document review needs; such as Figure 7 The business architecture diagram of the documentary credit review method provided is that the "intelligent document review" part can be used to process the documents and the credit to obtain the review result (also called the document review result); the document review expert can also manually review the documents and the credit. It should be noted that the execution principle of the "intelligent document review" part can be found in the above-mentioned embodiment of the present invention. Figure 1 The contents in will not be repeated here.

[0097] Corresponding to the method for reviewing a documentary credit provided by the above embodiment of the present invention, see Figure 8 , the embodiment of the present invention also provides a structural block diagram of a documentary credit review system, the review system comprising: an acquisition unit 801, a first processing unit 802, a second processing unit 803 and a review unit 804;

[0098] The acquisition unit 801 is used to acquire the letter of credit message corresponding to the letter of credit to be reviewed, and to acquire the document data corresponding to the document to be reviewed.

[0099] The first processing unit 802 is used to perform a first natural language processing on the letter of credit message to obtain letter of credit instantiation data.

[0100] In some embodiments, the first natural language processing includes at least: message splicing, extraction processing, processing of semi-structured fields, structuring processing of order review-related terms, structuring processing of product information, classification and understanding processing of order review-irrelevant terms, standardization processing and knowledge mapping processing.

[0101] The second processing unit 803 is used to perform a second natural language processing on the document data to obtain document instantiation data.

[0102] The review unit 804 is used to input the instantiation data of the letter of credit and the instantiation data of the document into the review rule engine, so that the review rule engine reviews the instantiation data of the letter of credit and the instantiation data of the document based on the review rules of the preset review knowledge graph, and obtains and outputs the review results of the letter of credit and the document to be reviewed.

[0103] In some embodiments, the audit unit 804 for outputting the audit results of the pending credit and the pending documents is specifically used to: display the audit results of the pending credit and the pending documents on the audit result display page.

[0104] In the embodiment of the present invention, the instantiation data of the letter of credit and the instantiation data of the document corresponding to the audited letter of credit and the document to be audited are obtained respectively. The instantiation data of the letter of credit and the instantiation data of the document are input into the audit rule engine configured with the audit rule for auditing, and the audit results of the audited letter of credit and the document to be audited are obtained and output. There is no need to manually audit the letter of credit and the document, which reduces the audit cost and improves the audit efficiency.

[0105] Preferably, combined Figure 8 The second processing unit 803 includes: an extraction module and a processing module. The execution principle of each module is as follows:

[0106] The extraction module is used to extract the elements to be reviewed and the values ​​of the elements to be reviewed from the document data.

[0107] In the specific implementation, the extraction module is specifically used to: perform preliminary element recognition processing on the document data to extract the elements to be reviewed; use the preset summary rules, product information or PICK model to process the elements to be reviewed to obtain the values ​​of the elements to be reviewed, and the PICK model is based on the graph neural network and conditional random field model.

[0108] The processing module is used to sort, standardize and knowledge map the elements to be reviewed and their values ​​to obtain document instantiation data.

[0109] Preferably, combined Figure 8 The content shown, the review system also includes:

[0110] The switching unit is used to switch to a page for executing a traditional document review business operation process when an abandonment instruction for abandoning the review result is detected.

[0111] Preferably, combined Figure 8 The content shown, the review system also includes:

[0112] The viewing unit is used to display a letter of credit display page or a bill display page in response to a viewing instruction. The letter of credit display page displays the segmentation results and classification results of the letter of credit to be reviewed, and the bill display page displays the bill classification results and elements to be reviewed of the document to be reviewed.

[0113] In summary, the embodiment of the present invention provides a method and system for reviewing documentary letters of credit, which respectively obtains letter of credit instantiation data and document instantiation data corresponding to the reviewed letter of credit and the document to be reviewed. The letter of credit instantiation data and the document instantiation data are input into a document review rule engine configured with document review rules for review, and the review results of the letter of credit to be reviewed and the document to be reviewed are obtained and output. There is no need for manual review of the letter of credit and the document, which reduces the review cost and improves the review efficiency.

[0114] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.

[0115] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0116] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for reviewing a documentary letter of credit, characterized in that: The method comprises: Obtain the letter of credit message corresponding to the letter of credit to be reviewed, and obtain the document data corresponding to the document to be reviewed; Performing a first natural language processing on the letter of credit message to obtain instantiation data of the letter of credit; the first natural language processing at least includes: message splicing, extraction processing, processing of semi-structured fields, structuring processing of clauses related to document review, structuring processing of commodity information, classification and understanding processing of clauses irrelevant to document review, standardization processing and knowledge mapping processing; Performing a second natural language processing on the document data to obtain document instantiation data; The instantiation data of the letter of credit and the instantiation data of the document are input into a document review rule engine, so that the document review rule engine reviews the instantiation data of the letter of credit and the instantiation data of the document based on the review rules of a preset document review knowledge graph, and displays the review results of the letter of credit to be reviewed and the document to be reviewed on a review result display page; all modules of the document review rule engine are deployed and run on a big data cloud.

2. The method according to claim 1, characterized in that Performing a second natural language processing on the document data to obtain document instantiation data includes: Extracting the elements to be reviewed and the values ​​of the elements to be reviewed from the document data; The elements to be reviewed and their values ​​are sorted, standardized and knowledge mapped to obtain document instantiation data.

3. The method according to claim 2, characterized in that Extracting the elements to be reviewed and the values ​​of the elements to be reviewed from the document data includes: Performing preliminary element recognition processing on the document data to extract the elements to be reviewed; The elements to be reviewed are processed using preset summary rules, product information or a PICK model to obtain the values ​​of the elements to be reviewed. The PICK model is based on a graph neural network and a conditional random field model.

4. The method according to claim 1, characterized in that: After the audit results of the pending letter of credit and the pending document are displayed on the audit result display page, it also includes: When an abandonment instruction to abandon the review result is detected, switch to a page for executing a traditional document review business operation process.

5. The method according to claim 2, characterized in that: The method further comprises: In response to the viewing instruction, a letter of credit display page or a bill display page is displayed, wherein the letter of credit display page displays the segmentation result and classification result of the letter of credit to be reviewed, and the bill display page displays the bill classification result and the elements to be reviewed of the document to be reviewed.

6. A documentary credit review system, characterized in that: The system comprises: An acquisition unit, used to acquire the letter of credit message corresponding to the letter of credit to be reviewed, and to acquire the document data corresponding to the document to be reviewed; a first processing unit, configured to perform a first natural language processing on the letter of credit message to obtain instantiation data of the letter of credit; the first natural language processing at least includes: message splicing, extraction processing, processing of semi-structured fields, structuring processing of clauses related to document examination, structuring processing of commodity information, classification and understanding processing of clauses irrelevant to document examination, standardization processing and knowledge mapping processing; A second processing unit, configured to perform a second natural language processing on the document data to obtain document instantiation data; The review unit is used to input the instantiation data of the letter of credit and the instantiation data of the document into the review rule engine, so that the review rule engine reviews the instantiation data of the letter of credit and the instantiation data of the document based on the review rules of the preset review knowledge graph, and displays the review results of the letter of credit to be reviewed and the document to be reviewed in the review result display page; all modules of the review rule engine are deployed and run on the big data cloud.

7. The system according to claim 6, characterized in that The second processing unit comprises: An extraction module, used for extracting the elements to be reviewed and the values ​​of the elements to be reviewed from the document data; The processing module is used to perform element sorting, standardization processing and knowledge mapping processing on the elements to be reviewed and the values ​​of the elements to be reviewed, so as to obtain document instantiation data.

Images