Order BOM differentiation intelligent checking method and system

Abstract

The application discloses an order BOM differentiation intelligent checking method and system, relates to the technical field of data processing and supply chain management, and comprises the following steps: checking the preliminary structured supplier BOM data based on the BOM detailed amount relationship and the order main amount relationship, positioning and correcting the abnormal fields in the checking, and obtaining target structured supplier BOM data; taking the target structured supplier BOM data as external actual BOM data, unifying the checking caliber with internal standard order BOM data, obtaining internal BOM checking units and external BOM checking units; extracting material description features, key attribute features and hierarchical path features of each BOM checking unit, and fusing and generating corresponding comprehensive features. The application realizes full-link automatic checking from order BOM structured extraction to internal and external BOM accurate matching and field-level difference attribution, and improves the reliability and difference positioning accuracy of order BOM checking.

Description

Technical Field

[0001] This invention relates to the fields of data processing and supply chain management technology, and in particular to an intelligent method and system for checking order BOM differences. Background Technology

[0002] With the digital development of enterprise procurement management and supply chain collaboration systems, order BOM reconciliation is gradually shifting from manual reconciliation to assisted reconciliation based on ERP, email parsing, and automated rules. In actual business operations, enterprises typically maintain standard purchase orders and standard BOM data using the ERP system, while suppliers provide feedback on actual executed orders, BOM details, quantities, unit prices, amounts, and cost descriptions via email bodies, PDFs, images, Excel spreadsheets, or text attachments. Existing order BOM reconciliation methods usually begin by extracting the order master information and BOM detail fields from the supplier's order file, and then compare them with the internal standard BOM according to material code, material name, or amount fields to determine if there are any discrepancies in the order execution content.

[0003] However, existing order BOM verification technologies still have the following shortcomings: First, the supplier order file formats vary greatly, and traditional fixed templates, single OCR, or regular expression extraction methods are difficult to reliably identify non-standard tables, mixed text, and cost descriptions. Furthermore, they lack a reverse verification mechanism based on the relationship between BOM details and the order master amount, which can easily lead to errors in field extraction in subsequent verifications. Second, the granularity of matching internal and external BOM nodes and locating differences is relatively coarse. Existing methods mostly rely on material codes or string similarity for line-by-line matching, which is difficult to handle situations such as inconsistencies in material descriptions between internal and external systems, conflicts in key attributes, and differences in BOM hierarchy relationships. This results in only providing macro-level prompts for inconsistent amounts or mismatched line items, making it difficult to further locate the specific BOM node, difference field, quantity deviation, unit price deviation, or additional cost deviation. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent verification method and system for order BOM differences, which solves the problems of insufficient reliability in parsing heterogeneous supplier order data, insufficient ability to match internal and external BOMs, and insufficient ability to attribute field-level differences.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an intelligent verification method for order BOM differentiation, comprising, Obtain supplier order data and corresponding internal standard order BOM data, perform layout area recognition and semantic extraction on supplier order data, and obtain preliminary structured supplier BOM data containing order master information and BOM detail lines; The preliminary structured supplier BOM data is validated based on the BOM detail amount relationship and the order master amount relationship, and the abnormal fields are located and corrected to obtain the target structured supplier BOM data. The target structured supplier BOM data is used as the external actual BOM data, and the verification standards are unified with the internal standard order BOM data to obtain the internal BOM verification unit and the external BOM verification unit. Extract the material description features, key attribute features, and hierarchical path features of each BOM verification unit, and merge them to generate corresponding comprehensive features; The matching cost between internal and external BOM verification units is calculated based on comprehensive features. The optimal matching relationship between internal and external BOM verification units is determined by combining material category, key attributes, amount tolerance and hierarchical path constraints. Based on the optimal matching relationship, the matched BOM verification units are compared at the field level, and an order BOM difference positioning and attribution report is generated by combining the unmatched BOM verification units.

[0007] As a preferred embodiment of the intelligent verification method for order BOM differentiation described in this invention, the supplier order data includes the supplier email body and / or supplier email attachments, as well as order identification information, order master information and BOM detail information identified from the supplier email body and / or email attachments. The internal standard order BOM data includes internal purchase order information corresponding to the order identification information, standard BOM node information, BOM hierarchy information, and order amount information.

[0008] As a preferred embodiment of the intelligent verification method for order BOM differentiation described in this invention, the step of obtaining preliminary structured supplier BOM data containing order master information and BOM detail rows includes: preprocessing the supplier email body and / or supplier email attachments under the same verification task to obtain preprocessed document data containing text content and table structure; identifying table areas and text areas in the preprocessed document data through a layout analysis model, and determining the order master information area and BOM detail table area; inputting the order master information area and BOM detail table area into a large language model for semantic extraction to obtain order master information, BOM detail row information and BOM hierarchy relationship, and generating preliminary structured supplier BOM data.

[0009] As a preferred embodiment of the order BOM differentiation intelligent verification method described in this invention, the step of obtaining the target structured supplier BOM data includes: performing row-level amount verification based on the quantity field, unit price field, and amount field in each BOM detail row, and performing order master-level amount verification based on the sum of amounts in each BOM detail row and the amount of the master order; when the row-level amount verification or the order master-level amount verification is abnormal, the abnormal field is located according to the abnormal amount relationship and the corresponding original context, and the abnormal field is extracted and corrected a second time; when the verification still fails after the second extraction and correction, the abnormal field is partially completed or marked as pending review, and the target structured supplier BOM data is generated.

[0010] As a preferred embodiment of the order BOM differentiation intelligent verification method of the present invention, the step of obtaining the internal BOM verification unit and the external BOM verification unit includes: determining the target structured supplier BOM data as the external actual BOM data, and unifying the field scope and BOM level scope with the internal standard order BOM data under the same verification task; based on the unified field scope and BOM level scope, converting the internal standard BOM node and the external actual BOM node into the internal BOM verification unit and the external BOM verification unit, respectively.

[0011] As a preferred embodiment of the order BOM differentiation intelligent verification method of the present invention, the step of extracting the material description features, key attribute features, and hierarchical path features of each BOM verification unit and fusing them to generate corresponding comprehensive features includes: generating dense semantic vectors based on the material descriptions of each BOM verification unit, and extracting at least one key attribute from specifications, dimensions, materials, and drawing numbers to generate sparse attribute vectors; generating hierarchical path features according to the BOM hierarchical relationship of each BOM verification unit, and associating and fusing the dense semantic vectors, sparse attribute vectors, and hierarchical path features to obtain the comprehensive features of the corresponding BOM verification unit.

[0012] As a preferred embodiment of the order BOM differentiation intelligent verification method of the present invention, the step of calculating the matching cost between the internal BOM verification unit and the external BOM verification unit based on comprehensive features includes: combining each internal BOM verification unit and each external BOM verification unit in pairs to form a pair of nodes to be matched based on the comprehensive features of each BOM verification unit, and calculating semantic similarity based on the cosine similarity of dense semantic vectors and the Jaccard similarity of sparse attribute vectors; calculating the corresponding matching cost according to the semantic similarity, relative difference in quantity, and relative difference in unit price of the pair of nodes to be matched according to preset weights, and rejecting or downgrading the pair of nodes to be matched that have hard attribute conflicts; and writing the matching cost of all pairs of nodes to be matched into a multidimensional cost matrix.

[0013] As a preferred embodiment of the order BOM differentiation intelligent verification method of the present invention, the step of determining the optimal matching relationship between internal and external BOM verification units includes: judging the material category, key attributes, amount tolerance and BOM level path constraints of the matching node pairs in the multidimensional cost matrix, and setting the cost value of the matching node pairs that do not meet the constraints to a preset maximum value; solving the matching relationship of internal and external BOM verification units with the minimum total cost value based on the constrained multidimensional cost matrix, and marking the BOM verification units that have not entered the matching relationship as isolated BOM verification units.

[0014] As a preferred embodiment of the intelligent verification method for order BOM differentiation described in this invention, the step of performing field-level difference comparison on the matched BOM verification units according to the optimal matching relationship, and generating an order BOM differentiation location and attribution report in conjunction with the unmatched BOM verification units, includes: The quantity difference, unit price difference, and amount difference of each matching node pair are calculated based on the optimal matching relationship, and the difference nodes are determined based on the preset quantity threshold, preset unit price threshold, and preset amount threshold. By using a pre-set difference decomposition decision tree, the difference nodes are attributed to the BOM quantity deviation factor and / or BOM unit price deviation factor, and the additional cost deviation factor is identified based on the difference between the amount of the main order on the supplier side and the sum of the amounts of the matched external BOM verification units. Based on the BOM quantity deviation factor, BOM unit price deviation factor, additional cost deviation factor, and the missing or added BOM information corresponding to the isolated BOM verification unit, generate an order BOM differentiation positioning attribution report.

[0015] Secondly, this invention provides an intelligent verification system for order BOM differentiation, comprising, The parsing and reconstruction module is used to obtain supplier order data, perform layout area recognition, semantic extraction and reverse error correction of the supplier order data, and generate target structured supplier BOM data. The cross-domain semantic alignment module is used to extract material description features, key attribute features, and hierarchical path features from internal BOM verification units and external BOM verification units, and then fuse them to generate comprehensive features. The nested optimization matching module is used to calculate the matching cost between internal and external BOM verification units based on comprehensive features, and determine the optimal matching relationship in combination with business constraints; The difference attribution module is used to perform field-level difference comparisons based on the optimal matching relationship, and generate an order BOM difference positioning attribution report in conjunction with the isolated BOM verification unit.

[0016] The beneficial effects of this invention are as follows: By performing layout area recognition, semantic extraction, and reverse verification of monetary relationships on the body and attachments of supplier emails, non-standard tables, scanned documents, and mixed text orders can be stably converted into target structured supplier BOM data, reducing the impact of field extraction errors on subsequent verification; at the same time, by constructing internal and external BOM verification units, integrating material descriptions, key attributes, and hierarchical path features, and combining matching costs and business constraints to determine the optimal matching relationship, the matching accuracy of internal and external BOM nodes under conditions of inconsistent names, attribute differences, and complex hierarchies is improved; furthermore, through field-level difference comparison and isolated node attribution, order differences can be located to specific BOM nodes and the reasons for quantity, unit price, additional costs, or missing or added items, thereby improving the efficiency of order BOM verification and the accuracy of difference tracing. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the intelligent verification method for order BOM differentiation in Example 1.

[0019] Figure 2 This is a structural diagram of the intelligent verification system for order BOM differentiation in Example 1. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, referring to Figure 1This is the first embodiment of the present invention, which provides an intelligent verification method for order BOM differentiation, including the following steps: S1. Obtain supplier order data and corresponding internal standard order BOM data, perform layout area recognition and semantic extraction on supplier order data, and obtain preliminary structured supplier BOM data containing order master information and BOM detail lines. S1.1: Obtain supplier order data and corresponding internal standard order BOM data; Specifically, the system reads order emails sent by suppliers through the enterprise email gateway interface, obtains the email body and / or attachments, and uses the email body and / or attachments as supplier order data. It then identifies order identification information from the supplier order data, including at least one of the following: purchase order number, supplier identifier, project number, or order batch number. Based on the order identification information, it retrieves the corresponding internal purchase order record in the enterprise's internal ERP system and reads the standard BOM node information, BOM hierarchy information, and order amount information under that internal purchase order record to obtain the internal standard order BOM data corresponding to the supplier order data.

[0024] S1.2: Confirm the correspondence between supplier order data and internal standard order BOM data; Specifically, when order identification information corresponds to multiple internal purchase order records or multiple BOM versions, the system filters according to the following order: consistent purchase order number, consistent supplier identification, consistent project number, and BOM version in effect, to determine the unique corresponding internal standard order BOM data. If the unique corresponding internal standard order BOM data cannot be determined, the current supplier order data is marked as pending confirmation and will not enter the subsequent automatic verification process. If a unique correspondence is determined, the supplier order data and the internal standard order BOM data are used as input data for the same order BOM verification task.

[0025] S1.3: Read supplier order data under the same verification task and preprocess it to obtain preprocessed document data; Specifically, after determining that the supplier order data and the internal standard order BOM data belong to the same order BOM verification task, the system reads the supplier email body and / or email attachments under that verification task and uses them as the objects to be parsed. Email attachments include PDF files, image files, non-standard Excel files, or text files. When the object to be parsed is a scanned PDF or image file, the system performs denoising, binarization, and text recognition processing on the scanned PDF or image file to obtain the corresponding text content and image area information. When the object to be parsed is a spreadsheet, HTML table, or text file, the system extracts the text content, table row and column structure, and cell content. The system associates the text content, table structure, and their regional positions in the original document to form preprocessed document data, which is then used as input for layout area recognition.

[0026] S1.4: Identify table and text areas in preprocessed document data using a layout analysis model; Specifically, the layout analysis model is invoked to divide the preprocessed document data into regions, identifying table regions and text regions. The table region is used to carry at least the BOM detail lines in the supplier's order, and the text region is used to carry at least the order number, supplier name, total order amount, and cost description or remarks. Based on the content attributes of the table region and the text region, the order master information region, BOM detail table region, and cost description region are further determined, and the original region information, text content, or table structure corresponding to each region is preserved.

[0027] Furthermore, the layout analysis model can be implemented using existing document layout analysis networks or object detection networks. Its input is preprocessed supplier order document data, including image regions corresponding to scanned PDFs or images, table structures corresponding to spreadsheets or HTML tables, and text layout information corresponding to text files. Its output is the region category and region location of each layout region in the document. The region category includes at least table regions and text regions, and can be further labeled as order master information region, BOM detail table region, or cost description region based on the region content. The layout analysis model is trained or fine-tuned based on historical supplier order document samples. The training samples pre-label table regions, text regions, and corresponding boundary ranges, enabling the model to identify non-standard tables, body descriptions, and cost description locations in different supplier email attachments. In actual operation, the layout analysis model does not directly generate BOM field values, but outputs regionalized document content for subsequent large language model extraction, thereby reducing field confusion caused by direct extraction of the entire document.

[0028] It should be noted that the order master information area is used to extract the order identifier and master order amount later, the BOM details area is used to extract the material name, specifications, quantity, unit price and amount later, and the cost description area is used to identify non-BOM details such as freight, packaging fees, and taxes later.

[0029] S1.5: Based on the page layout area recognition results, perform semantic extraction to generate preliminary structured supplier BOM data.

[0030] Specifically, the region type, region text, table row and column content, and context headings of the order master information area, BOM detail table area, and cost description area are input into the large language model. The extraction fields are limited by targeted extraction prompts, enabling the large language model to extract order master information, BOM detail row information, and non-BOM cost description information respectively. For BOM detail tables with multi-level structures, the hierarchical relationship between BOM detail rows is determined based on table indentation, level number, parent-child row relationship, or context headings. The extracted order master information, BOM detail row information, BOM hierarchical relationship information, and non-BOM cost description information are organized according to a preset structured format to obtain preliminary structured supplier BOM data.

[0031] Furthermore, the large language model can be implemented using existing text-based large language models or multimodal large language models. Its input is the regionalized document content output by the layout analysis model, including the text in the order master information area, the header and row / column content in the BOM details area, the text in the cost description area, and corresponding targeted extraction prompts. Its output is preliminary structured supplier BOM data containing order master information, BOM details row information, BOM hierarchical relationship information, and non-BOM cost description information. The large language model can directly call pre-trained models with information extraction capabilities, or it can use historical order emails, manual verification results, and standard structured BOM samples for fine-tuning, enabling the model to learn to map non-standard supplier descriptions to preset field formats. During actual extraction, targeted extraction prompts limit the output fields, field formats, and hierarchical relationships to avoid misidentifying remarks, headers, or additional costs as BOM details rows, and provide structured input for subsequent business logic verification and reverse error correction.

[0032] It should be noted that the order master information includes one or more of the following: order identifier, supplier information, master order amount, and order date. The BOM detail line information includes one or more of the following: BOM line number, material name, component description, specifications, quantity, unit price, amount, and remarks.

[0033] S2. Based on the BOM detail amount relationship and the order master amount relationship, the preliminary structured supplier BOM data is validated, and the abnormal fields are located and corrected to obtain the target structured supplier BOM data. S2.1: Read the preliminary structured supplier BOM data and determine the amount verification object; Specifically, the system reads the order master information, BOM detail line information, BOM hierarchy information, and non-BOM cost description information from the preliminary structured supplier BOM data; it uses the quantity field, unit price field, and amount field in each BOM detail line as the BOM detail amount verification object, and uses the sum of the amount fields of all BOM detail lines and the master order amount field in the order master information as the order master amount verification object.

[0034] S2.2: Perform row-level validation on BOM detail rows based on the BOM detail amount relationship; Specifically, the quantity, unit price, and amount fields in each BOM detail line are read line by line according to the BOM line number, and it is determined whether each BOM detail line satisfies the business amount relationship of "amount = quantity × unit price". When the amount field of a BOM detail line is consistent with the product of the quantity field and the unit price field, or the difference between the two is within the preset amount tolerance range, the BOM detail line is marked as passing the line-level verification. When the difference between the two exceeds the preset amount tolerance range, the BOM detail line is marked as failing the line-level verification, and the abnormal BOM line number, abnormal amount relationship, and the original area information corresponding to the BOM detail line are recorded.

[0035] The tolerance range for the amount is determined based on the accuracy of the order amount, financial settlement rules, or the terms of the purchase contract.

[0036] S2.3: Perform master-level validation on supplier orders based on the master-amount relationship of the order; Specifically, the amount fields of all BOM detail lines are summed to obtain the total BOM detail amount, and then compared with the main order amount in the order master information. When the total BOM detail amount matches the main order amount, or the difference between the two is within the preset amount tolerance range, the order master level verification is deemed to have passed. When the difference exceeds the preset amount tolerance range, the order master level verification is deemed to have failed, and the main order amount, the total BOM detail amount, the amount difference, and non-BOM expense descriptions that may be related to the amount difference are recorded.

[0037] S2.4: Locate the abnormal field based on the verification anomaly result and perform reverse error correction; Specifically, based on the "Amount = Quantity × Unit Price" validation result of the BOM detail rows with row-level validation anomalies, the system locates potentially abnormal quantity, unit price, or amount fields and generates an error location prompt with the abnormal BOM row number, abnormal field type, abnormal amount relationship, and corresponding original context. For data with order master-level validation anomalies, based on the amount difference between the BOM detail amount summary value and the master order amount, the system locates potentially abnormal order master amount field, BOM detail amount field, or non-BOM expense description field and generates corresponding error location prompts. The error location prompts, the original text or table area content corresponding to the abnormal fields, and the context information are then input into the large language model again, allowing the large language model to perform secondary targeted extraction and correction of the abnormal fields. When the secondary extraction result passes the validation of the corresponding BOM detail amount relationship or order master amount relationship, the corresponding abnormal field in the initial structured supplier BOM data is replaced with the secondary extraction result.

[0038] S2.5: Perform partial fallback completion on fields that still fail validation and generate target structured supplier BOM data.

[0039] Specifically, if the large language model fails validation after secondary extraction and correction, local regular expressions are used for fallback completion of the original local area containing the abnormal field. For quantity, unit price, and amount fields, completion is performed based on number format, currency symbol, and adjacent text. For the order master amount field, completion is performed based on the amount keyword in the order master information area and its adjacent values. If the completion result passes the corresponding amount relationship validation, the completion result is written to the corresponding field. If the completion result still fails validation, the original extraction result is retained, and the field is marked as a field to be reviewed. The fields that pass validation, corrected fields, completed fields, and fields to be reviewed are summarized to form the target structured supplier BOM data.

[0040] S3. Use the target structured supplier BOM data as the external actual BOM data and verify it against the internal standard order BOM data to obtain the internal BOM verification unit and the external BOM verification unit. S3.1: Receive the target structured supplier BOM data and call the internal standard order BOM data under the same verification task; Specifically, the target structured supplier BOM data is identified as the supplier's actual execution BOM data; the internal standard order BOM data with confirmed corresponding relationships is called up based on the same order BOM verification task.

[0041] It should be noted that the BOM data actually executed by the supplier includes at least the order master information, BOM detail line information, and BOM hierarchy information on the supplier side, while the BOM data of the internal standard order includes at least the standard BOM node information, BOM hierarchy information, and order amount information on the internal side. This forms the BOM data of the internal and external sides that need to be standardized.

[0042] S3.2: Standardize the field definitions for the supplier's actual BOM data and the internal standard order BOM data; Specifically, the material name, component description, specifications, quantity, unit price, and amount fields in the supplier's actual BOM data are mapped to the material code, material description, material category, standard quantity, standard unit price, and standard amount fields in the internal standard order BOM data. This ensures that both internal and external BOM data are grouped into a unified verification field structure that includes material description, key attributes, material category, quantity, unit price, and amount. Fields with different names but the same business meaning are grouped into the same verification field according to a preset field correspondence. Non-BOM detailed expense descriptions such as freight, packaging fees, and taxes in the supplier's actual BOM data are not included as BOM detailed points in this step's BOM node generation but are retained as the basis for subsequent identification of additional cost deviation factors.

[0043] S3.3: Standardize the hierarchical definitions of the supplier's actual BOM data and the internal standard order BOM data; Specifically, based on the BOM hierarchy in the internal standard order BOM data, the parent-child or sibling relationships between internal standard BOM nodes are determined. Simultaneously, based on the identified BOM hierarchy information in the target structured supplier BOM data, the parent-child or sibling relationships between actual supplier BOM nodes are determined. For BOM detail lines in the supplier's actual execution BOM data that already have hierarchy numbers, indentation relationships, or parent-child line relationships, they are retained according to their identified hierarchy. For BOM detail lines that do not explicitly show a hierarchy but belong to the same order master information, they are treated as sibling BOM nodes under the same order master information, thus enabling both internal and external BOM data to enter the subsequent verification process in a comparable BOM node format.

[0044] S3.4: Generate internal BOM verification units and external BOM verification units based on the unified field and hierarchical definitions; Specifically, based on the unified field and hierarchical definitions, each standard BOM node in the internal standard order BOM data is converted into an internal BOM verification unit, and each actual BOM node in the supplier's actual execution BOM data is converted into an external BOM verification unit. Each BOM verification unit includes at least node identifier, material description, key attributes, material category, quantity, unit price, amount, and hierarchical relationship information. This results in a set of comparable internal BOM verification units and a set of comparable external BOM verification units.

[0045] S4. Extract the material description features, key attribute features, and hierarchical path features of each BOM verification unit, and merge them to generate the corresponding comprehensive features; S4.1: Read the internal BOM verification unit and the external BOM verification unit; Specifically, the system receives an internal BOM verification unit set and an external BOM verification unit set, and uses each BOM verification unit as a feature extraction object. Each BOM verification unit already has unified material description, key attributes, material category, quantity, unit price, amount, and hierarchical relationship information, which serve as the basic data for generating comprehensive features in this step.

[0046] S4.2: Extract the material description features of each BOM verification unit; Specifically, the material description field in each BOM verification unit is read, and the text used to characterize the material content in the material name, component description, specification model description or remarks description is merged into component description text; the component description text is converted into dense semantic vectors through a pre-trained text embedding model to obtain the material description features of the BOM verification unit.

[0047] Dense semantic vectors are used to characterize the semantic similarity between material text descriptions in different systems, in order to accommodate situations where internal systems and supplier systems have different naming methods for the same material.

[0048] S4.3: Extract the key attribute features of each BOM verification unit; Specifically, one or more key attributes, such as specifications, dimensions, materials, and drawing numbers, are extracted from the component description text and existing key attribute fields using regular expressions or named entity recognition technology. The extraction results are then encoded into sparse attribute vectors to obtain the key attribute features of the BOM verification unit. Specifications, dimensions, materials, and drawing numbers are used to define the verifiable attributes of the materials. For materials, drawing numbers, or core specifications that are preset as hard attributes, their attribute values ​​are retained.

[0049] S4.4: Generate the hierarchical path characteristics of each BOM verification unit; Specifically, based on the unified BOM hierarchy, the parent affiliation, node hierarchy, or sibling affiliation of each BOM verification unit in its respective order BOM is determined, and this is used as the hierarchical path feature.

[0050] Hierarchical path features do not change the semantics and attribute representation of the material itself, but are used to determine whether the internal BOM verification unit and the external BOM verification unit are within the range of corresponding BOM structures during subsequent matching, thereby reducing the possibility of cross-level mismatch.

[0051] S4.5: Integrate and generate the comprehensive characteristics of each BOM check unit; Specifically, the dense semantic vector, sparse attribute vector, and hierarchical path features of the same BOM core unit are associated to form the comprehensive features of that BOM core unit.

[0052] S5. Calculate the matching cost between internal and external BOM verification units based on comprehensive features, and determine the optimal matching relationship between internal and external BOM verification units by combining material category, key attributes, amount tolerance and hierarchical path constraints. S5.1: Read the comprehensive characteristics of the internal and external BOM check units and form a pair of nodes to be matched; Specifically, it receives the set of internal BOM verification units, the set of external BOM verification units, and the comprehensive features corresponding to each BOM verification unit; it combines each internal BOM verification unit with each external BOM verification unit in pairs to form a pair of nodes to be matched, which are used as the objects for matching cost calculation.

[0053] S5.2: Calculate the semantic similarity of the pairs of nodes to be matched; Specifically, for any internal BOM verification unit X and external BOM verification unit Y, cosine similarity is calculated based on the dense semantic vector in their combined features, and Jaccard similarity is calculated based on the sparse attribute vector in their combined features. The material, drawing number, core specifications, and dimensions of X and Y are standardized and compared. If the same hard attribute field has valid values ​​in both X and Y, and the values ​​are different or the difference exceeds the preset tolerance range, it is determined to be a hard attribute conflict. For drawing number, material, or core specification conflicts, the semantic similarity of the node pair is set to 0 or the matching cost is set to the maximum value. For tolerable size conflicts, the semantic similarity of the node pair is multiplied by a preset weighting coefficient. If there is no hard attribute conflict, the cosine similarity and Jaccard similarity are weighted and fused to obtain the semantic similarity between X and Y.

[0054] It should be noted that the preset tolerance range should be no less than 0 and not exceed the upper limit of error allowed by the purchase contract, financial settlement accuracy, or enterprise BOM verification rules. The quantity tolerance is determined based on the smallest unit of measurement of materials or the shortfall or surplus range allowed by the contract. The unit price and amount tolerance are determined based on the smallest settlement unit of currency, tax rounding rules, or historical manual verification through samples. The preset weighting coefficient should be between 0 and 1 and not equal to 1. A value of 0 indicates that the match is directly rejected when there is a conflict in hard attributes. A decimal value between 0 and 1 indicates that the semantic similarity is reduced. The specific value is determined based on the degree of influence of hard attributes such as material, drawing number, and core specifications on the consistency of materials and the calibration of historical misjudgment samples.

[0055] S5.3: Calculate the matching cost of the node pair to be matched by combining the quantity deviation and the unit price deviation; Specifically, the quantity field and unit price field corresponding to the node pair X and Y to be matched are read, the relative difference in quantity and the relative difference in unit price are calculated respectively, and the semantic difference, the relative difference in quantity and the relative difference in unit price are unified into the matching cost; Matching cost Calculated using the following formula:

[0056] In the formula, These are semantic weight, quantity deviation weight, and unit price deviation weight, respectively, and satisfy the following conditions: , It is the semantic difference cost. It is the cost of quantity difference. It is the unit price difference cost.

[0057] It should be noted that, preferably, The value is 0.5. The value is 0.3. The value is 0.2; The value of 0.5 is based on the fact that order BOM verification first needs to determine whether the internal and external BOM nodes are the same or similar materials. Semantic similarity has the greatest impact on the correctness of matching, so the semantic weight is set to the highest. The basis for the value of 0.3 is that the quantity difference directly reflects the delivery deviation between the supplier's actual BOM and the internal standard BOM, which has a high impact on the positioning of the difference, but should be lower than the material semantic matching itself. The value of 0.2 is based on the fact that the unit price difference is mainly used to help judge the price deviation under the same BOM node. Its role is weaker than material matching and quantity consistency, so it is set to a relatively low weight. The semantic difference sub-cost is obtained by first calculating the cosine similarity between the integrated feature vectors of the internal BOM node and the supplier BOM node, and then subtracting the similarity from one; if there is a hard conflict between the key sparse attributes of the two, the semantic similarity is directly set to zero. The cost of quantity discrepancy is obtained by calculating the absolute difference between the internal standard quantity and the supplier's actual quantity, and then normalizing the relative deviation by dividing by the larger of the two values ​​and the maximum of the zero constant. The unit price difference sub-cost is obtained by calculating the absolute difference between the internal standard unit price and the supplier's actual unit price, and then normalizing the relative deviation by dividing by the larger of the two values ​​and the maximum of the zero constant.

[0058] S5.4: Construct a multidimensional cost matrix between internal and external BOM verification units; Specifically, the matching cost of all pairs of nodes to be matched is calculated in sequence, and each matching cost is written into a multidimensional cost matrix. The rows of the multidimensional cost matrix correspond to the internal BOM check unit, the columns correspond to the external BOM check unit, and the matrix elements correspond to the matching cost between the corresponding internal and external BOM check units.

[0059] S5.5: Perform business constraint processing on the multidimensional cost matrix; Specifically, a multidimensional cost matrix is ​​received, and business constraints are determined based on the corresponding material category, key attributes, amount fields, and BOM hierarchy relationships of each node pair to be matched. When the material categories of the node pair to be matched are inconsistent, there is a hard attribute conflict in the key attributes, the amount deviation exceeds the preset amount tolerance range, or the two do not belong to the corresponding BOM hierarchy range, the cost value of the node pair to be matched in the multidimensional cost matrix is ​​set to the preset maximum value. When the node pair to be matched satisfies the above business constraints, its original matching cost is retained.

[0060] It should be noted that the preset maximum value is a value greater than the sum of all valid cost values ​​in the multidimensional cost matrix, or a system-preset unmatchable mark value, used to prevent the node pair to be matched from being selected in the subsequent minimum cost solution; the corresponding BOM level range can be determined according to the parent-child or sibling relationship determined in the aforementioned steps.

[0061] S5.6: Solving for the optimal matching relationship based on the constrained multidimensional cost matrix; Specifically, the Hungarian algorithm is used to solve the constrained multidimensional cost matrix. The goal is to minimize the total cost of all matching node pairs with established mapping relationships, thus determining the matching relationships between internal and external BOM verification units. During the solution process, each internal BOM verification unit is only allowed to establish a mapping relationship with one external BOM verification unit, and vice versa. If the number of internal and external BOM verification units is inconsistent, the multidimensional cost matrix is ​​expanded into a square matrix by supplementing with virtual nodes, and the real BOM verification units matched with the virtual nodes are treated as unmatched nodes.

[0062] S5.7: Output the optimal matching relationship and isolated BOM check unit; Specifically, the real node pairs obtained from the solution are taken as the optimal matching relationship between the internal and external BOM verification units; internal BOM verification units and external BOM verification units that do not enter the optimal matching relationship, or real BOM verification units that match virtual nodes, are marked as isolated BOM verification units.

[0063] S6. Based on the optimal matching relationship, perform field-level difference comparison on the matched BOM verification units, and generate an order BOM difference positioning and attribution report in combination with the unmatched BOM verification units.

[0064] S6.1: Read the optimal matching relationship and calculate the field-level difference value; Specifically, the system receives the optimal matching relationship and isolated BOM verification units, and uses each matching node pair in the optimal matching relationship as a difference comparison object. For any matching node pair, it reads the standard quantity, standard unit price, and standard amount of the internal BOM verification unit, and the actual quantity, actual unit price, and actual amount of the external BOM verification unit, and calculates the quantity difference value, unit price difference value, and amount difference value. The quantity difference value is the difference between the actual quantity and the standard quantity; the unit price difference value is the difference between the actual unit price and the standard unit price; and the amount difference value is the difference between the actual amount and the standard amount. This yields the field-level difference results for each matching node pair.

[0065] S6.2: Determine the difference field based on a preset threshold; Specifically, the quantity difference value, unit price difference value, and amount difference value are compared with preset quantity thresholds, preset unit price thresholds, and preset amount thresholds, respectively. When none of the quantity difference value, unit price difference value, and amount difference value exceed the corresponding threshold, the matching node pair is marked as consistent. When any difference value exceeds the corresponding threshold, the matching node pair is marked as a difference node and enters the difference attribution process.

[0066] It should be noted that the preset quantity threshold is set based on the allowable delivery error of the purchase order and the accuracy of the material measurement unit. The value range is from 0 to 5% of the standard quantity. An example value is 1% of the standard quantity or the minimum distinguishable quantity of the corresponding measurement unit. The basis for the value is to ensure that minor quantity differences caused by packaging, measurement or rounding are not misjudged as BOM quantity deviations. The preset unit price threshold is based on the purchase contract base price fluctuation rules, the minimum settlement accuracy of the currency, and the historical price verification tolerance setting. The value range is from 0 to 5% of the standard unit price. An example value is 1% of the standard unit price or 0.01 yuan. The value is determined by distinguishing between normal price rounding errors and changes in the supplier's actual unit price. The preset amount threshold is set based on the accuracy of financial settlement, the rounding error of the product of quantity and unit price, and the order line amount verification rules. The value range is from 0.01 yuan to 1% of the corresponding BOM line amount. The example value is 0.1 yuan. The basis for the value is to ensure that the amount field verification can tolerate the normal decimal rounding error, while identifying the real line amount anomalies.

[0067] S6.3: Generate quantity deviation factor and unit price deviation factor by decomposing the decision tree based on differences; Specifically, for discrepancy nodes, a judgment is made according to a preset discrepancy decomposition decision tree: when the quantity discrepancy value exceeds a preset quantity threshold and the unit price discrepancy value does not exceed a preset unit price threshold, the discrepancy is attributed to the BOM quantity deviation factor; when the unit price discrepancy value exceeds a preset unit price threshold and the quantity discrepancy value does not exceed a preset quantity threshold, the discrepancy is attributed to the BOM unit price deviation factor; when both the quantity discrepancy value and the unit price discrepancy value exceed their respective thresholds, a BOM quantity deviation factor and a BOM unit price deviation factor are generated respectively, and the corresponding quantity discrepancy value, unit price discrepancy value, and amount discrepancy value are recorded under the same discrepancy node, so that the quantity anomaly and price anomaly of the discrepancy node can be located separately.

[0068] It should be noted that the preset difference decomposition decision tree is constructed by pre-selecting quantity difference value, unit price difference value, amount difference value, and remaining difference of the main order amount as judgment nodes, and using preset quantity threshold, preset unit price threshold, preset amount threshold, and preset additional fee threshold as splitting conditions for each judgment node. Among them, the quantity difference value corresponds to the BOM quantity deviation branch, the unit price difference value corresponds to the BOM unit price deviation branch, the remaining difference of the main order amount corresponds to the additional fee deviation branch, and the amount difference value is used to verify whether the attribution results of the quantity deviation branch and the unit price deviation branch cover the difference amount of the current matching node pair.

[0069] S6.4: Identify additional cost deviation factors; Specifically, after completing the attribution of differences for matched nodes, the first difference between the supplier's master order amount and the sum of the amounts of all matched external BOM verification units is calculated. When the first difference exceeds a preset additional cost threshold, non-BOM cost description information is read, and it is determined whether there are cost descriptions corresponding to freight, packaging fees, or taxes in the non-BOM cost description information. If there are corresponding cost descriptions, the first difference is attributed to the additional cost deviation factor, and the corresponding cost category is marked. If there are no corresponding cost descriptions, the first difference is recorded in the report as a difference in the order master amount to be reviewed.

[0070] It should be noted that the preset additional fee threshold is based on the company's review rules for non-BOM expenses such as freight, packaging fees, and taxes, and is set as a percentage of the total order amount. The value ranges from 0.01 yuan to 5% of the main order amount. An example value is 1% of the main order amount or the company's minimum audit amount. The basis for the value is that additional fee deviation identification is only triggered when the difference between the main order amount and the sum of the BOM details reaches the auditable fee scale.

[0071] S6.5: Generate missing or added BOM information based on isolated BOM verification units; Specifically, isolated BOM verification units are read. If the isolated BOM verification unit originates from an internal BOM verification unit set, it is marked as a missing BOM detail on the supplier side. If the isolated BOM verification unit originates from an external BOM verification unit set, it is marked as a newly added BOM detail on the supplier side. This process includes BOM nodes that cannot establish a matching relationship within the scope of difference attribution, avoiding the omission of missing or added line items by only comparing the quantity and unit price of matched nodes.

[0072] S6.6: Generate an order BOM differentiation positioning attribution report; Specifically, the system summarizes BOM quantity deviation factors, BOM unit price deviation factors, additional cost deviation factors, order master amount discrepancies to be reviewed, and missing or added BOM information. It then generates a structured order BOM discrepancy location and attribution report based on order identifier, internal BOM node, external BOM node, discrepancy field, discrepancy value, discrepancy amount, discrepancy type, and attribution results. The report is used to locate inconsistencies in amount or line items in order BOM reconciliation to specific BOM nodes, specific fields, and specific reasons for discrepancies.

[0073] Reference Figure 2 This embodiment also provides an intelligent verification system for order BOM differentiation, including: The parsing and reconstruction module is used to obtain supplier order data, perform layout area recognition, semantic extraction and reverse error correction of the supplier order data, and generate target structured supplier BOM data. The cross-domain semantic alignment module is used to extract material description features, key attribute features, and hierarchical path features from internal BOM verification units and external BOM verification units, and then fuse them to generate comprehensive features. The nested optimization matching module is used to calculate the matching cost between internal and external BOM verification units based on comprehensive features, and determine the optimal matching relationship in combination with business constraints; The difference attribution module is used to perform field-level difference comparisons based on the optimal matching relationship, and generate an order BOM difference positioning attribution report in conjunction with the isolated BOM verification unit.

[0074] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent verification of order BOM differences, characterized in that, include: Obtain supplier order data and corresponding internal standard order BOM data, perform layout area recognition and semantic extraction on supplier order data, and obtain preliminary structured supplier BOM data containing order master information and BOM detail lines; The preliminary structured supplier BOM data is validated based on the BOM detail amount relationship and the order master amount relationship, and the abnormal fields are located and corrected to obtain the target structured supplier BOM data. The target structured supplier BOM data is used as the external actual BOM data, and the verification standards are unified with the internal standard order BOM data to obtain the internal BOM verification unit and the external BOM verification unit. Extract material description features, key attribute features, and hierarchical path features from each BOM verification unit, and fuse them to generate corresponding comprehensive features. This includes: generating dense semantic vectors based on the material descriptions of each BOM verification unit, and extracting at least one key attribute from specifications, dimensions, materials, and drawing numbers to generate sparse attribute vectors; generating hierarchical path features based on the BOM hierarchy of each BOM verification unit, and associating and fusing the dense semantic vectors, sparse attribute vectors, and hierarchical path features to obtain the comprehensive features of the corresponding BOM verification unit. The matching cost between internal and external BOM verification units is calculated based on comprehensive characteristics. The optimal matching relationship between internal and external BOM verification units is determined by combining material category, key attributes, amount tolerance and hierarchical path constraints. Based on the optimal matching relationship, the matched BOM verification units are compared at the field level, and an order BOM difference positioning and attribution report is generated by combining the unmatched BOM verification units.

2. The intelligent verification method for order BOM differentiation as described in claim 1, characterized in that: The supplier order data includes the supplier email body and / or supplier email attachments, as well as order identification information, order master information and BOM details identified from the supplier email body and / or email attachments; The internal standard order BOM data includes internal purchase order information corresponding to the order identification information, standard BOM node information, BOM hierarchy information, and order amount information.

3. The intelligent verification method for order BOM differentiation as described in claim 2, characterized in that: The process of obtaining preliminary structured supplier BOM data containing order master information and BOM detail lines includes: preprocessing the body and / or attachments of supplier emails under the same verification task to obtain preprocessed document data containing text content and table structure; identifying table areas and text areas in the preprocessed document data through a layout analysis model, and determining the order master information area and BOM detail table area; inputting the order master information area and BOM detail table area into a large language model for semantic extraction to obtain order master information, BOM detail line information and BOM hierarchy relationship, and generating preliminary structured supplier BOM data.

4. The intelligent verification method for order BOM differentiation as described in claim 3, characterized in that: The process of obtaining the target structured supplier BOM data includes: performing row-level amount verification based on the quantity, unit price, and amount fields in each BOM detail row, and performing order master-level amount verification based on the sum of amounts in each BOM detail row and the master order amount; when the row-level amount verification or the order master-level amount verification is abnormal, the abnormal field is located according to the abnormal amount relationship and the corresponding original context, and the abnormal field is extracted and corrected a second time; when the verification still fails after the second extraction and correction, the abnormal field is partially completed or marked as pending review, and the target structured supplier BOM data is generated.

5. The intelligent verification method for order BOM differentiation as described in claim 4, characterized in that: The process of obtaining the internal BOM verification unit and the external BOM verification unit includes: determining the target structured supplier BOM data as the external actual BOM data, and unifying the field definitions and BOM hierarchy definitions with the internal standard order BOM data under the same verification task; based on the unified field definitions and BOM hierarchy definitions, converting the internal standard BOM nodes and the external actual BOM nodes into internal BOM verification units and external BOM verification units, respectively.

6. The intelligent verification method for order BOM differentiation as described in claim 5, characterized in that: The method of calculating the matching cost between internal and external BOM verification units based on comprehensive features includes: combining each internal BOM verification unit with each external BOM verification unit to form a pair of nodes to be matched based on the comprehensive features of each BOM verification unit; calculating semantic similarity based on the cosine similarity of dense semantic vectors and the Jaccard similarity of sparse attribute vectors; calculating the corresponding matching cost according to the semantic similarity, relative difference in quantity, and relative difference in unit price of the node pairs to be matched, and rejecting or downgrading node pairs to be matched with hard attribute conflicts; and writing the matching cost of all node pairs to be matched into a multidimensional cost matrix. The term "hard attribute conflict" refers to the following: when a standardized comparison is made between the material, drawing number, core specifications, and dimensions of any internal BOM verification unit X and external BOM verification unit Y, if the same hard attribute field has a valid value in both internal BOM verification unit X and external BOM verification unit Y, and the two values ​​are different or the difference exceeds the preset tolerance range, then it is determined to be a hard attribute conflict.

7. The intelligent verification method for order BOM differentiation as described in claim 6, characterized in that: Determining the optimal matching relationship between internal and external BOM verification units includes: judging the material category, key attributes, amount tolerance, and BOM level path constraints of the node pairs to be matched in the multidimensional cost matrix, and setting the cost value of the node pairs to be matched that do not meet the constraints to a preset maximum value; solving the matching relationship of internal and external BOM verification units with the minimum total cost value based on the constrained multidimensional cost matrix, and marking the BOM verification units that have not entered the matching relationship as isolated BOM verification units.

8. The intelligent verification method for order BOM differentiation as described in claim 7, characterized in that: The step of performing field-level difference comparison on the matched BOM verification units based on the optimal matching relationship, and generating an order BOM difference positioning and attribution report in combination with the unmatched BOM verification units, includes: The quantity difference, unit price difference, and amount difference of each matching node pair are calculated based on the optimal matching relationship, and the difference nodes are determined based on the preset quantity threshold, preset unit price threshold, and preset amount threshold. By using a pre-set difference decomposition decision tree, the difference nodes are attributed to the BOM quantity deviation factor and / or BOM unit price deviation factor, and the additional cost deviation factor is identified based on the difference between the amount of the main order on the supplier side and the sum of the amounts of the matched external BOM verification units. Based on the BOM quantity deviation factor, BOM unit price deviation factor, additional cost deviation factor, and the missing or added BOM information corresponding to the isolated BOM verification unit, generate an order BOM differentiation positioning attribution report.

9. An intelligent verification system for order BOM differentiation, based on the intelligent verification method for order BOM differentiation according to any one of claims 1 to 8, characterized in that, include: The parsing and reconstruction module is used to obtain supplier order data, perform layout area recognition, semantic extraction and reverse error correction of the supplier order data, and generate target structured supplier BOM data. The cross-domain semantic alignment module is used to extract material description features, key attribute features, and hierarchical path features from internal BOM verification units and external BOM verification units, and then fuse them to generate comprehensive features. The nested optimization matching module is used to calculate the matching cost between internal and external BOM verification units based on comprehensive features, and determine the optimal matching relationship in combination with business constraints; The difference attribution module is used to perform field-level difference comparisons based on the optimal matching relationship, and generate an order BOM difference positioning attribution report in conjunction with the isolated BOM verification unit.

Images