Multilingual trade document information extraction method and system based on artificial intelligence
By using an AI-based approach, the table structure of multilingual trade documents is identified and reconstructed. By combining text matching and semantic analysis, the problems of poor adaptability of table structure and insufficient semantic understanding in traditional methods are solved, and efficient information extraction is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUCHAN ZHONGDA INTERNATIONAL TRADING GROUP CO LTD
- Filing Date
- 2026-03-27
- Publication Date
- 2026-06-26
Smart Images
Figure CN122290155A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of document extraction technology, specifically to a method and system for extracting multilingual trade document information based on artificial intelligence. Background Technology
[0002] In international trade, a large number of documents, contracts, customs declarations, invoices, and other documents exist in tabular form and often involve multiple languages. Traditional information extraction methods mainly rely on two approaches: manual reading and data entry, which is inefficient and costly; and rule-based automated extraction, such as writing scripts for fixed-template tables to extract data from specific rows and columns. However, these traditional methods have the following technical shortcomings: 1. Poor adaptability to table structures: They struggle to handle borderless, incomplete, or non-standardized tables, easily leading to incorrect cell division and extraction failures. 2. Insufficient semantic understanding: Simple keyword matching cannot handle situations where the user's extracted items do not completely match the table labels, such as incomplete or scattered table label descriptions for extracted items. Therefore, there is a need to provide an AI-based multilingual trade document information extraction method and system to address the above problems. Summary of the Invention
[0003] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a method and system for extracting multilingual trade document information based on artificial intelligence, so as to solve the problems existing in the above-mentioned background technology.
[0004] This invention is implemented as follows: a method for extracting multilingual trade document information based on artificial intelligence, the method comprising the following steps: Translate the table titles to obtain the title content; Identify displayed and undisplayed border lines in the table, determine each cell, translate the information in the cells, and obtain the content of each cell; The extracted items are matched with the text content of each cell to determine the target cell with the highest matching degree. Determine whether the content of the target cell completely matches the extracted item. If it completely matches, determine the content of the right and bottom adjacent cells of the target cell as the content to be extracted. When there is no complete match, the extracted items are reduced according to the title content. It is then determined whether the extracted items completely match the content of the target cell. When there is a complete match, the contents of the right and bottom adjacent cells of the target cell are determined as the content to be extracted. If the match is still not perfect, the extracted items are broken down into words, the cells corresponding to each word are determined, the intersection area is determined, and the content to be extracted is obtained. Semantic analysis is performed on the content to be extracted to determine the final extracted information.
[0005] As a further aspect of the present invention, the step of identifying whether a table has displayed border lines or not specifically includes: Perform image parsing on the table document to obtain the text blocks and their coordinate information within the table area; Detect the line segments in the image and use the detected line segments as the displayed border lines; For text blocks without displayed borders, the text blocks are grouped by rows and columns according to their coordinate distribution, and the positions of the row separators and column separators are determined to obtain the text blocks without displayed borders.
[0006] As a further aspect of the present invention, the step of matching the extracted items with the text content of each cell specifically includes: The extracted items and the content of each cell are preprocessed, including word segmentation and stop word removal; The extracted items are expanded with synonyms to obtain an item lexicon, with the original extracted items located in the first position of the item lexicon; This process sequentially matches the extracted items from the project thesaurus with the text content of each cell.
[0007] As a further aspect of the present invention, the step of reducing the extracted items based on the title content specifically includes: The title content is compared with the extracted items to identify the words contained in the title content; Remove the aforementioned words from the extracted items to obtain the reduced extracted items.
[0008] As a further aspect of the present invention, the steps of determining the cell corresponding to each decomposed word, determining the intersection area, and obtaining the content to be extracted specifically include: For each word to be broken down, iterate through all cells, determine the candidate cells corresponding to the word to be broken down, and record the row index and column index of the candidate cells; The intersection area is determined based on the row and column indices, and the content of the cells containing the intersection area is determined as the content to be extracted.
[0009] As a further aspect of the present invention, the step of performing semantic analysis on the content to be extracted to determine the final extracted information specifically includes: Semantic parsing is performed on the extracted items to determine the type and specific category of the target information to be extracted; Semantic annotation is performed on each piece of content to be extracted to identify its actual information type; The content to be extracted that matches the actual information type with the target information type is taken as candidate content. If there is only one candidate content, that candidate content will be directly selected as the final extracted information; if there are multiple candidate contents, the candidate content that matches the specific category will be selected as the final extracted information.
[0010] Another objective of this invention is to provide a multilingual trade document information extraction system based on artificial intelligence, the system comprising: The title content determination module is used to translate table titles to obtain title content; The border line recognition module is used to identify displayed and undisplayed border lines in a table, determine each cell, translate the information in the cell, and obtain the content of each cell. The project content matching module is used to perform text matching between the extracted project and the content of each cell to determine the target cell with the highest matching degree. The first pending content module is used to determine whether the content of the target cell completely matches the extracted items. When they completely match, the content of the right and bottom adjacent cells of the target cell is determined as pending extraction content. The second pending content module is used to reduce the extracted items based on the title content when there is no complete match, and to determine whether the extracted items completely match the content of the target cell. When there is a complete match, the content of the right adjacent cell and the bottom adjacent cell of the target cell are determined as pending extraction content. The third pending content module is used to break down the extracted items into words when there is still no complete match, determine the cell corresponding to each word, determine the intersection area, and obtain the pending extraction content. The extraction information determination module is used to perform semantic analysis on the content to be extracted and determine the final extraction information.
[0011] As a further embodiment of the present invention, the border line recognition module includes: The text block recognition unit is used to perform image parsing on the table document and obtain the text blocks and their coordinate information in the table area. The displayed border unit is used to detect straight line segments in the image and use the detected straight line segments as the displayed border lines; The "No Border" unit is used to group text blocks by rows and columns based on their coordinate distribution using a clustering algorithm, thereby determining the positions of row and column separators and obtaining the "No Border" unit.
[0012] As a further aspect of the present invention, the project content matching module includes: The text preprocessing unit is used to preprocess the extracted items and the content of each cell. The preprocessing includes word segmentation and stop word removal. The project terminology determination unit is used to expand the extracted items with synonyms to obtain the project terminology. The original extracted items are located in the first position of the project terminology. The text matching unit is used to sequentially match the extracted items from the item thesaurus with the content of each cell.
[0013] As a further aspect of the present invention, the second undetermined content module includes: The content comparison unit is used to compare the title content with the extracted items and identify the words contained in the title content; The word removal unit is used to remove the words from the extraction items to obtain the reduced extraction items.
[0014] Compared with the prior art, the beneficial effects of the present invention are: By combining the detection of displayed border lines and the inference of undisplayed border lines, the cell structure of any formatted table can be accurately reconstructed, even when table borders are missing or incomplete, ensuring precise location of each cell. Employing a layered and progressive matching strategy, from "exact match" to "title-assisted deletion matching," and then to "word decomposition and cross-location," it adaptively handles various matching scenarios from simple to complex. This fault-tolerant mechanism significantly improves the extraction success rate when table labels are incomplete or information is scattered. Attached Figure Description
[0015] Figure 1 This is a flowchart of a method for extracting information from multilingual trade documents based on artificial intelligence.
[0016] Figure 2 This is a flowchart illustrating the process of identifying table border lines in an AI-based multilingual trade document information extraction method.
[0017] Figure 3 This is a flowchart illustrating text matching in an AI-based multilingual trade document information extraction method.
[0018] Figure 4 This is a flowchart illustrating the process of removing items from a multilingual trade document information extraction method based on artificial intelligence.
[0019] Figure 5 This is a flowchart for determining the intersection area in an AI-based multilingual trade document information extraction method.
[0020] Figure 6 This is a flowchart illustrating semantic analysis in a multilingual trade document information extraction method based on artificial intelligence.
[0021] Figure 7 This is a schematic diagram of the structure of a multilingual trade document information extraction system based on artificial intelligence. Detailed Implementation
[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0023] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.
[0024] like Figure 1 As shown in the figure, this invention provides a method for extracting multilingual trade document information based on artificial intelligence. The method includes the following steps: S100, Translate the table title to obtain the title content; S200 identifies displayed and undisplayed border lines in the table, determines each cell, translates the information in the cell, and obtains the content of each cell; S300 performs text matching between the extracted items and the content of each cell to determine the target cell with the highest matching degree; S400: Determine whether the content of the target cell completely matches the extracted item. If it completely matches, determine the content of the right and bottom adjacent cells of the target cell as the content to be extracted. S500: When there is no complete match, the extracted items are reduced according to the title content. It is then determined whether the extracted items completely match the content of the target cell. When there is a complete match, the contents of the right and bottom adjacent cells of the target cell are determined as the content to be extracted. S600, when there is still no complete match, the extracted items are broken down into words, the cells corresponding to each word are determined, the intersection area is determined, and the content to be extracted is obtained; S700, perform semantic analysis on the content to be extracted to determine the final extraction information.
[0025] In this embodiment of the invention, since the document is multilingual, the table titles first need to be translated into a unified processing language, such as Chinese. Tables in multilingual documents may have non-standard formats and missing borders. This step uses intelligent recognition (including explicit borders and implicit layouts) to accurately locate the physical boundaries of each cell. After determining the independent area of each cell, the text information within it is translated, laying the foundation for subsequent matching operations. In this embodiment of the invention, many extraction items are pre-set. An extraction item (such as "July strawberry purchase volume") is a query condition. A text matching algorithm is used to find the cell in the entire table content that is semantically closest to this item. This found cell is called the target cell, which is usually a label in the table describing the item name (for example, a cell that says "strawberry purchase volume"). Next, it is determined whether the content of the target cell completely matches the extraction item. For example, the extraction item "July strawberry purchase volume" does not completely match the target cell "strawberry purchase volume". If a perfect match is found, the cells to the right and below the target cell are highly likely to contain the data to be extracted. The contents of these adjacent cells are treated as "pending extraction content" and proceed directly to the final confirmation stage. If the match is incomplete, the extraction items are first reduced based on the title content. Using the title content "July Trade Information Table" obtained in step S100, redundant information (such as July) is removed from the extraction items, resulting in "Strawberry Purchase Quantity". At this point, the reduced extraction item perfectly matches the target cell content, and adjacent cells are treated as pending extraction content. This solves the matching problem caused by uneven information distribution (time information in the title, item information in the header). If the extraction items still cannot perfectly match after reduction (for example, "Strawberry" and "Purchase Quantity" are in different cells in the table), the extraction items are broken down into smaller word units ("Strawberry", "Purchase Quantity"). Then, the cells containing each word are located, and the "intersection area" they point to is locked based on the row and column coordinates of these cells. For example, if "Strawberry" is in the second row and "Purchase Quantity" is in the third column, then the cell at their intersection (second row, third column) contains the target data. The content of this intersection area is considered as the content to be extracted. Through the preceding steps, one or more pieces of content to be extracted may be obtained. These candidate contents are then subjected to deep semantic understanding, and finally, the one that best matches the original semantics of the "extraction item" is selected as the final output.
[0026] like Figure 2 As shown, in a preferred embodiment of the present invention, the step of identifying whether a table has displayed border lines or not specifically includes: S201, Perform image parsing on the table document to obtain the text blocks and their coordinate information of the table area; S202, Detect the straight line segments existing in the image, and regard the detected straight line segments as the displayed border lines; S203, For the text blocks without displayed border lines, according to the coordinate distribution of the text blocks, group the text blocks by rows and columns through a clustering algorithm, determine the positions of the row separation lines and column separation lines, and obtain the non-displayed border lines.
[0027] In the embodiments of the present invention, for a document in the form of a scanned copy or a picture, all text fragments belonging to the table area in the document are located through OCR (Optical Character Recognition), and the exact positions of each fragment on the page are recorded, for example, represented by the coordinates of the upper left corner and the lower right corner. Then, image processing techniques (such as the Hough transform, etc.) are used to detect straight lines in the table image. The successfully detected straight lines are considered as the existing and clear border lines of the table, and these lines directly divide the cells. For areas without clear border lines, only rely on the text content and its coordinates to infer the structure. Using a clustering algorithm (such as K-Means), the text blocks are aggregated into "rows" according to the similarity of the vertical coordinates, and aggregated into "columns" according to the similarity of the horizontal coordinates. Then, virtual separation lines are drawn in the blank areas between rows and between columns, thereby defining the cell boundaries of the borderless table.
[0028] As Figure 3 shown, as a preferred embodiment of the present invention, the step of performing text matching between the extracted items and the content of each cell specifically includes: S301, Perform preprocessing on the extracted items and the content of each cell. The preprocessing includes word segmentation and removal of stop words; S302, Expand the extracted items with near synonyms to obtain an item thesaurus, and the original extracted item is in the first position of the item thesaurus; S303, Sequentially perform text matching between the extracted items in the item thesaurus and the content of each cell.
[0029] In the embodiments of the present invention, preprocessing will be performed to cut continuous text strings (such as "July strawberry purchase quantity") into meaningful word units ("July", "strawberry", "purchase quantity"). At the same time, filter out stop words without practical meaning such as "of" and "is" to perform more accurate matching. To improve the robustness of the matching, handle synonyms or idiomatic differences. For example, expand the near synonyms of "purchase quantity" to generate words such as "procurement quantity" and "warehousing quantity", forming an "item thesaurus" containing multiple candidate words, and the original word "purchase quantity" has the highest priority. In the order of priority, first use the original word (the first position) to match all cells. If an exact match can be found, it is directly successful. If not, then use the second near synonym to match, and so on. This ensures both accurate matching and fault tolerance.
[0030] like Figure 4 As shown in the preferred embodiment of the present invention, the step of reducing the extracted items according to the title content specifically includes: S501, compare the title content with the extracted items to identify the words contained in the title content; S502, Remove the words from the extraction items to obtain the reduced extraction items.
[0031] In this embodiment of the invention, the title content obtained in step S100 (such as "July Trade Information Table") is compared with the extracted item ("July Strawberry Purchase Volume") to find the common word, namely "July". Then, a deletion operation is performed to remove the words that already appear in the title from the extracted item, resulting in a more refined deleted item ("Strawberry Purchase Volume").
[0032] like Figure 5 As shown, in a preferred embodiment of the present invention, the steps of determining the cell corresponding to each decomposed word, determining the intersection area, and obtaining the content to be extracted specifically include: S601, for each word to be decomposed, traverse all cells, determine the candidate cell corresponding to the word to be decomposed, and record the row index and column index of the candidate cell; S602 determines the intersection area based on the row index and column index, and identifies the content of the cell containing the intersection area as the content to be extracted.
[0033] In this embodiment of the invention, words need to be mapped to coordinates. For example, first, the word "strawberry" is traversed, and the cell containing "strawberry" is found in the table. Its row number (e.g., row 2) and column number (e.g., column B) are recorded. Then, the word "purchase volume" is traversed, and the cell containing it is found, and its row number (e.g., row 1) and column number (e.g., column C) are recorded. Then, the intersection area is determined based on the row index and column index. The content of the cell containing the intersection area is determined as the content to be extracted. It is easy to understand that the intersection area may contain multiple cells, which will generate multiple contents to be extracted.
[0034] like Figure 6 As shown, in a preferred embodiment of the present invention, the step of performing semantic analysis on the content to be extracted to determine the final extracted information specifically includes: S701, perform semantic parsing on the extracted items to determine the type and specific category of the target information to be extracted; S702, semantically label each piece of content to be extracted to identify its actual information type; S703, take the undetermined content that matches the actual information type with the target information type as candidate content; S704 If there is only one candidate content, directly determine that candidate content as the final extracted information; if there are multiple candidate contents, determine the candidate content that matches the specific category as the final extracted information.
[0035] In this embodiment of the invention, the user's needs are clearly defined, the extraction item ("July strawberry purchase volume") is analyzed, and it is determined whether the user wants the target information type (numerical) and the specific category (purchase data). Then, each pending extraction content is semantically labeled to identify the attributes of the candidate content. Each pending content (e.g., "500kg" or "2023-07-15") is analyzed to determine whether it is a number, date, currency, or other type, and is labeled accordingly. If the target information type is "quantity", then only those pending content labeled as "number" or "quantity" are retained, and those dates or text are filtered out. If there is only one candidate content, it is directly determined as the final extraction information; if there are multiple candidate contents, such as two quantities, "500kg" and "100 boxes", combined with the previously parsed "specific category" (which is "strawberry purchase volume" rather than "strawberry shipment volume"), 500kg is determined as the final extraction information.
[0036] like Figure 7 As shown, this embodiment of the invention also provides a multilingual trade document information extraction system based on artificial intelligence, the system comprising: The title content determination module 100 is used to translate the table title to obtain the title content; The border line recognition module 200 is used to identify displayed and undisplayed border lines in a table, determine each cell, translate the information in the cell, and obtain the content of each cell. The project content matching module 300 is used to match the extracted project with the content of each cell to determine the target cell with the highest matching degree; The first pending content module 400 is used to determine whether the content of the target cell completely matches the extracted items. When they completely match, the content of the right adjacent cell and the bottom adjacent cell of the target cell are determined as pending extraction content. The second pending content module 500 is used to reduce the extracted items according to the title content when there is no complete match, and to determine whether the extracted items after reduction match the content of the target cell. When there is a complete match, the content of the right adjacent cell and the bottom adjacent cell of the target cell are determined as pending extraction content. The third pending content module 600 is used to break down the extracted items into words when there is still no complete match, determine the cell corresponding to each word, determine the intersection area, and obtain the pending extracted content. The extraction information determination module 700 is used to perform semantic analysis on the content to be extracted and determine the final extraction information.
[0037] In a preferred embodiment of the present invention, the border line recognition module 200 includes: The text block recognition unit is used to perform image parsing on the table document and obtain the text blocks and their coordinate information in the table area. The displayed border unit is used to detect straight line segments in the image and use the detected straight line segments as the displayed border lines; The "No Border" unit is used to group text blocks by rows and columns based on their coordinate distribution using a clustering algorithm, thereby determining the positions of row and column separators and obtaining the "No Border" unit.
[0038] In a preferred embodiment of the present invention, the project content matching module 300 includes: The text preprocessing unit is used to preprocess the extracted items and the content of each cell. The preprocessing includes word segmentation and stop word removal. The project terminology determination unit is used to expand the extracted items with synonyms to obtain the project terminology. The original extracted items are located in the first position of the project terminology. The text matching unit is used to sequentially match the extracted items from the item thesaurus with the content of each cell.
[0039] In a preferred embodiment of the present invention, the second undetermined content module 500 includes: The content comparison unit is used to compare the title content with the extracted items and identify the words contained in the title content; The word removal unit is used to remove the words from the extraction items to obtain the reduced extraction items.
[0040] The above description only details the preferred embodiments of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
[0041] It should be understood that although the steps in the flowcharts of the various embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the various embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.
[0042] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the disclosure in the specification and embodiments. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.
Claims
1. A method for multilingual trade document information extraction based on artificial intelligence, characterized in that, The method includes the following steps: Translate the table titles to obtain the title content; Identify whether border lines are displayed or not in the table, determine each cell, translate the information in the cell, and obtain the content of each cell; The extracted items are matched with the text content of each cell to determine the target cell with the highest matching degree. Determine whether the content of the target cell completely matches the extracted item. If it completely matches, determine the content of the right and bottom adjacent cells of the target cell as the content to be extracted. When there is no complete match, the extracted items are reduced according to the title content. It is then determined whether the extracted items completely match the content of the target cell. When there is a complete match, the contents of the right and bottom adjacent cells of the target cell are determined as the content to be extracted. If the match is still not perfect, the extracted items are broken down into words, the cells corresponding to each word are determined, the intersection area is determined, and the content to be extracted is obtained. Semantic analysis is performed on the content to be extracted to determine the final extracted information.
2. The method for extracting multilingual trade document information based on artificial intelligence according to claim 1, characterized in that, The steps for identifying whether a table has displayed border lines or not specifically include: Perform image parsing on the table document to obtain the text blocks and their coordinate information within the table area; Detect the line segments in the image and use the detected line segments as the displayed border lines; For text blocks without displayed borders, the text blocks are grouped by rows and columns according to their coordinate distribution, and the positions of the row separators and column separators are determined to obtain the text blocks without displayed borders.
3. The method for extracting multilingual trade document information based on artificial intelligence according to claim 1, characterized in that, The step of matching the extracted items with the text content of each cell specifically includes: The extracted items and the content of each cell are preprocessed, including word segmentation and stop word removal; The extracted items are expanded with synonyms to obtain an item lexicon, with the original extracted items located in the first position of the item lexicon; This process sequentially matches the extracted items from the project thesaurus with the text content of each cell.
4. The method for extracting multilingual trade document information based on artificial intelligence according to claim 1, characterized in that, The step of reducing the extracted items based on the title content specifically includes: The title content is compared with the extracted items to identify the words contained in the title content; Remove the aforementioned words from the extracted items to obtain the reduced extracted items.
5. The method for extracting multilingual trade document information based on artificial intelligence according to claim 1, characterized in that, The steps to determine the cell corresponding to each decomposed word, identify the intersection area, and obtain the content to be extracted include: For each word to be broken down, iterate through all cells, determine the candidate cells corresponding to the word to be broken down, and record the row index and column index of the candidate cells; The intersection area is determined based on the row and column indices, and the content of the cells containing the intersection area is determined as the content to be extracted. 6.The method of claim 1, wherein, The steps of performing semantic analysis on the content to be extracted to determine the final extracted information specifically include: Semantic parsing is performed on the extracted items to determine the type and specific category of the target information to be extracted; Semantic annotation is performed on each piece of content to be extracted to identify its actual information type; The content to be extracted that matches the actual information type with the target information type is taken as candidate content. If there is only one candidate content, that candidate content will be directly selected as the final extracted information; if there are multiple candidate contents, the candidate content that matches the specific category will be selected as the final extracted information.
7. A multilingual trade document information extraction system based on artificial intelligence, characterized in that: The system includes: The title content determination module is used to translate table titles to obtain title content; The border line recognition module is used to identify displayed and undisplayed border lines in a table, determine each cell, translate the information in the cell, and obtain the content of each cell. The project content matching module is used to perform text matching between the extracted project and the content of each cell to determine the target cell with the highest matching degree. The first pending content module is used to determine whether the content of the target cell completely matches the extracted items. When they completely match, the content of the right and bottom adjacent cells of the target cell is determined as pending extraction content. The second pending content module is used to reduce the extracted items based on the title content when there is no complete match, and to determine whether the extracted items completely match the content of the target cell. When there is a complete match, the content of the right adjacent cell and the bottom adjacent cell of the target cell are determined as pending extraction content. The third pending content module is used to break down the extracted items into words when there is still no complete match, determine the cell corresponding to each word, determine the intersection area, and obtain the pending extraction content. The extraction information determination module is used to perform semantic analysis on the content to be extracted and determine the final extraction information.
8. The multilingual trade document information extraction system based on artificial intelligence according to claim 7, characterized in that, The border line recognition module includes: The text block recognition unit is used to perform image parsing on the table document and obtain the text blocks and their coordinate information in the table area. The displayed border unit is used to detect straight line segments in the image and use the detected straight line segments as the displayed border lines; The "No Border" unit is used to group text blocks by rows and columns based on their coordinate distribution using a clustering algorithm, thereby determining the positions of row and column separators and obtaining the "No Border" unit.
9. The multilingual trade document information extraction system based on artificial intelligence according to claim 7, characterized in that, The project content matching module includes: The text preprocessing unit is used to preprocess the extracted items and the content of each cell. The preprocessing includes word segmentation and stop word removal. The project terminology determination unit is used to expand the extracted items with synonyms to obtain the project terminology. The original extracted items are located in the first position of the project terminology. The text matching unit is used to sequentially match the extracted items from the item thesaurus with the content of each cell.
10. The multilingual trade document information extraction system based on artificial intelligence according to claim 7, characterized in that, The second pending content module includes: The content comparison unit is used to compare the title content with the extracted items and identify the words contained in the title content; The word removal unit is used to remove the words from the extraction items to obtain the reduced extraction items.