Intelligent identification and import method for building industry communication project contract scanning copy

Abstract

The invention relates to an intelligent identification and import method for a building industry communication project contract scanning copy. The method comprises the steps of obtaining a target template and a to-be-processed contract scanning copy, preprocessing the contract scanning copy, performing structured analysis on the target template, extracting a field standard value, marking a field type, and generating a field type mapping table; inputting the preprocessed contract scanning copy into a multi-modal feature analysis model to obtain a contract information extraction set; performing semantic matching on the contract information extraction set and a field type mapping table to generate a field matching result, and inputting successfully matched contract information into a target template; and if the necessary fields are missing, retrieving associated historical information from the dynamic knowledge base through an entity identification and link technology, obtaining contract information of missing field mapping, and inputting the contract information of missing field mapping into the target template. According to the invention, the recognition accuracy is improved, and the labor cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of construction technology, and in particular to a method for intelligently identifying and importing scanned copies of communication project contracts in the construction industry. Background Art

[0002] Currently, existing systems for project contract management typically use optical character recognition (OCR) technology to extract contract text and match key clauses using predefined rule templates. However, due to the presence of unstructured data such as handwritten annotations and tables spanning multiple pages, OCR engines often encounter problems such as misaligned text and table nesting failures when processing these complex documents, resulting in a key element extraction error rate generally exceeding 28%.

[0003] Furthermore, existing industry systems generally rely on manually pre-set validation rule bases (e.g., checking the range of monetary values ​​and the consistency of contracting party names). Because project contracts involve dynamically changing enterprise qualifications (e.g., the validity period of construction permits) and industry standards, static rule bases require frequent manual maintenance and updates, increasing system maintenance costs by over 40% and creating the risk of rule failure during the transition period (typically 3-6 months).

[0004] Therefore, there is an urgent need for an intelligent recognition and import method for scanned copies of communication project contracts in the construction industry with high accuracy and low labor costs. Summary of the Invention

[0005] (1) Technical issues to be solved

[0006] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a method for intelligent recognition and importing scanned copies of communication project contracts in the construction industry, which solves the technical problems of high recognition error rate and high manual review cost in the prior art.

[0007] (2) Technical solution

[0008] In order to achieve the above objectives, the main technical solutions adopted by the present invention include:

[0009] An embodiment of the present invention provides a method for intelligently identifying and importing scanned copies of construction industry communication project contracts, which is applied to construction industry communication projects. The method includes:

[0010] S100: Obtain a target template and a scanned copy of a contract to be processed, perform image preprocessing on the scanned copy of the contract, and obtain a preprocessed scanned copy of the contract; and perform structured parsing on the target template, extracting standard field values ​​and marking field types, and generating a field type mapping table;

[0011] The field type mapping table includes a required field mapping table and an optional field mapping table;

[0012] S200: Inputting the pre-processed contract scan into a pre-built multimodal feature parsing model to extract visual structure features and text language features, and obtaining a contract information extraction set based on the visual structure features and text language features;

[0013] S300: semantically match the contract information extraction set with the field type mapping table to generate a field matching result, and input the successfully matched contract information into the target template;

[0014] S400. If a required field is missing in the field matching result, related historical information is retrieved from the dynamic knowledge base through entity recognition and linking technology to obtain the contract information mapped by the missing field, and the contract information mapped by the missing field is input into the target template.

[0015] Optionally, in S200, the multimodal feature parsing model includes:

[0016] The visual branch layer, based on the Faster R-CNN architecture, extracts visual structural features of tables, titles, and signatures;

[0017] The text language branch layer uses the BERT model, combined with a contract-specific corpus, to extract clause semantics and entity relationship features to obtain text language features;

[0018] The feature fusion layer combines visual structure features and text language features through the Transformer decoder to obtain the contract information extraction set.

[0019] Optionally, the contract information extraction set includes contract information fields and corresponding field contents;

[0020] The contract information fields include: project information, Party A information, construction unit information and time information; the project information includes project name, project category, professional category, project year, project location information, contract number and contract amount; the Party A information includes Party A's witness and Party A's contact number; the construction unit information includes construction unit category, construction unit address, construction unit name and project manager; the time information includes signing date, start date and completion date.

[0021] Optionally, in S100, the target template is subjected to structured parsing, field standard values ​​are extracted, field types are marked, and a field type mapping table is generated;

[0022] S110, building a regular expression rule library based on the construction industry communication project contract specification, performing a full text scan on the target template, and identifying a set containing field identifiers as an initial field candidate set;

[0023] S120. Based on the national construction project demonstration text classification standard, the initial field candidate set is type-labeled and hierarchically divided to obtain a required field mapping table and an optional field mapping table.

[0024] Optionally, the required field mapping table includes project information, Party A information, construction unit information and time information, and the optional field mapping table includes supplementary agreement terms and customized additional instructions.

[0025] Optionally, the S300 includes:

[0026] S310: Using an edit distance algorithm to obtain a similarity between each contract information field and a corresponding target template field type as a first similarity;

[0027] S320, extracting the semantic similarity between the content of each contract information field and the standard value of the target template field type by cosine similarity calculation as a second similarity;

[0028] S330: Generate a field matching result based on the first similarity and the second similarity, and input the successfully matched contract information into the target template.

[0029] Optionally, the S330 includes:

[0030] When the first similarity of the required field is ≥0.7 and the second similarity is ≥0.75, the field matching result is successful and the corresponding contract information is entered into the target template;

[0031] When the first similarity of the optional field is ≥0.6 or the second similarity is ≥0.6, the field matching result is a successful match, and the corresponding contract information is input into the target template.

[0032] Optionally, the S400 includes:

[0033] S410. Based on the field matching results, identify missing segments in the required fields that failed to match;

[0034] S420: Retrieve historical information associated with the missing fields from the dynamic knowledge base, perform multi-dimensional scoring on the historical information, and obtain the historical information with the highest score as the contract information for each missing field;

[0035] S430: Input the contract information of each missing segment into the target template.

[0036] Optionally, in S420, performing multi-dimensional scoring on the historical information includes:

[0037] Obtain the data source credibility, time coefficient, and semantic similarity of each piece of historical information, and input the data source credibility, time coefficient, and semantic similarity of each piece of historical information into the following formula to obtain the score of each piece of historical information:

[0038] Score = Data source credibility weight × time coefficient × semantic similarity 2 ;

[0039] Among them, the credibility weight of government-registered data is 0.9, and the credibility weight of enterprise-filled data is 0.6; the time coefficient of data in the past two years is 1.2, the time coefficient of data between 2 and 3 years is 1.0, and the time coefficient of data over 3 years is 0.5; the semantic similarity is the similarity between the context of missing fields and historical information, and the semantic similarity is obtained using the BERT semantic matching model.

[0040] Optionally, the dynamic knowledge base includes but is not limited to: a filing database of the housing and construction department, an enterprise registration information database of the market supervision department, an enterprise historical contract database, a project cost information database, and a legal and regulatory knowledge base.

[0041] (3) Beneficial effects

[0042] The beneficial effects of the present invention are: a method for intelligently identifying and importing scanned copies of construction industry communication project contracts of the present invention adopts a combination of multimodal feature analysis models, combines visual structure features and text language features to extract contract information, and performs semantic matching and dynamic knowledge base completion through field type mapping tables. Compared with the existing technology, it can effectively improve the accuracy and automation level of information extraction from scanned copies of construction industry communication project contracts, and achieves the technical effect of quickly and accurately structuring paper contract scans and importing them into target templates. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 The present invention is a flowchart of a method for intelligently identifying and importing scanned copies of communication project contracts in the construction industry according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation methods in conjunction with the accompanying drawings.

[0045] Construction industry communication projects often involve numerous contract documents, encompassing key information ranging from project costs and schedules to quality assurance clauses. Due to the numerous project participants and complex contract structures, contract management often faces issues such as fragmented information, confusing versions, and inefficient review, severely impacting project efficiency and risk management. This invention addresses these industry challenges by providing a method for intelligently identifying and importing scanned contracts for construction industry communication projects.

[0046] The present invention provides a method for intelligently identifying and importing scanned copies of construction industry communication project contracts. By incorporating the multimodal feature parsing model of Faster R-CNN and BERT, it achieves efficient parsing of the complex typesetting and specialized terminology in contract scans. It also combines a regular expression rule library with the national construction project demonstration text classification standard to construct required and optional field mapping tables, improving the standardization and adaptability of field recognition. Furthermore, through a dual evaluation mechanism of edit distance and cosine similarity, as well as a missing field completion mechanism based on a dynamic knowledge base, the integrity of key information and system availability are further guaranteed, significantly reducing the need for manual intervention, making it suitable for intelligent applications in large-scale project contract management scenarios.

[0047] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. Rather, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0048] Example 1

[0049] See also Figure 1 A method for intelligently identifying and importing scanned copies of construction industry communication project contracts according to an embodiment of the present invention is applied to construction industry communication projects, comprising:

[0050] Step S100: Obtain a target template and a scanned copy of a contract to be processed, perform image preprocessing on the scanned copy of the contract, and simultaneously perform structured parsing on the target template, extracting standard field values ​​and marking field types, and generating a field type mapping table;

[0051] The field type mapping table includes a required field mapping table and an optional field mapping table;

[0052] Specifically, image preprocessing of contract scans includes:

[0053] Grayscale processing: convert color images into grayscale images to remove color information interference and simplify subsequent processing;

[0054] Binarization: Convert the grayscale image into a black and white binary image, separate the text from the background by setting a threshold, and improve the contrast to facilitate subsequent extraction;

[0055] Denoising: Use median filtering, Gaussian filtering or morphological operations to remove noise points, spots and line interference in the image;

[0056] Tilt correction: Retrieve the rotation angle in the image and perform rotation correction on the image through affine transformation or perspective transformation to keep the document content horizontal and avoid misalignment or recognition failure during subsequent recognition.

[0057] Step S200: Input the preprocessed contract scan into a pre-built multimodal feature parsing model to extract visual structural features and textual language features. Based on these features, a contract information extraction set is obtained. The visual structural features are primarily used through a visual branch layer based on the Faster R-CNN architecture to identify content layout information such as table areas, titles, and signatures. The textual language features are extracted through a BERT model combined with a textual language branch layer from a contract-specific corpus to extract language-level information such as clause semantics and entity relationships. Finally, a feature fusion layer composed of a Transformer decoder effectively fuses the visual and textual features to form a contract information extraction set containing rich semantic information, including multiple contract information fields and their corresponding content.

[0058] Step S300: semantically match the contract information extraction set with the field type mapping table to generate a field matching result, and input the successfully matched contract information into the target template;

[0059] Step S400: If a required field is missing in the field matching result, the related historical information is retrieved from the dynamic knowledge base through entity recognition and linking technology to obtain the contract information mapped by the missing field, and the contract information mapped by the missing field is input into the target template.

[0060] The intelligent recognition and import method of scanned copies of construction industry communication project contracts in this embodiment can significantly improve the accuracy and automation level of contract information recognition, reduce manual intervention, and achieve the technical effect of quickly and accurately converting unstructured paper contract scans into structured data and importing them into the target template.

[0061] Example 2

[0062] The present embodiment provides a method for intelligently identifying and importing scanned copies of construction industry communication project contracts, which is applied to construction industry communication projects and includes:

[0063] Step S100: Obtain a target template and a scanned copy of a contract to be processed, perform image preprocessing on the scanned copy of the contract, and obtain a preprocessed scanned copy of the contract; perform structured parsing on the target template, extract standard field values, mark field types, and generate a field type mapping table;

[0064] The field type mapping table includes a required field mapping table and an optional field mapping table;

[0065] Step S200: Input the pre-processed contract scan into a pre-built multimodal feature parsing model to extract visual structure features and text language features, and obtain a contract information extraction set based on the visual structure features and text language features;

[0066] Step S300: semantically match the contract information extraction set with the field type mapping table to generate a field matching result, and input the successfully matched contract information into the target template;

[0067] Step S400: If a required field is missing in the field matching result, the associated historical information is retrieved from the dynamic knowledge base through entity recognition and linking technology to obtain the contract information mapped by the missing field, and the contract information mapped by the missing field is input into the target template.

[0068] The multimodal feature parsing model in step S200 includes:

[0069] The visual branch layer, based on the Faster R-CNN architecture, extracts visual structural features of tables, titles, and signatures;

[0070] The text language branch layer uses the BERT model, combined with a contract-specific corpus, to extract clause semantics and entity relationship features to obtain text language features;

[0071] The feature fusion layer combines visual structure features and text language features through the Transformer decoder to obtain the contract information extraction set.

[0072] Specifically, the visual branching layer in this embodiment accurately locates and identifies the boundaries and content of complex table structures, such as budget tables and project schedules, commonly found in construction industry communication project contracts. Furthermore, the visual branching layer automatically distinguishes between titles of different sections, such as project overview, payment terms, and body content, segmenting and classifying the multi-level title structures commonly found in construction industry communication project contracts. Furthermore, during the contract signing process, the visual branching layer, utilizing the Faster R-CNN architecture, accurately identifies and extracts panoramic images from the contract, providing a foundation for further authenticity verification.

[0073] In this embodiment, the Faster R-CNN architecture within the visual branch layer improves its recognition accuracy for specific format tables by training a construction industry communication project contract set with a large number of construction industry communication project-related table images.

[0074] Furthermore, the text language branching layer in this embodiment focuses on deeply understanding and parsing the specific terms in the contract. At the same time, it is fine-tuned using a contract-specific corpus containing a rich vocabulary of construction industry communication projects, enabling the text language branching layer to accurately capture the text language features in the contract text. The contract-specific corpus containing a rich vocabulary of professional industry projects is constructed through the following steps:

[0075] Access internal contract archives. Large construction companies or general contractors often have a large number of historical contract documents, including construction contracts, design contracts, supervision contracts, and procurement contracts. These contracts themselves contain a large number of professional terms, clause templates, and fixed expressions used in construction industry communications projects.

[0076] Obtain contracts filed with government or industry authorities, such as public or authorized contract data stored by housing and construction departments, public resource trading centers, and bidding platforms. These contracts are highly standardized, well-structured, and have standardized terminology.

[0077] We perform OCR recognition on the internal contract archives of the above-mentioned enterprises and the contracts filed with the government or industry authorities, extract the text, clean the text, segment it, and annotate the key fields, and then build a professional corpus that can be used for BERT model training as a contract-specific corpus.

[0078] The feature fusion layer combines the visual structure features obtained from the visual branch layer with the text language features obtained from the text language branch layer through the Transformer decoder. It comprehensively considers visual elements and text content, thereby generating a more comprehensive and accurate contract information extraction set.

[0079] In this embodiment, the contract information extraction set includes contract information fields and corresponding field contents;

[0080] The contract information fields include: project information, Party A information, construction unit information and time information; project information includes project name, project category, professional category, project year, project location information, contract number and contract amount; Party A information includes Party A's witness and Party A's contact number; construction unit information includes construction unit category, construction unit address, construction unit name and project manager; the time information includes signing date, start date and completion date.

[0081] In the specific implementation process, in step S100, the target template is structurally parsed, the field standard values ​​are extracted and the field types are marked, and a field type mapping table is generated;

[0082] Step S110: construct a regular expression rule library based on the construction industry communication project contract specification, perform a full text scan on the target template, and identify a set containing field identifiers as an initial field candidate set;

[0083] Step S120: Based on the national construction project demonstration text classification standard, the initial field candidate set is type-labeled and hierarchically divided to obtain a required field mapping table and an optional field mapping table.

[0084] The required field mapping table includes project information, Party A information, construction unit information, and time information. The optional field mapping table includes supplementary agreement terms and customized additional instructions.

[0085] Specifically, a target template is structurally parsed, and the target template may be a Word document, a PDF document, or a scanned document processed by OCR.

[0086] Optionally, in step S110, performing full text scanning on the target template according to the constructed regular expression rule library further includes:

[0087] (1) Target template preprocessing: standardize the target template format, including unifying character encoding, removing scanned watermarks, and repairing damaged formats;

[0088] Convert the target template content into plain text format to remove formatting interference; use the established policy expression rule library to scan the entire contract text line by line to match all possible field keywords, field values, and their contextual information;

[0089] (2) Generate an initial field candidate set: organize all identified field keywords and corresponding contents into a preliminary field candidate set, each field candidate contains the field type and the field type standard value.

[0090] Furthermore, after type annotation and hierarchical division, it is determined whether the field is a required field or an optional field based on the field type;

[0091] In the optional field mapping table, supplementary agreement terms include change terms, breach of contract liability, etc., and customized additional requirements include environmental protection requirements, energy-saving measures, etc.

[0092] In this embodiment, step S300 includes:

[0093] Step S310: Using an edit distance algorithm, obtain the similarity between each contract information field and the corresponding target template field type as a first similarity;

[0094] The edit distance algorithm is the minimum number of single-character editing operations required to convert two strings from one to another, that is, the minimum number of single-character editing operations required to convert the contract information field into the target template field type.

[0095] For example, if the contract information field A is "Project Name" and the target template field type B is "Project Name", and both have a length of 4, then create a matrix of size (m+1)×(n+1), where m is the length of A and n is the length of B. The first row and the first type of the matrix represent the number of operations from the string to the target string.

[0096] For each position (i, j) in the matrix, it represents the edit distance between the first i characters of A and the first j characters of B. If A(i) = B(j), no operation is required and the value in the upper left corner is directly integrated. Otherwise, the minimum of the following three directions is taken + 1:

[0097] Left: Insert operation;

[0098] Above: delete operation;

[0099] Upper left corner: Replace operation;

[0100] The value in the lower right corner of the matrix is ​​the edit distance between the two strings. From this, we can see that the edit distance from "project name" to "project name" is 1;

[0101] Enter the above edit distance into the following formula to obtain the first similarity:

[0102]

[0103] From this we can see that the similarity between "project name" and "project name" is 0.75.

[0104] Step S320: extracting the semantic similarity between the content of each contract information field and the standard value of the target template field type by cosine similarity calculation as a second similarity;

[0105] Specifically, it includes: using the BERT model to encode the contract information field content and the target template field type respectively, generating corresponding high-dimensional vectors, substituting the two vectors into the cosine similarity formula to obtain the semantic similarity score between them, namely the second similarity.

[0106] The cosine similarity formula is:

[0107]

[0108] Where C is the cosine similarity, i.e., the second similarity, ai represents the i-th element of the contract information field, bi represents the i-th element of the target template field type standard value, and n is the number of elements shared by both vectors.

[0109] Step S330: Generate a field matching result based on the first similarity and the second similarity, and input the successfully matched contract information into the target template.

[0110] Furthermore, step S330 includes:

[0111] When the first similarity of the required field is ≥0.7 and the second similarity is ≥0.75, the field matching result is successful and the corresponding contract information is entered into the target template;

[0112] When the first similarity of the optional field is ≥0.6 or the second similarity is ≥0.6, the field matching result is a successful match, and the corresponding contract information is input into the target template.

[0113] In this embodiment, step S400 includes:

[0114] Step S410: Based on the field matching results, identify the missing segments in the required fields that failed to match;

[0115] Step S420: Retrieve historical information associated with the missing fields in the dynamic knowledge base, perform multi-dimensional scoring on the historical information, and obtain the historical information with the highest score as the contract information for each missing field;

[0116] Step S430: Input the contract information of each missing field into the target template.

[0117] In step S420, performing multi-dimensional scoring on historical information includes:

[0118] Obtain the data source credibility, time coefficient, and semantic similarity of each historical information, and input the data source credibility, time coefficient, and semantic similarity of each historical information into the following formula to obtain the score of each historical information:

[0119] Score = Data source credibility weight × time coefficient × semantic similarity 2 ;

[0120] Among them, the credibility weight of government-registered data is 0.9, and the credibility weight of enterprise-filled data is 0.6; the time coefficient of data in the past two years is 1.2, the time coefficient of data between 2 and 3 years is 1.0, and the time coefficient of data over 3 years is 0.5; the semantic similarity is the similarity between the context of missing fields and historical information, and the semantic similarity is obtained using the BERT semantic matching model.

[0121] Among them, the dynamic knowledge base includes but is not limited to: the registration database of the housing and construction department, the enterprise registration information database of the market supervision department, the enterprise historical contract database, the engineering cost information database and the legal and regulatory knowledge base.

[0122] Furthermore, the method for intelligently identifying and importing scanned copies of communication project contracts in the construction industry according to this embodiment further includes:

[0123] After obtaining the contract information extract set, it is automatically desensitized, using a reversible encryption algorithm or masking replacement to hide the original information. Specifically, this includes automatically desensitizing fields involving personal privacy information, sensitive corporate data, project cost amounts, and Party A's contact information.

[0124] At the same time, after the contract information is imported, the temporary cache data is automatically cleared, and key operations are digitally signed and recorded to prevent information leakage and illegal tampering.

[0125] The present embodiment provides a method for intelligently identifying and importing scanned copies of construction industry communication project contracts, which is specially designed and optimized for the actual needs of construction industry communication project contract management. The present invention deeply integrates visual structural features and text language features, and can effectively cope with the challenges of information extraction and management of various contract documents (such as construction contracts, design contracts, supervision contracts, etc.) in construction industry communication projects, improves the processing efficiency of contract information, strengthens the accuracy of contract content recognition, and enhances adaptability and flexibility. It not only significantly improves the efficiency and accuracy of construction industry communication project contract management, but also provides strong support for the company's informatization construction and intelligent management level.

[0126] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0127] In the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," "connect," "fixed," etc. should be understood broadly. For example, they may refer to fixed connection, detachable connection, or integration; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; and internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0128] In the present invention, unless otherwise expressly specified or limited, when a first feature is "above" or "below" a second feature, it may mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediate medium. Furthermore, when a first feature is "above," "above," or "above" a second feature, it may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is at a higher level than the second feature. When a first feature is "below," "below," or "below" a second feature, it may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is at a lower level than the second feature.

[0129] In the description of this specification, the terms "one embodiment", "some embodiments", "embodiments", "examples", "specific examples" or "some examples" refer to the specific features, structures, materials or characteristics described in conjunction with the embodiment or example and included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and features of different embodiments or examples, unless they are mutually inconsistent.

[0130] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may alter, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for intelligent recognition and importing scanned copies of communication project contracts in the construction industry, characterized in that: Applied to a construction industry communication project, the method includes: S100: Obtain a target template and a scanned copy of a contract to be processed, perform image preprocessing on the scanned copy of the contract, and obtain a preprocessed scanned copy of the contract; and perform structured parsing on the target template, extracting standard field values ​​and marking field types, and generating a field type mapping table; The field type mapping table includes a required field mapping table and an optional field mapping table; S200: Inputting the pre-processed contract scan into a pre-built multimodal feature parsing model to extract visual structure features and text language features, and obtaining a contract information extraction set based on the visual structure features and text language features; S300: semantically match the contract information extraction set with the field type mapping table to generate a field matching result, and input the successfully matched contract information into the target template; S400. If a required field is missing in the field matching result, relevant historical information is retrieved from the dynamic knowledge base through entity recognition and linking technology to obtain the contract information mapped by the missing field, and the contract information mapped by the missing field is input into the target template.

2. The intelligent recognition and import method for scanned copies of communication project contracts in the construction industry according to claim 1 is characterized in that: In S200, the multimodal feature parsing model includes: The visual branch layer, based on the Faster R-CNN architecture, extracts visual structural features of tables, titles, and signatures; The text language branch layer uses the BERT model, combined with a contract-specific corpus, to extract clause semantics and entity relationship features to obtain text language features; The feature fusion layer combines visual structure features and text language features through the Transformer decoder to obtain the contract information extraction set.

3. The intelligent recognition and import method for scanned copies of communication project contracts in the construction industry according to claim 2 is characterized in that: The contract information extraction set includes contract information fields and corresponding field contents; The contract information fields include: project information, Party A information, construction unit information and time information; the project information includes project name, project category, professional category, project year, project location information, contract number and contract amount; the Party A information includes Party A's witness and Party A's contact number; the construction unit information includes construction unit category, construction unit address, construction unit name and project manager; the time information includes signing date, start date and completion date.

4. The intelligent recognition and import method for scanned copies of communication project contracts in the construction industry according to claim 1 is characterized in that: In S100, the target template is structurally parsed, field standard values ​​are extracted, field types are marked, and a field type mapping table is generated; S110, building a regular expression rule library based on the construction industry communication project contract specification, performing a full text scan on the target template, and identifying a set containing field identifiers as an initial field candidate set; S120. Based on the national construction project demonstration text classification standard, the initial field candidate set is type-labeled and hierarchically divided to obtain a required field mapping table and an optional field mapping table.

5. The intelligent recognition and import method for scanned copies of communication project contracts in the construction industry according to claim 4 is characterized in that: The required field mapping table includes project information, Party A information, construction unit information and time information, and the optional field mapping table includes supplementary agreement terms and customized additional instructions.

6. The intelligent recognition and import method for scanned copies of communication project contracts in the construction industry according to claim 1 is characterized in that: The S300 includes: S310: Using an edit distance algorithm to obtain a similarity between each contract information field and a corresponding target template field type as a first similarity; S320, extracting the semantic similarity between the content of each contract information field and the standard value of the target template field type by cosine similarity calculation as a second similarity; S330: Generate a field matching result based on the first similarity and the second similarity, and input the successfully matched contract information into the target template.

7. The intelligent recognition and import method for scanned copies of communication project contracts in the construction industry according to claim 6 is characterized in that: The S330 includes: When the first similarity of the required field is ≥0.7 and the second similarity is ≥0.75, the field matching result is successful and the corresponding contract information is entered into the target template; When the first similarity of the optional field is ≥0.6 or the second similarity is ≥0.6, the field matching result is a successful match, and the corresponding contract information is input into the target template.

8. The intelligent recognition and import method for scanned copies of communication project contracts in the construction industry according to claim 1 is characterized in that: The S400 includes: S410. Based on the field matching results, identify missing segments in the required fields that failed to match; S420: Retrieve historical information associated with the missing fields in the dynamic knowledge base, perform multi-dimensional scoring on the historical information, and obtain the historical information with the highest score as the contract information for each missing field; S430: Input the contract information of each missing field into the target template.

9. The intelligent recognition and import method for scanned copies of communication project contracts in the construction industry according to claim 8 is characterized in that: In S420, performing multi-dimensional scoring on the historical information includes: Obtain the data source credibility, time coefficient, and semantic similarity of each piece of historical information, and input the data source credibility, time coefficient, and semantic similarity of each piece of historical information into the following formula to obtain the score of each piece of historical information: Score = Data source credibility weight × time coefficient × semantic similarity 2 ; Among them, the credibility weight of government-registered data is 0.9, and the credibility weight of enterprise-filled data is 0.6; the time coefficient of data in the past two years is 1.2, the time coefficient of data between 2 and 3 years is 1.0, and the time coefficient of data over 3 years is 0.5; the semantic similarity is the similarity between the context of missing fields and historical information, and the semantic similarity is obtained using the BERT semantic matching model.

10. The intelligent recognition and import method for scanned copies of communication project contracts in the construction industry according to claim 8 is characterized in that: The dynamic knowledge base includes but is not limited to: a filing database of the housing and construction department, an enterprise registration information database of the market supervision department, an enterprise historical contract database, an engineering cost information database, and a legal and regulatory knowledge database.

Images