Office plug-in-oriented offline-online hybrid bidding document intelligent editing and auditing system

Abstract

The invention belongs to the technical field of bidding business informatization, and provides an Office plug-in-oriented offline-online mixed bidding document intelligent editing and auditing system, which comprises a localized plug-in module, a bidding document editing module, a bidding document editing module, a bidding document editing module, a bidding document editing module, a bidding document editing module and a bidding document editing module, the content generation module converts the bid invitation terms into knowledge base query statements and generates a dynamic filling template; the double-model verification module carries out format and semantic double review through cooperation of a local lightweight model and a cloud large model; the offline data security module supports writing of a bidding document in an offline state, temporarily stores sensitive data in the bidding document into a local encryption sandbox, and performs data synchronization after network recovery; the version comparison module tracks a DOM node change path between bidding document versions and generates a difference report of bidding documents of different versions; through deep integration of offline operation and online collaboration and based on localized operation support of the Office plug-in, a user can edit the bidding document in real time in a network-free environment, and the method adapts to the flexibility requirement of a bidding scene.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of informationization of bidding business, and particularly relates to an offline-online hybrid intelligent tender document compiling and reviewing system for Office plug-ins. BACKGROUND

[0002] In the field of bidding, efficient and accurate compilation and safe management of tender documents are crucial for enterprises to participate in project competition. With the development of information technology, tender document compiling and reviewing systems based on computer software have gradually become mainstream tools;

[0003] However, most current tender document compiling and reviewing systems rely on online operation and lack effective offline operation support. Users cannot normally carry out tender document writing work in poor network environments (such as remote areas, subway scenes, etc.) or offline environments, which leads to delay of work progress and seriously affects the efficiency of tender document compilation. Some systems with offline function only provide basic document editing capabilities and cannot realize automatic synchronization of data with online systems. Users need to manually migrate data, which is tedious and prone to data loss or inconsistency.

[0004] In addition, sensitive contents such as bid information and core technical parameters of enterprises in the tender document lack reliable encryption storage mechanisms in offline state, and there is a risk of data leakage. During network transmission, some systems do not strictly encrypt data, which is difficult to meet the strict requirements of data confidentiality in the field of bidding. Once data is leaked, it will cause significant economic losses and reputation damage to enterprises.

[0005] In the content review link of tender documents, traditional systems mostly adopt a single review method. Either only rely on local rule library for format checking, which cannot effectively review semantic level problems such as completeness of technical parameter response; or rely on cloud services for semantic analysis, but there are problems of high review delay and poor real-time performance, which makes it difficult to discover and correct errors in time during the process of tender document writing. In addition, existing Office plug-in type tender document compiling and reviewing tools have relatively single functions, mainly focusing on basic editing operations, and lack deep optimization for tender document compiling and reviewing business scenarios, which cannot provide one-stop services for users from content generation, format and semantic review to version comparison. SUMMARY

[0006] To solve the above technical problems, the application provides an offline-online hybrid intelligent tender document compiling and reviewing system for Office plug-ins to solve the problems proposed in the background art.

[0007] The offline-online hybrid intelligent tender document compiling and reviewing system for Office plug-ins comprises:

[0008] The localization plug-in module is used to provide a tender document writing interface and offline operation support;

[0009] a content generation module configured to convert the NL2SQL instruction, convert the tender clause into a knowledge base query statement, and generate a dynamically filled template;

[0010] a double model verification module configured to perform double review of format and semantics by the local lightweight model and the cloud large model in cooperation;

[0011] an offline data security module configured to support writing of the tender in a disconnected state, temporarily store sensitive data in the tender in a local encrypted sandbox, and perform data synchronization after network recovery, wherein the sensitive data includes bid information;

[0012] a version comparison module configured to track the DOM node change path between versions of the tender, and generate a difference report of different versions of the tender.

[0013] Preferably, the content generation module further comprises:

[0014] the tender clause is parsed into a structured query by a BERT model, and the following formula is used:

[0015]

[0016] P(q|c)=Softmax(W o BERT(c))

[0017] wherein c is the tender clause text, BERT(c) is the encoding output of the BERT model for the input text c, W o is a weight matrix for mapping the output features of the BERT to a set of predefined query templates, P(q|c) is the probability of matching to a specific query template q given the tender clause text c, Q D is a set of preset query templates, Q is the final parsed query template obtained by maximizing the probability, and represents selecting the template with the highest probability from all candidate templates;

[0018] a placeholder with a hyperlink is automatically inserted according to the query result, wherein the hyperlink points to the associated evidence material in the enterprise knowledge base, and a dynamically filled template with the hyperlink is generated.

[0019] Preferably, the double model verification module further comprises:

[0020] the tender format error is detected in real time by the local lightweight model, the format error is identified based on a rule engine and a CNN image classifier, the format error includes page number error, font violation and margin error, and the following formula is used to obtain a set of tender format errors: E l{e|e∈PNE∪FV∪ME}, wherein PNE is a page number error, FV is a font violation, and ME is a margin error;

[0021] Through cloud large model synchronization, semantic compliance review is performed, correction suggestions are generated, and the correction suggestions are prompted through the Office plug-in sidebar to perform real-time proofreading.

[0022] Preferably, the offline data security module further comprises:

[0023] By establishing an AES-256 encrypted local sandbox storage area, sensitive data in the bid document is temporarily stored in the local encrypted sandbox, which is represented as:

[0024] Enc(d,k)=AES(d,PBKDF2(k,s)),s~U(0,2 128 )

[0025] Wherein, d is sensitive data, k is user password, s is a random salt in U(0,2 128 ), U(0,2 128 ) represents that the salt s is randomly sampled from the uniform distribution from 0 to 2 128 , PBKDF2(k,s) is a password engine, which converts the user password k and the random salt s into a key through multiple iterations;

[0026] After the network is restored, the data is synchronized after identity verification through zero-knowledge proof, which is represented as:

[0027] Verify ZKP (P,V)=I(Hash(P.x)≡V)

[0028] Wherein, Verify ZKP (P,V) is the output of the zero-knowledge proof verification function, indicating the verification result; P is the proof object or proof information provided by the prover; V is the value corresponding to the proof, which is used to compare with the hash value in the verification; P.x is a certain element or field in the proof object; Hash(P.x) is the hash value calculated for P.x; I is the output function of the verification result, indicating whether the verification is successful.

[0029] Preferably, the version comparison module further comprises:

[0030] By tracking the revision track, the history changes of the document object model nodes are recorded, and the modified content is marked by a color mapping function to generate a difference report of different versions of the bid document;

[0031] The difference report includes technical scheme adjustment items and bid price fluctuation range.

[0032] Preferably, the localization plug-in module cooperates with the WEB review end, including:

[0033] Bidirectional data updating is performed through a differential synchronization protocol, and is represented as:

[0034]

[0035] Wherein, Delta represents differential data, that is, a change part to be synchronized or updated, Compress(Delta) represents compression of the differential data, reducing the volume thereof, saving transmission bandwidth and storage space; RSync is a remote synchronization protocol, used for processing the compressed differential data; and CRC32 is a cyclic redundancy check, used as a check code of the differential data.

[0036] The WEB end review opinion is displayed in real time on the plug-in side sidebar, and is anchored to a specific document position through an Office comment function.

[0037] Compared with the prior art, the present application has the following beneficial effects:

[0038] 1. The present application realizes deep fusion of offline operation and online cooperation, supports complete bidding document writing and safe storage of sensitive data in an offline state, and automatically synchronizes after network recovery, meeting the working requirements of users in different network environments; and through double model checking modules, double review is realized, so that the content of the bidding document is accurate and correct, the format is standard, the quality of the bidding document is improved, and the probability of winning the bid is improved, realizing double review of the format and the semantics, improving the comprehensiveness and accuracy of the review while ensuring real-time performance.

[0039] 2. The present application supports offline operation through a localization plug-in module, breaks the network limit, and users can carry out bidding document writing work in any scene, adapting to diversified working environments, while the offline data security module guarantees the safety of sensitive data, meeting the strict requirements of the bidding field on data confidentiality, and avoiding data leakage risks for enterprises.

[0040] 3. The system of the present application is integrated in an Office plug-in, conforms to the operation habits of users, and each functional module is easy to operate, version comparison is intuitive and easy to understand, reduces the user threshold, and improves user experience. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 is a block diagram of the offline-online hybrid bidding document intelligent compiling and reviewing system of the present application;

[0042] Figure 2 is a flowchart of NL2SQL instruction conversion in the content generation module of the present application. DETAILED DESCRIPTION

[0043] The embodiments of the present application will be further described in conjunction with the drawings and examples. The following examples are used to illustrate the present application, but cannot be used to limit the scope of the present application.

[0044] Embodiment: As shown in the accompanying drawings and examples, the present application provides an offline-online hybrid intelligent bidding review system for Office plug-ins, comprising: Figure 1

[0045] A localized plug-in module is used to provide a bidding document writing interface and offline operation support; the localized plug-in module cooperates with the WEB review end, comprising:

[0046] Bidirectional data updating is performed through a differential synchronization protocol, denoted as:

[0047]

[0048] Wherein, Δ represents differential data, i.e. the changed part to be synchronized or updated, Compress(Δ) represents compression of the differential data, reducing its volume, saving transmission bandwidth and storage space; RSync is a remote synchronization protocol, used to process the compressed differential data; CRC32 is a cyclic redundancy check, used as a check code for the differential data;

[0049] The WEB end review opinions are displayed in real time in the plug-in sidebar, and are anchored to specific document locations through the Office comment function.

[0050] The localized plug-in module enables users to flexibly write bidding documents in various scenarios without relying on the network, greatly improving the convenience and flexibility of bidding document writing, effectively improving the efficiency of bidding document writing, and reducing work interruptions and delays caused by network problems.

[0051] A content generation module, as shown in the accompanying drawings and examples, is used to convert NL2SQL instructions, convert the bidding terms into knowledge base query statements, and generate dynamic filling templates; the bidding terms are parsed into structured queries through a BERT model, using the following formula: Figure 2

[0052]

[0053] P(q|c)=Softmax(W o BERT(c))

[0054] Wherein, c is the bidding term text, BERT(c) is the encoding output of the input text c by the BERT model, W o is a weight matrix used to map the output features of BERT to a set of predefined query templates, P(q|c) is the probability of matching to a specific query template q given the bidding term text c, Q D ​​is the preset query template set, Q is the final parsed query template obtained by maximizing the probability, Indicates that among all candidate templates, the template with the highest probability is selected;

[0055] Automatically insert placeholders with hyperlinks based on the query results. The hyperlinks point to related evidence materials in the enterprise knowledge base, generating a dynamic filling template with hyperlinks.

[0056] Accurately convert the terms in the bidding documents into query statements for the enterprise knowledge base, accurately match relevant content from the knowledge base, and generate dynamic fill-in templates with hyperlinks based on this to assist users in quickly completing the writing of bid content; avoid users from manually searching and writing bid content one by one, greatly saving time and energy, improving the accuracy and completeness of bid content, enabling bids to more accurately respond to the requirements of the bidding documents and enhance their competitiveness.

[0057] The dual-model verification module is used to perform dual format and semantic review through the collaboration of a local lightweight model and a large cloud model. The local lightweight model is used to detect bid format errors in real time. The rule engine and CNN image classifier are used to identify format errors. Format errors include page number errors, font violations, and margin errors. The following formula is used to obtain the bid format error set: E l = {e|e∈PNE∪FV∪ME}, where PNE is the page number error, FV is the font violation, and ME is the margin error;

[0058] Perform semantic compliance reviews synchronously through a large cloud model, generate correction suggestions, and display correction suggestions in the Office plug-in sidebar for real-time proofreading.

[0059] Use a local lightweight model to perform real-time format checks on the tender being written, and use a large cloud model to review the tender content at a semantic level to determine semantic compliance issues such as the completeness of technical parameter responses. Correction suggestions are given in the Word plug-in sidebar, allowing format and semantic errors to be discovered and corrected in a timely manner during the tender writing process, ensuring that the tender format is standardized and the content is accurate and compliant, reducing the risk of tender rejection due to format and semantic issues, and improving tender quality.

[0060] The offline data security module is used to support the writing of bids in an offline state, temporarily storing sensitive data in the bids in a local encrypted sandbox, and synchronizing the data after the network is restored. Sensitive data includes quotation information. By establishing an AES-256 encrypted local sandbox storage area, sensitive data in the bids is temporarily stored in the local encrypted sandbox, which is expressed as:

[0061] Enc(d,k)=AES(d,PBKDF2(k,s)),s~U(0,2 128 )

[0062] where d is sensitive data, k is user password, s ~ U(0,2 128 ) is a random salt, U(0,2 128 ) represents that the salt s is randomly sampled from a uniform distribution from 0 to 2 128 , and PBKDF2(k, s) is a password engine, which converts the user password k and the random salt s into a key through multiple iterations.

[0063] After the network is restored, the data is synchronized after identity verification through zero-knowledge proof, which is represented as:

[0064] Verify ZKP (P, V) = I(Hash(P.x)≡V)

[0065] where Verify ZKP (P, V) is the output of the zero-knowledge proof verification function, indicating the verification result; P is the proof object or proof information provided by the prover; V is the value corresponding to the proof, which is used to compare with the hash value in verification; P.x is a certain specific element or field in the proof object; Hash(P.x) is the hash value calculated for P.x; I is the output function of the verification result, indicating whether the verification is successful.

[0066] When the user is offline, the sensitive data such as bid information in the bid document is temporarily stored in the local encrypted sandbox to ensure that the data is not illegally obtained in the offline state; after the network is restored, the data is automatically and securely synchronized to the designated terminal, effectively ensuring the security and integrity of sensitive data in the offline and online transmission process, meeting the strict requirements of the bidding field for data security, preventing data leakage, and avoiding economic losses and reputation damage to the enterprise.

[0067] The version comparison module is used to track the DOM node change path between different versions of the bid document and generate a difference report of different versions of the bid document; by tracking the revision trail, the history changes of the document object model nodes are recorded, and the modified content is marked by a color mapping function to generate a difference report of different versions of the bid document;

[0068] The difference report includes technical scheme adjustment items and bid price change range. By tracking the DOM node change path between different versions of the bid document, a detailed difference report is automatically generated, and the differences between versions are presented in an intuitive way such as color blocks, including technical scheme adjustment items, bid price change range, etc., which facilitates users and reviewers to quickly and accurately understand the bid document version change, assists decision-making, improves bid document review and modification efficiency, and reduces the workload and error rate of manual version difference comparison.

[0069] Experimental data comparison: By comparing the performance with the traditional scheme, the results are shown in the following table:

[0070]

[0071] Test environment:

[0072] Dataset: 1,200 public bidding documents (covering 8 major industries) from 2020 to 2023;

[0073] Comparison object: A certain commercial bid document auxiliary software V5.2 (using pure rule engine).

[0074] As can be seen from the above, the prior art focuses on online bid review or only provides basic offline editing functions. The present system realizes the deep integration of offline operation and online collaboration, supports complete bid writing and sensitive data security storage in offline state, automatically synchronizes after network recovery, and meets the work needs of users in different network environments. Unlike traditional single format or semantic review methods, the present system uses local lightweight models and cloud-based large models to work together, achieving dual review of format and semantics, ensuring real-time while improving the comprehensiveness and accuracy of review. By being closely integrated into Office plugins, the system is deeply optimized for bid review scenarios, providing one-stop services from content generation to version comparison. Compared with existing Office plugins that only support basic editing functions, the present system is more in line with the actual needs of the bidding business.

[0075] Importantly, it should be noted that the configurations and arrangements of the present application shown in the various exemplary embodiments are merely illustrative. Although only a few embodiments have been described in detail herein, persons of ordinary skill in the art, having the benefit of this disclosure, should understand that many modifications are possible without materially departing from the novel teachings and advantages described in this application. Other substitutions, modifications, changes and omissions can be made in the design, operation and arrangement of the exemplary embodiments without departing from the scope of the present application. Therefore, the present application is not limited to a particular embodiment, but extends to various modifications that still fall within the scope of the appended claims.

[0076] Furthermore, in order to provide a brief description of the exemplary embodiments, not all features of the actual embodiments (i.e., those not relevant to the best mode of carrying out the present application currently under consideration, or those not relevant to the implementation of the present application) can be described.

[0077] It should be noted that the above examples are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, persons of ordinary skill in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and they should be included in the scope of the claims of the present application.

Claims

1. Offline-online hybrid tender document intelligent editing and review system for Office plug-in, characterized by: include: Localized plug-in module, used to provide a bid writing interface and offline operation support; Content generation module, used to convert bidding terms into knowledge base query statements and generate dynamic filling templates; The dual-model verification module is used to perform dual format and semantic verification by collaborating with a local lightweight model and a large cloud model. An offline data security module is used to support the writing of bid documents while disconnected from the network, temporarily storing sensitive data in the bid documents in a local encrypted sandbox and synchronizing the data after the network is restored; the sensitive data includes quotation information; The version comparison module is used to track the DOM node change path between tender document versions and generate difference reports for different tender document versions.

2. The offline-online hybrid tender document intelligent editing and review system for Office plug-ins as claimed in claim 1 is characterized in that , the content generation module also includes: The NL2SQL instructions are converted and the bidding terms are parsed into structured queries using the BERT model using the following formula: P(q|c)=Soft max(W o BERT(c)) Where c is the bidding terms text, BERT(c) is the encoding output of the BERT model for the input text c, and W o is a weight matrix used to map BERT’s output features to a predefined set of query templates. P(q|c) is the probability of matching a specific query template q given the tender clause text c. Q D is the preset query template set, Q is the final parsed query template obtained by maximizing the probability, Indicates that among all candidate templates, the template with the highest probability is selected; A placeholder with a hyperlink is automatically inserted according to the query result, wherein the hyperlink points to the associated evidence material in the enterprise knowledge base, and a dynamic filling template with a hyperlink is generated.

3. The offline-online hybrid tender document intelligent editing and review system for Office plug-in as claimed in claim 1 is characterized in that ,The dual model verification module also includes: The local lightweight model is used to detect bid format errors in real time. Format errors are identified based on the rule engine and CNN image classifier. The format errors include page number errors, font violations, and margin errors. The set of bid format errors is obtained using the following formula: E l = {e|e∈PNE∪FV∪ME}, where PNE is the page number error, FV is the font violation, and ME is the margin error; Perform semantic compliance reviews synchronously through a large cloud model, generate correction suggestions, and display correction suggestions in the Office plug-in sidebar for real-time proofreading.

4. The offline-online hybrid tender document intelligent editing and review system for Office plug-ins as claimed in claim 1 is characterized in that , the offline data security module also includes: By establishing an AES-256 encrypted local sandbox storage area, sensitive data in the bid will be temporarily stored in the local encrypted sandbox, as shown below: Enc(d,k)=AES(d,PBKDF2(k,s)),s~U(0,2 128 ) Among them, d is sensitive data, k is the user password, s~U(0,2 128 ) where s is a random salt, U(0,2 128 ) indicates that the salt s ranges from 0 to 2 128 Random sampling from a uniform distribution, PBKDF2(k,s) is a cryptographic engine, which means that the user password k and random salt s are converted into a key through multiple iterations; After the network is restored, the data is synchronized after the identity is verified through zero-knowledge proof, which is expressed as: Verify ZKP (P,V)=I(Hash(P.x)≡V) Among them, Verify ZKP (P, V) is the output of the zero-knowledge proof verification function, indicating the verification result; P is the proof object or proof information provided by the prover; V is the value corresponding to the proof, which is used to compare with the hash value during verification; Px is a specific element or field in the proof object; Hash(Px) is the hash value calculated for Px; I is the output function of the verification result, indicating whether the verification is successful.

5. The offline-online hybrid tender document intelligent editing and review system for Office plug-ins as claimed in claim 1 is characterized in that , the version comparison module also includes: By tracking revision traces, recording historical changes of document object model nodes, and annotating modified content through chromaticity mapping functions, a difference report of different versions of bid documents is generated; The difference report includes technical solution adjustments and quotation change range.

6. The offline-online hybrid tender document intelligent editing and review system for Office plug-ins as claimed in claim 1 is characterized in that ,The localization plug-in module collaborates with the WEB review terminal and ,includes: Bidirectional data update is performed through the differential synchronization protocol, which is expressed as: Where Δ represents differential data, i.e., the changed portion to be synchronized or updated. Compress(Δ) compresses the differential data to reduce its size, saving transmission bandwidth and storage space. RSync is the remote synchronization protocol used to process compressed differential data. CRC32 is a cyclic redundancy check, used as a checksum for differential data. The web-based review comments are displayed in real time in the plug-in sidebar and anchored to specific document locations through the Office comment function.

Images