Project document data processing method and system for intelligent review

Abstract

The present invention discloses a project document data processing method and system for intelligent review, the method comprising: obtaining an initial review model based on LLM algorithm and pre-training data set training; obtaining the project establishment document to be reviewed, inputting the project establishment document into the initial review model to obtain the output document review result; sending the document review result to the manual review end, and receiving the manual correction annotation of the document review result sent by the manual review end; inputting the project establishment document, the document review result and the manual correction annotation into the initial review model for iterative training to obtain an optimized review model. It can be seen that the present invention can realize efficient and automated intelligent review of project establishment documents, improve the effect and accuracy of the review, reduce errors, and reduce the cost of review.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a project document data processing method and system for intelligent review. Background Art

[0002] With the strengthening of scientific and technological innovation capabilities, many scientific and technological innovation entities have emerged in the market. Effective review of scientific and technological innovation projects is conducive to discovering truly innovative technologies to enhance market activity. However, when implementing the review of scientific and technological project documents, for example, when reviewing project documents based on the information innovation standards, most of the existing technologies still use manual review methods, which are inefficient, prone to errors, and have high labor costs. It is also difficult to unify the review standards. It can be seen that the existing technology has defects that need to be solved urgently. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a project document data processing method and system for intelligent review, which can realize efficient and automated intelligent review of project establishment documents, improve the effect and accuracy of the review, reduce errors, and reduce the cost of review.

[0004] In order to solve the above technical problems, the first aspect of the present invention discloses a project document data processing method for intelligent review, the method comprising:

[0005] The initial review model is trained based on the LLM algorithm and the pre-training dataset;

[0006] Obtaining a project establishment document to be reviewed, and inputting the project establishment document into the initial review model to obtain an output document review result;

[0007] Sending the document review result to a manual review end, and receiving manual correction annotations on the document review result sent by the manual review end;

[0008] The project establishment document, the document review results and the manual correction annotations are input into the initial review model for iterative training to obtain an optimized review model.

[0009] As an optional implementation, in the first aspect of the present invention, the pre-training data set includes a plurality of training project establishment documents and corresponding document review annotations.

[0010] As an optional implementation, in the first aspect of the present invention, the document review annotations include necessary content annotations, prohibited content annotations and erroneous content annotations; the necessary content annotations include at least one of current situation summary content annotations, infrastructure requirement content annotations, technology and product requirement content annotations, original technology implementation route content annotations, information innovation feasibility analysis content annotations and technology implementation route design content annotations.

[0011] As an optional implementation, in the first aspect of the present invention, the initial review model is obtained by training based on the LLM algorithm and the pre-training data set, comprising:

[0012] Obtain multiple training project establishment documents and corresponding manual review feedback results;

[0013] Based on the preset review criteria, the manual review feedback results are analyzed and summarized into a must-use word library and a prohibited word library;

[0014] According to the required word library and the prohibited word library, and based on the prompt word construction rules, each of the training project establishment documents is annotated to obtain corresponding document review annotations;

[0015] The pre-training data set including multiple training project establishment documents and corresponding document review annotations is input into the constructed LLM algorithm model for training to obtain the initial review model.

[0016] As an optional embodiment, in the first aspect of the present invention, the initial review model includes a first CNN model, a second CNN model and a Chinese corpus LLM model connected in sequence; the first CNN model is used to identify the document part and correction category corresponding to the manual correction annotation; the second CNN model is used to update the required vocabulary and the prohibited vocabulary according to the document part and the correction category, and to update and annotate the project establishment document; the Chinese corpus LLM model is an LLM model obtained by pre-training with Chinese corpus data, and is used to iteratively train according to the updated and annotated project establishment document, the document review results, the manual correction annotation and the updated required vocabulary and prohibited vocabulary to obtain an optimized review model.

[0017] As an optional embodiment, in the first aspect of the present invention, the first CNN model is used to perform the following steps:

[0018] Inputting the manual correction annotation into a pre-trained text classification CNN neural network to obtain a plurality of correction part texts and corresponding position indication texts in the manual correction annotation; the text classification CNN neural network is trained by a training data set including a plurality of training texts and corresponding correction part text annotations and position indication text annotations;

[0019] For each of the modified partial texts, the modified partial text is input into a trained classifier model to obtain a modified category corresponding to the modified partial text;

[0020] Calculate a first similarity between the position indication text corresponding to the revised part text and each preset document part in the project establishment document;

[0021] Filter out all the preset document parts whose first similarity is higher than a preset first similarity threshold to obtain a plurality of candidate document parts;

[0022] Calculating a second similarity between the corrected partial text and a corresponding partial result of each of the candidate document parts in the document review result;

[0023] Determine the candidate document portion having the second highest similarity as the document portion corresponding to the revised partial text;

[0024] All the correction part texts and the corresponding correction categories and document parts corresponding to the manual correction annotations are input into the second CNN model.

[0025] As an optional embodiment, in the first aspect of the present invention, the second CNN model is used to perform the following steps:

[0026] For each of the modified partial texts, according to the modification category corresponding to the modified partial text, a text word library corresponding to the modified partial text is determined; the text word library is the required word library or the prohibited word library;

[0027] Input the revised text into a trained keyword prediction CNN network to obtain keywords corresponding to the revised text; the keyword prediction CNN network is trained by a plurality of training revised texts and a training data set annotated with corresponding key revised words;

[0028] Calculating a third similarity between the keyword corresponding to the modified partial text and each text word in the text word library;

[0029] Replacing the text word with the highest third similarity with the keyword corresponding to the modified text portion, so as to update the text word library;

[0030] Determine the partial text corresponding to the text portion in the project establishment document according to the document portion corresponding to the revised partial text;

[0031] Calculating a fourth similarity between the keyword corresponding to the modified partial text and each text object in the partial text; the text object is a character, word, word combination, paragraph or text block RUN object;

[0032] The revised part of the text is annotated to the text object with the fourth highest similarity to achieve update annotation of the project establishment document.

[0033] As an optional embodiment, in the first aspect of the present invention, the method further comprises:

[0034] After the optimal optimization review model is obtained through iterative training, the formal project establishment document to be reviewed sent by the client is obtained;

[0035] Inputting the formal project establishment document into the optimization review model to obtain the output document formal review result;

[0036] The formal review results of the document are sent to the client for display.

[0037] A second aspect of an embodiment of the present invention discloses a project document data processing system for intelligent review, the system comprising:

[0038] The pre-training module is used to train the initial review model based on the LLM algorithm and the pre-training dataset;

[0039] A review module, used to obtain the project establishment document to be reviewed, and input the project establishment document into the initial review model to obtain an output document review result;

[0040] A correction module, used to send the document review result to the manual review end, and receive manual correction annotations on the document review result sent by the manual review end;

[0041] The optimization module is used to input the project establishment document, the document review results and the manual correction annotations into the initial review model for iterative training to obtain an optimized review model.

[0042] As an optional implementation, in the second aspect of the present invention, the pre-training data set includes a plurality of training project establishment documents and corresponding document review annotations.

[0043] As an optional implementation, in the second aspect of the present invention, the document review annotations include necessary content annotations, prohibited content annotations and erroneous content annotations; the necessary content annotations include at least one of current situation summary content annotations, infrastructure requirement content annotations, technology and product requirement content annotations, original technology implementation route content annotations, information innovation feasibility analysis content annotations and technology implementation route design content annotations.

[0044] As an optional implementation, in the second aspect of the present invention, the specific manner in which the pre-training module obtains the initial review model based on the LLM algorithm and the pre-training data set includes:

[0045] Obtain multiple training project establishment documents and corresponding manual review feedback results;

[0046] Based on the preset review criteria, the manual review feedback results are analyzed and summarized into a must-use word library and a prohibited word library;

[0047] According to the required word library and the prohibited word library, and based on the prompt word construction rules, each of the training project establishment documents is annotated to obtain corresponding document review annotations;

[0048] The pre-training data set including multiple training project establishment documents and corresponding document review annotations is input into the constructed LLM algorithm model for training to obtain the initial review model.

[0049] As an optional embodiment, in the second aspect of the present invention, the initial review model includes a first CNN model, a second CNN model and a Chinese corpus LLM model connected in sequence; the first CNN model is used to identify the document part and correction category corresponding to the manual correction annotation; the second CNN model is used to update the required vocabulary and the prohibited vocabulary according to the document part and the correction category, and to update and annotate the project establishment document; the Chinese corpus LLM model is an LLM model obtained by pre-training with Chinese corpus data, and is used to iteratively train according to the updated and annotated project establishment document, the document review results, the manual correction annotation and the updated required vocabulary and prohibited vocabulary to obtain an optimized review model.

[0050] As an optional embodiment, in the second aspect of the present invention, the first CNN model is used to perform the following steps:

[0051] Inputting the manual correction annotation into a pre-trained text classification CNN neural network to obtain a plurality of correction part texts and corresponding position indication texts in the manual correction annotation; the text classification CNN neural network is trained by a training data set including a plurality of training texts and corresponding correction part text annotations and position indication text annotations;

[0052] For each of the modified partial texts, the modified partial text is input into a trained classifier model to obtain a modified category corresponding to the modified partial text;

[0053] Calculate a first similarity between the position indication text corresponding to the revised part text and each preset document part in the project establishment document;

[0054] Filter out all the preset document parts whose first similarity is higher than a preset first similarity threshold to obtain a plurality of candidate document parts;

[0055] Calculating a second similarity between the corrected partial text and a corresponding partial result of each of the candidate document parts in the document review result;

[0056] Determine the candidate document portion having the second highest similarity as the document portion corresponding to the revised partial text;

[0057] All the correction part texts and the corresponding correction categories and document parts corresponding to the manual correction annotations are input into the second CNN model.

[0058] As an optional embodiment, in the second aspect of the present invention, the second CNN model is used to perform the following steps:

[0059] For each of the modified partial texts, according to the modification category corresponding to the modified partial text, a text word library corresponding to the modified partial text is determined; the text word library is the required word library or the prohibited word library;

[0060] Input the revised text into a trained keyword prediction CNN network to obtain keywords corresponding to the revised text; the keyword prediction CNN network is trained by a plurality of training revised texts and a training data set annotated with corresponding key revised words;

[0061] Calculating a third similarity between the keyword corresponding to the modified partial text and each text word in the text word library;

[0062] Replacing the text word with the highest third similarity with the keyword corresponding to the modified text portion, so as to update the text word library;

[0063] Determine the partial text corresponding to the text portion in the project establishment document according to the document portion corresponding to the revised partial text;

[0064] Calculating a fourth similarity between the keyword corresponding to the modified partial text and each text object in the partial text; the text object is a character, word, word combination, paragraph or text block RUN object;

[0065] The revised part of the text is annotated to the text object with the fourth highest similarity to achieve update annotation of the project establishment document.

[0066] As an optional embodiment, in the second aspect of the present invention, the system is further configured to perform the following steps:

[0067] After the optimal optimization review model is obtained through iterative training, the formal project establishment document to be reviewed sent by the client is obtained;

[0068] Inputting the formal project establishment document into the optimization review model to obtain the output document formal review result;

[0069] The formal review results of the document are sent to the client for display.

[0070] The third aspect of the present invention discloses another project document data processing system for intelligent review, the system comprising:

[0071] A memory storing executable program code;

[0072] a processor coupled to the memory;

[0073] The processor calls the executable program code stored in the memory to execute part or all of the steps in the project document data processing method for intelligent review disclosed in the first aspect of the present invention.

[0074] The fourth aspect of the present invention discloses a computer storage medium, which stores computer instructions. When the computer instructions are called, they are used to execute some or all of the steps in the project document data processing method for intelligent review disclosed in the first aspect of the present invention.

[0075] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0076] The present invention can train an initial review model according to the LLM algorithm and a pre-training data set to achieve a preliminary review of documents, and then iteratively train the initial review model based on the project establishment document, the document review results and the corresponding manual correction annotations to obtain an optimized review model, thereby achieving efficient and automated intelligent review of project establishment documents, improving the effect and accuracy of the review, reducing errors and lowering the review cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0078] Figure 1 It is a flowchart of a project document data processing method for intelligent review disclosed in an embodiment of the present invention.

[0079] Figure 2 It is a structural schematic diagram of a project document data processing system for intelligent review disclosed in an embodiment of the present invention.

[0080] Figure 3 It is a structural schematic diagram of another project document data processing system for intelligent review disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0081] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0082] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, device, product or equipment that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or equipment.

[0083] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0084] The present invention discloses a project document data processing method and system for intelligent review, which can obtain an initial review model according to the LLM algorithm and pre-training data set training to achieve preliminary review of the document, and then iteratively train the initial review model based on the project establishment document, the document review results and the corresponding manual correction annotations to obtain an optimized review model, thereby achieving efficient and automated intelligent review of the project establishment document, improving the effect and accuracy of the review, reducing errors and reducing the review cost. The following are detailed descriptions.

[0085] Embodiment 1

[0086] See also Figure 1 , Figure 1This is a flowchart of a project document data processing method for intelligent review disclosed in an embodiment of the present invention. Figure 1 The project document data processing method for intelligent review described above can be applied to a data processing system / data processing device / data processing server (wherein the server includes a local processing server or a cloud processing server). Figure 1 As shown, the project document data processing method for intelligent review may include the following operations:

[0087] 101. The initial review model is trained based on the LLM algorithm and the pre-training dataset.

[0088] 102. Obtain the project establishment document to be reviewed, and input the project establishment document into the initial review model to obtain the output document review result.

[0089] 103. Send the document review result to the manual review end, and receive the manual correction annotation of the document review result sent by the manual review end.

[0090] 104. Input the project establishment document, document review results and manual correction annotations into the initial review model for iterative training to obtain an optimized review model.

[0091] Specifically, the LLM (Large Language Model) model refers to a type of artificial intelligence model trained on large-scale corpus, which has strong natural language understanding and generation capabilities. The main features of the LLM model include: large model scale, strong learning ability, strong generalization ability and multi-task learning. Among them, the available open source LLM models include: MiLM-6B (Xiaomi AI large model), Code Llama (AI code generation large model based on Llama 2), etc.

[0092] Specifically, in an implementation scenario in which the present invention can be applied, the project establishment ICT review refers to the project party submitting the project establishment document (usually a word document) to the ICT evaluation center, which reviews the document content in accordance with the ICT requirements, points out the missing or inappropriate content, and provides rectification suggestions for the document. Currently, it is mainly manual review.

[0093] Among them, one review requirement can be briefly described as: the document content should include a summary of the current situation (such as the infrastructure, operating system, database, middleware used, etc.), infrastructure requirements, technology and product requirements, feasibility analysis of information and communication technology, technology implementation route design, etc., and should not contain content that does not conform to the information and communication technology route (such as: the system uses foreign middleware, database, etc.).

[0094] Specifically, the existing project document review rules can be briefly described as follows:

[0095] 1. Review whether the project establishment document describes the following contents. If not, provide relevant supplementary comments. The following contents must be included:

[0096] Summary of current situation: Describe the infrastructure, operating system, database, middleware, general software, and professional software brands used (must comply with the information technology route).

[0097] Infrastructure requirements: Whether the public infrastructure planned to be used comes from the provincial government cloud, and whether other infrastructure complies with the information and communication technology route.

[0098] Technology and product requirements: Whether the technology and software and hardware product application requirements for trusted computing are proposed.

[0099] Original technical implementation route: Whether the built information system has been written in accordance with the requirements of the original technical route and application scenarios.

[0100] Feasibility analysis of ICT: The platform (system) technology route and key infrastructure adopted by this project are ICT technology routes and software and hardware products.

[0101] Technical implementation route design: the adopted technical route, application scenarios, common components, development environment and security protection measures. If high availability and high concurrency are required, please explain the implementation method.

[0102] 2. Review whether the project establishment document describes any prohibited content. If so, indicate whether it should be modified or deleted:

[0103] The diagrams, text descriptions, etc. contain expressions that do not conform to the technology route of the information and communication technology innovation.

[0104] 3. Review the project documents to see if they contain typos, incorrect sentences or ambiguous meanings.

[0105] It can be seen that the above-mentioned embodiment of the invention can obtain an initial review model based on the LLM algorithm and the pre-training data set to achieve a preliminary review of the document, and then iteratively train the initial review model based on the project establishment document, the document review results and the corresponding manual correction annotations to obtain an optimized review model, thereby achieving efficient and automated intelligent review of the project establishment document, improving the effectiveness and accuracy of the review, reducing errors, and reducing the cost of the review.

[0106] As an optional embodiment, in the above steps, the pre-training data set includes a plurality of training project establishment documents and corresponding document review annotations.

[0107] It can be seen that through the above-mentioned optional embodiments, the content of the pre-training data set is limited, and an initial review model capable of performing preliminary document review tasks can be fully pre-trained, thereby assisting in realizing efficient and automated intelligent review of project establishment documents, improving the effectiveness and accuracy of the review, reducing errors, and reducing review costs.

[0108] As an optional embodiment, in the above steps, the document review annotations include necessary content annotations, prohibited content annotations and erroneous content annotations; the necessary content annotations include at least one of current situation summary content annotations, infrastructure requirement content annotations, technology and product requirement content annotations, original technology implementation route content annotations, information innovation feasibility analysis content annotations and technology implementation route design content annotations.

[0109] It can be seen that through the above optional embodiments, the content of the document review annotation is limited, which enables the trained review model to conduct a fully automated review of the document, assist in realizing efficient and automated intelligent review of project establishment documents, improve the effectiveness and accuracy of the review, reduce errors, and reduce review costs.

[0110] As an optional embodiment, in the above steps, the initial review model is obtained by training based on the LLM algorithm and the pre-training data set, including:

[0111] Obtain multiple training project establishment documents and corresponding manual review feedback results;

[0112] Based on the preset review criteria, the manual review feedback results are analyzed and summarized into the required and prohibited word libraries;

[0113] According to the required word library and the prohibited word library, rules are constructed based on prompt words, and each training project establishment document is annotated to obtain the corresponding document review annotation;

[0114] The pre-training data set including multiple training project establishment documents and corresponding document review annotations is input into the constructed LLM algorithm model for training to obtain the initial review model.

[0115] Optionally, the required vocabulary may contain words and phrases such as: operating system, database, middleware, application scenarios, security protection measures, etc.; the prohibited vocabulary may contain words and phrases such as: Tomcat, windows, x86, etc.

[0116] Specifically, the prompt words constructed by the prompt word construction rule can be the following examples, which are used for pre-training the LLM algorithm model:

[0117] Please review and provide feedback on the input document according to the following requirements. This document describes the project establishment information of an information system, and is required to include: descriptions of the operating system, database, middleware, application scenarios, security protection measures, etc. If not, please indicate; the information system is required not to contain the use of software and hardware such as Jetty, Tomcat, Mysql, Oracle, Windows, Linux, etc. If so, please indicate; and provide modification suggestions for typos, incorrect sentences and unreasonable places in the document.

[0118] It can be seen that through the above-mentioned optional embodiments, the manual review feedback results can be analyzed based on the preset review standards to summarize the required vocabulary and the banned vocabulary, and each training project establishment document can be annotated based on the prompt word construction rules to obtain the corresponding document review annotations, which are then input into the constructed LLM model for training to obtain the initial review model, which is subsequently used for preliminary document review and iterative training, to assist in the realization of efficient and automated intelligent review of project establishment documents, improve the review effect and accuracy, reduce errors, and reduce review costs.

[0119] As an optional embodiment, in the above steps, the initial review model includes a first CNN model, a second CNN model and a Chinese corpus LLM model connected in sequence; the first CNN model is used to identify the document part and correction category corresponding to the manual correction annotation; the second CNN model is used to update the required vocabulary and the prohibited vocabulary according to the document part and the correction category, and to update the annotation of the project establishment document; the Chinese corpus LLM model is an LLM model obtained by pre-training with Chinese corpus data, and is used to iteratively train according to the updated and annotated project establishment documents, document review results, manual correction annotations, and the updated required vocabulary and prohibited vocabulary to obtain an optimized review model.

[0120] Specifically, some examples of manual correction annotations are given below:

[0121] This is a new system plan. It is recommended to consider XXX plan, XXX evaluation, and XXX construction.

[0122] The document's table of contents page number is wrong and needs to be regenerated.

[0123] Chapter 2.1.5 recommends adding relevant plans for XXX.

[0124] Appendix 1 recommends that XXX adopt domestic alternative products instead.

[0125] It can be seen that through the above-mentioned optional embodiments, the overall training architecture of the initial review model is limited, which can respectively realize the analysis of manually corrected annotations and the updating of the vocabulary and documents, and iterative training is performed based on the updated results to obtain a better optimized review model, so as to realize efficient and automated intelligent review of project establishment documents, improve the effect and accuracy of the review, reduce errors, and reduce review costs.

[0126] As an optional embodiment, in the above steps, the first CNN model is used to perform the following steps:

[0127] Inputting the manual correction annotations into a pre-trained text classification CNN neural network to obtain a plurality of correction part texts and corresponding position indication texts in the manual correction annotations; optionally, the text classification CNN neural network is trained by a training data set including a plurality of training texts and corresponding correction part text annotations and position indication text annotations;

[0128] For each revised text portion, the revised text portion is input into a trained classifier model to obtain a revised category corresponding to the revised text portion;

[0129] Calculate the first similarity between the position indication text corresponding to the revised part text and each preset document part in the project establishment document;

[0130] Screening out all preset document parts whose first similarity is higher than a preset first similarity threshold to obtain a plurality of candidate document parts;

[0131] Calculating a second similarity between the corrected partial text and a partial result corresponding to each candidate document portion in the document review result;

[0132] Determine the candidate document portion with the second highest similarity as the document portion corresponding to the revised partial text;

[0133] All the correction part texts corresponding to the manual correction annotations and the corresponding correction categories and document parts are input into the second CNN model.

[0134] It can be seen that through the above optional embodiments, the pre-trained text classification CNN neural network can be used to recognize the correction part text and the corresponding position indication text in the manual correction annotation, and based on this, the accurate document part and correction category can be predicted, thereby assisting in the realization of efficient and automated intelligent review of project establishment documents, improving the review effect and accuracy, reducing errors, and reducing review costs.

[0135] As an optional embodiment, in the above steps, the second CNN model is used to perform the following steps:

[0136] For each revised text portion, according to the revision category corresponding to the revised text portion, determine the text word library corresponding to the revised text portion; optionally, the text word library is a mandatory word library or a prohibited word library;

[0137] Input the corrected text into a trained keyword prediction CNN network to obtain keywords corresponding to the corrected text; optionally, the keyword prediction CNN network is trained by using a plurality of training corrected texts and a training data set annotated with corresponding key corrected words;

[0138] Calculating a third similarity between the keyword corresponding to the revised partial text and each text word in the text word library;

[0139] Replacing the text word with the third highest similarity with the keyword corresponding to the revised text portion to update the text word library;

[0140] According to the document portion corresponding to the revised partial text, determine the partial text corresponding to the text portion in the project establishment document;

[0141] Calculating a fourth similarity between the keyword corresponding to the modified partial text and each text object in the partial text; optionally, the text object is a character, word, word combination, paragraph or text block RUN object;

[0142] The revised part of the text is annotated to the text object with the fourth highest similarity, so as to realize the update annotation of the project establishment document.

[0143] It can be seen that through the above optional embodiments, it is possible to accurately update the vocabulary according to the document parts and correction categories and update the project establishment documents through the trained keyword prediction CNN network and similarity algorithm, thereby assisting in the realization of efficient and automated intelligent review of project establishment documents, improving the effectiveness and accuracy of the review, reducing errors, and reducing review costs.

[0144] As an optional embodiment, in the above steps, the method further includes:

[0145] After iterative training to obtain the best optimization review model, obtain the formal project establishment document to be reviewed sent by the client;

[0146] Input the formal project establishment document into the optimization review model to obtain the output document formal review results;

[0147] Send the formal review results of the document to the client for display.

[0148] It can be seen that through the above-mentioned optional embodiments, the optimized review model obtained by the final training can be used to implement the formal project establishment document review, so as to ultimately achieve efficient and automated intelligent review of the project establishment documents, improve the review effect and accuracy, reduce errors, and reduce review costs.

[0149] Embodiment 2

[0150] See also Figure 2 , Figure 2 This is a schematic diagram of the structure of a project document data processing system for intelligent review disclosed in an embodiment of the present invention. Figure 2 The project document data processing system for intelligent review described above can be applied to a data processing system / data processing device / data processing server (wherein the server includes a local processing server or a cloud processing server). Figure 2 As shown, the project document data processing system for intelligent review may include:

[0151] The pre-training module 201 is used to train an initial review model based on the LLM algorithm and the pre-training data set.

[0152] The review module 202 is used to obtain the project establishment document to be reviewed, and input the project establishment document into the initial review model to obtain the output document review result.

[0153] The correction module 203 is used to send the document review result to the manual review end, and receive the manual correction annotation of the document review result sent by the manual review end.

[0154] The optimization module 204 is used to input the project establishment document, document review results and manual correction annotations into the initial review model for iterative training to obtain an optimized review model.

[0155] It can be seen that the above-mentioned embodiment of the invention can obtain an initial review model based on the LLM algorithm and the pre-training data set to achieve a preliminary review of the document, and then iteratively train the initial review model based on the project establishment document, the document review results and the corresponding manual correction annotations to obtain an optimized review model, thereby achieving efficient and automated intelligent review of the project establishment document, improving the effectiveness and accuracy of the review, reducing errors, and reducing the cost of the review.

[0156] As an optional embodiment, the pre-training data set includes a plurality of training project establishment documents and corresponding document review annotations.

[0157] It can be seen that through the above-mentioned optional embodiments, the content of the pre-training data set is limited, and an initial review model capable of performing preliminary document review tasks can be fully pre-trained, thereby assisting in realizing efficient and automated intelligent review of project establishment documents, improving the effectiveness and accuracy of the review, reducing errors, and reducing review costs.

[0158] As an optional embodiment, the document review annotations include necessary content annotations, prohibited content annotations, and erroneous content annotations; necessary content annotations include at least one of current situation summary content annotations, infrastructure requirement content annotations, technology and product requirement content annotations, original technology implementation route content annotations, information innovation feasibility analysis content annotations, and technology implementation route design content annotations.

[0159] It can be seen that through the above optional embodiments, the content of the document review annotation is limited, which enables the trained review model to conduct a fully automated review of the document, assist in realizing efficient and automated intelligent review of project establishment documents, improve the effectiveness and accuracy of the review, reduce errors, and reduce review costs.

[0160] As an optional embodiment, the specific method of obtaining the initial review model by the pre-training module based on the LLM algorithm and the pre-training data set includes:

[0161] Obtain multiple training project establishment documents and corresponding manual review feedback results;

[0162] Based on the preset review criteria, the manual review feedback results are analyzed and summarized into the required and prohibited word libraries;

[0163] According to the required word library and the prohibited word library, rules are constructed based on prompt words, and each training project establishment document is annotated to obtain the corresponding document review annotation;

[0164] The pre-training data set including multiple training project establishment documents and corresponding document review annotations is input into the constructed LLM algorithm model for training to obtain the initial review model.

[0165] It can be seen that through the above-mentioned optional embodiments, the manual review feedback results can be analyzed based on the preset review standards to summarize the required vocabulary and the banned vocabulary, and each training project establishment document can be annotated based on the prompt word construction rules to obtain the corresponding document review annotations, which are then input into the constructed LLM model for training to obtain the initial review model, which is subsequently used for preliminary document review and iterative training, to assist in the realization of efficient and automated intelligent review of project establishment documents, improve the review effect and accuracy, reduce errors, and reduce review costs.

[0166] As an optional embodiment, the initial review model includes a first CNN model, a second CNN model and a Chinese corpus LLM model connected in sequence; the first CNN model is used to identify the document part and correction category corresponding to the manual correction annotation; the second CNN model is used to update the required vocabulary and the banned vocabulary according to the document part and the correction category, and to update the annotation of the project establishment document; the Chinese corpus LLM model is an LLM model obtained by pre-training with Chinese corpus data, and is used to iteratively train according to the updated and annotated project establishment documents, document review results, manual correction annotations and the updated required vocabulary and banned vocabulary to obtain an optimized review model.

[0167] It can be seen that through the above-mentioned optional embodiments, the overall training architecture of the initial review model is limited, which can respectively realize the analysis of manually corrected annotations and the updating of the vocabulary and documents, and iterative training is performed based on the updated results to obtain a better optimized review model, so as to realize efficient and automated intelligent review of project establishment documents, improve the effect and accuracy of the review, reduce errors, and reduce review costs.

[0168] As an optional embodiment, the first CNN model is used to perform the following steps:

[0169] Inputting the manual correction annotations into a pre-trained text classification CNN neural network to obtain a plurality of correction part texts and corresponding position indication texts in the manual correction annotations; optionally, the text classification CNN neural network is trained by a training data set including a plurality of training texts and corresponding correction part text annotations and position indication text annotations;

[0170] For each revised text portion, the revised text portion is input into a trained classifier model to obtain a revised category corresponding to the revised text portion;

[0171] Calculate the first similarity between the position indication text corresponding to the revised part text and each preset document part in the project establishment document;

[0172] Screening out all preset document parts whose first similarity is higher than a preset first similarity threshold to obtain a plurality of candidate document parts;

[0173] Calculating a second similarity between the corrected partial text and a partial result corresponding to each candidate document portion in the document review result;

[0174] Determine the candidate document portion with the second highest similarity as the document portion corresponding to the revised partial text;

[0175] All the correction part texts corresponding to the manual correction annotations and the corresponding correction categories and document parts are input into the second CNN model.

[0176] It can be seen that through the above optional embodiments, the pre-trained text classification CNN neural network can be used to recognize the correction part text and the corresponding position indication text in the manual correction annotation, and based on this, the accurate document part and correction category can be predicted, thereby assisting in the realization of efficient and automated intelligent review of project establishment documents, improving the review effect and accuracy, reducing errors, and reducing review costs.

[0177] As an optional embodiment, a second CNN model is used to perform the following steps:

[0178] For each revised text portion, according to the revision category corresponding to the revised text portion, determine the text word library corresponding to the revised text portion; optionally, the text word library is a mandatory word library or a prohibited word library;

[0179] Input the corrected text into a trained keyword prediction CNN network to obtain keywords corresponding to the corrected text; optionally, the keyword prediction CNN network is trained by using a plurality of training corrected texts and a training data set annotated with corresponding key corrected words;

[0180] Calculating a third similarity between the keyword corresponding to the revised partial text and each text word in the text word library;

[0181] Replacing the text word with the third highest similarity with the keyword corresponding to the revised text portion to update the text word library;

[0182] According to the document portion corresponding to the revised partial text, determine the partial text corresponding to the text portion in the project establishment document;

[0183] Calculating a fourth similarity between the keyword corresponding to the modified partial text and each text object in the partial text; optionally, the text object is a character, word, word combination, paragraph or text block RUN object;

[0184] The revised part of the text is annotated to the text object with the fourth highest similarity, so as to realize the update annotation of the project establishment document.

[0185] It can be seen that through the above optional embodiments, it is possible to accurately update the vocabulary according to the document parts and correction categories and update the project establishment documents through the trained keyword prediction CNN network and similarity algorithm, thereby assisting in the realization of efficient and automated intelligent review of project establishment documents, improving the effectiveness and accuracy of the review, reducing errors, and reducing review costs.

[0186] As an optional embodiment, the system is further configured to perform the following steps:

[0187] After iterative training to obtain the best optimization review model, obtain the formal project establishment document to be reviewed sent by the client;

[0188] Input the formal project establishment document into the optimization review model to obtain the output document formal review results;

[0189] Send the formal review results of the document to the client for display.

[0190] It can be seen that through the above-mentioned optional embodiments, the optimized review model obtained by the final training can be used to implement the formal project establishment document review, so as to ultimately achieve efficient and automated intelligent review of the project establishment documents, improve the review effect and accuracy, reduce errors, and reduce review costs.

[0191] Embodiment 3

[0192] See also Figure 3 , Figure 3 It is another project document data processing system for intelligent review disclosed in an embodiment of the present invention. Figure 3 The project document data processing system for intelligent review described is applied to a data processing system / data processing device / data processing server (wherein the server includes a local processing server or a cloud processing server). Figure 3 As shown, the project document data processing system for intelligent review may include:

[0193] A memory 301 storing executable program codes;

[0194] a processor 302 coupled to the memory 301;

[0195] The processor 302 calls the executable program code stored in the memory 301 to execute the steps of the project document data processing method for intelligent review described in the first embodiment.

[0196] Embodiment 4

[0197] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program for electronic data exchange, wherein the computer program enables a computer to execute the steps of the project document data processing method for intelligent review described in the first embodiment.

[0198] Embodiment 5

[0199] An embodiment of the present invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to enable a computer to execute the steps of the project document data processing method for intelligent review described in the first embodiment.

[0200] The above describes specific embodiments of the present specification, and other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily have to be performed in the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0201] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0202] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0203] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification may be in the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the embodiments of this specification may be in the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0204] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0205] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0206] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0207] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0208] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0209] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0210] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0211] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0212] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0213] Finally, it should be noted that the project document data processing method and system for intelligent review disclosed in the embodiment of the present invention discloses only the preferred embodiments of the present invention, which are only used to illustrate the technical solution of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A project document data processing method for intelligent review, characterized in that: The method comprises: The initial review model is trained based on the LLM algorithm and the pre-training dataset, including: Obtain multiple training project establishment documents and corresponding manual review feedback results; Based on the preset review criteria, the manual review feedback results are analyzed and summarized into a must-use word library and a prohibited word library; According to the required word library and the prohibited word library, and based on the prompt word construction rules, each training project establishment document is annotated to obtain the corresponding document review annotation; Input a pre-training data set including multiple training project establishment documents and corresponding document review annotations into the constructed LLM algorithm model for training to obtain an initial review model; Obtaining a project establishment document to be reviewed, and inputting the project establishment document into the initial review model to obtain an output document review result; Sending the document review result to a manual review end, and receiving manual correction annotations on the document review result sent by the manual review end; The project establishment document, the document review results and the manual correction annotations are input into the initial review model for iterative training to obtain an optimized review model; the initial review model includes a first CNN model, a second CNN model and a Chinese corpus LLM model connected in sequence; the first CNN model is used to identify the document part and the correction category corresponding to the manual correction annotation; the second CNN model is used to update the required vocabulary and the prohibited vocabulary according to the document part and the correction category, and to update the annotation of the project establishment document; the Chinese corpus LLM model is an LLM model obtained by pre-training with Chinese corpus data, and is used to iteratively train the updated and annotated project establishment document, the document review results, the manual correction annotations and the updated required vocabulary and prohibited vocabulary to obtain an optimized review model.

2. The project document data processing method for intelligent review according to claim 1 is characterized in that: The document review annotations include necessary content annotations, prohibited content annotations and erroneous content annotations; the necessary content annotations include at least one of current situation summary content annotations, infrastructure requirement content annotations, technology and product requirement content annotations, original technology implementation route content annotations, information innovation feasibility analysis content annotations and technology implementation route design content annotations.

3. The project document data processing method for intelligent review according to claim 1, characterized in that: The first CNN model is used to perform the following steps: Inputting the manual correction annotation into a pre-trained text classification CNN neural network to obtain a plurality of correction part texts and corresponding position indication texts in the manual correction annotation; the text classification CNN neural network is trained by a training data set including a plurality of training texts and corresponding correction part text annotations and position indication text annotations; For each of the modified partial texts, the modified partial text is input into a trained classifier model to obtain a modified category corresponding to the modified partial text; Calculate a first similarity between the position indication text corresponding to the revised partial text and each preset document part in the project establishment document; Filter out all the preset document parts whose first similarity is higher than a preset first similarity threshold to obtain a plurality of candidate document parts; Calculating a second similarity between the corrected partial text and a corresponding partial result of each of the candidate document parts in the document review result; Determine the candidate document portion having the second highest similarity as the document portion corresponding to the revised partial text; All the correction part texts and the corresponding correction categories and document parts corresponding to the manual correction annotations are input into the second CNN model.

4. The project document data processing method for intelligent review according to claim 3 is characterized in that: The second CNN model is used to perform the following steps: For each of the modified partial texts, according to the modification category corresponding to the modified partial text, a text word library corresponding to the modified partial text is determined; the text word library is the required word library or the prohibited word library; Input the revised text into a trained keyword prediction CNN network to obtain keywords corresponding to the revised text; the keyword prediction CNN network is trained by a plurality of training revised texts and a training data set annotated with corresponding key revised words; Calculating a third similarity between the keyword corresponding to the modified partial text and each text word in the text word library; Replacing the text word with the highest third similarity with the keyword corresponding to the modified text portion, so as to update the text word library; Determine the partial text corresponding to the text portion in the project establishment document according to the document portion corresponding to the revised partial text; Calculating a fourth similarity between the keyword corresponding to the modified partial text and each text object in the partial text; the text object is a character, word, word combination, paragraph or text block RUN object; The revised part of the text is annotated to the text object with the fourth highest similarity to achieve update annotation of the project establishment document.

5. The project document data processing method for intelligent review according to claim 4 is characterized in that: The method further comprises: After the optimal optimization review model is obtained through iterative training, the formal project establishment document to be reviewed sent by the client is obtained; Inputting the formal project establishment document into the optimization review model to obtain the output document formal review result; The formal review results of the document are sent to the client for display.

6. A project document data processing system for intelligent review, characterized in that: The system comprises: The pre-training module is used to train the initial review model based on the LLM algorithm and the pre-training dataset, including: Obtain multiple training project establishment documents and corresponding manual review feedback results; Based on the preset review criteria, the manual review feedback results are analyzed and summarized into a must-use word library and a prohibited word library; According to the required word library and the prohibited word library, and based on the prompt word construction rules, each training project establishment document is annotated to obtain the corresponding document review annotation; Input a pre-training data set including multiple training project establishment documents and corresponding document review annotations into the constructed LLM algorithm model for training to obtain an initial review model; The review module is used to obtain the project establishment document to be reviewed, and input the project establishment document into the initial review model to obtain the output document review result; A correction module, used to send the document review result to the manual review end, and receive manual correction annotations on the document review result sent by the manual review end; The optimization module is used to input the project establishment document, the document review results and the manual correction annotations into the initial review model for iterative training to obtain an optimized review model; the initial review model includes a first CNN model, a second CNN model and a Chinese corpus LLM model connected in sequence; the first CNN model is used to identify the document part and the correction category corresponding to the manual correction annotation; the second CNN model is used to update the required vocabulary and the prohibited vocabulary according to the document part and the correction category, and to update the project establishment document; the Chinese corpus LLM model is an LLM model obtained by pre-training with Chinese corpus data, and is used to iteratively train the updated and annotated project establishment document, the document review results, the manual correction annotations and the updated required vocabulary and prohibited vocabulary to obtain an optimized review model.

7. A project document data processing system for intelligent review, characterized in that: The system comprises: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the project document data processing method for intelligent review as described in any one of claims 1-5.

Images