Document checking method, device and system and server
By adding format tags to documents and combining document slicing and keyword retrieval, the problem of poor user experience in existing AI large-scale model review methods is solved, realizing an efficient and accurate document review process and simplifying user operations.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-18
- Publication Date
- 2026-03-20
AI Technical Summary
Existing document review methods based on large AI models require users to manually locate and correct errors in the document, resulting in a poor user experience, especially when the document contains a lot of content, which increases the difficulty of finding errors.
After adding format tags to the original document and reviewing it using an AI model, the document is restored to the same format as the original document by using the format tags. The document is then compared with the original document to generate a document containing revision marks. The review process is optimized by combining document slicing technology and keyword retrieval.
It simplifies user operations, improves user experience, and eliminates the need for users to manually find and correct errors in documents, thus improving the efficiency and accuracy of proofreading.
Smart Images
Figure CN121706741A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of document review using AI technology, and in particular to a document review method, apparatus, system and server. Background Technology
[0002] During the document writing process, errors can be introduced to varying degrees due to author's typos, misunderstandings of meaning, etc., affecting the user's document reading experience. Proofreading systems are needed to identify these problems and improve document quality.
[0003] With the widespread application of AI (Artificial Intelligence) big data models, such as AIGC (Artificial Intelligence Generated Content) big data models, the ability of generative big data models to understand text, especially Chinese, is constantly improving. This has led to the emergence of methods for reviewing documents based on AI big data models, which are capable of reviewing Chinese documents.
[0004] Existing AI-based large-scale model review methods mainly use technologies such as natural language processing, machine learning, and deep learning to identify grammatical errors, logical errors, and semantic errors in documents, and provide corresponding correction suggestions.
[0005] The final output of AI-based large-scale model-based proofreading typically marks the location of errors in the document being reviewed and provides corresponding correction suggestions through annotations or lists. Users need to manually locate the errors in the document and modify it according to the suggestions. Moreover, the proofreading document output by the model is usually continuous plain text, which increases the difficulty for users to find the error location when the document content is large. It is clear that the existing AI-based large-scale model-based proofreading methods have a poor user experience. Summary of the Invention
[0006] The purpose of this application is to provide a document review method, apparatus, system, and server to improve user experience. The specific technical solution is as follows:
[0007] In a first aspect, embodiments of this application provide a document review method, including:
[0008] Obtain the original document submitted by the user;
[0009] The document content and format are parsed, and corresponding format tags are added to content with different formats.
[0010] The original document with added formatting tags and the pre-set proofreading prompt template are input into the AI model to obtain the AI-proofed document returned by the AI model based on the proofreading prompts; the proofreading prompt template includes at least: prompts for modifying a variety of preset low-level errors, and prompts for modifying a variety of preset basic logic.
[0011] The document content and format of the AI-reviewed document are analyzed, and the document is restored based on different format tags to obtain a reviewed document with the same format as the original document.
[0012] The reviewed document is compared with the original document to generate a reviewed document containing revision marks.
[0013] In one possible implementation, if the number of words in the original document exceeds a preset threshold, before inputting the original document with added formatting tags and the pre-set proofreading prompt template into the AI model to obtain the AI-proofread document returned by the AI model based on the proofreading prompt, the method further includes: logically merging the natural paragraphs of the document according to preset merging rules, splicing out logical paragraphs, and using the logical paragraphs as the smallest slice unit to slice the document content of the original document to obtain at least two sliced documents;
[0014] The process involves inputting the original document with added format tags and a pre-set proofreading prompt template into the AI model to obtain the AI-proofed document returned by the AI model based on the proofreading prompts. This includes: inputting each slice of document and the proofreading prompt template into the AI model in stages to obtain each AI-proofed slice of document returned by the AI model based on the proofreading prompts.
[0015] The process of parsing the content and format of the AI-reviewed document, restoring the AI-reviewed document based on different format tags, and obtaining a reviewed document with the same format as the original document includes: merging the various AI-reviewed slice documents, parsing the content and format of the merged document, restoring the merged document based on different format tags, and obtaining a reviewed document with the same format as the original document.
[0016] In one possible implementation, the step of logically merging natural paragraphs of a document according to a preset merging rule to create logical paragraphs, and then using these logical paragraphs as the smallest slicing unit to slice the original document content to obtain at least two sliced documents, includes: logically merging natural paragraphs of a document whose paragraph styles conform to the preset merging rule to create logical paragraphs; using these logical paragraphs as the smallest slicing unit, counting from the initial logical paragraph, and when the cumulative word count reaches a preset threshold, backtracking to find the most recent logical paragraph as the slicing point, and performing content slicing at this point to obtain sliced documents; if the remaining word count is greater than the preset threshold, then continuing the above slicing operation until slicing is complete.
[0017] In one possible implementation, before inputting the original document with added formatting tags and the pre-set review prompt word template into the AI big model to obtain the AI-reviewed document returned by the AI big model based on the review prompt words, the method further includes: retrieving preset sensitive information and preset erroneous words from the original document; saving the retrieved sensitive information and its original text in the paragraph as a sensitive information review conclusion; and replacing the retrieved erroneous words with the preset revision words corresponding to the erroneous words.
[0018] In one possible implementation, the document review method also includes:
[0019] AI review conclusions are obtained by utilizing the revised content in the reviewed document containing revision marks; the AI review conclusions include at least: the modification points, the content before modification, and the content after modification;
[0020] The reviewed document, the AI review conclusion, the sensitive information review conclusion, and the pre-set quality scoring prompts are input into the AI model to obtain the document quality score returned by the AI model based on the quality scoring prompts; the quality scoring prompts contain the scoring principles.
[0021] In one possible implementation, the document review method also includes:
[0022] A review report is generated and sent to the user who submitted the original document; the review report includes: AI review conclusion, sensitive information review conclusion, document quality score, original document, and final reviewed document.
[0023] In one possible implementation, after comparing the reviewed document with the original document to generate a reviewed document containing revision marks, the method further includes: based on the reviewed document containing revision marks, matching the content before modification with preset allowed words; if a match is found, rejecting the modification and restoring the content before modification.
[0024] Secondly, embodiments of this application also provide a document review device, including:
[0025] The original document retrieval module is used to retrieve the original documents submitted by the user.
[0026] The format tags module is added to parse the document content and format of the original document and add corresponding format tags to content with different formats;
[0027] The model review module is used to input the original document with added format tags and the pre-set review prompt word template into the AI big model, and obtain the AI-reviewed document returned by the AI big model based on the review prompt words; the review prompt word template includes at least: prompt words for modifying a variety of preset low-level errors, and prompt words for modifying a variety of preset basic logics;
[0028] The document restoration module is used to parse the content and format of AI-reviewed documents. Based on different format tags, it restores the AI-reviewed documents to obtain a reviewed document with the same format as the original document.
[0029] The document comparison module is used to compare the reviewed document with the original document and generate a reviewed document containing revision marks.
[0030] Thirdly, embodiments of this application also provide a document review system, including: a document review server;
[0031] The document review server provides a web interface for document review and can interact with large AI models.
[0032] Users log in to the document review webpage interface via their terminal devices;
[0033] The document review server is used to receive original documents submitted by users through the document review web interface; parse the content and format of the original documents, and add corresponding format tags to content with different formats; input the original document with added format tags and a pre-set review prompt word template into the AI model to obtain the AI-reviewed document returned by the AI model based on the review prompt words; the review prompt word template includes at least: prompt words for modifying a variety of preset low-level errors, and prompt words for modifying a variety of preset basic logic; parse the content and format of the AI-reviewed document, restore the AI-reviewed document based on different format tags, and obtain a reviewed document with the same format as the original document; compare the reviewed document with the original document to generate a reviewed document containing revision marks; and display the reviewed document containing revision marks to the user through the document review web interface.
[0034] Fourthly, embodiments of this application also provide a server, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor is used to implement the above-described document review method steps when executing the program stored in the memory.
[0035] Fifthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method steps of the document review method described above.
[0036] Beneficial effects of the embodiments in this application:
[0037] This application provides a document review method, apparatus, system, and server. The document review method adds format tags to the original document before inputting it into an AI model. After obtaining the AI-reviewed document returned by the AI model based on review prompts, the format tags can be used to restore the AI-reviewed document to a format identical to the original document. Then, by comparing the reviewed document with the original document, a reviewed document containing revision marks is generated. Thus, users only need to submit the original document to directly obtain a reviewed document with the same format as the original and containing revision marks. This process eliminates the need for users to manually locate and correct errors in the document, simplifying the user operation, meeting review expectations, and improving the user experience.
[0038] In some embodiments, if the number of words in the original document exceeds a preset threshold, the original document can be sliced before being input into the AI model for review. This way, the AI model reviews only one slice at a time, avoiding the problem of decreased attention due to a large number of words in a single dialogue.
[0039] Of course, implementing any product or method of this application does not necessarily require achieving all of the advantages described above at the same time. Attached Figure Description
[0040] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other embodiments can be obtained based on these drawings.
[0041] Figure 1 An exemplary flowchart of the document review method provided in this application embodiment;
[0042] Figure 2 Another exemplary flowchart of the document review method provided in the embodiments of this application;
[0043] Figure 3 This is an exemplary flowchart of slicing the document content of the original document in an embodiment of this application;
[0044] Figure 4 An exemplary structural diagram of the document review apparatus provided in this application embodiment;
[0045] Figure 5 This is a schematic diagram of the structure of the document review system provided in the embodiments of this application;
[0046] Figure 6 A schematic diagram of an example process for reviewing documents provided in this application's embodiments;
[0047] Figure 7 A schematic diagram of the first state of the web interface of the document review system provided in this application embodiment;
[0048] Figure 8 for Figure 7 The diagram shows the second state of the webpage interface.
[0049] Figure 9 for Figure 7 The diagram shows the third state of the webpage interface.
[0050] Figure 10 This is an exemplary structural diagram of a server provided in an embodiment of this application. Detailed Implementation
[0051] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art based on this application are within the scope of protection of this application.
[0052] This application provides a document review method, apparatus, system, and server to improve user experience. These will be described in detail below.
[0053] See Figure 1 , Figure 1 An exemplary flowchart of the document review method provided in this application embodiment; such as Figure 1 The document review methods shown include:
[0054] Step S101: Obtain the original document submitted by the user;
[0055] Step S102: Parse the document content and format of the original document, and add corresponding format tags to the content of different formats;
[0056] Step S103: Input the original document with added format tags and the pre-set review prompt word template into the AI big model to obtain the AI-reviewed document returned by the AI big model based on the review prompt words; the review prompt word template includes at least: prompt words for modifying a variety of preset low-level errors, and prompt words for modifying a variety of preset basic logics;
[0057] Step S104: Analyze the document content and format of the AI-reviewed document, and restore the AI-reviewed document based on different format tags to obtain a reviewed document with the same format as the original document.
[0058] Step S105: Compare the reviewed document with the original document, and generate a reviewed document containing revision marks based on the comparison results.
[0059] Depend on Figure 1 As can be seen, in the document review method provided in this application, format tags are added to the original document before it is input into the AI model. This way, after obtaining the AI-reviewed document returned by the AI model based on review prompts, the format tags can be used to restore the AI-reviewed document to a format identical to the original document. Then, by comparing the reviewed document with the original document, a reviewed document containing revision marks is generated. In this way, users only need to submit the original document to directly obtain a reviewed document with the same format as the original document and containing revision marks. This process eliminates the need for users to manually find and correct errors in the document, simplifying the user operation, meeting review expectations, and improving the user experience.
[0060] Since large AI models can only recognize document content and not its corresponding format, this embodiment uses a Python-docx parser (the python-docx library) to parse the document content and format. Different formats are tagged with corresponding XML (Extensible Markup Language) tags. During restoration, these XML tags are mapped to the format, ensuring that all content formats in the input original document and the output reviewed document remain consistent.
[0061] In some embodiments, the AI big model can be an AIGC big model or other AI big models with text understanding capabilities.
[0062] The proofreading prompt template used in this embodiment may include the following content:
[0063] The system includes a "Problem Checklist," identification criteria for the presence of problems listed in the "Problem Checklist," rules for correcting problems, output requirements, and modification examples.
[0064] This demonstrates that the AI model reviews the original document based on the review prompt template, and the returned AI-reviewed document is a document modified according to the review prompt template. Thus, by restoring the AI-reviewed document and obtaining a reviewed document with the same format as the original document, it can be compared with the original document to generate a reviewed document containing revision marks.
[0065] In this embodiment, the reviewed document can be in Word format and can be processed using diff software. Specifically, the Microsoft.Office.Interop library is used to diff the original document and the reviewed document, and the corrections in the reviewed document are reflected in the original document through revision styles in Word, making it easier for users to quickly review and process the corrections.
[0066] In some embodiments, if the number of words in the original document exceeds a preset threshold, the original document can be sliced first, and then input into an AI model for review. For details, see [link to relevant documentation]. Figure 2 , Figure 2 Another exemplary flowchart of the document review method provided in this application embodiment; the process includes:
[0067] Step S201: Obtain the original document submitted by the user;
[0068] Step S202: Parse the document content and format of the original document, and add corresponding format tags to the content of different formats;
[0069] Step S203: According to the preset merging rules, logically merge the natural paragraphs of the document to splice out logical paragraphs. Using the logical paragraphs as the smallest slice unit, slice the document content of the original document to obtain at least two slice documents.
[0070] Step S204: Input each sliced document and the review prompt word template into the AI big model in batches to obtain each AI-reviewed sliced document returned by the AI big model based on the review prompt words;
[0071] Step S205: Merge the various AI-reviewed sliced documents, parse the content and format of the merged document, restore the merged document based on different format tags, and obtain a reviewed document with the same format as the original document.
[0072] Step S206: Compare the reviewed document with the original document, and generate a reviewed document containing revision marks based on the comparison results.
[0073] In this embodiment, if the number of words in the original document exceeds a preset threshold, the original document can be sliced first, and then input into the AI model for review. Since there is a word limit to the number of words in a single dialogue (Token) for the AI model, and the AI model experiences a decline in attention (manifested as a significant decrease in response quality), the larger the number of words in a single dialogue, the more pronounced the attention decline. Therefore, in this embodiment, the AI model reviews only one slice of document at a time, avoiding the problem of attention decline caused by an excessively large number of words in a single dialogue.
[0074] In this embodiment, step S220 may specifically include: logically merging natural paragraphs of a document whose paragraph styles conform to the merging rules according to preset merging rules, and splicing them into logical paragraphs; using the logical paragraph as the smallest slice unit, counting from the starting logical paragraph, and when the cumulative number of words reaches a preset threshold, backtracking to find the previous nearest logical paragraph as the slice point, and performing content slicing at this point to obtain the sliced document; if the remaining number of words is greater than the preset threshold, then the above slicing operation is continued until slicing is completed.
[0075] In some embodiments, the preset threshold can be set according to the different upper limits of the number of characters in a single dialogue of different large AI models and actual needs, for example, it can be 3000 characters or 2000 characters, etc. The following uses a preset threshold of 3000 characters as an example to explain the slicing process in detail. In practical applications, the slicing process can be implemented by software. For details, see [link to relevant documentation]. Figure 3 , Figure 3 This is an exemplary flowchart illustrating the process of slicing the content of an original document in an embodiment of this application; the process includes:
[0076] Step S301, document input.
[0077] Step S302: Determine if the number of characters is greater than 3000. If yes, proceed to step S303; otherwise, proceed to step S307 to end the slicing process.
[0078] Step S303: Logically merge all paragraphs in the text according to the preset merging rules.
[0079] Step S304: Calculate the logical paragraph value corresponding to the 3000th character accumulated.
[0080] Step S305: Subtract 1 from the logical paragraph value to obtain the paragraph slicing point, and then perform slicing.
[0081] Step S306: Determine if the number of remaining characters is greater than 3000. If not, proceed to step S307 to end the slicing process; if yes, return to step S304.
[0082] In this embodiment, step S303 involves logically merging the paragraphs of the entire text according to preset merging rules. Specifically, this may include the following two steps:
[0083] 1. Determine whether the current paragraph style and the next paragraph style conform to the splicing relationship in Table 1 below. If they do, splice the two paragraphs into one logical paragraph. The style of the spliced logical paragraph is based on the style of the last paragraph in the splicing unit, and the process continues to merge downwards until the splicing relationship in Table 1 is no longer satisfied.
[0084] Second, start anew from the paragraph that does not satisfy the splicing relationship as the first paragraph, and repeat the splicing operation above. After splicing all the natural paragraphs of the entire document, you will get the logical paragraphs of the entire document.
[0085] Table 1
[0086]
[0087] For example: A document contains three levels of headings, formatted as follows;
[0088] I. First-level headings
[0089] 1.1 The first heading in the second level
[0090] The text content under this title is A
[0091] 1.2 The second heading in the second level of headings
[0092] The text under this title is B
[0093] 1.3 The third heading in the second level
[0094] 1.3.1 The first heading in a third-level heading
[0095] The text under this title is C
[0096] 1.3.2 The second heading in the third-level heading
[0097] The text under this title is D
[0098] 1.3.3 The third heading in the third-level heading
[0099] The text under this title is E
[0100] 1.4 The fourth heading in the second level heading
[0101] The text content under this title is F.
[0102] For this document, based on the current paragraph style, first logically merge text content A in 1.1 with text content B in 1.2. If the merged content is less than 3000 words, then logically merge it with text content C in 1.3.1 of 1.3. If the merged content exceeds 3000 words, then the first logical paragraph is formed from the beginning of the document to the end of text content B. Next, logically merge text content C in 1.3.1 of 1.3 with text content D in 1.3.2. If the merged content is less than 3000 words, then logically merge it with text content E in 1.3.3. If the merged content is still less than 3000 words, then logically merge it with text content F in 1.4. If the merged content exceeds 3000 words, then the second logical paragraph is formed from the beginning of text content C to the end of text content E; text content F is the third logical paragraph.
[0103] In this embodiment, natural paragraphs in the original document can be logically merged, and then sliced based on the logical paragraphs. This can maintain semantic coherence and alleviate the attention decay of large models.
[0104] In some implementations, before inputting the original document with added formatting tags and pre-set review prompt templates into the AI model, special styles in the original document can be preprocessed. If the original document contains special styles such as revisions and comments, these might lead to misprocessing during AI model review, affecting the accuracy of the review. Therefore, the Microsoft.Office.Interop library can be used to clean up special styles such as revisions and comments in the document, mainly by accepting revisions and deleting comments, thus ensuring that special styles in the original document do not affect the review results.
[0105] In some embodiments, before inputting the original document with added format tags and the pre-set proofreading prompt template into the AI big model, the software can also use keyword retrieval to search for preset sensitive information and preset erroneous words, and modify the preset erroneous words.
[0106] For example, it is possible to traverse and retrieve preset sensitive information in the original document, and save the retrieved sensitive information and the original text of the corresponding paragraphs as the sensitive information review conclusions, and add them to the final review report for output to the user. The sensitive information includes general political and legal sensitive information, and may also include sensitive information customized based on application scenarios. This application's embodiments are not limited in scope.
[0107] Specifically, preset incorrect words can be added to a blacklist. The blacklist is then used to match and correct the corresponding incorrect words. Users can customize the blacklist, for example, adding common incorrect words based on user writing guidelines. Each record in the blacklist includes: the word to be searched, the issue category, and suggested corrections. By iterating through the search terms in the blacklist and directly replacing them with suggested corrections, the preset incorrect words can be corrected.
[0108] In some embodiments, after generating a revised document containing revision marks, software retrieval can be used to match the original content with preset allowed words based on the revised document containing revision marks; if a match is found, the modification is rejected and the original content is restored.
[0109] Specifically, preset allowed words can be added to a whitelist. Matching the whitelist with the corresponding allowed words allows for restoration. Users can customize the whitelist by adding allowed words. Each record in the whitelist includes: the word to be searched and the processing method (usually rejecting the modification). By traversing the reviewed documents containing revision marks, the whitelist's search term list is retrieved. If the modified content exactly matches a word in the whitelist, the modification is rejected, and the content reverts to its original state. Because large AI models are generative, erroneous modifications may occur. Using software to retrieve the whitelist helps minimize the risk of erroneous modifications.
[0110] Because AI large-scale models can be unstable, using them alone to review documents typically results in a low problem detection rate, and some common issues are not consistently detected. Therefore, the software uses keyword retrieval to search for and correct preset sensitive information and erroneous words before the AI large-scale model's review. After the AI large-scale model's review, based on the reviewed document containing revision marks, the software matches the original content with preset allowed words. If a match is found, the modification is rejected, and the original content is restored. This approach compensates for the instability inherent in AI large-scale models.
[0111] The review method in this embodiment utilizes an AI large-scale model to focus on reviewing low-level errors and basic logical problems; it uses software search keywords (blacklist) to focus on retrieving and correcting sensitive information and commonly misspelled words; and it uses software search keywords (whitelist) to focus on restoring some invalid modifications introduced by the AI large-scale model. This combined method significantly improves the problem detection rate during review. The test comparison results are shown in Table 2:
[0112] Table 2
[0113]
[0114] In some embodiments, an AI model can be used to score the quality of the reviewed document. Specifically, the AI review conclusion can be obtained first using the revised content in the reviewed document containing revision marks; the AI review conclusion includes at least: the modification points, the content before modification, and the content after modification; then, the reviewed document, the AI review conclusion, the sensitive information review conclusion, and pre-set quality scoring prompts are input into the AI model to obtain a document quality score returned by the AI model based on the quality scoring prompts; the quality scoring prompts include scoring principles. It should be noted that the model used to score the quality of the reviewed document can be different from the model used during review, as long as it can complete the quality scoring.
[0115] In this embodiment, the pre-set quality rating prompts can be set as another prompt template. This prompt template may specifically include:
[0116] The scoring criteria include reference content, scores (out of 100, 90 for excellent, 80 for good, 60 for passing, with the smallest unit being 1 point), known issues, and scoring requirements.
[0117] The scoring requirements may include: 1. When scoring, please consider the known issues, including the number of issues, the density of issues, the impact of issues, and the severity of issues; 2. In addition to the known issues, a comprehensive evaluation of the entire text is also required when scoring; 3. The final scoring result will be displayed in the format of "Overall Score:".
[0118] Specifically, software can be used to summarize the reviewed documents containing revision marks to obtain AI review conclusions, allowing users to directly view the changes through these conclusions. See Table 3, which provides an example of an AI review conclusion.
[0119] Table 3
[0120]
[0121] In some embodiments, a review report can also be generated and sent to the user who submitted the original document. The review report may include: AI review conclusions, sensitive information review conclusions, document quality score, the original document, and the final reviewed document. Specifically, it can be sent to the user via email using the user's registered email address.
[0122] Thus, in some embodiments, a combined review method of "AI review" + "software keyword retrieval" + "software diff comparison and software summary of review conclusions" can be used to review the original document. This three-stage progressive processing can improve the accuracy of review.
[0123] In some embodiments, by combining the review methods, it is possible to support the review of 9 types of sensitive information, 13 types of low-level errors, and 3 types of basic logic problems, as shown in Table 4:
[0124] Table 4
[0125]
[0126] This application also provides a document review device, such as... Figure 4 As shown, it includes:
[0127] The original document retrieval module 401 is used to retrieve the original document submitted by the user.
[0128] The format tag module 402 is added to parse the document content and format of the original document and add corresponding format tags to the content of different formats;
[0129] The model review module 403 is used to input the original document with added format tags and the pre-set review prompt word template into the AI big model, and obtain the AI-reviewed document returned by the AI big model based on the review prompt words; the review prompt word template includes at least: prompt words for modifying a variety of preset low-level errors, and prompt words for modifying a variety of preset basic logics;
[0130] The document restoration module 404 is used to parse the document content and format of the AI-reviewed document, and restore the AI-reviewed document based on different format tags to obtain a reviewed document with the same format as the original document.
[0131] The document comparison module 405 is used to compare the reviewed document with the original document and generate a reviewed document containing revision marks.
[0132] Using the document review apparatus provided in this application, users only need to submit the original document to directly obtain a reviewed document with the same format as the original document and containing revision marks. This process eliminates the need for users to manually locate and correct errors in the document, simplifying the user operation, meeting review expectations, and improving the user experience.
[0133] This application also provides a document review system for implementing the above-described document review method, such as... Figure 5 As shown, the document review system includes: a document review server 200; users access the document review server 200 via terminal device 100 and the network. The document review server 200 provides a document review web interface and can interact with large AI models. Users log in to the document review web interface via terminal device 100.
[0134] The document review server 200 is used to receive original documents submitted by users through a document review web interface; parse the content and format of the original documents, and add corresponding format tags to content with different formats; input the original document with added format tags and a pre-set review prompt word template into the AI model to obtain an AI-reviewed document returned by the AI model based on the review prompt words; the review prompt word template includes at least: prompt words for modifying various preset low-level errors, and prompt words for modifying various preset basic logics; parse the content and format of the AI-reviewed document, restore the AI-reviewed document based on different format tags, and obtain a reviewed document with the same format as the original document; compare the reviewed document with the original document to generate a reviewed document containing revision marks; and display the reviewed document containing revision marks to the user through the document review web interface.
[0135] In this way, users only need to submit the original document to directly obtain a reviewed document with the same format as the original and including revision marks. This process eliminates the need for users to manually locate and correct errors in the document, simplifying the user experience, meeting review expectations, and improving the user experience.
[0136] In some embodiments, the document review system can achieve the aforementioned document slicing function and implement a combined review method of "AI review" + "software keyword retrieval" + "software diff comparison and software summary review conclusions". This will be explained in detail below.
[0137] In some embodiments, the document review server 200 provides document review services to users through document review platform software. For details, see [link to documentation]. Figure 6 , Figure 6 This is a schematic diagram illustrating an example workflow for the document review system provided in this application embodiment. The review process includes:
[0138] Step S601, User Login.
[0139] In this step, the user logs into the document review platform's Web (World Wide Web) terminal via a terminal device.
[0140] In step S602, the user selects the review model via the web interface.
[0141] In this step, the user selects a review model through the document review webpage interface. See also... Figure 7 , Figure 7 This is a first-state schematic diagram of the web interface of the document review system provided in this application embodiment. The web interface mainly includes two columns, left and right. The left column includes a model selection control, a function to select whether to review a document or text, and a function to view review records. The right column is mainly used to display the review results, including quality scores, review conclusions, and correction suggestions. After the review is completed, when the review conclusion is selected by the user, the modification points, the content before modification, and the content after modification will be displayed.
[0142] like Figure 7 As shown, users can select different large AI models using the model selection control. For example, they can choose commonly used AIGC models such as GPT or DEEPSEEK. There are no restrictions here.
[0143] Step S603: Select the docx / txt document and submit it for review.
[0144] In this step, users need to review the document. This can be done by selecting the "Review Document" function in the left column of the webpage. The webpage will then display two selectable icons, such as... Figure 7 The system displays DOCX and TXT icons and prompts users to submit the original document with the message "Click to upload a single document or drag and drop a single document here".
[0145] In some implementations, users can also submit documents in other formats, such as PDF documents. If the document is in another format, the document review platform can first convert the document to docx or txt format before reviewing it.
[0146] After a user submits the original document, the document review platform begins the review process, at which point a relevant prompt window may pop up. For example, see... Figure 8 , Figure 8 for Figure 7 The diagram shows the second state of the webpage interface; in this state, a pop-up window includes reminders for the review time and reminders for sending the review report. Specifically, as shown... Figure 8The system displays the message: "Friendly reminder: The review process may take some time. Once the review is complete, the system will send you a detailed review report via email. You can also check the review progress at any time in the review log."
[0147] Step S604, document format parsing.
[0148] When the document review platform begins reviewing, it first parses the content and format of the original document and adds corresponding format tags to content of different formats.
[0149] Step S605: Preprocess special styles such as annotations / revisions / controls.
[0150] In this step, you can use the Microsoft.Office.Interop library to pre-clean up special styles such as revisions and comments in the document, mainly for accepting revisions, deleting comments and controls, etc.
[0151] Step S606: Determine if the number of words in the original document is less than or equal to 3000. If yes, proceed to step S608; if no, proceed to step S607, slice the document using the slicing tool, and then proceed to step S608.
[0152] In this embodiment, the slicing tool refers to a software program that performs document slicing functionality. For the specific slicing process, please refer to... Figure 3 This will not be repeated here.
[0153] Step S608: The software searches for "sensitive information".
[0154] In this step, software can be used to search for pre-defined sensitive information in the original document using keywords, and the found sensitive information is saved as the sensitive information review conclusion. This is for summarizing the review conclusion in the subsequent step S616.
[0155] Step S609: The software searches for and processes the "blacklist".
[0156] This step involves creating a blacklist containing commonly used incorrect words based on user writing guidelines. Each entry in the blacklist includes: the word to be searched, the issue category, and suggested modifications. By iterating through the blacklist and retrieving the search terms, and directly replacing them with suggested modifications, the preset incorrect words are corrected.
[0157] Step S610: Correct the sliced content based on the large model and preset prompt word template.
[0158] Specifically, this step involves inputting the original document with added formatting tags and a pre-set proofreading prompt template into the AIGC large model, and obtaining the AI-proofread document returned by the AIGC large model based on the proofreading prompts.
[0159] Steps S608-S610 above constitute the document review and processing stage.
[0160] Step S611: Determine if there are any more slices. If yes, return to step S608 to perform document review on the next slice; otherwise, proceed to step S612.
[0161] Step S612: Merge the corrected sliced content and restore the Word style.
[0162] In this embodiment, when the original document is sliced, each step of the document review process described above is performed on each slice to obtain the reviewed sliced document returned by the AIGC large model. This step involves merging the various reviewed sliced documents, parsing the content and format of the merged document, and restoring the merged document based on different format tags to obtain a reviewed document with the same format as the original document.
[0163] Step S613: Perform diff processing on the original document and the corrected document after merging the slices.
[0164] This step involves comparing the reviewed document with the original document to generate a revised document containing revision marks. See the previous section for details. It will not be repeated here.
[0165] Step S614: The software searches for and processes the "whitelist".
[0166] In this step, software can be used to search and match the original content with the pre-defined allowed words in the whitelist based on the reviewed document containing revision marks. If a match is found, the modification is rejected and the original content is restored.
[0167] Step S615: The software summarizes the AI review conclusions, including "modification points", "content before modification", and "content after modification".
[0168] In this step, the software summarizes the revisions in the reviewed document containing revision marks to obtain the AI review conclusion; the AI review conclusion includes at least: the modification points, the content before modification, and the content after modification.
[0169] Step S616: Summarize the review conclusions (including AI review conclusions + sensitive information).
[0170] In this step, the AI review conclusion obtained in step S615 is summarized with the sensitive information review conclusion saved in step S608 to generate a review report.
[0171] Step S617, AI quality scoring.
[0172] In this step, the reviewed document, AI review conclusion, sensitive information review conclusion, and pre-set quality scoring prompts can be input into the AIGC large model to obtain the document quality score returned by the AIGC large model based on the quality scoring prompts; the quality scoring prompts contain the scoring principles. The specifics are as described above and will not be repeated here.
[0173] Step S618: Send the report (including review conclusions, quality score, original document, and corrected document) via email.
[0174] The corrected document in this step is the same as the reviewed document.
[0175] Steps S612-S618 above constitute the review result processing stage.
[0176] Step S619, review complete.
[0177] In this embodiment, after the review is completed, the reviewed document, including revision marks, can be displayed in the right column of the webpage interface. See also Figure 9 , Figure 9 for Figure 7 The diagram illustrates the third state of the webpage interface. In this state, the user selects a review record. The left column displays the document name, word count, creation time, review status, and an export button. After clicking the export button, the reviewed document, including revision marks, will be downloaded and saved to the user's device. In this state, if the user selects correction suggestions, the review issues, including revision marks, will be displayed in the right column.
[0178] In this embodiment, a combined review method of "AI review" + "software keyword retrieval" + "software diff comparison and software summary of review conclusions" is used to review the original document. This three-stage progressive processing can improve the accuracy of review.
[0179] This application also provides a server, such as... Figure 10 As shown, it includes a processor 1001, a communication interface 1002, a memory 1003, and a communication bus 1004, wherein the processor 1001, the communication interface 1002, and the memory 1003 communicate with each other through the communication bus 1004.
[0180] Memory 1003 is used to store computer programs;
[0181] When processor 1001 executes a program stored in memory 1003, it performs the following steps:
[0182] Obtain the original document submitted by the user;
[0183] The document content and format are parsed, and corresponding format tags are added to content with different formats.
[0184] The original document with added formatting tags and the pre-set proofreading prompt template are input into the AI model to obtain the AI-proofed document returned by the AI model based on the proofreading prompts; the proofreading prompt template includes at least: prompts for modifying a variety of preset low-level errors, and prompts for modifying a variety of preset basic logic.
[0185] The document content and format of the AI-reviewed document are analyzed, and the document is restored based on different format tags to obtain a reviewed document with the same format as the original document.
[0186] The reviewed document is compared with the original document to generate a reviewed document containing revision marks.
[0187] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.
[0188] The communication interface is used for communication between the aforementioned electronic devices and other devices.
[0189] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.
[0190] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0191] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor, implements the steps of any of the above document review methods.
[0192] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform any of the document review methods described above.
[0193] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).
[0194] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0195] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.
[0196] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application are included within the scope of protection of this application.
Claims
1. A document review method, characterized in that, include: Obtain the original document submitted by the user; The document content and format are parsed, and corresponding format tags are added to content with different formats. The original document with added formatting tags and the pre-set proofreading prompt template are input into the AI model to obtain the AI-proofed document returned by the AI model based on the proofreading prompt template; The review prompt template includes at least: prompts for modifying various preset low-level errors, and prompts for modifying various preset basic logics; The document content and format of the AI-reviewed document are analyzed, and the document is restored based on different format tags to obtain a reviewed document with the same format as the original document. The reviewed document is compared with the original document to generate a reviewed document containing revision marks.
2. The document review method according to claim 1, characterized in that, If the word count of the original document exceeds a preset threshold, before inputting the original document with added formatting tags and the pre-set proofreading prompt template into the AI model to obtain the AI-proofread document returned by the AI model based on the proofreading prompts, the following steps are also included: According to the preset merging rules, the natural paragraphs of the document are logically merged and spliced into logical paragraphs. The logical paragraphs are used as the smallest slice unit to slice the content of the original document, resulting in at least two slice documents. The process involves inputting the original document with formatted tags and a pre-set proofreading prompt template into the AI model, and obtaining the AI-proofread document returned by the AI model based on the proofreading prompts, including: Each sliced document and the proofreading prompt word template are input into the AI big model in stages to obtain each AI-proofed sliced document returned by the AI big model based on the proofreading prompt words; The process of parsing the content and format of the AI-reviewed document, restoring the document based on different format tags, and obtaining a reviewed document with the same format as the original document includes: The various AI-reviewed sliced documents are merged, and the content and format of the merged document are analyzed. Based on different format tags, the merged document is restored to obtain a reviewed document with the same format as the original document.
3. The document review method according to claim 2, characterized in that, The document's natural paragraphs are logically merged according to preset merging rules to create logical paragraphs. These logical paragraphs are then used as the smallest slicing unit to slice the original document's content, resulting in at least two sliced documents, including: According to the preset merging rules, the document's natural paragraphs whose paragraph styles conform to the merging rules are logically merged and spliced into logical paragraphs. Using logical paragraphs as the smallest slice unit, starting from the initial logical paragraph, when the cumulative word count reaches a preset threshold, backtracking to find the most recent logical paragraph as the slice point, and performing content slicing at this point to obtain the sliced document; If the number of remaining characters exceeds a preset threshold, the above slicing operation will continue until slicing is complete.
4. The document review method according to claim 1, characterized in that, Before inputting the original document with added formatting tags and the pre-set proofreading prompt template into the AI model to obtain the AI-proofread document returned by the AI model based on the proofreading prompts, the following is also included: The original document is searched for pre-defined sensitive information and pre-defined incorrect words; The retrieved sensitive information and the original text of the paragraph containing it are saved as the sensitive information review conclusion; For any erroneous terms found during retrieval, replace them with the corresponding pre-defined revised terms.
5. The document review method according to claim 4, characterized in that, Also includes: AI review conclusions are obtained by utilizing the revisions in the reviewed document that includes revision marks. The AI review conclusion shall include at least: the points to be modified, the content before modification, and the content after modification; The reviewed document, the AI review conclusion, the sensitive information review conclusion, and the pre-set quality scoring prompts are input into the AI model to obtain the document quality score returned by the AI model based on the quality scoring prompts; the quality scoring prompts contain the scoring principles.
6. The document review method according to claim 5, characterized in that, Also includes: Generate a review report and send it to the user who submitted the original document; The review report includes: AI review conclusions, sensitive information review conclusions, document quality score, original document, and final reviewed document.
7. The document review method according to claim 1, characterized in that, After comparing the reviewed document with the original document to generate a reviewed document containing revision marks, the process further includes: Based on the reviewed document containing revision marks, the original content is matched against preset allowed words; if a match is found, the modification is rejected and the original content is restored.
8. A document review device, characterized in that, include: The original document retrieval module is used to retrieve the original documents submitted by the user. The format tags module is added to parse the document content and format of the original document and add corresponding format tags to content with different formats; The model review module is used to input the original document with added format tags and the pre-set review prompt word template into the AI big model, and obtain the AI-reviewed document returned by the AI big model based on the review prompt words; The review prompt template includes at least: prompts for modifying various preset low-level errors, and prompts for modifying various preset basic logics; The document restoration module is used to parse the content and format of AI-reviewed documents. Based on different format tags, it restores the AI-reviewed documents to obtain a reviewed document with the same format as the original document. The document comparison module is used to compare the reviewed document with the original document and generate a reviewed document containing revision marks.
9. A document review system, characterized in that, include: Document review server; The document review server provides a web interface for document review and can interact with large AI models. Users log in to the document review webpage interface via their terminal devices; The document review server is used to receive original documents submitted by users through the document review web interface; parse the document content and format of the original documents, add corresponding format tags to content of different formats; input the original documents with added format tags and the pre-set review prompt word templates into the AI big model, and obtain the AI-reviewed document returned by the AI big model based on the review prompt words; The review prompt template includes at least: prompts for correcting various preset low-level errors and prompts for correcting various preset basic logics; parsing the document content and format of the AI-reviewed document, restoring the AI-reviewed document based on different format tags to obtain a reviewed document with the same format as the original document; comparing the reviewed document with the original document to generate a reviewed document containing revision marks; and displaying the reviewed document containing revision marks to the user through the document review webpage interface.
10. A server, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the method steps of any one of claims 1-7.
11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method described in any one of claims 1-7.