Method and system for assisting text writing

Abstract

The invention discloses a method for assisting text writing. The method comprises the following steps: (i) providing a text and a first comment and a second comment related to the text; (ii) defining a plurality of portions of the text using a neural network; (iii) associating each of the first comment and the second comment with one or more respective portions of the text to which the comment relates through a neural network; for the first comment and the second comment, one or more corresponding portions of the text are the same or different; and (iv) for each portion of the text, determining, via a neural network, whether the first comment contradicts with the second comment, and outputting any contradiction found between the first and second comments.

Description

Technical Field

[0001] The present invention relates to a writing assistance tool, and in particular to a method and system for assisting text writing using a neural network. background

[0002] In recent years, the application of artificial intelligence (AI) in the field of text writing has increased significantly. This includes generating articles based on user needs, correcting grammatical errors in articles, and providing improved writing. In addition, some applications use AI to provide feedback and suggestions to improve the grammar or content of text, thereby improving the quality of writing.

[0003] In professional fields like work and education, it's common practice for texts to undergo review by multiple experienced staff members in addition to AI-powered revisions before finalization. During the revision process, different authors may offer varying and sometimes contradictory comments on the same topic. Currently, there are no systems to identify and resolve these inconsistencies. Therefore, manual review is required to identify and address these inconsistencies, which can be a significant workload.

[0004] Existing writing assistance review systems have not yet implemented AI technology for comparative analysis and still require manual inspection to verify whether the expected modifications have been applied, which is inefficient.

[0005] The present invention aims to alleviate or at least mitigate such problems or drawbacks by providing a new or improved method and system for assisting text writing. Summary of the Invention

[0006] According to a first aspect of the present invention, a method for assisting text writing is provided, comprising the following steps: (i) providing a text and a first comment and a second comment associated with the text; (ii) defining a plurality of parts of the text using a neural network; (iii) associating each of the first comment and the second comment with one or more corresponding parts of the text to which the comment relates through the neural network; for the first comment and the second comment, the one or more corresponding parts of the text are the same or different; and (iv) for each part of the text, determining through the neural network whether the first comment contradicts the second comment, and outputting any contradiction found between the first and second comments.

[0007] In one embodiment, the method further comprises the step of providing an explanation as to why the first and second reviews are considered to be contradictory.

[0008] In one embodiment, the first comment or the second comment includes a suggestion to modify the corresponding portion of the text.

[0009] In one embodiment, the step (iii) of associating the first comment with the corresponding part of the text further includes the following steps: (i) filtering out one or more parts of the text that contain the keywords of the first comment; (ii) obtaining the embedding similarity between the first comment and the one or more parts, and providing a score for each of the one or more parts based on the embedding similarity; and (iii) determining the corresponding part of the first comment in the one or more parts with the highest score through a neural network.

[0010] In one embodiment, text is generated by a neural network based on user-specified input requirements.

[0011] In one embodiment, generating text includes the following steps: (i) receiving a user-specified input requirement; (ii) wrapping the input requirement with a first text template to provide a first input prompt; and (iii) generating text based on the first input prompt by a neural network.

[0012] In one embodiment, the form of the input requirement is selected from any one of text, image, audio and a combination thereof.

[0013] In one embodiment, the first comment or the second comment is generated by a neural network according to comment requirements specified by a user.

[0014] In one embodiment, generating the first comment or the second comment includes the following steps: (i) receiving a comment requirement specified by a user; (ii) wrapping the comment requirement with a second text template to provide a second input prompt; and (iii) generating the first comment or the second comment based on the second input prompt by a neural network.

[0015] In one embodiment, the review requirement is selected from the group consisting of a review target number, a review purpose, and a combination thereof.

[0016] In one embodiment, the method further comprises the step of ignoring at least a portion of either the first review or the second review if the first review and the second review contradict each other, so as to provide a final review set comprising one or more non-contradictory reviews.

[0017] In one embodiment, the method further comprises the step of modifying the text via a neural network to provide a revised text based on the final set of comments.

[0018] In one embodiment, the method includes the following steps: checking whether the revised text has been modified based on the final set of comments, wherein checking the revised text includes the following steps: (i) providing text, a final set of comments, and the revised text; (ii) associating each part of the text with one or more corresponding parts of the revised text that are most similar to the part of the text through a neural network; and (iii) determining whether each part of the revised text has been modified based on the corresponding comments in the final set of comments through the neural network.

[0019] In one embodiment, the method further includes the step of checking whether the revised text has been modified based on the first comment or the second comment, wherein checking the revised text includes the following steps: (i) providing text, a first comment and a second comment related to the text, and the revised text; (ii) associating each part of the text with one or more corresponding parts of the revised text that are most similar to the part of the text through a neural network; and (iii) determining whether each part of the revised text has been modified based on the first comment or the second comment through the neural network.

[0020] In one embodiment, the neural network is a natural language processing (NLP) model or a large language model (LLM).

[0021] According to a second aspect of the present invention, a system for assisting text writing is provided, comprising a revision module adapted to determine whether first and second comments associated with a text contradict each other, wherein the revision module comprises: (i) a first unit adapted to provide a text and first and second comments associated with the text; and (ii) a neural network adapted to define a plurality of parts of the text, associate each of the first and second comments with one or more corresponding parts of the text to which the comment relates, determine for each part of the text whether the first comment contradicts the second comment, and output any contradiction found between the first and second comments; wherein, for the first comment and the second comment, the one or more corresponding parts of the text are the same or different.

[0022] In one embodiment, the system further includes a text generation module communicatively connected to the revision module, wherein the text generation module is adapted to generate text according to input requirements specified by a user, wherein the text generation module includes: (i) a second unit adapted to receive input requirements specified by a user; (ii) a processing unit adapted to wrap the input requirements with a first text template to provide a first input prompt; and (iii) a neural network adapted to generate text according to the first input prompt.

[0023] In one embodiment, the method further comprises a comparison module in communication with the revision module, the comparison module being adapted to check whether the revised text has been modified based on the first comment or the second comment, wherein the comparison module comprises: (i) a third unit adapted to provide the text, the first and second comments associated with the text, and the revised text; and (ii) a neural network adapted to associate each portion of the text with one or more corresponding portions of the revised text that are most similar to the portion of the text, and to determine whether each portion of the revised text has been modified based on the first comment or the second comment.

[0024] In one embodiment, the revision module also includes a comment generation unit, which is suitable for generating a first comment or a second comment according to the comment requirements specified by the user, wherein the comment generation unit includes: (i) a fourth unit, which is suitable for receiving the comment requirements specified by the user; (ii) a processing unit, which is suitable for packaging the comment requirements using a second predetermined text template to provide a second input prompt; and (iii) a neural network, which is suitable for generating the first comment or the second comment based on the second input prompt.

[0025] In one embodiment, the neural network is a natural language processing (NLP) model or a large language model (LLM).

[0026] According to a third aspect of the present invention, a method for article and comment generation and comment verification is provided, comprising: (a) an article generation method that stores domain knowledge and is capable of generating articles according to user instructions; (b) a revision method that generates comments for a given article and verifies the consistency between the comments; and (c) a comparison method that checks whether the modification requirements have been implemented in the new version.

[0027] In one embodiment, the method further comprises the step of generating a text output based on the input using a large language model.

[0028] In one embodiment, the method further comprises the step of converting various types of input into text using an input conversion system, including audio, .doc, and .pdf.

[0029] In one embodiment, the comparison method summarizes the differences between two documents and can receive and validate a list of suggested changes.

[0030] In one embodiment, the method further comprises the step of calculating the connection between the two parts using a scoring system.

[0031] In one embodiment, the method further comprises the step of assigning a vector to each character string using a sentence embedding system.

[0032] Embodiments of the present invention provide a method for using a neural network to review and determine whether there are inconsistencies between comments provided by two or more reviewers, thereby improving the efficiency of existing writing assistance tools and avoiding any tedious work of manually checking for inconsistencies. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The present invention will now be described in more detail, by way of example only, with reference to the accompanying drawings, in which:

[0034] Figure 1 A flowchart showing a method for assisting text writing according to an embodiment of the present invention;

[0035] Figure 2a A schematic diagram showing the steps of determining a contradiction between two reviews according to the method of the present invention;

[0036] Figure 2b Show due to Figure 2a Schematic diagram of the steps involved in invalidating one of the two comments that are shown to be inconsistent;

[0037] Figure 3 A flowchart showing the steps of generating text based on user needs according to the method of the present invention;

[0038] Figure 4 A flow chart showing the steps of providing comments and modification suggestions for a text according to the method of the present invention;

[0039] Figure 5 A flow chart showing the steps of determining whether two sets of reviews contradict each other according to the method of the present invention;

[0040] Figure 6 A flowchart showing the steps of confirming whether a text has been modified according to a modification suggestion according to a method of the present invention;

[0041] Figure 7a A screenshot showing an application utilizing an assistive writing tool according to the present invention, wherein there are no contradictions between the comments;

[0042] Figure 7b yes Figure 7a Screenshots of an app with conflicting reviews;

[0043] Figure 8 A flow chart showing the operation of an embodiment of a system according to the present invention is shown. DETAILED DESCRIPTION

[0044] One embodiment of the present invention provides a method and system for assisting text writing to achieve more efficient writing, thereby reducing the need for manual review of conflicting comments about the text. Figure 1An embodiment of an assisted text writing method 100 according to the present invention is described in detail, in which the implementation of a neural network 160 is essential for integrating and reviewing any existing comments related to a specific part of the text. The initial step 110 of the method 100 includes receiving a text (which can be an article or a paragraph of text) and comments containing revision suggestions related to the text. When the comments are generated by different authors, such as between two reviewers or between a reviewer and an AI engine, differences will arise between the two or more groups of comments. Although sometimes such differences are not important in generating a revised text because they may propose two different revisions that can coexist in the article, sometimes the two opinions are for the same part of the article and contradict each other.

[0045] This leads to the second step 120 of the method 100 of the present invention, which is to define multiple parts of the text through a neural network 160 (such as a large language model (LLM)). In a third step 130, also performed by the neural network 160, each comment provided by the user or the AI engine is associated with the corresponding part of the text related to the comment. Then, in a fourth step 140, the neural network 160 will check whether there are multiple comments related to the same part of the text, and if so, determine whether these comments conflict with each other. If two or more comments related to the same part of the text do conflict with each other, then in a fifth step 150, these comments are output to inform the user of the inconsistency so that the user can review and choose to continue processing one of the comments.

[0046] Figure 2a and 2b The following further illustrates the logic behind step 140 in which the neural network 160 determines a discrepancy between two reviews. In one embodiment of the present invention, a discrepancy between two or more reviews is defined as a non-empty overlap of their related textual portions, and the reviews cannot be processed simultaneously. When the reviews are input into the neural network, metadata is created for each review, including the following components: (i) the review content, (ii) the author, and (iii) the textual portion associated with the review (the relevant portion). Figure 2a and 2bAn example is shown in comments 1 210 and 220 in Figure 2. Both comments 210 and 220 are related to the third and fourth sections of the text. Comment 1 210 suggests "adding a picture to the examples in Chapter 1," while comment 220, written by a different author, makes the opposite suggestion: "delete the examples in Chapter 1 to make it shorter." The neural network determines that there is a contradiction 230 between comments 210 and 220 because they cannot be processed simultaneously in the same text. That is, if an example is to be deleted, an image cannot be added for it. Inconsistency 230 is then output to the user for consideration as to whether to keep comment 210 or 220. In one embodiment, this consideration is performed by a neural network.

[0047] like Figure 2b As shown, a text with a conflict 230 between comments 210 and 220 is passed to a user for review, and the user creates a second version of the text, incorporating data from the previous version. The user can add a new comment C4 to the comments of the previous version, or choose to ignore or not use comment 220 (indicated by a slash symbol 225), retaining only comment 1 210 to avoid conflict 230.

[0048] In one embodiment, if a contradiction 230 occurs between comments, a report is generated and output to the user. The report summarizes all contradictions in the text and provides explanatory notes. The following example illustrates a contradiction / conflict: Conflict 1 Comment ID Comment content Related parts Comment 3 Deletion of unnecessary data in Section 3 10,11 Comments 5 Elaborate on the details of Section 3 11 LLM explanation: There is a conflict between Comment 3 and Comment 5. In the same place in the document, there is a contradiction between "delete unnecessary data" and "elaborate on the details."

[0049] In one embodiment, each user who provides a comment is assigned a designated level. If two or more users provide comments that contradict each other, only the comment provided by the user with the highest level is retained. Other comments from users with lower levels are ignored.

[0050] The present invention also provides a system that provides an integrated service for generating text, providing comments on the text, checking any inconsistencies between the comments, and revising the text based on the comments.

[0051] In one embodiment, the system of the present invention includes three modules: a text generation module, a revision module, and a comparison module, which are interconnected to provide text and improve its writing according to user requirements. The text generation module is used to receive various types of input and generate text according to user requirements. The revision module is used to provide comments on the text including modification suggestions and check the consistency of the comments. The comparison module is used to compare two texts and display their differences. It also accepts a list of suggested modifications as input and checks whether the suggested modifications are applied to the new version of the text.

[0052] Figure 3 The steps of the text generation module are shown. First, input is entered into the module, specifying the user's requirements for the generated text. The module accepts various types of input, including text (including .doc and .pdf), images, and audio. Non-text input is first converted to text and then input into the text generation module. In one embodiment, the text generation module includes a transcription system for converting audio files into text, a transcription system for converting audio into text, a PDF extraction method for converting PDF files into text, and an OCR engine for extracting text from images. Next, the input text is wrapped with a predefined template (machine language) to obtain an input prompt. The input prompt is then input into a natural language processing (NLP) engine to generate an article based on the input. The basic working principles of the NLP engine include: (i) using a tokenizer to convert the input prompt into tokens; (ii) passing the token list into a neural network layer; (iii) generating a list of possible next words and their possibilities through the neural network; and (iv) selecting the most likely word. This process is repeated to generate an article.

[0053] After the text generation module generates an article, the article will be input into the revision module to provide comments and check the consistency between the comments. Alternatively, the user can skip Figure 3 The text generation process shown in the figure directly inputs an article written by yourself to obtain comments on the article.

[0054] Figure 4The workflow of an embodiment of a revision module according to the present invention is shown. The revision module receives text as input, where the text can be an AI-generated article provided by the text generation module or a user-written article provided by the user. In one embodiment, the text can be a single article or a project containing multiple articles. After importing the text, the user can specify the requirements and goals for the comments they want to receive, including the number of comments. The article, along with the specified requirements, will then be organized into an input prompt of a predetermined template, and the input prompt will be sent to the NLP engine. The revision module also uses NLP technology similar to that of the text generation module. Finally, the NLP engine generates comments and modification suggestions.

[0055] Figure 5 The steps involved in aggregating comments and checking for inconsistencies are shown. As previously described, the checking process includes the steps of dividing the article into multiple sections and associating each comment with the corresponding section of the article to which it is associated. In one embodiment, the redaction module uses two methods to determine which section of the document is associated with the comment. The first method is a hierarchical screening method, which includes the following steps: 1. Initial screening (keyword matching): Quickly filter out the parts of the comments that contain clear keywords. 2. Semantic Filtering (Embedding): Use the embedding model to calculate the cosine similarity between the review and the filtered part. The nature of the embedding vector is that the distance between two or more similar texts is small. 3. Final Verification (LLM Verification): Use LLM to verify the relevance of the top k parts with the highest similarity scores. In one embodiment, the user can control the method of selecting the best part, including modifying the value of "k" or setting a threshold.

[0056] The second approach is to apply the LLM directly. For each part of the document, the LLM is asked whether it is relevant to the review. To provide a more comprehensive evaluation, the evaluation can be facilitated by incorporating context (e.g., the previous and next three parts).

[0057] The revision module then checks each section of the document for two or more comments and, if so, whether these comments contradict each other. Contradictory comments are brought to the attention of the user, who then decides which comment to retain and which to discard. Alternatively, the LLM can be trained to select comments. This process repeats until all conflicting comments have been processed and no more conflicting comments remain. The remaining comments and text are then sent to the comparison module to generate the revised document.

[0058] The operation of the comparison module is as follows Figure 6As shown in Figure 2. The comparison module receives as input the article and a set of non-conflicting final reviews. In one embodiment, the set of reviews is provided by the revision module. Alternatively, the reviews can be provided by the user. If the reviews are indeed provided by the user rather than the revision module, the comparison module must first perform the same chunking steps as in the revision module, dividing the article into multiple parts and associating each review in the set with its relevant part. However, if the text and reviews are passed from the revision module, these steps can be omitted because they have already been performed in the revision module.

[0059] The comparison module ensures that all comments, including suggested changes, are considered in the revised article. To do this, it first determines whether relevant sections from the original article are mentioned in the revised article. To do this, it embeds all sections from the revised article and the original article into vectors and then finds the section with the highest similarity to the revised article's embedding, thus obtaining the relevant sections from the original article.

[0060] The comparison module then considers whether the modified section is mentioned in the new article. This is done by determining the embedding similarity between each section in the new document and the modified content, assigning a new score to each section based on the positional distance, and then determining the modified section in the new version based on the embedding similarity and positional distance, resulting in the edited section of the new article that meets the modification requirements. Finally, an NLP engine is used to compare the two sections to check whether the modification requirements are met. Alternatively, the user can manually associate the modified section of the revised article with the relevant section of the original article.

[0061] In one embodiment, for "change" and "delete" modifications, it is assumed that the location of the relevant part in the new version is more likely to be close to the corresponding part in the old version. Both the location of the part and the embedding similarity are taken into account. The comparison module then finds the most relevant part in the new version of the document. In one embodiment, the user can participate in checking whether the correct part is selected or specify the correct part related to the comment. After finding two parts, both the relevant part and the expected modification are imported as input prompts, and then the NLP engine is used to compare whether the given change is applied. The comparison results are presented to the user in a visual manner.

[0062] In one embodiment, according to the present invention, the NLP engine has two roles in the system: as a writer, it generates written articles based on user-specified requirements and comments; and as a reviewer, it reviews the document and provides comments, checks all user comments and verifies their consistency, and checks whether the latest version of the article contains the required changes. This cycle can be repeated multiple times and can be terminated at the user's discretion.

[0063] The following is an application example of the present invention, comprising the following steps: 1. The user enters the requirements for the article to be generated. (For example, enter "Please write an article about product feature <feature>. Detailed information is attached here: <technical details>. My requirements are: <requirements>"). 2. The text generation module generates articles. 3. The revision module reviews the article and generates comments. 4. Generate a list of comments and submit it to the user. 5. After reviewing the document, the user provides additional comments. 6. The revision module analyzes all available comments. If there are any conflicts between the comments, they are submitted to the user for revision. 7. After receiving the list of conflicting comments, the user updates the comments to make them consistent. 8. The revision module analyzes all comments. If it determines that there are no conflicts between the comments, the review engine rewrites the document based on the comments. 9. Compare Module Compare and determine whether all comments have been reflected in the new version of the document. 10. If it determines that certain comments are not reflected in the new version of the document, the comparison module rewrites the document. 11. Generate new revision document. 12. The Compare module compares and confirms that all comments have been reflected in the new version of the document. The number of repetitions can be limited to prevent the process from getting stuck in an endless loop. 13. If all comments have been processed correctly, return to the user.

[0064] After reviewing the article, if the user has a new comment, it returns to step 3 or 5, otherwise the workflow is completed.

[0065] Figure 7a and 7b A screenshot of an actual application using the text assistance system of the present invention is shown. In the application window 700, the comments 710 provided by the user or the revision module are displayed on the left, and the processed text 725 is displayed on the right. For each comment, the system will try to find the relevant part in the text, highlight it, and check whether there are other comments that may conflict with it. For example, Figure 7a The selected comment 711 corresponds to the highlighted portion 725, and it is shown in the consistency check panel 730 that there are no other conflicting comments regarding the highlighted portion 725. However, Figure 7bAnother highlighted section 726 relates to multiple reviews that the system has determined to be conflicting. Therefore, the consistency check panel 730 displays "Review 3 and Review 5 conflict," further explaining why the two reviews conflict. Upon seeing such reviews, the user is obligated to review the reviews and decide which review (in this case, Review 3 or Review 5) should be disregarded.

[0066] Figure 8 The process of generating and optimizing articles by a single user using the system according to an embodiment of the present invention is shown. The process includes the following steps: 1. User-designed task templates 2. The system uses the text generation module to generate articles based on templates 3. The system uses the revision module to generate article comments 4. Users can view and modify comments or add new comments 5. Systematically check for consistency between reviews 6. The system uses the text generation module to rewrite the article based on the comments 7. Repeat steps 3 to 6 until you generate a satisfactory article.

[0067] The same process applies to interactions between authors and reviewers, and the specific steps include: 1. The author designs the task template 2. The system uses the text generation module to generate articles based on templates 3. The author checks the article and makes minor adjustments before sending it to reviewers 4. Reviewers use the system to generate comments through the revision module and can integrate them with their own comments 5. Systematically check for consistency between reviews 6. The author modifies the article based on the comments 7. Reviewers check whether the comments are adopted through the comparison module 8. Repeat steps 3 to 7 to optimize the article.

[0068] In one embodiment, the NLP or LLM engines used in the system of the present invention include llama3.3 and deepseek-r1.

[0069] In one embodiment, when comments are generated, they are tagged with the corresponding location (e.g., section ID), so that the comparison module does not need to check the location. The prompt can be modified so that the NLP engine can output the relevant location. This can improve the accuracy of the comparison module.

[0070] One embodiment of the present invention provides an automated and transparent generation method that tracks different versions and corresponding comments. Unlike traditional LLMs, which can directly generate and optimize articles using LLMs, which may require multiple interactions between humans and AI engines, this system maintains automation and provides users with a visual representation of the generation process.

[0071] The present invention has been presented by way of example only, and those skilled in the art may make various other modifications and / or changes to the described embodiments without departing from the scope of the invention as defined in the appended claims.

Claims

1. A method for assisting text writing, comprising the following steps: (i) providing a text and a first comment and a second comment related to the text; (ii) defining a plurality of portions of the text using a neural network; (iii) associating, by a neural network, each of the first comment and the second comment with one or more corresponding portions of text to which the comment relates; the one or more corresponding portions of text being the same or different for the first comment and the second comment; as well as (iv) for each portion of the text, determining, by the neural network, whether the first comment contradicts the second comment, and outputting any contradiction found between the first comment and the second comment.

2. The method of claim 1, further comprising the step of providing an explanation as to why the first comment and the second comment are considered to be contradictory. The method according to claim 1 , wherein the first comment or the second comment includes a suggestion for modifying the corresponding portion of the text.

4. The method according to claim 1, wherein The step (iii) of associating the first comment with the corresponding portion of the text further comprises the following steps: (i) filtering out one or more parts of the text that contain keywords of the first comment; (ii) obtaining embedding similarities between the first comment and the one or more parts, and providing a score for each of the one or more parts according to the embedding similarities; and (iii) determining, by the neural network, a corresponding part of the first review from one or more parts with the highest scores.

5. The method according to claim 1, wherein The text is generated by the neural network according to input requirements specified by the user.

6. The method according to claim 5, wherein the generating of the text comprises the following steps: (i) receiving the input requirement specified by the user; (ii) wrapping the input requirement with a first text template to provide a first input prompt; as well as (iii) generating the text by the neural network based on the first input prompt.

7. The method according to claim 5, wherein: The form of the input requirement is selected from any one of text, image, audio and a combination thereof.

8. The method according to claim 1, wherein the first comment or the second comment is generated by the neural network according to a comment requirement specified by a user.

9. The method according to claim 8, wherein Generating the first comment or the second comment includes the following steps: (i) receiving the review request specified by the user; (ii) wrapping the comment request with a second text template to provide a second input prompt; and (iii) generating, by the neural network, the first comment or the second comment according to the second input prompt.

10. The method according to claim 9, wherein the review requirement is selected from any one of a target number of reviews, a purpose of reviews, and a combination thereof.

11. The method according to claim 1 , further comprising the steps of: If the first review and the second review contradict each other, at least a portion of either the first review or the second review is ignored to provide a final review set including one or more non-contradictory reviews.

12. The method of claim 11 further comprising the step of modifying the text via the neural network to provide a revised text based on the final set of comments.

13. The method according to claim 11, further comprising the step of checking whether the revised text has been modified based on the final review set, wherein: Checking the revised text includes the following steps: (i) provide the text, the final set of comments and the revised text; (ii) associating, by the neural network, each portion of the text with one or more corresponding portions of the revised text that are most similar to that portion of the text; and (iii) determining, by the neural network, whether each portion of the revised text has been modified based on the corresponding review in the final review set.

14. The method according to claim 1, further comprising the step of checking whether the revised text has been modified based on the first comment or the second comment, wherein: Checking the revised text includes the following steps: (i) providing the text, the first comment and the second comment related to the text, and the revised text; (ii) associating, by the neural network, each portion of the text with one or more corresponding portions of the revised text that are most similar to that portion of the text; (iii) determining, by the neural network, whether each portion of the revised text has been modified based on the first comment or the second comment.

15. The method of claim 1, wherein the neural network is a natural language processing (NLP) model or a large language model (LLM).

16. A system for assisting text writing, comprising a revision module adapted to determine whether a first comment and a second comment related to the text contradict each other, wherein: The revision module includes: (i) a first unit adapted to provide a text and a first comment and a second comment related to the text; (ii) a neural network adapted to define a plurality of portions of the text; associate each of the first comment and the second comment with one or more corresponding portions of the text to which the comment relates, wherein the one or more corresponding portions of the text are the same or different for the first comment and the second comment; for each portion of the text, determine whether the first comment contradicts the second comment; and output any contradiction found between the first comment and the second comment.

17. The system according to claim 16, further comprising a text generation module in communication with the revision module, wherein the text generation module is adapted to generate the text according to input requirements specified by a user, wherein The text generation module includes: (i) a second unit, adapted to receive the input requirement specified by the user; (ii) a processing unit, adapted to wrap the input requirement with a first text template to provide a first input prompt; and (iii) a neural network adapted to generate the text based on the first input prompt.

18. The system according to claim 16, further comprising a comparison module in communication with the revision module, the comparison module being adapted to check whether the revised text is modified based on the first comment or the second comment, wherein: The comparison module includes: (i) a third unit adapted to provide the text, the first comment and the second comment related to the text, and the revised text; (ii) a neural network adapted to associate each portion of the text with one or more corresponding portions of the revised text that are most similar to the portion of the text; and to determine whether each portion of the revised text has been modified based on the first review or the second review.

19. The system according to claim 16, wherein: The revision module further includes a comment generation unit, which is adapted to generate the first comment or the second comment according to a comment requirement specified by a user. The comment generation unit includes: (i) a fourth unit, adapted to receive the comment request specified by the user; (ii) a processing unit, adapted to package the comment request using a second predetermined text template to provide a second input prompt; and (iii) a neural network adapted to generate the first comment or the second comment according to the second input prompt.

20. The system of claim 16, wherein the neural network is a natural language processing (NLP) model or a large language model (LLM).

Images