Intelligent writing system based on large model

Abstract

The invention discloses an intelligent writing system based on a large model, relates to the technical field of intelligent writing, and solves the technical problem that a written text is not personalized enough as writing requirements of a user are not considered in the prior art. According to the method, user writing requirements and historical user portraits of users are collected by designing a user interface, the user writing requirements are input into a large model to obtain a primary text, and the primary text is input into a natural language processing model for semantic check to obtain a standard text; performing knowledge point supplementary optimization on the large model through the knowledge graph, and inputting the standard text into the supplemented and optimized large model to regenerate a corrected text; sending the corrected text to a user through a user interface; updating the historical user portrait according to the feedback information to obtain an updated user portrait, and adjusting the corrected text according to the updated user portrait to obtain a high-quality text; the reply condition of the user continues to be detected, and the high-quality text is further adjusted and updated according to the reply condition of the user.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent writing and relates to AI big model technology, specifically to an intelligent writing system based on a big model. Background Art

[0002] With the rapid development of artificial intelligence technology, the field of natural language processing (NLP) has made significant progress. In particular, large language models based on deep learning (such as OpenAI's GPT series models, Google's BERT, etc.) have demonstrated strong capabilities in tasks such as text generation, translation, summarization, and question answering. These models are trained on large-scale corpora and are able to generate high-quality, context-related texts, making them show great potential in the application of intelligent writing.

[0003] The invention patent with application number CN2019113255533 discloses an intelligent writing system and an intelligent writing method. The invention extracts keywords from the writing data through the artistic conception extraction module, matches the writing artistic conception according to the keywords, extracts the corresponding related phrases and related sentences from the scenario database, and inputs the keywords, related phrases and related sentences into the writing module to write the article; when writing the article, the invention analyzes the article through keywords, and can only analyze the writing artistic conception corresponding to the text, but different users use and understand the text differently, so it is impossible to perform personalized text analysis and writing for the user.

[0004] The present invention provides an intelligent writing system based on a large model to solve the above technical problems. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an intelligent writing system based on a large model, which is used to solve the technical problem in the prior art that the writing text is not personalized enough because the user's own writing requirements are not taken into consideration.

[0006] To achieve the above-mentioned object, the first aspect of the present invention provides an intelligent writing system based on a large model, comprising: a text collection module, a semantic checking module, a text correction module and a real-time interaction module;

[0007] Text collection module: used to collect user writing requirements and historical user portraits;

[0008] Semantic checking module: used to input the user's writing requirements into the language model to generate the initial text; the initial text is semantically checked by the natural language processing model to obtain the standard text;

[0009] Text correction module: used to optimize the language model through the knowledge graph and regenerate the corrected text;

[0010] Real-time interaction module: used to collect user feedback information, supplement the historical user portrait based on the feedback information to obtain an updated user portrait; and personalize the revised text based on the updated user portrait to obtain high-quality text.

[0011] Preferably, the process of performing semantic checking on the initial text by a natural language processing model to obtain a standard text includes:

[0012] Extracting initial text;

[0013] The initial text is input into the natural language processing model to detect and correct semantic errors and spelling errors, and the corrected initial text is refined to remove redundant words and generate a standard text; wherein, redundant words are words without actual meaning, including modal particles, emotional words and other words.

[0014] Preferably, the optimizing the language model through the knowledge graph and regenerating the corrected text includes:

[0015] Extract knowledge graph and standard text;

[0016] Identify and extract professional knowledge points in standard texts, and match the professional knowledge points with knowledge graphs;

[0017] Determine whether the professional knowledge point is the same as the knowledge point in the knowledge graph; if yes, mark the knowledge point in the knowledge graph as a matching knowledge point, and mark the professional knowledge corresponding to the matching knowledge point as matching knowledge; if no, mark the corresponding knowledge point as an unmatched knowledge point;

[0018] The matching knowledge is input into the large model for supplementary optimization, and the standard text is input into the large model to regenerate the revised text.

[0019] Preferably, the step of supplementing the historical user portrait according to the feedback information to obtain an updated user portrait includes:

[0020] Extracting feedback information and historical user portraits; the feedback information includes users' revision opinions and supplementary directions for high-quality texts; historical user portraits include historical personality portraits and historical writing portraits;

[0021] The revision opinions are refined through the natural language processing model to obtain revision feedback, and the revision feedback is added to the historical character portrait to obtain an updated character portrait; the supplementary writing requirements in the supplementary direction are analyzed through the natural language processing model, and the historical writing portrait is updated according to the supplementary writing requirements to obtain an updated writing portrait;

[0022] The updated personality portrait is combined with the updated writing portrait to obtain the updated user portrait.

[0023] Preferably, the step of performing personalized adjustment on the revised text according to the updated user portrait to obtain high-quality text includes:

[0024] Extract revised text and update user profile;

[0025] Match the updated user portrait with the matching knowledge points, and divide the updated user portrait into domain portrait and non-domain portrait according to the matching results;

[0026] Adjust the language style of the revised text according to the domain portrait to generate high-quality text.

[0027] Preferably, the processing of the high-quality text according to the user's reply includes:

[0028] Extracting user response information; the user response information includes user response text and user response status; the user response status includes replied and unreplied;

[0029] Determine whether the user's reply status is replied; if yes, continue to update the user portrait according to the user's reply text, and continue to adjust the high-quality text for replying to the user; if no, wait for the user to reply.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. The present invention collects the user writing requirements and historical user portraits of users by designing a user interface, inputs the user writing requirements into the big model to obtain the initial text, inputs the initial text into the natural language processing model to detect and correct the semantic errors and spelling errors in the initial text, and obtains the standard text, thereby ensuring the coherence and logic of the text content, reducing the workload of manual proofreading, and improving the accuracy of the text; the big model is supplemented and optimized with knowledge points through the knowledge graph, and the standard text is input into the supplemented and optimized big model to regenerate the corrected text. The knowledge graph provides a more comprehensive knowledge background for the big model, making the generated text content more accurate and complete, thereby improving the accuracy of the text. Professionalism and authority; sending the revised text to the user interface and collecting user feedback on the revised text, supplementing and optimizing the historical user portrait through the feedback information to obtain an updated user portrait, and obtaining the domain portrait by matching the updated user portrait with the matching knowledge points. Adjusting the revised text according to the domain portrait to generate high-quality text for replying to users; detecting user replies, and judging whether it is necessary to continue processing the high-quality text according to the user replies, which can improve user satisfaction and continue to serve users until they are satisfied. The user portrait can be continuously optimized according to user feedback to form a closed-loop system of continuous learning and improvement.

[0032] 2. The present invention supplements and optimizes the big model through the knowledge graph, and adds the matching knowledge corresponding to the professional knowledge points in the user's writing requirements in the knowledge graph to the big model, which can better help the big model complete the writing task. The professional terms and concepts contained in the knowledge graph can enhance the professionalism of the text, making it more in line with the standards of specific fields, and can quickly provide the model with the required knowledge points, thereby speeding up the text generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1 The following is a flowchart of an embodiment of the present invention.

[0035] Figure 2 It is a system composition diagram of the present invention.

[0036] Figure 3 This is a complete flow chart for supplementing historical user portraits based on feedback information in one embodiment of the present invention. DETAILED DESCRIPTION

[0037] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] See also Figure 1-Figure 3 ,The first aspect of the present invention provides an intelligent writing system based on a large model, including: a text collection module, a semantic checking module, a text correction module and a real-time interaction module;

[0039] Text collection module: used to collect user writing requirements and historical user portraits.

[0040] Exemplarily, user writing requirements sent by users are collected through the user interface, and the user writing requirements include information such as writing topics, writing styles, and target readers; historical user portraits of users are collected, and the historical user portraits include the user's writing preferences, language styles, and areas of interest in historical usage records.

[0041] In some other embodiments, if the user has no historical usage record, the user is marked as a new user and a blank user portrait is created.

[0042] In this embodiment, the user's writing requirement is to write a Chinese composition related to the phenomenon of robot development, which is required to be narrated in combination with the history of scientific and technological development. The target readers are newspaper readers. The user is an old user, and the historical user portrait includes a historical character portrait that focuses on detailed description and likes narrative texts. The historical writing portrait is the field of science and technology, with a focus on the history of scientific and technological development.

[0043] Semantic checking module: used to input the user's writing requirements into the language model to generate the initial text; the initial text is semantically checked through the natural language processing model to obtain the standard text.

[0044] Exemplarily, an initial large model is constructed based on a deep learning framework, and the user's writing requirements are input into the language large model to generate an initial text. The initial text is input into a natural language processing model (such as a GPT model) to check for typos, semantic errors, and spelling errors in the initial text. The inspection shows that there are two typos, two semantic errors, and no spelling errors in the initial text. After correction, refined sentences in the initial text are extracted and combined into an article abstract. The article abstract is combined with the corrected initial text to generate a standard text.

[0045] Text correction module: used to optimize the language model through the knowledge graph and regenerate the corrected text.

[0046] Exemplarily, professional knowledge points in the standard text are identified and refined, and the professional knowledge points identified in the standard text include robots, technological development, and robot performance testing; it is determined whether the professional knowledge points are the same as the knowledge points in the knowledge graph. In this embodiment, the knowledge graph stores a large amount of professional knowledge in different fields, and can provide relevant information in combination with user writing requirements and professional knowledge points; robot-related information, technological development-related information, and robot performance testing-related information are matched, and robots, technological development, and robot performance testing are marked as matching knowledge points, and robot-related information, technological development-related information, and robot performance testing-related information are marked as matching knowledge.

[0047] The matching knowledge is input into the big model for supplementary optimization, and the standard text is input into the big model to regenerate the revised text. Compared with the standard text, the revised text adds more matching knowledge related to the matching knowledge points, which can better supplement the text required by the user and make the text more credible and scientific based on professional knowledge.

[0048] Real-time interaction module: used to collect user feedback information, supplement the historical user portrait based on the feedback information to obtain an updated user portrait; and personalize the revised text based on the updated user portrait to obtain high-quality text.

[0049] Exemplarily, the revised text is sent to a user interface, and feedback information on the revised text collected through the user interface includes users' opinions that "the original text is too lengthy, and I want to simplify it, add more technical details, make the text more concise and clear, and add more professional vocabulary." The revised feedback is extracted from the feedback information through a natural language processing model, focusing on the need for detailed descriptions and simplifications, and the supplementary writing requirements are to add technical details and professional vocabulary.

[0050] Add the revision feedback to the historical character portrait, and the updated character portrait is a detailed description of the key points but concise and clear description of other unimportant texts; according to the supplementary writing requirements, the historical writing portrait is updated to obtain an updated writing portrait that focuses on the science and technology field and the history of science and technology development, and technical details and professional vocabulary need to be added when describing; the updated character portrait is combined with the updated writing portrait to obtain an updated user portrait.

[0051] In some other embodiments, if the user is a new user, no supplement is needed, and the revision feedback is used as an updated personality portrait, and the supplementary writing requirements are used as an updated writing portrait, and the user portrait is obtained by combining them.

[0052] The updated user portrait is matched with the matching knowledge points. If the matching result is that the key technology fields and technical details related content in the updated user portrait match the matching knowledge points, the key technology fields and technical details related content in the updated user portrait are marked as domain portraits, and the other parts of the updated user portrait are marked as non-domain portraits.

[0053] The domain portrait and the revised text are input into the big model, and the language style, text writing method and involved fields in the revised text are adjusted and updated according to the corresponding domain portrait to obtain high-quality text; the high-quality text is sent to the user interface to reply to the user.

[0054] Continue to detect whether the user continues to reply. In this embodiment, if the user does not continue to reply, the user reply status is not replied, and the user reply is waited.

[0055] In some other embodiments, if the user's reply status is replied, the user portrait is continuously updated according to the newly sent reply text of the user, and after the update is completed, the high-quality text is adjusted to reply to the user, and the user's reply status is continuously detected until the user's reply status remains unreplied.

[0056] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0057] Working principle of the present invention:

[0058] The present invention collects user writing requirements and historical user portraits through a designed user interface, inputs the user writing requirements into a language macro model to generate an initial text, inputs the initial text into a natural language processing model to detect and eliminate semantic errors and spelling errors in the initial text to obtain a standard text; extracts professional knowledge points in the user writing requirements, matches the corresponding matching knowledge in the knowledge graph, optimizes the macro model through matching knowledge and regenerates a revised text, and sends the revised text to the user through the user interface; collects user feedback information, and updates the historical user portrait according to the feedback information to obtain an updated user portrait, and adjusts the revised text according to the updated user portrait to obtain a high-quality text; continues to detect user replies, and if the user replies, continues to update the user portrait and adjust the high-quality text.

[0059] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. Intelligent writing system based on large model, characterized by: It includes: text collection module, semantic checking module, text correction module and real-time interaction module; Text collection module: used to collect user writing requirements and historical user portraits; Semantic checking module: used to input the user's writing requirements into the language model to generate the initial text; The semantics of the initial text is checked by a natural language processing model to obtain a standard text; Text correction module: used to optimize the language model through the knowledge graph and regenerate the corrected text; Real-time interaction module: used to collect user feedback information, supplement the historical user portrait based on the feedback information to obtain an updated user portrait; personalize the revised text based on the updated user portrait to obtain high-quality text; detect user responses and process the high-quality text based on the user responses.

2. The intelligent writing system based on a large model according to claim 1, characterized in that: The process of performing semantic checking on the initial text by a natural language processing model to obtain a standard text includes: Extracting initial text; The initial text is input into the natural language processing model to detect and correct semantic errors and spelling errors, and the corrected initial text is refined to remove redundant words and generate a standard text.

3. The large model-based intelligent writing system according to claim 1, characterized in that: The optimization of the language model through the knowledge graph and regeneration of the revised text include: Extract knowledge graph and standard text; Identify and extract professional knowledge points in standard texts, and match the professional knowledge points with knowledge graphs; Determine whether the professional knowledge point is the same as the knowledge point in the knowledge graph; if yes, mark the knowledge point in the knowledge graph as a matching knowledge point, and mark the professional knowledge corresponding to the matching knowledge point as matching knowledge; if no, mark the corresponding knowledge point as an unmatched knowledge point; The matching knowledge is input into the large model for supplementary optimization, and the standard text is input into the large model to regenerate the revised text.

4. The large model-based intelligent writing system according to claim 1, characterized in that: The method of supplementing the historical user portrait according to the feedback information to obtain an updated user portrait includes: Extracting feedback information and historical user portraits; the feedback information includes users' revision opinions and supplementary directions for high-quality texts; historical user portraits include historical personality portraits and historical writing portraits; The revision opinions are refined through the natural language processing model to obtain revision feedback, and the revision feedback is added to the historical character portrait to obtain an updated character portrait; the supplementary writing requirements in the supplementary direction are analyzed through the natural language processing model, and the historical writing portrait is updated according to the supplementary writing requirements to obtain an updated writing portrait; The updated personality portrait is combined with the updated writing portrait to obtain the updated user portrait.

5. The large model-based intelligent writing system according to claim 1, characterized in that: The step of performing personalized adjustment on the revised text according to the updated user portrait to obtain high-quality text includes: Extract revised text and update user profile; Match the updated user portrait with the matching knowledge points, and divide the updated user portrait into domain portrait and non-domain portrait according to the matching results; Adjust the language style of the revised text according to the domain portrait to generate high-quality text.

6. The intelligent writing system based on a large model according to claim 1, characterized in that: The processing of high-quality text according to the user's response includes: Extracting user response information; the user response information includes user response text and user response status; the user response status includes replied and unreplied; Determine whether the user's reply status is replied; if yes, continue to update the user portrait according to the user's reply text, and continue to adjust the high-quality text for replying to the user; if no, wait for the user to reply.

Images