Automatic copywriting generation system based on AIGC model

Abstract

The invention provides an automatic copywriting generation system based on an AIGC model, which relates to the field of copywriting generation and comprises a style selection module, an information input module and a copywriting generation module. The selection and reservation module comprises an identification module, a statement generation module and a confirmation module, the identification module is used for identifying key statements, and the statement generation module generates a plurality of statements which have the same semantics with the identification statements but different expression forms with the identification statements for the identification statements; when the identification statement is selected, a plurality of statements in different expression forms and requirement options are popped up, and the confirmation module selects to use the original identification statement or use one popped-up statement or use the combination of the plurality of statements. According to the automatic copywriting generation system based on the AIGC model, the unsatisfied statement part in the copywriting can be independently adjusted and revised, the use flexibility of the automatic copywriting generation system is improved, and the finalizing efficiency of the copywriting is improved.

Description

Technical Field

[0001] This invention relates to the field of copywriting generation, and more particularly to an automatic copywriting generation system based on an AIGC model. Background Technology

[0002] AIGC (Artificial Intelligence Generated Content) refers to the technology and applications that use artificial intelligence technologies such as deep learning and large language models to automatically generate various types of content, including text, images, audio, video, and code, by learning the patterns of human creation from massive amounts of data.

[0003] As the market demand for efficient content creation continues to increase, users will choose tools such as AIGC to assist in text generation, thereby improving the efficiency of copywriting. Currently, after users generate copy using software tools such as AIGC, if they are not satisfied with the content of the copy, they can choose to regenerate or add key descriptive information before regenerating the entire copy. At this time, there may still be unsatisfactory parts in the newly generated copy. In this case, it may be necessary to extract the satisfactory parts from two or more copy articles and combine them, or users can actively write and modify the unsatisfactory parts of the copy themselves. This revision method is relatively troublesome and still has room for improvement.

[0004] Therefore, it is necessary to provide an AIGC-based automatic copywriting generation system to solve the above-mentioned technical problems. Summary of the Invention

[0005] This invention provides an automatic copywriting generation system based on the AIGC model, which solves the problem that when unsatisfactory sentences are generated by existing copywriting generation systems, it is necessary to add key information and regenerate the entire document, or for the user to actively write and modify the unsatisfactory sentences, which is a relatively cumbersome revision process.

[0006] To address the aforementioned technical problems, the present invention provides an automatic copywriting generation system based on an AIGC model, comprising: A style selection module, which is used to select the type of copy and the writing style; An information input module is used to input key information about the subject of the writing. The copywriting generation module is used to generate copywriting content; The selection module includes an identification module, a statement generation module, and a confirmation module. The identification module is used to identify key statements. For the identified statements, the statement generation module generates multiple statements that have the same semantics as the identified statements but have different expressions. When the identifier statement is selected, multiple statements with different expressions and requirement options pop up. The confirmation module can choose to use the original identifier statement or use one or more statements that pop up. The requirement options are used to select or input the generation requirements for the identifier statement. The identification module includes an automatic identification module and an active identification module. The automatic identification module identifies sentences containing key information when the copy is generated, while the active identification module allows users to actively select sentences to be identified after the copy is generated.

[0007] Preferably, the AIGC-based automatic text generation system further includes a graphics module, which includes an automatic generation module and an insertion module. The automatic generation module generates one or more corresponding images based on the semantics of the identified statement. When multiple images are generated, one or more images are selected and retained. The insertion module is used by staff to actively insert images from the local disk into the text.

[0008] Preferably, the AIGC-based automatic copywriting generation system further includes a reading switching module, which includes an image reading module. When a user reads the copywriting and clicks to view an image in the copywriting, the image pops up and covers the text portion of the copywriting. The image reading module displays the copywriting statement portion corresponding to the content of the image within a preset area of ​​the image. When the copywriting statement portion exceeds the preset area of ​​the image, the copywriting statement portion scrolls through the text.

[0009] Preferably, the reading switching module further includes an information extraction module. When a user clicks to view an image in the text, the information extraction module is used to extract key information of the text statement corresponding to the image content, and the key information is displayed in a preset area of ​​the image.

[0010] Preferably, the reading switching module further includes a playback module, which is used to combine multiple images in the text into a video and play them sequentially, wherein a preset area of ​​each image displays the corresponding text statement.

[0011] Preferably, the construction of the style selection module includes the following steps: S11. Data Receipts: Collection of multi-source textual data and preprocessing of the data; S12. Data Processing: Style labeling and quantification of preprocessed copy data; S13. Model Architecture: The Transformer architecture is selected and a style control module is introduced. The style control module includes: a conditional embedding layer, which concatenates style vectors in the input layer to guide the generation direction; a dynamic adapter, which is a lightweight module inserted between Transformer layers and dynamically weights style features through an attention mechanism; and knowledge graph fusion, which combines domain knowledge graphs to ensure the professionalism of the content. S14. Data Training: Train the text data using the model architecture; S15. Evaluation and Optimization: Evaluate and optimize the trained model in multiple dimensions.

[0012] Preferably, the AIGC-based automatic copywriting generation system further includes a copywriting template management module, which provides various types of preset templates.

[0013] Preferably, the AIGC-based automatic copywriting generation system further includes a user feedback management module, which includes a user information recording module, a classification module, a content recognition module, and a statistics module. The classification module is used to categorize copywriting of the same type and mark the writing style of each piece of copywriting. The user information recording module is used to collect the user's gender and age; The content recognition module is used to obtain user feedback information and identify whether the content of the feedback information is the object of the comment copywriting or the writing level of the comment copywriting. When the writing level of the comment copywriting is identified, the content of the comment is judged to be a positive or negative review, and the number of positive and negative reviews is counted. The statistics module is used to analyze the positive review rate of different copywriting works of the same type under different age groups, genders and writing styles, and to analyze the types of negative reviews.

[0014] Preferably, the AIGC-based automatic copywriting generation system further includes a record generation module, which is used to record historical copywriting information.

[0015] Preferably, the AIGC-based automatic copywriting generation system further includes a multilingual generation module, which includes a dialect generation module and a foreign language generation module. The dialect generation module is used to select a dialect to generate corresponding copywriting, and the foreign language generation module is used to select a foreign language to generate corresponding copywriting.

[0016] Compared with related technologies, the AIGC-based automatic copywriting generation system provided by this invention has the following advantages: This invention provides an automatic copywriting generation system based on an AIGC model. By setting up an option module, after the copywriting generation module generates the copy content, the system automatically marks statements containing key information. When a user is reviewing the copy and is dissatisfied with a marked statement or wants to see a better expression, they can click on the marked statement. This will display multiple different expressions of the same statement—meaning the same but expressed differently. The user can then choose the better expression. If all statements are unsatisfactory, the user can select other styles through the requirements option, generating multiple statements in the corresponding style for selection, or add requirements to regenerate the statement. This allows for individual adjustment and revision of unsatisfactory parts of the copy, improving the flexibility of the automatic copywriting generation system and increasing the efficiency of finalizing the copy. Attached Figure Description

[0017] Figure 1 A flowchart illustrating the steps of the AIGC-based automatic copywriting generation system provided by this invention; Figure 2 A block diagram illustrating the composition of the AIGC-based automatic copywriting generation system provided by this invention; Figure 3 A schematic diagram of the selection module provided by the present invention; Figure 4 A schematic diagram of the reading switching module provided by the present invention; Figure 5 A flowchart illustrating the construction steps of the style selection module provided by this invention; Figure 6 This is a schematic diagram of the user feedback management module provided by the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] This invention provides an automatic copywriting generation system based on the AIGC model.

[0020] Please refer to the following: Figures 1 to 3 In one embodiment of the present invention, the copywriting automatic generation system based on the AIGC model includes: A style selection module, which is used to select the type of copy and the writing style; An information input module is used to input key information about the subject of the writing. The copywriting generation module is used to generate copywriting content; The selection module includes an identification module, a statement generation module, and a confirmation module. The identification module is used to identify key statements. For the identified statements, the statement generation module generates multiple statements that have the same semantics as the identified statements but have different expressions. When the identifier statement is selected, multiple statements with different expressions and requirement options pop up. The confirmation module can choose to use the original identifier statement or use one or more statements that pop up. The requirement options are used to select or input the generation requirements for the identifier statement. The identification module includes an automatic identification module and an active identification module. The automatic identification module identifies sentences containing key information when the copy is generated, while the active identification module allows users to actively select sentences to be identified after the copy is generated.

[0021] By setting up an option module, after the copywriting generation module generates the copy content, the system automatically marks sentences containing key information. When users are reviewing the copy to see if it meets their requirements, and if they are dissatisfied with the marked sentences or want to see if there are better expressions, they can click on the marked sentences. This will bring up multiple different expressions of the same sentence, i.e., the same meaning but different ways of expression. Users can then choose the expression they think is better. If they are not satisfied with all the sentences, they can select other styles through the requirements option, generate multiple sentences of the corresponding style, select one, or add requirements to regenerate the sentence. This allows for individual adjustment and revision of unsatisfactory sentences in the copy, improving the flexibility of the automatic copywriting generation system and increasing the efficiency of finalizing the copy.

[0022] In this system, after marking statements containing key information, users can actively mark the sentences before and after the marked statements according to the context. At this time, the statement generation module generates multiple statements together with the actively marked parts.

[0023] Since the writing style has been selected before the copy is generated, the initial sentences generated by the sentence generation module are sentences of the same style but different expressions. Users can choose different styles in the requirements options. For example, if the initial style is formal, a humorous style can be selected for the marked sentences later.

[0024] When multiple statements are selected to be retained, a conjunction is automatically added between adjacent statements. For example, if one statement is "The water cup on the table replenishes the energy of the employees" and another statement is "The energy cup in the overtime scenario", when they are combined into one sentence, "The water cup on the table replenishes the energy of the employees" is more like "The energy cup in the overtime scenario".

[0025] In this embodiment, the copywriting type may include education, tourism, food, services, technology, product sales, advertising, etc. Writing styles can include creative, formal, humorous, ornate, concise, technological, and emotional; The subjects of writing can include products, places, buildings, works, etc.

[0026] The description of key information includes the input of keywords and the input of key information descriptions. For example, when generating copy for a product, keywords can be high quality and cost-effectiveness, and key information descriptions can include the product name, main data and functions of the product, etc.

[0027] Please see Figure 2 The AIGC-based automatic text generation system also includes a graphics module, which comprises an automatic generation module and an insertion module. The automatic generation module generates one or more corresponding images based on the semantics of the identified statements. When multiple images are generated, one or more images are selected and retained. The insertion module is used by staff to actively insert images from the local disk into the text.

[0028] By setting up a graphics module, graphics can be emoticons, pictures, or GIFs. By inserting pictures into the text, the content of the text can be enriched, and some descriptive statements can be made more concrete. The image can be an image automatically generated by the system based on semantics. Users can choose to directly replace the statement with the image or place the image after the statement. For example, when describing a water cup in a work setting, an ordinary water cup can be described as an "energy cup for overtime work." At this time, an emoticon or GIF can be generated based on the semantics of the following statement to increase the fun of the copy.

[0029] When the copy describes a product or a tourist attraction, images of the corresponding parts of the product can be inserted. Each image can be placed after the description of the corresponding part. For example, in the description of a car product, when describing multiple parts of the car, the corresponding images can be placed after the description. For example, when describing a car door, a picture of the car door can be inserted in a separate paragraph after the description of the car door.

[0030] Similarly, you can insert pictures of multiple locations within the tourist attraction, placing the pictures after the description of each location.

[0031] Please see Figure 4 As a preferred embodiment, the AIGC-based automatic copywriting generation system further includes a reading switching module, which includes an image reading module. When a user reads the copywriting and clicks to view an image in the copywriting, the image pops up and covers the text portion of the copywriting. The image reading module displays the copywriting statement portion corresponding to the content of the image in a preset area of ​​the image. When the copywriting statement portion exceeds the preset area of ​​the image, the copywriting statement portion scrolls.

[0032] When viewers read product or tourist destination descriptions, they often need to use inserted images to aid their understanding. Currently, when viewers click on an image to view it, the image usually zooms in and covers the text for a clearer view. When they need to read the text description, they have to click on the image again to return it to its initial state. This requires switching back and forth between viewing images and text, making reading cumbersome. By setting up the image reading module, clicking on an image will zoom in and display a description of the parts or places in the image in a preset area, thus eliminating the need to switch between images and text descriptions multiple times.

[0033] The preset area can be one side, top, or bottom of the image without obscuring the image content. When the length of the sentence exceeds the range that the preset area can display, the sentence can be scrolled and played.

[0034] Please see Figure 4 As a preferred embodiment, the reading switching module further includes an information extraction module. When a user clicks to view an image in the text, the information extraction module is used to extract key information of the text statement corresponding to the image content, and the key information is displayed in a preset area of ​​the image.

[0035] Since the descriptions of the corresponding images may be quite long, an information extraction module can be set up to extract key information from that section and display it in a preset area of ​​the image. This allows viewers to flip through the images sequentially to understand the basic information of the entire text, and then read the text again for more details.

[0036] For example, in a vehicle description, the image of the rearview mirror might be displayed in a pre-defined area with the message: "Electric rearview mirror, wide viewing angle, clear image quality." Similarly, the image of the car door might be displayed in a pre-defined area with the message: "Power-closing door, frameless." Other parts could be described concisely, allowing users to quickly connect to the core of the description. After understanding the key features, users can then read further to learn more about the specific functional parameters and other details if they are interested.

[0037] Please refer to it again. Figure 4 As an optional embodiment, the reading switching module further includes a playback module, which is used to combine multiple images in the text into a video and play them sequentially, wherein a preset area of ​​each image displays the corresponding text statement.

[0038] By setting up a playback module, when an image in the text is clicked, the images will automatically play in the order they are placed, eliminating the need for manual swiping to switch images. This expands the playback scenarios and is suitable for users who read quickly or whose hands are not convenient for frequent operation.

[0039] Please see Figure 5 In this embodiment, the construction of the style selection module includes the following steps: S11. Data Receipts: Collection of multi-source textual data and preprocessing of the data; S12. Data Processing: Style labeling and quantification of preprocessed copy data; S13. Model Architecture: The Transformer architecture is selected and a style control module is introduced. The style control module includes: a conditional embedding layer, which concatenates style vectors in the input layer to guide the generation direction; a dynamic adapter, which is a lightweight module inserted between Transformer layers and dynamically weights style features through an attention mechanism; and knowledge graph fusion, which combines domain knowledge graphs to ensure the professionalism of the content. S14. Data Training: Train the text data using the model architecture; S15. Evaluation and Optimization: Evaluate and optimize the trained model in multiple dimensions.

[0040] The data collection sources include social media (Xiaohongshu, Weibo), advertising copy, press releases, literary works, and official document template libraries, covering target styles (such as tech style, emotional style, and humorous style).

[0041] Data preprocessing includes removing duplicate text, garbled characters, and irrelevant symbols (such as URLs and emoticons); standardizing punctuation formats (such as converting full-width characters to half-width characters); filtering low-quality data: deleting logically disordered and inconsistently styled samples; standardizing length: limiting text length according to application scenarios (such as controlling e-commerce text to 50-200 words) to avoid excessively long or short texts affecting model learning; and expanding data through back-translation (Chinese-English translation) to enhance generalization. Batch labeling is performed using pre-trained style classifiers such as BERT-like models to extract text style vectors and quantify abstract styles. Backbone network of the model architecture: Use Transformer architecture (such as GPT-4, LLaMA2) or encoder-decoder structure to support long text generation.

[0042] The specific training strategies include: the three-stage training method. Pre-training: Learn the language basics using a general corpus (1 billion+ tokens); Multi-style fine-tuning: Train on labeled datasets and add style classification loss (enhancing style capture); Few-sample adaptation: For new styles or brands, quickly adapt using 5-10 samples through meta-learning (MAML).

[0043] Loss function design: Style consistency loss: calculate the cosine similarity between the generated text and the target style vector; Content preservation loss: use BERT to measure the semantic difference between the original text and the generated text (threshold > 0.85).

[0044] Diversity enhancement: Increase the temperature parameter (Temperature=0.7~1.0) or use Top-p kernel sampling (p=0.9) during the sampling phase; introduce a discriminator in adversarial training (GAN) to evaluate style consistency.

[0045] The multi-dimensional evaluation includes automatic metrics such as BLEU-4 (content accuracy) and Style Accuracy; and manual evaluation metrics such as semantic fluency, creativity, and style fit.

[0046] If the generated copy style is ambiguous, it is necessary to check whether the style vector is decoupled from the semantics, or to increase the number of adversarial training rounds.

[0047] As an optional approach in this embodiment, the AIGC-based automatic copywriting generation system also includes a multi-dimensional weighted popularity algorithm to reflect the latest user preference trends.

[0048] The algorithm constructs a dynamic weighted scoring model by comprehensively analyzing three core fields: styleName (style name), isActive (whether it is enabled), and creationDate (creation date).

[0049] Specifically, the following steps are included: a1: Basic popularity calculation; calculate the time decay factor based on the creationDate (creation date) field, using an exponential decay model, with the formula 1 / (1+ln(current date-creation date)), to ensure that newly created styles receive a higher initial score; at the same time, combine the isActive (whether it is enabled) field for binary judgment, and the enabled state automatically receives a 30% base score bonus; a2: Semantic feature extraction; NLP processing is performed on styleName (style name), and a 512-dimensional semantic vector is extracted using the BERT model. The cosine similarity between styleName and mainstream styles is calculated using a pre-trained copywriting style classifier. A 5% score bonus is awarded for every 10% increase in similarity. A3: User behavior analysis; Establish the association between the creator field and the system user profile, and assign different weights based on the creator's historical activity (including login frequency, number of operations, etc.); the style created by VIP users automatically receives a 15% weight boost, while the style created by ordinary users receives the standard weight; a4: Tag diffusion calculation; parse the multi-tag data of the tags field, construct the tag co-occurrence matrix, and calculate the global influence score of each tag using the PageRank algorithm; the final tag score of the style is the geometric mean of the scores of the tags it contains, ensuring the synergistic effect of multiple tags; a5: Dynamic weighted fusion; after the scores of the above four dimensions are normalized by min-max, an adaptive weight allocation mechanism is adopted; the system will monitor the variance changes of each dimension in real time and automatically adjust the weight allocation ratio to ensure that the scoring system always reflects the latest user preference trends.

[0050] The final score range is controlled between 0 and 100 points, making it easy for business systems to use directly.

[0051] As an optional embodiment, the copywriting automatic generation system based on the AIGC model also includes a style competitiveness index algorithm; This algorithm constructs a differentiated competition evaluation model between styles by deeply mining the content features of description and example text.

[0052] Specifically, the steps include: b1: Text complexity analysis; using the Flesch-Kincaid readability test to score the difficulty of the description (style description), and calculating the information density index in combination with the text length; at the same time, using the TF-IDF algorithm to extract the keyword distribution in exampleText, and constructing an n-gram language model to evaluate the text's innovativeness; b2: Visual feature fusion; When the styleImage field exists, the ResNet50 convolutional neural network is used to extract image feature vectors, and the image-text matching degree is calculated through a pre-trained visual-text cross-modal model; every 10% increase in the matching degree brings a 3% increase to the competitiveness index; b3: Custom style analysis; Considering the special characteristics of the customStyle field, a recursive feature extraction process is designed; First, the style parameters in the JSON structure are parsed, and then the style transfer network is used to map the custom parameters to the standard style space, and the Hausdorff distance between it and the existing style set is calculated as a uniqueness indicator. b4: Competitive Landscape Modeling; Establish a Markov Chain Monte Carlo model to simulate the competitive relationship between different styles during the user selection process; Calculate the steady-state probability of each style being selected through 100,000 Monte Carlo simulations, and convert it into a competitiveness index between 0 and 1. This index will be automatically updated weekly to reflect the dynamic changes in the style library.

[0053] As an optional embodiment, the AIGC-based automatic copywriting generation system also includes a cross-platform compatibility algorithm. This algorithm innovatively combines the system attribute of dataId (unique data identifier) ​​with the actual use scenario to predict the performance adaptability of the style on different platforms.

[0054] Specifically, the steps include: c1: ID feature decoding; decomposing the snowflake algorithm ID of dataId (data unique identifier) ​​into three parts: timestamp, work node, and sequence number; by analyzing the creation of time distribution and work node topology, constructing style regional and departmental distribution heatmaps, and identifying potential platform usage preference patterns; c2: Multi-platform feature engineering; interface with external platform APIs to obtain content style requirements of each platform in real time; use contrastive learning techniques to encode description (style description) and exampleText (example text) into platform-independent feature vectors, and then calculate their KL divergence with the standard style of each platform as the basic fit.

[0055] c3: Dynamic environment modeling; establish a reinforcement learning environment, constructing a three-dimensional state space from style features, platform characteristics, and user feedback; train the agent using the DQN algorithm to predict the optimal strategy for style modification under different platform environments; finally, the fitness score is the value function output of the agent's strategy, ranging from 0 to 5 stars.

[0056] The system retrains the model every 24 hours to ensure the timeliness of predictions.

[0057] As an optional embodiment, the AIGC-based automatic copywriting generation system further includes a copywriting template management module, which provides various types of preset templates.

[0058] The copywriting template management feature aims to provide users with an efficient and flexible foundation for copywriting creation. Users can create, edit, and delete copywriting templates according to different needs. Templates include various industries and scenarios, ensuring users can quickly find suitable copywriting formats.

[0059] Users can customize the structure and content of the template.

[0060] The system offers a variety of preset templates covering multiple areas such as advertising, social media, and product descriptions. Users can modify these templates to save creation time. Similarly, a multi-dimensional weighted scoring algorithm can be used to determine whether a new template can be liked by users; For the copywriting template management module, a semantic sentiment-structural complexity composite algorithm is also provided to reflect the actual value of the template; Specifically, the following steps are included: d1: Semantic sentiment analysis; NLP processing is performed on templateContent, and sentiment analysis models are used to calculate sentiment polarity values, ranging from [-1, 1], which are mapped to the 0-1 score interval; at the same time, the sentiment richness of the text is calculated, and a supplementary score is given by sentiment word density (number of sentiment words / total number of words); d2: Structural complexity calculation; Analyze the text structure of templateContent and calculate indicators such as the number of paragraphs, sentence variation, and cohesive word density; 0.1 points are added for each additional paragraph, sentence variation is scored by counting the number of different sentence types, and 0.05 points are added for each additional sentence type, and 1 point is awarded for cohesive word density in the range of 0.3-0.5.

[0061] d3: File complexity assessment; Analyze the format complexity of templateFile (template file): HTML format gets 1 point, Markdown gets 0.8 points, and plain text gets 0.6 points; d4: Composite calculation; the semantic sentiment score (weight 0.4), structural complexity score (weight 0.4), and document complexity score (weight 0.2) are weighted and summed, and then the result is geometrically averaged with the result of the multi-dimensional weighted scoring algorithm to finally obtain a standardized template comprehensive score of 0-10.

[0062] This algorithm fully considers the content quality and formal complexity of the template, and can comprehensively reflect the actual value of the template.

[0063] Please see Figure 6As a preferred embodiment, the copywriting automatic generation system based on the AIGC model further includes a user feedback management module. The user feedback management module includes a user information recording module, a classification module, a content recognition module, and a statistics module. The classification module is used to summarize copywriting of the same type and mark the writing style of each copywriting. The user information recording module is used to collect the user's gender and age; The content recognition module is used to obtain user feedback information and identify whether the content of the feedback information is the object of the comment copywriting or the writing level of the comment copywriting. When the writing level of the comment copywriting is identified, the content of the comment is judged to be a positive or negative review, and the number of positive and negative reviews is counted. The statistics module is used to analyze the positive review rate of different copywriting works of the same type under different age groups, genders and writing styles, and to analyze the types of negative reviews.

[0064] By collecting and analyzing comments on the copywriting quality in the comment section, we can understand the audience's evaluation of the copy after it is published. Based on the comments, we can further optimize and adjust the system, and we can also statistically analyze the approval rate of different styles of copywriting. We can see which styles of copywriting are preferred by different age groups and genders, and thus use more popular styles to write copy for the target audience of the product.

[0065] Among them, based on the negative comments, such as unclear sentence logic, typos, and punctuation errors, the article can be optimized and adjusted according to the comments.

[0066] The comment section includes content on the writing level of comments, which may be either solely on the writing level of the comments themselves or contain elements of comment writing.

[0067] The user feedback management module also includes feedback from users who use the system for copywriting.

[0068] As a preferred embodiment, the AIGC-based automatic copywriting generation system further includes a record generation module, which is used to record historical copywriting information.

[0069] The generation history module provides users with a comprehensive record of copywriting generation. Users can view previously generated copywriting content, and each piece of copywriting includes key information such as generation time, keywords, and copywriting style. Users can quickly retrieve specific generation records.

[0070] This can improve the user experience and supports exporting records, making it convenient for users to conduct further analysis and organization.

[0071] As a preferred embodiment, the AIGC-based automatic copywriting generation system further includes a multilingual generation module, which includes a dialect generation module and a foreign language generation module. The dialect generation module is used to select a dialect to generate corresponding copywriting, and the foreign language generation module is used to select a foreign language to generate corresponding copywriting.

[0072] By setting up a multilingual generation module, the copy can be generated in different languages. The foreign language generation module can be used for foreign users, while the dialect generation module can generate copy in the corresponding regional dialects for different regions, thereby increasing the fun and relatability of the copy.

[0073] The foreign language generation module includes languages ​​from known countries such as English, Japanese, Chinese, and Russian, and dialects from known regions such as Cantonese, Sichuanese, and Minnanese.

[0074] The AIGC-based automatic copywriting generation system also includes a content review and management module, which is used to detect whether the generated copy contains sensitive content such as politics, pornography, or violence.

[0075] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A copywriting automatic generation system based on an AIGC model, characterized in that, include: A style selection module, which is used to select the type of copy and the writing style; An information input module is used to input key information about the subject of the writing. The copywriting generation module is used to generate copywriting content; The selection module includes an identification module, a statement generation module, and a confirmation module. The identification module is used to identify key statements. For the identified statements, the statement generation module generates multiple statements that have the same semantics as the identified statements but have different expressions. When the identifier statement is selected, multiple statements with different expressions and requirement options pop up. The confirmation module can choose to use the original identifier statement or use one or more statements that pop up. The requirement options are used to select or input the generation requirements for the identifier statement. The identification module includes an automatic identification module and an active identification module. The automatic identification module identifies sentences containing key information when the copy is generated, while the active identification module allows users to actively select sentences to be identified after the copy is generated.

2. The automatic copywriting generation system based on the AIGC model according to claim 1, characterized in that, The AIGC-based automatic text generation system also includes a graphics module, which comprises an automatic generation module and an insertion module. The automatic generation module generates one or more corresponding images based on the semantics of the identified statements. When multiple images are generated, one or more images are selected and retained. The insertion module is used by staff to actively insert images from the local disk into the text.

3. The automatic copywriting generation system based on the AIGC model according to claim 2, characterized in that, The AIGC-based automatic copywriting generation system also includes a reading switching module, which includes an image reading module. When a user reads the copywriting and clicks to view an image in the copywriting, the image pops up and covers the text portion of the copywriting. The image reading module displays the copywriting statement portion corresponding to the content of the image within a preset area of ​​the image. When the copywriting statement portion exceeds the preset area of ​​the image, the copywriting statement portion scrolls.

4. The automatic copywriting generation system based on the AIGC model according to claim 3, characterized in that, The reading switching module also includes an information extraction module. When a user clicks to view an image in the text, the information extraction module is used to extract key information of the text corresponding to the image content, and the key information is displayed in a preset area of ​​the image.

5. The automatic copywriting generation system based on the AIGC model according to claim 4, characterized in that, The reading switching module also includes a playback module, which is used to combine multiple images in the text into a video and play them sequentially. Each image has a preset area displaying the corresponding text statement.

6. The automatic copywriting generation system based on the AIGC model according to claim 1, characterized in that, The construction of the style selection module includes the following steps: S11. Data Receipts: Collection of multi-source textual data and preprocessing of the data; S12. Data Processing: Style labeling and quantification of preprocessed copy data; S13. Model Architecture: The Transformer architecture is selected and a style control module is introduced. The style control module includes: a conditional embedding layer, which concatenates style vectors in the input layer to guide the generation direction; a dynamic adapter, which is a lightweight module inserted between Transformer layers and dynamically weights style features through an attention mechanism; and knowledge graph fusion, which combines domain knowledge graphs to ensure the professionalism of the content. S14. Data Training: Train the text data using the model architecture; S15. Evaluation and Optimization: Evaluate and optimize the trained model in multiple dimensions.

7. The automatic copywriting generation system based on the AIGC model according to claim 1, characterized in that, The AIGC-based automatic copywriting generation system also includes a copywriting template management module, which provides various types of preset templates.

8. The automatic copywriting generation system based on the AIGC model according to claim 1, characterized in that, The AIGC-based automatic copywriting generation system also includes a user feedback management module, which includes a user information recording module, a classification module, a content recognition module, and a statistics module. The classification module is used to categorize copywriting of the same type and mark the writing style of each piece of copywriting. The user information recording module is used to collect the user's gender and age; The content recognition module is used to obtain user feedback information and identify whether the content of the feedback information is the object of the comment copywriting or the writing level of the comment copywriting. When the writing level of the comment copywriting is identified, the content of the comment is judged to be a positive or negative review, and the number of positive and negative reviews is counted. The statistics module is used to analyze the positive review rate of different copywriting works of the same type under different age groups, genders and writing styles, and to analyze the types of negative reviews.

9. The automatic copywriting generation system based on the AIGC model according to claim 1, characterized in that, The AIGC-based automatic copywriting generation system also includes a record generation module, which is used to record historical copywriting information.

10. The automatic copywriting generation system based on the AIGC model according to claim 1, characterized in that, The AIGC-based automatic copywriting generation system also includes a multilingual generation module, which comprises a dialect generation module and a foreign language generation module. The dialect generation module is used to select a dialect to generate corresponding copywriting, and the foreign language generation module is used to select a foreign language to generate corresponding copywriting.

Images