Information display method based on AI algorithm

Abstract

The invention discloses an information display method based on an AI algorithm, and relates to the technical field of AI.The method comprises the steps that an app end obtains voice information of a picture to be generated, and the voice information is converted into text information through the voice recognition technology; analyzing the character information by using a natural language processing technology, obtaining a grammar structure and a keyword, and performing text enhancement; according to the technical key points, the matching degree of each picture and original character information is calculated by utilizing a multi-modal alignment model, the generated picture is ensured to be highly consistent with the description of a user, and then by setting the S score and the maximum value of the same-screen display quantity of an app end and combining the calculation of the introduced picture quality and the picture popularity, the user experience is improved. According to the method, the limitation that sorting is carried out only depending on the matching degree in a traditional picture generation scheme is solved, when the number of the pictures displayed on the same screen does not meet the requirement, sorting of the recommended pictures can be automatically carried out again according to the actual situation, and it is ensured that the content of the generated pictures conforms to user description and the quality reaches the standard.

Description

Technical Field

[0001] The present invention relates to the field of AI technology, and in particular to an information display method based on AI algorithm. Background Art

[0002] The existing application number is 202310484946.9, and the document named "A method for AR space annotation and display integrating AI general assistant" points out: first, the smart terminal identifies each target object in the scene through the AR engine; based on the target detection algorithm, each target object is identified separately to obtain the feature information of the target object; the feature information is input into the AI ​​general assistant to obtain the output result; the output result is matched with the preset user needs, the objects related to the user needs in the scene are identified, and the corresponding feedback information is extracted with feature words and sentences, and annotated and displayed in the AR space, so as to achieve the effect of personalized recommendation of relevant content for users; the present invention combines AR space annotation technology with AI general assistant, which can not only meet the intuitive, rich and vivid user interaction experience, but also provide users with more personalized recommended content through AI general assistant, and has been improved from form to content; however, the invention does not further guarantee that the generated content meets user needs and meets the quality standards, and there are certain limitations.

[0003] In combination with the above documents and prior art: Traditionally, in the process of text-generated images, the user's voice is analyzed, the voice is converted into text, and the text content is input into the text-generated image function module of the AI ​​large model. Several pictures that meet the user's needs can be generated at the same time, but user satisfaction cannot be guaranteed. The traditional picture generation scheme only relies on the limitation of sorting by matching degree. The higher the matching degree, the higher the ranking of the picture. However, some situations will be ignored in actual use. If only a few pictures can be displayed on the screen of the display end, and the number of pictures generated based on the matching degree is as high as dozens, and the matching degree of each picture is similar, there may be pictures that meet the user's needs displayed at the end, affecting the user experience. At the same time, there are limitations in the sorting of picture display, which further affects the efficiency and satisfaction of picture selection. Summary of the invention

[0004] 1. Technical issues to be resolved In view of the shortcomings of the prior art, the present invention provides an information display method based on AI algorithm. Through the specific scheme design in the technical solution, the generated picture content conforms to the user description and the quality meets the standards, thus solving the problems raised in the background technology.

[0005] (II) Technical solution To achieve the above objectives, the present invention is implemented through the following technical solutions: The information display method based on AI algorithm includes the following steps: The app obtains the voice information of the image to be generated and converts the voice information into text information; Parse text information, obtain grammatical structure and keywords, and perform text enhancement. Detect whether there is dimension data in the enhanced text. If not, trigger the adjustment mechanism. Input the enhanced text into the AI ​​system to generate several images; Using the multimodal alignment model, several images are analyzed, the matching degree between each image and the original text information is calculated, and the priority determination mechanism is implemented according to the matching degree; the corresponding image with the maximum matching degree is extracted, and the number of images with a difference of S points or less from the maximum matching degree is obtained; If the number of pictures does not exceed the maximum number of pictures displayed on the same screen on the app side, the first determination strategy is triggered; If the number of pictures exceeds the maximum number of pictures that can be displayed on the same screen on the app, the second determination strategy is triggered; According to the determination result of the corresponding determination strategy, a number of images are sorted according to priority; Display a number of sorted pictures on the app for selection; If the selected picture exists, the selected picture will be pushed from the app to the device. If the selected picture does not exist, the feedback voice data will be obtained, the feedback voice data will be converted into text information again, and the optimization adjustment strategy will be executed.

[0006] Furthermore, the process of converting voice information into text information is as follows: Audio preprocessing: Preprocessing includes at least noise reduction, echo removal, and audio format conversion; Feature extraction: The preprocessed speech information is converted into a feature vector through a feature extraction algorithm; Speech recognition model processing: The extracted feature vectors are fed into the trained HMM-GMM model, which converts speech to text based on the input feature vector sequence; Text output and post-processing: Post-process the converted text information, including error correction and format adjustment.

[0007] Furthermore, text enhancement: performing detail completion operations based on grammatical structures and keywords, and synchronously optimizing text information, including at least: correcting grammatical errors and adjusting vocabulary order; The triggering mechanism is: Sample images of various sizes are pre-selected on the app for selection; the sample images of various sizes are displayed in order from small to large size specifications.

[0008] Furthermore, before text enhancement, recommendation recognition processing is also included: When the user uses the app for the first time or the amount of historical data under the corresponding user does not reach the set threshold, the emotion recognition action is performed to determine the emotional tendency, including positive, negative and neutral; when the amount of historical data under the corresponding user reaches the set threshold, the preference recommendation action is performed to determine the preferred option; The process of determining emotional tendency is as follows: Run the emotional index library, which records words representing positive, neutral, and negative emotional tendencies as keywords for emotional recognition, and is used to judge emotional tendencies and select matching picture styles accordingly; positive emotional tendencies correspond to bright tones; negative emotional tendencies correspond to dark tones; neutral emotional tendencies correspond to tones other than bright and dark; The process of determining the preferred option is: Obtain the historical selection and behavior records of the corresponding user, analyze them using machine learning algorithms, and build a preference model for the corresponding user.

[0009] Furthermore, after the enhanced text is input into the AI ​​system, it also includes combining the results of the recommended recognition processing to generate several pictures.

[0010] Furthermore, the value and calculation process of S score are as follows: Calculate the matching degree of all images and the original text information to obtain a matching degree set. According to the pre-built threshold setting engine, set S based on the standard deviation of the matching degree set. Use the standard deviation method: calculate the standard deviation σ of the matching degree set and set S = σ*Bs, where Bs represents the multiple value.

[0011] Furthermore, the calculation process of the maximum amount of content displayed on the same screen on the app side is as follows: Determine the display area: Get the width W and height H of the app screen size; subtract the layout space, including at least the space occupied by the screen edge, navigation bar, and toolbar, to get the actual display area width W_usable and height H_usable; Determine the grid unit size: According to the design requirements, determine the width w_unit and height h_unit of each grid unit; Calculate the same-screen display volume: In the width direction, the number of grid units placed is: W_usable / w_unit; in the height direction, the number of grid units placed is: H_usable / h_unit; Then, the calculation formula for the maximum value N_max of the same-screen display volume is as follows: N_max = floor(W_usable / w_unit) * floor(H_usable / h_unit) Among them, the floor() function means rounding down.

[0012] Furthermore, the first determination strategy triggered is as follows: The priority is determined based on the matching degree of each image: ; Among them, P represents priority, and Em represents matching degree; The second decision strategy triggered is as follows: Calculate and obtain the image quality estimation and image heat estimation of each image, combine the matching degree, build a comprehensive calculation model based on product and exponential function, generate the comprehensive value of each image, and determine the priority based on the comprehensive value of each image: ; Among them, Zs represents the comprehensive value; The process of calculating the image quality estimation and image heat estimation of each image is as follows: Obtain the clarity and color reproduction of each image, and perform weighted calculations to obtain an estimated image quality; Obtain the popularity score and time decay score of each image, and perform weighted calculation to obtain the image heat estimation; Build a comprehensive calculation model based on product and exponential functions: The image quality estimation, image heat estimation, and matching degree of each image are normalized, and the comprehensive value of each image is calculated using the following formula: ; Where λ represents the attenuation coefficient, which ranges from [0, 1], Qp represents the image quality estimation, and Phv represents the image heat estimation.

[0013] Furthermore, the corresponding index of clarity is: average gradient; the corresponding index of color reproduction is: structural similarity index; The popularity score is calculated by the number of likes, shares, and comments on the corresponding image under the current display number. The formula is: ; Where FS represents the popularity score, h1, h2 and h3 are weight coefficients, all ranging from [0, 1], dz represents the number of likes, fs represents the number of shares, pl represents the number of comments, and zs represents the number of impressions; The time decay score indicates the heat decay of the corresponding image over time after it is released. It is calculated using the time decay function. The formula is: ; Where Fj represents the time decay fraction, e represents the natural constant, k represents the cooling coefficient, and its value range is [0, 1]. (T-T0) represents the time interval, where T is the current time and T0 is the time when the picture is released.

[0014] Furthermore, the content of the optimization and adjustment strategy is as follows: Keyword extraction technology is used to obtain keywords in the feedback voice information conversion feedback text information, and corresponding parameter optimization is made according to the content of the keywords. The interval of parameter optimization is a preset fixed value, and at least three types of pictures with optimized parameters are displayed on the app for secondary selection; when the corresponding user selects any picture, the parameters under the corresponding picture are recorded and used as standard parameters.

[0015] An information display system based on AI algorithm, which includes: Voice collection and conversion module: The app obtains the voice information of the image to be generated and converts the voice information into text information; Text enhancement module: parses text information, obtains grammatical structure and keywords, and performs text enhancement. It detects whether there is dimension data in the enhanced text, and triggers the adjustment mechanism if it is not detected. Text-to-image module: input enhanced text into the AI ​​system to generate several images; Alignment and screening module: Use the multimodal alignment model to analyze several images, calculate the matching degree between each image and the original text information, and implement the priority determination mechanism based on the matching degree; extract the corresponding image with the maximum matching degree, and obtain the number of images with a difference of S points or less from the maximum matching degree; If the number of pictures does not exceed the maximum number of pictures displayed on the same screen on the app side, the first determination strategy is triggered; If the number of pictures exceeds the maximum number of pictures that can be displayed on the same screen on the app, the second determination strategy is triggered; According to the determination result of the corresponding determination strategy, a number of images are sorted according to priority; Pre-display module: displays a number of sorted pictures on the app for selection; Display adjustment module: If the selected picture exists, the selected picture will be pushed from the app to the device. If the selected picture does not exist, the feedback voice data will be obtained, the feedback voice data will be converted into text information again, and the optimization adjustment strategy will be executed.

[0016] (III) Beneficial effects The present invention provides an information display method based on AI algorithm, which has the following beneficial effects: (1) By introducing emotion recognition and preference recommendation mechanisms, this solution can intelligently recommend image styles that suit the user's taste based on the user's emotional tendencies and historical preferences. This not only improves the personalization and customization of image generation, but also makes the generated images more in line with the user's actual needs and aesthetic preferences, thereby improving user satisfaction and ensuring the display effect of subsequent images to a certain extent; (2) This solution uses a multimodal alignment model to calculate the matching degree between each image and the original text information, ensuring that the generated image is highly consistent with the user's description. Then, by setting the maximum value of the S score and the amount of images displayed on the same screen on the app side, combined with the introduction of the calculation of image quality and image popularity, this solves the limitation of traditional image generation solutions that only rely on matching degree for sorting. When the number of images displayed on the same screen does not meet the requirements, the recommended images can be automatically re-sorted according to the actual situation to ensure that the generated image content meets the user's description and the quality meets the standards. (3) This solution implements the function of dynamically optimizing image display parameters based on real-time user feedback. It extracts keywords from user feedback to accurately adjust image parameters, such as color and brightness, and displays multiple images with optimized parameters on the app for users to choose from. This not only significantly improves the user experience, but also makes it easier for users to select satisfactory images through personalized adjustments. In addition, by recording user-preferred parameters as system standard parameters, it effectively solves the problem of users having to flip through pages to select images multiple times or being unable to select an image because the image display does not meet their expectations. This greatly improves the efficiency and satisfaction of image selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flowchart of the overall steps of the information display method in Example 1 of the present invention; Figure 2 This is a schematic diagram of the overall process of executing S4 in Example 2 of the present invention; Figure 3 It is a modular schematic diagram of the information display system based on AI algorithm in the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Research and development concept: With the continuous development of AI technology, it is now necessary to build an information display application that integrates AI technology. The core is to realize the automatic generation of high-quality pictures from user voice. The solution will focus on optimizing the accuracy of voice recognition, enhancing the richness of text descriptions, improving the accuracy of AI picture generation and the effectiveness of matching calculation. Current issues to be solved: 1. Text enhancement requires more sophisticated semantic understanding and completion; 2. The expressiveness of AI image generation in complex scenes and styles; 3. The matching calculation model’s sensitivity to details and style, etc. Solution Description: 1. Introduce more advanced speech recognition models and combine filtering technology to improve recognition accuracy; use deep learning technology to enhance the depth of text understanding, or combine knowledge graphs to complete description details; 2. Upgrade the AI ​​image generation model, introduce more style and scene data, and improve the diversity and realism of the images; 3. Optimize multimodal alignment models, such as CLIP, to enhance the ability to recognize image details and styles and improve the accuracy of matching calculations; at the same time, consider introducing user behavior data to further personalize matching results; The problems and solutions given above are just one direction or basic concept. The specific solution designs given later may be implemented based on one direction or several concepts. This solution aims to create an efficient, accurate and user-friendly image generation and display system to meet users' diverse image needs.

[0020] Example 1

[0021] See also Figure 1 This embodiment provides an information display method based on an AI algorithm, and the steps of the method are as follows: S1. The app obtains the voice information of the image to be generated and converts the voice information into text information using voice recognition technology; Among them, the app user says the information such as the content and size of the picture he wants to generate; S2. Use natural language processing technology to parse text information, obtain grammatical structure and keywords, and perform text enhancement; Among them, after the process given in S1: the app submits the voice information to the backend, the backend converts the voice information into text information, and then sends it to the "text enhancement module" for further processing; after the "text enhancement module" parses the input text information, it obtains the grammatical structure and keywords, completes the description details, and ensures that the enhanced text matches the requirements of the generated image model (i.e., AI system); S3, the "text enhancement module" accesses the AI ​​interface and submits the task of generating pictures to the AI ​​system; S4, after receiving the task, the AI ​​system generates a series of pictures of a specific style; S5, the backend uses CLIP or a similar multimodal alignment model to calculate the matching degree between each image and the original text description (i.e., the text information in S2), and displays images with higher matching degrees to users first; S6. The user selects from the app according to the needs. If selected, the picture is pushed out from the app for display on the target device; if not selected, the operation of S1 is executed again.

[0022] The above technical solution realizes an efficient and personalized generation process from voice to image, bringing significant beneficial effects. First, through voice recognition technology, users can express the information such as the content and size of the picture they want to generate in the most natural way - voice input, which greatly improves the user experience and convenience. Secondly, the application of natural language processing technology enables the system to accurately analyze the user's intention, refine and enrich the description details through the text enhancement module, and ensure that the generated pictures are more in line with user needs. In addition, the seamless connection with the AI ​​system enables text information to be quickly converted into pictures of a specific style, which not only enriches the diversity and creativity of picture generation, but also improves the generation efficiency. The matching degree between the picture and the original text description is evaluated through CLIP or similar models, ensuring that the picture displayed to the user is the most consistent with its description, further improving user satisfaction. Finally, users can select pictures on the app side according to actual needs, and if they are not selected, they can be regenerated. This flexible interaction mechanism gives users higher participation and decision-making power, making the entire picture generation process more personalized and intelligent. In summary, the technical solution has achieved remarkable results in improving user experience and enhancing the accuracy and personalization of generated pictures.

[0023] Embodiment 2: Based on Example 1, please refer to Figure 2 This embodiment also provides a further optimized information display method based on AI algorithm (this embodiment provides a detailed description and explanation of some contents and optimized contents in embodiment 1). The specific steps of this method are as follows: S1. The app obtains the voice information of the image to be generated and converts the voice information into text information using voice recognition technology; The user speaks the content and size of the image to be generated through the app. The content and size together constitute the original voice information. Before the user speaks the original voice information, the app will give an information prompt, for example: reminding the user that the content to be spoken needs to include the content and size, but of course it is not limited to the content and size. At this time, user A may say: Example 1: "I want to generate a landscape painting with a pleasant atmosphere and a beautiful environment, in a larger size"; Or Example 2: “I want to generate a landscape painting, of moderate size”; Or Example 3: "I want to generate a minimalist landscape painting with a size of 1920*1080"; It should be noted that not all users will say specific sizes. Most users will say vague and colloquial words such as larger, medium, and smaller. Therefore, it is necessary to implement a supply and adjustment mechanism in the subsequent S2. Please refer to the content expressed in S2 for details. The conversion process of voice information to text information is as follows: Audio preprocessing (i.e., speech information): Preprocessing includes at least: noise reduction, echo removal, and audio format conversion to improve the accuracy of speech recognition; audio segmentation is also performed as needed to cut continuous speech into single words or phrases for subsequent processing; Feature extraction: The preprocessed speech information is converted into a series of feature vectors through feature extraction algorithms (such as MFCC, FBank, etc.). These feature vectors are the representation of speech signals in the frequency domain and time domain, reflecting the acoustic characteristics of speech; Speech recognition model processing: The extracted feature vectors are fed into a trained speech recognition model (such as a deep learning model, HMM-GMM model, etc.); the model converts speech to text based on the input feature vector sequence, combined with the language model and acoustic model; the language model is used to predict possible vocabulary sequences, and the acoustic model is used to evaluate the degree of match between these sequences and the input speech; Text output and post-processing: The text output by the speech recognition model may contain some errors or uncertain words and requires post-processing; post-processing includes error correction, format adjustment, special word replacement, etc. to ensure that the output text is accurate, clear and consistent with the user's intention.

[0024] S2. Analyze text information using natural language processing technology, obtain grammatical structure and keywords, and perform text enhancement; preliminarily detect whether there is dimension data in the enhanced text, and trigger the supply and adjustment mechanism if it is not detected; Among them, before text enhancement, it also includes recommendation recognition processing; When the user uses it for the first time or the amount of historical data (i.e., voice information) of the corresponding user does not reach the set threshold, the emotion recognition action is performed to determine the emotional tendency, including positive, negative, and neutral; When the amount of historical data under the corresponding user reaches the set threshold, the preference recommendation action is executed to determine the preference option; It should be noted that the amount of voice information refers to the amount of memory occupied by the original voice information spoken by the corresponding user, expressed in kb. For example, if the threshold is set to 1000kb, after the current app obtains the voice information of the corresponding user, it detects that the current amount of voice information of the corresponding user reaches 1200kb, which means that the corresponding user has entered the option of executing the recommended action; however, if the corresponding user is using it for the first time, the amount of historical data of the user will not be considered; Of course, if the original voice information converted into text information already contains a style description, such as the content in Example 3, it has been clearly stated that a simple style picture is required, but most users do not know or have forgotten the required style, so a recommendation recognition process is required; The process of performing emotion recognition and determining emotional tendency is as follows: Run the emotional index library, which records: words representing positive emotional tendencies include at least "happiness", "joy", "excitement", "thrilled", "satisfaction", "relieved" and "optimism", which are usually associated with positive emotions and pleasant feelings; words representing negative emotional tendencies include at least "sadness", "depression", "disappointment", "anxiety", "pain", "anger" and "depression", which are often associated with negative emotions and unpleasant feelings; after excluding words representing positive and negative emotional tendencies, the remaining words are words representing neutral emotional tendencies; these words can be used as keywords for emotion recognition to help the system judge the user's emotional tendencies and select matching picture styles accordingly; Match the corresponding picture style according to the emotional tendency; Positive affective tendencies correspond to bright (cheerful) tones; Negative affective tendencies correspond to dark (deep) tones; Neutral affective tendency corresponds to tones other than bright and dark, i.e. regular or normal tones; The process of executing the preference recommendation action and determining the preference options is as follows: Obtain the historical selection and behavior records of the corresponding user, analyze them using machine learning algorithms, and build a preference model for the corresponding user. This model is used to reflect the user's strong preference for different styles and form the corresponding user's personalized preferences; For example, if a user selects simple-style images multiple times, the system will give priority to simple-style images in subsequent generation. In the subsequent image generation process, the system will give priority to images that suit the user's taste based on this preference model. Specifically, the system will adjust the parameters of the generated images so that the newly generated images are closer to the simple style preferred by the user in terms of color, composition, and theme. In this way, when users browse images again, they can more easily find the type they like, thereby improving user experience and satisfaction. In general, after the recommendation recognition process, the results obtained are: corresponding to the user's determination of the picture style; When using natural language processing technology to parse text information, the grammatical structure of the text information is first analyzed to identify language units such as sentences, phrases and words, and understand the association and hierarchical relationship between them; this helps to more accurately grasp the user's intention and the focus of expression; for the acquisition of keywords, keyword extraction technology is used to identify words with key meanings from the text information, including at least nouns, verbs and adjectives; these keywords are usually closely related to the specific information such as the content, style, size, etc. of the picture that the user wants to generate; Text enhancement: Perform detail completion operations based on grammatical structure and keywords, and optimize text information synchronously, including at least: correcting grammatical errors, adjusting word order, and adding necessary modifiers to ensure that it is more fluent, accurate, and meets the requirements of the generated image model; among them, the detail completion operation means completing the details of the text information; For example, if the user only mentioned "I want a landscape painting", based on the grammatical structure and keywords, and according to common landscape elements, the detail completion operation will automatically complete it to "I want a landscape painting with blue sky and white clouds, green mountains and green water, and small bridges and flowing water"; Examples of integrity are as follows: Suppose the user inputs via voice: "I want a picture of a retro-style living room with a sofa, coffee table, and bookshelf." After text enhancement and parsing, the "Text Enhancement Module" may enhance this text into: "I want a picture of a retro living room with a brown leather sofa and a wooden coffee table in front of it, on which are placed several decorative books and a vase. In the corner of the living room, there is a retro bookshelf full of books. The material of the bookshelf is dark solid wood, which is in harmony with the overall style of the living room. The color tone of the whole living room is mainly warm, creating a warm and comfortable atmosphere." The enhanced text not only contains the key information originally input by the user, but also provides more detailed descriptions, such as the color and material of the sofa, the items on the coffee table, the style and location of the bookshelf, etc., so as to more comprehensively meet the needs of the generated image model and generate images that better meet the user's expectations. The contents that trigger the supply and regulation mechanism are as follows: Sample images of various sizes are pre-selected on the app for users to choose from; the sample images of various sizes are displayed in order from small to large until the user selects one.

[0025] Specifically, by introducing emotion recognition and preference recommendation mechanisms, the solution can intelligently recommend image styles that suit the user's taste based on the user's emotional tendencies and historical preferences. This not only improves the personalization and customization of image generation, but also makes the generated images more in line with the user's actual needs and aesthetic preferences, thereby improving user satisfaction and ensuring the display effect of subsequent images to a certain extent. This solution also solves the problem of user input ambiguity. For ambiguous size descriptions input by users, such as colloquial expressions such as "bigger" and "medium", this solution can convert them into specific size parameters by triggering the adjustment mechanism, thereby ensuring the accuracy of image generation and meeting user expectations. In summary, the above solution not only realizes the intelligent generation process from voice to picture, but also improves the personalization, accuracy and quality of picture generation by introducing technologies such as emotion recognition, preference recommendation and text enhancement, bringing users a more convenient, efficient and personalized picture generation experience.

[0026] S3, input the enhanced text into the AI ​​system, and generate several pictures in combination with the results of the recommended recognition processing; Among them, the AI ​​system is a tool for Wenshengtu; AI systems specifically refer to comprehensive systems that integrate advanced text processing and image generation technologies. Such systems usually use deep learning algorithms, especially generative adversarial networks (GANs) and diffusion models (such as Stable Diffusion), to achieve the function of converting enhanced text input into high-quality images; Take AI-Chat as an example. It integrates multiple AI models, including GPT4.0 and Stable Diffusion painting models. Users can generate various anime, cartoon-style avatars or other works of art by inputting text descriptions. In addition, there are many AI image software on the market, such as Jasperart, Photosonic, Shutterstock, NightCafe, etc. They have their own characteristics and can provide users with a variety of image generation experiences. These systems not only improve the efficiency of image creation, but also provide powerful auxiliary tools for creative workers such as designers and illustrators. The AI ​​system produces more than one picture. Instead, it outputs all pictures that meet the text description. For some common text descriptions, the number of pictures that meet the requirements is as high as dozens or hundreds. For some uncommon or special text descriptions, the number of pictures that meet the requirements is as low as 1 and as many as 10. The above numbers are just examples, and the specific data can only be obtained based on actual conditions.

[0027] S4, using a multimodal alignment model to analyze a number of images, calculate the matching degree between each image and the original text information, and execute a priority determination mechanism based on the matching degree; Extract the corresponding picture with the maximum matching degree, obtain the number of pictures with a difference of S points or less from the maximum matching degree, and if the number of pictures does not exceed the maximum number of pictures displayed on the same screen on the app side, trigger the first judgment strategy; If the number of pictures exceeds the maximum number of pictures that can be displayed on the same screen on the app, the second determination strategy is triggered; According to the determination result of the corresponding determination strategy, a number of images are sorted according to priority; In this scenario, the system uses CLIP (Contrastive Language–Image Pre-training) or a similar multimodal alignment model to perform in-depth analysis on each image generated by the AI ​​system. The CLIP model (i.e., the multimodal alignment model used in this embodiment) is a powerful tool for learning the association between vision and language. It can encode images and text into the same high-dimensional space, so that the similarity between the two can be measured by their distance in this space. The matching degree calculation is as follows: In the multimodal alignment model, calculating the matching degree between each image and the original text description is a key step, which is achieved by calculating the distance between the image and the original text information in the encoding space; the smaller the distance, the higher the matching degree between the image and the text description; conversely, the lower the matching degree; therefore, the matching degree represents the inverse of the distance value between the corresponding image and the original text information in the encoding space, for example: 1 / distance value; In addition, in actual situations, in order to ensure the accuracy of the match, the model needs to comprehensively consider factors such as the content, style, and size of the image. For example, in terms of content, the model needs to determine whether the objects and scenes in the image are consistent with the text description; in terms of style, the model needs to evaluate whether the color, lines, composition, etc. of the image are consistent with the style of the text description; in terms of size, the model needs to ensure that the generated image meets the size requirements specified by the user; The value and calculation process of S score are as follows: Calculate the matching degree of all the images generated in S3 and the original text information to obtain a matching degree set. According to the pre-built threshold setting engine, S is set based on the standard deviation of the matching degree set. The standard deviation method is adopted: calculate the standard deviation σ of the matching degree set, and set S = σ*Bs, where Bs represents the multiple value, and the value of Bs is 1 or 2, and the value in this embodiment is 2; this method takes into account the distribution width of the matching degree, and is more dynamic and adaptable; The formula for calculating the standard deviation σ of the matching degree set is as follows: ;

[0028] In the formula, N represents the number of data points, that is, the number of each matching degree in the matching degree set; xi represents each matching degree value in the matching degree set; μ represents the average value of all matching degree values ​​in the matching degree set; Therefore, the formula for calculating the S score is: S = 2σ; Suppose we have the following set of matching degrees (expressed as 1 / distance value): [0.1, 0.2, 0.4, 0.5, 0.7, 0.8, 0.9, 1.0]; Among them, the maximum matching degree is 1.0; Using the standard deviation method: Assume that the standard deviation σ of this set is about 0.25 (this is an assumed value and should be calculated in practice); if we set S to be twice the standard deviation, then S = 2 * 0.25 = 0.5; then, the matching degrees that are within 0.5 of the maximum value are: [0.5, 0.7, 0.8, 0.9, 1.0]; in this example, the number of images that are within S points of the maximum matching value is 4; The calculation process of the maximum number of displays on the same screen on the app side is as follows: Determine available display area: Get the width W and height H of the app screen size; subtract the layout space, including at least the space occupied by interface elements such as the screen edge, navigation bar, and toolbar, to get the actual available display area width W_usable and height H_usable; Determine the grid unit size (the images are displayed on the app in a grid): According to the design requirements, determine the width w_unit and height h_unit of each grid unit; these dimensions are determined in advance according to the built-in program settings of the app; Calculate the same-screen display volume: In the width direction, the number of grid units placed is W_usable / w_unit (take the integer part); In the height direction, the number of grid cells that can be placed is approximately H_usable / h_unit (the integer part); Therefore, the maximum value of the same-screen display volume is (W_usable / w_unit) * (H_usable / h_unit) (take the integer part); therefore, the calculation formula for the maximum value of the same-screen display volume N_max is as follows: N_max = floor(W_usable / w_unit) * floor(H_usable / h_unit) Among them, the floor() function means rounding down; It should be noted that the purpose of comparing the number of pictures with the maximum number of pictures displayed on the same screen on the app side is to automatically select which judgment strategy to execute. If the number of pictures does not exceed the maximum number of pictures displayed on the same screen on the app side, it means that all pictures can be displayed on the same screen on the app side (they can be regarded as "no gap or the gap can be ignored" when displayed), and there is no need to slide down or switch to the next screen (if such operation is performed, it will not only waste time but also affect the user's selection judgment). At this time, only the matching degree needs to be considered to determine the picture priority; but if the number of pictures exceeds the maximum number of pictures displayed on the same screen on the app side, it means that all pictures cannot be fully displayed on one screen. In order to ensure that users can select the required pictures only on the first screen, it is necessary to consider not only the matching degree but also other factors (i.e. picture quality and popularity) to determine the picture priority; The first judgment strategy triggered is as follows: The priority is determined based on the matching degree of each image: ;

[0029] Among them, P represents priority, and Em represents matching degree; The content of the second judgment strategy triggered is as follows: Calculate and obtain the image quality estimation and image heat estimation of each image, combine the matching degree, build a comprehensive calculation model based on product and exponential function, generate the comprehensive value of each image, and determine the priority based on the comprehensive value of each image: ;

[0030] Among them, Zs represents the comprehensive value; Regardless of the first or second determination strategy, the images are sorted according to the priority, for example, they are sorted from left to right and from top to bottom on the screen of the app. The process of calculating the image quality estimation and image heat estimation of each image is described as follows: Obtain the clarity and color reproduction of each image, and perform weighted calculations to get an estimate of the image quality; Obtain the popularity score and time decay score of each image, and perform weighted calculation to get the image heat estimation; The explanation is as follows: Clarity is an important indicator in image quality assessment, which reflects the visibility of image details. Clarity assessment methods include edge detection-based methods and frequency domain analysis-based methods. Among them, the corresponding indicator of clarity is: Mean Gradient, which reflects the rate of detail contrast and texture change in the image; Color reproduction refers to the degree of closeness between the color of an image and the color of the real scene. The color reproduction evaluation methods include color space-based methods and histogram-based methods. Among them, the corresponding index of color reproduction is the Structural Similarity Index (SSIM), which reflects the color reproduction of an image. The image quality estimate is calculated based on the following formula: ;

[0031] Where Qp represents the image quality estimation, a1 and a2 are weight coefficients, both in the range of [0, 1], MG represents the clarity, and SM represents the color restoration; The popularity score is calculated by the number of likes, shares, and comments on the corresponding image under the current display number. The specific formula is: ;

[0032] Where FS represents the popularity score, h1, h2, and h3 are the single score items of their respective behaviors, which are set according to the platform strategy to reflect the contribution of different behaviors to the popularity score, that is, the weight coefficient, and the value range is [0, 1]. dz represents the number of likes, fs represents the number of shares, pl represents the number of comments, and zs represents the number of impressions. The time decay score indicates the heat decay of the corresponding image over time after it is released. It is calculated using a time decay function (such as Newton's law of cooling). The specific formula is: ;

[0033] In the formula, Fj represents the time decay fraction, e represents the natural constant, k represents the cooling coefficient, which reflects the speed at which the heat value decays over time, and its value range is [0, 1]. (T-T0) represents the time interval, where T is the current time and T0 is the time when the picture is released. The formula for calculating the estimated heat value of an image is as follows: ;

[0034] Where Phv represents the estimated heat value of the image, b1 and b2 are weight coefficients, and their value range is [0, 1]; The weight coefficient is determined by the coefficient of variation method, which is a method of assigning weights to each indicator according to the degree of variation between the current value of each evaluation indicator and the target value. If the numerical difference of a certain indicator is large, it can clearly distinguish the evaluated objects, indicating that the indicator has rich information for distinguishing, and thus the indicator should be given a larger weight. On the contrary, if the numerical difference of each evaluated object on a certain indicator is small, then the ability of this indicator to distinguish the evaluated objects is weak, and thus the indicator should be given a smaller weight. This method directly uses the information contained in each indicator to obtain the weight of the indicator through calculation, so it is objective. Build a comprehensive calculation model based on product and exponential functions: The image quality estimation, image heat estimation and matching degree of each image are normalized so that their value range is limited to between 0 and 1, and then the comprehensive value of each image is calculated using the following formula: ;

[0035] In the formula, λ represents the attenuation coefficient, which can be adjusted according to actual conditions and has a value range of [0, 1]; Logic description: Product part: Em*Qp*Phv reflects the combined effect of the three indicators of matching degree, image quality estimation, and image heat estimation. When any one of the indicators is close to 0, the entire product term will also be close to 0, indicating that the comprehensive value will be significantly affected; Denominator: 1-(1-Em)(1-Qp)(1-Phv) is an adjustment factor used to enhance the interaction between the three indicators. When all three indicators are high, the denominator is close to 1, and the impact of the product term is more significant. When at least one of the three indicators is low, the denominator increases, thereby reducing the impact of the product term on the comprehensive value. Index part: (-)^(1 / 3) is the square root of the product term to smooth out the extreme differences between the three indicators. This ensures that when one indicator is abnormally high or low, the combined value will not deviate too much from the normal range. Time decay part: It is an attenuation term based on an exponential function, which is used to consider the attenuation of the average values ​​of the three indicators. When the average value of Em+QP+Phv is close to 3 (that is, all three indicators are very high), the attenuation term is close to 1, and the impact on the comprehensive value is small; when the average value is low, the attenuation term will decrease, thereby reducing the comprehensive value; Comprehensive consideration: The entire formula takes into account the three indicators of matching degree, image quality and image popularity, and combines them through a non-linear relationship. This combination can not only reflect the individual effects of each indicator, but also reflect their interaction and overall effect.

[0036] First, the solution uses a multimodal alignment model (such as CLIP) to calculate the matching degree between each image and the original text information, ensuring that the generated image is highly consistent with the user's description. This not only improves the accuracy of image generation, but also makes it easier for users to find images that meet their needs. By setting the maximum value of the S score and the amount of images displayed on the same screen on the app side, the solution can flexibly adapt to different quantities and qualities of images, providing users with more accurate recommendations. Secondly, by introducing the calculation of image quality and image popularity, this solution solves the limitation of traditional image generation solutions that only rely on matching degree for sorting. When the number of images displayed on the same screen does not meet the requirement, the solution can automatically select images based on the matching degree, and also comprehensively consider image quality and image popularity. This not only re-sorts the recommended images according to the actual situation, but also ensures that the generated image content meets the user's description and is of qualified quality. The solution constructs a comprehensive calculation model based on product and exponential functions, organically combining the three indicators of matching degree, image quality and image popularity to generate a comprehensive value to evaluate the overall performance of each image. This comprehensive evaluation method not only considers the individual effects of each indicator, but also reflects their interactions and overall effects, making the ranking results more comprehensive, objective and in line with user expectations. Finally, by adopting the above scheme, this method not only improves the accuracy and quality of image generation, but also enhances user experience and satisfaction; users can more easily find pictures that meet their needs, and can browse more high-quality and popular pictures on one screen; at the same time, the scheme also has certain adaptability and scalability, and can adjust parameters and models according to actual conditions to meet the needs and scenarios of different users.

[0037] In summary, the above solution realizes comprehensive evaluation and sorting of AI-generated images by integrating multimodal alignment models, image quality assessment, image heat calculation and comprehensive calculation models, improves the accuracy and quality of image generation, and enhances user experience and satisfaction.

[0038] S5. Displaying a number of sorted pictures on the app for users to choose from; Among them, a number of pictures sorted by priority are arranged in a grid on the display screen of the app. The user selects on the display screen. If it is selected, the subsequent steps are performed. If it is not selected, the user continues to turn the page to select until it is selected. However, in the actual selection process, the user usually completes the selection on the homepage. If the user turns the page to select, there is a high probability that there will still be no selected picture; S6. If the selected picture exists, the selected picture is pushed from the app to the device. If the selected picture does not exist, the feedback voice data is obtained, S1 is re-executed, and the optimization adjustment strategy is executed according to the output result of S1. Among them, the device side refers to a specified user device, such as a large screen, so as to achieve the effect of information display; The feedback voice data represents the user's feedback voice, for example, the user feedbacks that the color of the picture is not bright enough; then, when S1 is re-executed later, the content of the optimization adjustment strategy executed according to the output result of S1 is as follows: The keyword extraction technology is used to obtain the keywords in the feedback voice information and the feedback text information, and the corresponding parameter optimization is performed according to the content of the keywords. The interval of parameter optimization is a preset fixed value, and at least three types of pictures with optimized parameters are displayed on the app for users to choose from. When the corresponding user selects any picture, the parameters of the corresponding picture are recorded and used as the standard parameters of the current system. Among them, if the user feedback is that the color of the picture is not bright enough, the keywords are "color" and "not bright enough", and the system optimizes and adjusts the color parameters. The original color parameter is U, and the interval value is 1, then pictures with three color parameters of U+1, U+2 and U+3 are displayed on the app (the three types of pictures only differ in color parameters). If the corresponding user selects a picture with U+3 color parameters, the color parameter is recorded. The next time the corresponding user operates again, the color parameter is retrieved, which can avoid the problem of not being able to select the picture at once to a certain extent and improve work efficiency.

[0039] Specifically, the above scheme is adopted to realize the function of dynamically optimizing image display parameters according to real-time user feedback. By extracting keywords in the user feedback voice, the image parameters such as color, brightness, etc. are accurately adjusted, and multiple parameter-optimized images are displayed on the APP for users to make a second choice. This scheme not only significantly improves the user experience, but also makes it easier for users to select satisfactory images through personalized adjustments. Moreover, by recording the user's preferred parameters as the system standard parameters, it effectively solves the problem that users have to flip pages to select multiple times or cannot select images because the image display does not meet their expectations, which greatly improves the efficiency and satisfaction of image selection.

[0040] Embodiment 3: See also Figure 3 , based on Example 2, this embodiment also provides an information display system based on AI algorithm, which includes a voice collection and conversion module, a text enhancement module, a text image module, an alignment and screening module, a pre-display module and a display adjustment module running in sequence; Voice collection and conversion module: The app obtains the voice information of the image to be generated and converts the voice information into text information; Text enhancement module: parses text information, obtains grammatical structure and keywords, and performs text enhancement. It detects whether there is dimension data in the enhanced text, and triggers the adjustment mechanism if it is not detected. Text-to-image module: input enhanced text into the AI ​​system to generate several images; Alignment and screening module: Use the multimodal alignment model to analyze several images, calculate the matching degree between each image and the original text information, and implement the priority determination mechanism based on the matching degree; extract the corresponding image with the maximum matching degree, and obtain the number of images with a difference of S points or less from the maximum matching degree; If the number of pictures does not exceed the maximum number of pictures displayed on the same screen on the app side, the first determination strategy is triggered; If the number of pictures exceeds the maximum number of pictures that can be displayed on the same screen on the app, the second determination strategy is triggered; According to the determination result of the corresponding determination strategy, a number of images are sorted according to priority; Pre-display module: displays a number of sorted pictures on the app for selection; Display adjustment module: If the selected picture exists, the selected picture will be pushed from the app to the device. If the selected picture does not exist, the feedback voice data will be obtained, the feedback voice data will be converted into text information again, and the optimization adjustment strategy will be executed.

[0041] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0042] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0043] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. The information display method based on AI algorithm is characterized by: The steps of this method are as follows: The app obtains the voice information of the image to be generated and converts the voice information into text information; Parse text information, obtain grammatical structure and keywords, and perform text enhancement. Detect whether there is dimension data in the enhanced text. If not, trigger the adjustment mechanism. Input the enhanced text into the AI ​​system to generate several images; Using the multimodal alignment model, several images are analyzed, the matching degree between each image and the original text information is calculated, and the priority determination mechanism is implemented according to the matching degree; the corresponding image with the maximum matching degree is extracted, and the number of images with a difference of S points or less from the maximum matching degree is obtained; If the number of pictures does not exceed the maximum number of pictures displayed on the same screen on the app side, the first determination strategy is triggered; If the number of pictures exceeds the maximum number of pictures that can be displayed on the same screen on the app, the second determination strategy is triggered; According to the determination result of the corresponding determination strategy, a number of images are sorted according to priority; Display a number of sorted pictures on the app for selection; If the selected picture exists, the selected picture will be pushed from the app to the device. If the selected picture does not exist, the feedback voice data will be obtained, the feedback voice data will be converted into text information again, and the optimization adjustment strategy will be executed.

2. The information display method based on AI algorithm according to claim 1, characterized in that: The process of converting voice messages to text messages is as follows: Audio preprocessing: Preprocessing includes at least noise reduction, echo removal, and audio format conversion; Feature extraction: The preprocessed speech information is converted into a feature vector through a feature extraction algorithm; Speech recognition model processing: The extracted feature vectors are fed into the trained HMM-GMM model, which converts speech to text based on the input feature vector sequence; Text output and post-processing: Post-process the converted text information, including error correction and format adjustment.

3. The information display method based on AI algorithm according to claim 1, characterized in that: Text enhancement: Perform detailed completion operations based on grammatical structure and keywords, and simultaneously optimize text information, including at least: correcting grammatical errors and adjusting vocabulary order; The triggering mechanism is: Sample images of various sizes are pre-selected on the app for selection; the sample images of various sizes are displayed in order from small to large size specifications.

4. The information display method based on AI algorithm according to claim 1, characterized in that: Before text enhancement, recommendation recognition processing is also included: When the user uses the app for the first time or the amount of historical data under the corresponding user does not reach the set threshold, the emotion recognition action is performed to determine the emotional tendency, including positive, negative and neutral; when the amount of historical data under the corresponding user reaches the set threshold, the preference recommendation action is performed to determine the preferred option; The process of determining emotional tendency is as follows: Run the emotional index library, which records words representing positive, neutral, and negative emotional tendencies as keywords for emotional recognition, and is used to judge emotional tendencies and select matching picture styles accordingly; positive emotional tendencies correspond to bright tones; negative emotional tendencies correspond to dark tones; neutral emotional tendencies correspond to tones other than bright and dark; The process of determining the preferred option is: Obtain the historical selection and behavior records of the corresponding user, analyze them using machine learning algorithms, and build a preference model for the corresponding user.

5. The information display method based on AI algorithm according to claim 4 is characterized in that: After the enhanced text is input into the AI ​​system, it is also combined with the results of the recommended recognition processing to generate several pictures.

6. The information display method based on AI algorithm according to claim 1, characterized in that: The value and calculation process of S score are as follows: Calculate the matching degree of all images and the original text information to obtain the matching degree set. According to the pre-built threshold setting engine, set S based on the standard deviation of the matching degree set. Use the standard deviation method: calculate the standard deviation σ of the matching degree set, and set S=σ*Bs, where Bs represents the multiple value.

7. The information display method based on AI algorithm according to claim 1, characterized in that: The calculation process of the maximum number of displays on the same screen on the app side is as follows: Determine the display area: Get the width W and height H of the app screen size; subtract the layout space, including at least the space occupied by the screen edge, navigation bar, and toolbar, to get the actual display area width W_usable and height H_usable; Determine the grid unit size: According to the design requirements, determine the width w_unit and height h_unit of each grid unit; Calculate the same-screen display volume: In the width direction, the number of grid units placed is: W_usable / w_unit; in the height direction, the number of grid units placed is: H_usable / h_unit; Then, the calculation formula for the maximum value N_max of the same-screen display volume is as follows: N_max = floor(W_usable / w_unit) * floor(H_usable / h_unit) Among them, the floor() function means rounding down.

8. The information display method based on AI algorithm according to claim 1, characterized in that: The first judgment strategy triggered is as follows: The priority is determined based on the matching degree of each image: ； Among them, P represents priority, and Em represents matching degree; The second decision strategy triggered is as follows: Calculate and obtain the image quality estimation and image heat estimation of each image, combine the matching degree, build a comprehensive calculation model based on product and exponential function, generate the comprehensive value of each image, and determine the priority based on the comprehensive value of each image: ； Among them, Zs represents the comprehensive value; The process of calculating the image quality estimation and image heat estimation of each image is as follows: Obtain the clarity and color reproduction of each image, and perform weighted calculations to obtain an estimated image quality; Obtain the popularity score and time decay score of each image, and perform weighted calculation to obtain the image heat estimation; Build a comprehensive calculation model based on product and exponential functions: The image quality estimation, image heat estimation, and matching degree of each image are normalized, and the comprehensive value of each image is calculated using the following formula: ； Where λ represents the attenuation coefficient, which ranges from [0, 1], Qp represents the image quality estimation, and Phv represents the image heat estimation.

9. The information display method based on AI algorithm according to claim 8, characterized in that: The corresponding index of clarity is: average gradient; the corresponding index of color reproduction is: structural similarity index; The popularity score is calculated by the number of likes, shares, and comments on the corresponding image under the current display number. The formula is: ； Where FS represents the popularity score, h1, h2 and h3 are weight coefficients, all ranging from [0, 1], dz represents the number of likes, fs represents the number of shares, pl represents the number of comments, and zs represents the number of impressions; The time decay score indicates the heat decay of the corresponding image over time after it is released. It is calculated using the time decay function. The formula is: ； Where Fj represents the time decay fraction, e represents the natural constant, k represents the cooling coefficient, and its value range is [0, 1]. (T-T0) represents the time interval, where T is the current time and T0 is the time when the picture is released.

10. The information display method based on AI algorithm according to claim 1, characterized in that: The content of the optimization and adjustment strategy is as follows: Keyword extraction technology is used to obtain keywords in the feedback voice information conversion feedback text information, and corresponding parameter optimization is made according to the content of the keywords. The interval of parameter optimization is a preset fixed value, and at least three types of pictures with optimized parameters are displayed on the app for secondary selection; when the corresponding user selects any picture, the parameters under the corresponding picture are recorded and used as standard parameters.

Images