A new product material design method and system based on AI

Abstract

The present invention discloses a new product material design method and system based on AI, which specifically relates to the technical field of image material design, and is used to solve the problem of low matching between material design and target audience; the present invention collects and cleans user behavior data, identifies user emotional state in combination with sentiment analysis, builds accurate user portraits and extracts interest preferences, generates personalized design versions based on user portraits using a generative adversarial network, optimizes design elements in combination with real-time user feedback and a reinforcement learning algorithm, evaluates design versions through A / B testing and a multi-armed bandit algorithm, selects the best version for adjustment and optimization, generates marketing copy using natural language generation technology in combination with user portraits and design elements, optimizes design drawings through image processing and a generative adversarial network, automatically adjusts size, tone and layout, improves the display effect of marketing materials, evaluates and optimizes design versions based on real-time market feedback, and improves marketing effects.

Description

Technical Field

[0001] The present invention relates to the field of image material technology, and more specifically, to an AI-based new product material design method and system. Background Art

[0002] With the rapid development of digital marketing and e-commerce platforms, the demand for personalized product design and advertising continues to increase. Users' purchasing decisions are not only affected by the product itself, but are also closely related to the visual effects of the advertisement, marketing copy, and promotion channels.

[0003] Deficiencies in existing technologies: Traditional design methods usually rely on fixed templates and design rules, and are often unable to be flexibly adjusted according to the diverse needs of users. Most design solutions fail to fully consider the personalized preferences of the target user group when creating them, resulting in the visual effects and content of the design possibly deviating from the actual needs of users, thereby reducing the matching degree and attractiveness between the design and the user. In addition, traditional design optimization and marketing material generation processes are usually static, and are mostly based on pre-set rules or periodic manual adjustments. Once the design and advertising image materials are generated, they cannot quickly respond to market changes and instant feedback on user behavior. The design optimization process mostly relies on manual analysis of market data, such as click-through rate, conversion rate, etc., for subsequent adjustments. This manually driven adjustment method is not only inefficient, but also difficult to promptly reflect the rapid changes in market trends, resulting in the inability to quickly improve and enhance marketing results. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides an AI-based new product material design method and system to solve the problem of low matching between material design and target audience in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A new product material design method based on AI includes the following steps:

[0007] Collect and clean user behavior data, and identify user emotional states through sentiment analysis of user behavior data, segment user groups, extract interest preferences, and generate user portraits;

[0008] Based on user portraits and preference information, we use generative adversarial networks to generate different design versions. Combined with user feedback data, we use reinforcement learning algorithms to optimize design elements. We also use A / B testing and multi-armed bandit algorithms to evaluate design versions and select the best version for adjustment.

[0009] Combine user portraits and design elements, use natural language generation technology to generate marketing copy, optimize design images through image processing and generative adversarial networks, and automatically adjust the size, color tone and layout to optimize the display effect of marketing materials;

[0010] Put marketing materials into use and collect real-time market feedback to evaluate, adjust and optimize the design version.

[0011] In a preferred embodiment, user behavior data is collected and cleaned, and the user behavior data is analyzed through sentiment analysis to identify the user's emotional state, segment the user groups and extract interest preferences to generate user portraits. The specific process is as follows:

[0012] Obtain user behavior data, including access records, click behaviors, purchase records, product browsing history, comment content, data timestamp, user ID, browsed product ID, purchased product ID, and review text;

[0013] Detect abnormal fluctuations in user behavior data and correct or delete them using box plots or distribution-based anomaly detection algorithms;

[0014] Segment the text in the user behavior data, remove stop words, and process special characters;

[0015] Extract the characteristics of user behavior data, including user visit frequency, purchase frequency, preferred product category, interaction behavior intensity, and sentiment tendency, and use word frequency inverse document frequency to extract user interest topics;

[0016] Use K-means clustering algorithm to cluster users according to their behavioral characteristics to obtain different user groups;

[0017] Use the BERT pre-trained language model to perform sentiment analysis on user reviews and generate each user’s sentiment score for the product design;

[0018] After completing the user portrait construction and sentiment analysis, all user behavior characteristics, sentiment analysis results, and preference information are used as user portraits.

[0019] In a preferred embodiment, the BERT pre-trained language model is used to perform sentiment analysis on user comments to generate each user's sentiment score for the product design. The specific process is as follows:

[0020] Each user’s comment text is input into the BERT model after word segmentation. BERT generates context-aware word vector representation based on the input.

[0021] The sentiment classification task is represented by minimizing the cross entropy loss: , where L is the loss function, It is a real emotional label. is the probability output by the BERT model;

[0022] Based on the BERT model, a sentiment category and a probability value are generated for each comment. The sentiment categories include positive and negative.

[0023] The BERT model is used to assign a sentiment label to each user comment and generate a sentiment score at the same time. The sentiment score is used to indicate the confidence level in the sentiment of the comment.

[0024] In a preferred embodiment, based on the preference information of the user portrait, a generative adversarial network is used to generate different design versions, and the design elements are optimized through a reinforcement learning algorithm in combination with user feedback data. The specific process is as follows:

[0025] Generate user portraits as input to the generative adversarial network to generate different design versions;

[0026] Generate a design graph through a generative adversarial network, which consists of two parts: a generator and a discriminator;

[0027] The generator is used to generate designs that meet the preferences of the user group. The discriminator is used to evaluate the quality of the design and optimize the generator generation process. The objective function of the generator is: , where G represents the generator and the parameters of the generator are is the variable that needs to be learned during the training process, D represents the discriminator, which is used to determine the true probability of the generated samples; z is the input random noise, E is the expected value, is the probability distribution of noise;

[0028] The user profile and preference information will be passed as input to the generator, which will generate different design solutions;

[0029] Based on the click-through rate, number of interactions, purchase conversions and preference information extracted from user portraits, the design elements of the design plan are adjusted, including color, font and layout.

[0030] In a preferred implementation, the A / B test and the multi-armed bandit algorithm are combined to evaluate the design versions and select the best version for adjustment. The specific process is as follows:

[0031] A / B testing compares the click-through rate, conversion rate, and user interaction indicators of different designs to select the design that is most popular with the target group;

[0032] The multi-armed bandit algorithm evaluates the performance of each design version and dynamically selects the design version;

[0033] Set the design version filter target: , where R(a) is the cumulative reward of the selected design version, T is the number of experimental rounds, is the reward for round t;

[0034] The best design version is selected according to the design version screening target, and preliminary design drawings are generated.

[0035] In a preferred embodiment, the marketing copy is generated by combining user portraits and design elements using natural language generation technology, and the design image is optimized through image processing and generative adversarial networks, and the size, tone and layout are automatically adjusted to optimize the display effect of the marketing material. The specific process is as follows:

[0036] Generate copy that matches user needs based on design elements and user sentiment analysis results combined with user portraits;

[0037] Use natural language generation technology to automatically generate copy that meets emotional tendencies and market needs;

[0038] Adjust the tone, vocabulary, and structure based on the generated copy;

[0039] Automatically generate marketing materials based on the generated design drawings and copywriting, combined with the visual preferences of the target user group and the requirements of marketing channels, using image processing and generative adversarial networks to generate marketing materials;

[0040] Optimize the color tone, layout, and size of marketing materials.

[0041] A new product material design system based on AI, used to implement the above-mentioned new product material design method based on AI, comprising:

[0042] User portrait generation module, which is used to collect and clean user behavior data, identify user emotional states through sentiment analysis of user behavior data, segment user groups, extract interest preferences, and generate user portraits;

[0043] The design analysis module is used to generate different design versions based on user portraits and preference information using a generative adversarial network, optimize design elements through a reinforcement learning algorithm based on user feedback data, and evaluate the design versions using an A / B test and a multi-armed bandit algorithm to select the best version for adjustment;

[0044] The marketing material generation module is used to combine user portraits and design elements, use natural language generation technology to generate marketing copy, optimize design drawings through image processing and generative adversarial networks, and automatically adjust the size, color tone and layout to optimize the display effect of marketing materials;

[0045] The delivery optimization module is used to deliver marketing materials and collect real-time market feedback to evaluate, adjust and optimize the design version.

[0046] Technical effects and advantages of the present invention:

[0047] The present invention generates diversified design versions by analyzing user preferences through generative adversarial networks, optimizes design elements through real-time user feedback and reinforcement learning, evaluates multiple design versions in combination with A / B testing and multi-armed bandit algorithms, and dynamically selects the best version for adjustment, thereby improving the design effect. Based on user portraits and design elements, the present invention uses natural language generation technology to automatically generate marketing copy that conforms to emotional tendencies, and optimizes the design drawings in combination with image processing and generative adversarial networks, automatically adjusting the image size, tone and layout to meet the display needs of different platforms and users. Finally, the marketing materials are continuously adjusted and optimized through real-time market feedback to ensure the maximum display effect and market conversion of advertisements, realize intelligent and personalized advertisement creation and delivery, and improve advertising effects and market responsiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The present invention is a flowchart of a new product material design method based on AI.

[0049] Figure 2 This is a structural schematic diagram of an AI-based new product material design system of the present invention. DETAILED DESCRIPTION

[0050] 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.

[0051] Example 1: Figure 1 As shown, a new product material design method based on AI includes the following steps:

[0052] Collect and clean user behavior data, and identify user emotional states through sentiment analysis of user behavior data, segment user groups, extract interest preferences, and generate user portraits;

[0053] Based on user portraits and preference information, we use generative adversarial networks to generate different design versions. Combined with user feedback data, we use reinforcement learning algorithms to optimize design elements. We also use A / B testing and multi-armed bandit algorithms to evaluate design versions and select the best version for adjustment.

[0054] Combine user portraits and design elements, use natural language generation technology to generate marketing copy, optimize design images through image processing and generative adversarial networks, and automatically adjust the size, color tone and layout to optimize the display effect of marketing materials;

[0055] Put marketing materials into use and collect real-time market feedback to evaluate, adjust and optimize the design version.

[0056] Step 1: Build user portraits and conduct preference analysis. Through data mining and analysis techniques, build fine-grained user portraits and conduct preference analysis. Collect, segment and analyze user data, and process and model it based on user behavior characteristics, sentiment analysis and specific preferences. The specific steps are as follows:

[0057] Collect and clean user behavior data from multiple data sources (such as e-commerce platforms, social media, etc.);

[0058] Obtain user access records, click behaviors, purchase records (such as purchase frequency), product browsing history (such as visit frequency), comment content and other data from the platform database, as well as information including data timestamp, user ID, browsed product ID, purchased product ID, review text, etc.;

[0059] Preprocess and clean the data, interpolate missing values, or delete records with serious missing values ​​to avoid the impact of missing data on the analysis results;

[0060] Detect abnormal fluctuations in user behavior data and correct or delete them using box plots or distribution-based anomaly detection algorithms;

[0061] The text in the user behavior data is segmented, stop words are removed, special characters are processed, sentiment information is extracted, and the sentiment information is analyzed to obtain sentiment tendencies, such as comment content, evaluation text, and other texts related to personal intentions.

[0062] The construction of user portraits and the extraction of behavioral features divide users into different groups through clustering and dimensionality reduction methods, and extract behavioral features from them. The specific steps are as follows:

[0063] Based on user behavior data, we extract multiple features, including user visit frequency, purchase frequency, preferred product categories, interaction intensity (such as click duration, page dwell time, etc.), sentiment tendency (sentiment analysis from comments), and use TF-IDF (term frequency-inverse document frequency) to extract user interest topics. For example, we can infer users' preferences for different design styles by analyzing user comments.

[0064] Cluster and segment user groups. Use the K-means clustering algorithm to cluster users according to their behavioral characteristics (such as purchase frequency, access preferences, etc.) to obtain different user groups. The clustering process is as follows: ,in, is the behavior feature vector of the i-th user, is the center of the k-th class of users, and the goal is to minimize the squared error from all users to the class center, and n is the total number of users;

[0065] Each user group represents a group of users with similar behavior patterns. Specifically, the groups may be divided into secondary groups based on users' purchasing power, purchasing cycle, design preferences, etc.

[0066] It should be noted that data such as user purchasing power, purchasing cycle, design preferences, click duration, page dwell time, etc. are all obtained from user behavior data, and the data in the user behavior data are not listed one by one here.

[0067] Use principal component analysis (PCA) to reduce dimensionality to reduce the complexity of data processing and retain the most informative features. PCA performs singular value decomposition (SVD) on the user behavior matrix to obtain the following optimization objectives: , where W is the feature matrix after dimensionality reduction and X is the original behavioral feature matrix. PCA reduces the data dimension while retaining as much variability as possible for subsequent modeling and analysis.

[0068] Perform sentiment analysis and preference extraction, that is, extract sentiment tendencies from user-generated text data and further infer user preferences for design elements. The specific steps are as follows:

[0069] Preference extraction infers users’ preferences for specific design elements (such as color, pattern, font, etc.) through behavioral and sentiment analysis results;

[0070] Use the BERT pre-trained language model to perform sentiment analysis on user comments and generate each user's sentiment score for product design (such as "positive" or "negative"). Through the sentiment analysis model, we can understand the intensity of the emotions expressed by users in their comments, and then reflect the user's design preferences. For example, if a user mentions in a comment that "I really like this modern style packaging", the user can be labeled "positive" and "modern style";

[0071] The sentiment analysis set is combined with the user behavior characteristics to infer the user's preference for specific design elements as preference information. For example, if the user's sentiment inclination towards modern minimalist style is positive, the system will give priority to modern minimalist elements in the design style generated for the user.

[0072] Specifically, each user's comment text is input into the BERT model after word segmentation, and BERT generates context-aware word vector representation based on the input;

[0073] The BERT model can be fine-tuned for sentiment analysis tasks. In sentiment analysis tasks, a classification layer (such as a fully connected layer) is usually added to the output layer of BERT to map the word vector representation generated by BERT to the probability distribution of sentiment categories (such as "positive", "negative", "neutral");

[0074] The goal of the sentiment classification task is to minimize the cross entropy loss: , where L is the loss function, is the true sentiment label (i.e., the positive or negative sentiment category), is the probability output by the BERT model;

[0075] Based on the trained BERT model, a sentiment category (such as "positive" or "negative") and a probability value are generated for each comment. For example, BERT may output a "positive" sentiment probability of 0.85, indicating that the comment is very likely to be a positive comment;

[0076] After inference through the BERT model, each user comment is assigned a sentiment label (such as "positive" or "negative"), and a sentiment score is generated, indicating the model's confidence in the sentiment of the comment. For example: for the comment "The design of this product is really great, simple and modern, I like it!" The BERT model may predict "positive" and the sentiment score is 0.9 (indicating that the model believes that this is a 90% probability of a positive comment).

[0077] After completing the user portrait construction and sentiment analysis, all user behavior characteristics, sentiment analysis results and preference information are used as user portraits and passed as input data to the subsequent design generation and optimization system.

[0078] Through accurate user portrait construction and preference analysis, not only can designs be tailored for each user group, but potential emotional needs can also be captured through sentiment analysis and preference extraction, thereby ensuring that AI can better meet users' personalized needs when generating designs.

[0079] Step 2: Perform preference-based AI design generation and personalized adjustment, that is, generate a design that meets the target user group based on known user needs and design preferences through AI, and adjust and optimize the generated design in real time. That is, use the user portrait and design preferences generated in step 1 as input, generate a preliminary design through a generative model (such as generative adversarial network GANs), and use an algorithm to fine-tune and optimize it in a personalized way, so as to obtain a more attractive product design, packaging and promotional poster that meets user needs. The specific steps are as follows:

[0080] Based on the generated user portraits, multiple design versions are generated using generative adversarial networks (GANs) to meet the personalized needs of different user groups;

[0081] Input data includes: user behavior characteristics (such as purchase frequency, click behavior, design preferences, etc.); user sentiment analysis results (for example, preference for dark colors, simple styles, modern designs, etc.); preference information extracted from user portraits (for example, specific requirements for design elements, such as font type, color tone, etc.);

[0082] Generate design images through generative adversarial networks (GANs). GANs consist of two parts: the generator and the discriminator. The generator is responsible for generating designs, while the discriminator guides the optimization of the generator by judging whether the design meets the given preference requirements.

[0083] In this process, the input data will be used to generate multiple design versions, each version corresponds to a different design style, element combination and color scheme. The goal of the generator is to generate designs that meet the preferences of the user group as much as possible, while the discriminator continuously evaluates the quality and diversity of the design to help the generator optimize the generation process. The objective function of the generator is: , where G represents the generator and the parameters of the generator are is the variable that needs to be learned during the training process, D represents the discriminator, which is used to determine the true probability of the generated samples; z is the input random noise (or the encoding of user preferences), and E is the expected value. is the probability distribution of noise, and the goal of the generator is to make the discriminator unable to distinguish the difference between the generated data and the real data;

[0084] The user profile and preference information are passed as input to the generator, which generates a variety of design solutions based on this information;

[0085] Input real-time user feedback (such as click-through rate, purchase conversion rate, user reviews, etc.). These feedback data can reflect the acceptance of the current design and user preferences;

[0086] Based on user preferences and real-time feedback, the generated preliminary design is personalized to ensure that the design better meets the needs of the target group;

[0087] Based on user feedback (such as click-through rate, number of interactions, purchase conversion, etc.) and preference information extracted from user portraits, the design solution is fine-tuned through an optimization algorithm. The optimization method can use an adjustment strategy based on reinforcement learning to fine-tune the design solution through a reinforcement learning algorithm. The goal of each optimization is to adjust design elements (such as color, font, layout, etc.) to meet user needs based on current user feedback;

[0088] For example, if a design has a low click-through rate, the AI ​​system will regenerate an optimized design by adjusting the color tone, font, or other design elements. This process can be optimized through the Q-learning model, where the design adjustment is regarded as an action. The goal of design optimization is to maximize user acceptance (i.e., the reward function). The reward function can be designed as: , where R(a) is the reward for design adjustment, a is the specific design adjustment operation, , are weight parameters, which respectively represent the importance of click-through rate (CR), purchase conversion rate (PR) and number of user evaluations (ET) in design optimization.

[0089] When optimizing, the AI ​​system will evaluate the fit between each design version and the target user group and select the appropriate design version;

[0090] After generating multiple design versions (including optimized and fine-tuned designs), the best design is selected through optimization algorithms for final promotion and release, using user behavior data (such as click volume, conversion rate, purchase data) and preference data as evaluation criteria for the selected designs;

[0091] For the multiple design versions generated, the A / B test and the multi-armed bandit algorithm are combined for evaluation to select the best design version; the A / B test is to present different design versions to the user group separately, and select the most expressive design through statistical analysis, and determine which design is most popular with the target group by comparing the click-through rate, conversion rate, user interaction and other indicators of different designs; the multi-armed bandit algorithm dynamically selects the most successful design version for promotion through the balance of exploration and utilization. The algorithm will evaluate the long-term performance of each design version;

[0092] The final design version screening goals are: , where R(a) is the cumulative reward of the selected design version, T is the number of experimental rounds, is the reward for round t (such as click-through rate, conversion rate, etc.).

[0093] Based on the evaluation results, the design version with the best performance is selected to generate preliminary design drawings, and then further optimized and applied.

[0094] Step 3: Generate and automatically optimize intelligent marketing materials, that is, generate intelligent marketing materials and automatically optimize them. Based on the personalized design generated previously, convert them into materials with market promotion value, and optimize the marketing effect through automated means. The specific steps are as follows:

[0095] Optimizing marketing effects involves not only the generation of design drawings, but also the automatic writing of copywriting, the adaptation of marketing channels, etc., to ensure the maximum effect of promotional content;

[0096] Obtain the design elements (such as color, font, pattern, etc.) passed in step 2, and generate copy that matches user needs based on the design elements and the user's sentiment analysis results, combined with the user portrait;

[0097] Use natural language generation (NLG) technology to automatically generate copy that meets emotional tendencies and market needs. The objective function of copy generation is: ,in, The loss function generated for the copywriting, is the generation probability of the model for the i-th word, is the i-th word in the copy, x is the input design features and user preferences, is a regularization term used to control the degree of matching between the generated copy and the target user’s preferences;

[0098] Based on the generated copy, the AI ​​system can further adjust the tone, vocabulary, structure, etc. to make it more in line with the communication habits of the target user group;

[0099] Automatically generate marketing materials (such as e-commerce ads, social media ads, promotional posters, etc.) based on the generated design drawings and copywriting, combined with the visual preferences of the target user group and the requirements of marketing channels;

[0100] The design drawings (including the preliminary design and the adjusted version) passed from step 2, as well as the generated marketing copy;

[0101] Using technologies such as image processing and generative adversarial networks (GANs), we can automatically generate marketing materials that meet advertising specifications based on platform requirements and design drawings. The AI ​​system will automatically adjust the design to suit the platform environment based on the display requirements of different platforms (such as aspect ratio, image size, color tone requirements, etc.);

[0102] Use image processing technology and generative adversarial networks to optimize generated marketing materials. GANs generates ad designs that meet visual preferences through the adversarial process between the generator and the discriminator, and optimizes them according to the platform's requirements. The generator generates preliminary materials based on the user's design preferences, while the discriminator helps the generator optimize the design by judging whether the materials meet the target standards.

[0103] Optimize color tones and adjust according to target user preferences to ensure that the design style and color meet the emotional needs of the audience (for example, young groups may prefer bright colors, while mature user groups may prefer soft colors);

[0104] Perform layout optimization, adjust the layout of images according to platform requirements, and ensure that important content (such as product display, brand logo, etc.) is properly displayed in the design;

[0105] Resize and automatically adjust the image size and proportion according to the specifications of different platforms to meet the platform display requirements;

[0106] AI generates the final marketing materials based on the optimized design drawings and copywriting, and formats them to suit the display requirements of different platforms.

[0107] Market and user demands are changing dynamically, so design and marketing materials also need to be highly adaptable. The real-time feedback mechanism can quickly adjust design plans, advertising materials, etc. through instant analysis of market reactions, thereby improving the match between design and the market.

[0108] Step 4: Conduct real-time feedback and design optimization, put marketing materials into use, and continuously optimize the generated design materials and marketing strategies through real-time market feedback. AI can adjust and optimize the design in real time based on these feedbacks. The specific steps are as follows:

[0109] Put marketing materials into use and obtain market feedback data corresponding to the generated marketing materials (including design drawings and copywriting), including ad click-through rate, purchase conversion rate, dwell time, comment content, social media interaction, etc.;

[0110] Clean the collected feedback data, remove irrelevant data and outliers, pre-process the data using methods based on frequency analysis and fluctuation detection, and extract features that are valuable for design optimization from the data. For example, an ad with a high click-through rate may indicate that the design has generated strong interest among users and is therefore more in line with user preferences.

[0111] According to the market feedback data, the design plan is optimized and adjusted in real time to improve the performance of the advertisement. The design plan (such as images, copywriting, advertising layout, etc.) generated from the previous steps is obtained through the processed real-time market feedback data.

[0112] Through reinforcement learning algorithms, AI can adjust the design based on market feedback data. Each design adjustment (such as color, layout, font, etc.) corresponds to an action. AI accumulates reward values ​​and optimizes the design by constantly trying different actions.

[0113] Optimize the performance of ads by adjusting design elements (such as color, layout, font, etc.). Adjust the color, pattern, text content, etc. of the design based on market feedback to increase the attractiveness of the ad. For example, if a design has a low click-through rate, the system will regenerate an optimized version of the design by changing the color or font. The goals of the optimization process are: ,in, is the optimized design loss, To adjust the operation for the design, Optimization goals for each operation;

[0114] The optimized design will be passed to the subsequent screening and selection stages to ensure that the design can be adjusted to the best state based on the feedback.

[0115] Use methods such as A / B testing to evaluate multiple design versions based on market feedback, select the best performing version for promotion, fine-tune and re-optimize the selected best design version to further improve the performance of the design. This can be done by readjusting some design details (such as fonts, colors, layout, etc.), and using the selected best design version as the final material to pass it on to the actual promotion and execution stage.

[0116] It should be noted that the thresholds involved in the embodiments can be determined according to specific scenarios and requirements.

[0117] The present invention generates diversified design versions by analyzing user preferences through generative adversarial networks, optimizes design elements through real-time user feedback and reinforcement learning, evaluates multiple design versions in combination with A / B testing and multi-armed bandit algorithms, and dynamically selects the best version for adjustment, thereby improving the design effect. Based on user portraits and design elements, the present invention uses natural language generation technology to automatically generate marketing copy that conforms to emotional tendencies, and optimizes the design drawings in combination with image processing and generative adversarial networks, automatically adjusting the image size, tone and layout to meet the display needs of different platforms and users. Finally, the marketing materials are continuously adjusted and optimized through real-time market feedback to ensure the maximum display effect and market conversion of marketing, realize intelligent and personalized advertising creation and delivery, and improve advertising effect and market responsiveness.

[0118] Embodiment 2: A new product material design system based on AI, such as Figure 2 As shown, specifically including:

[0119] User portrait generation module, which is used to collect and clean user behavior data, identify user emotional states through sentiment analysis of user behavior data, segment user groups, extract interest preferences, and generate user portraits;

[0120] The design analysis module is used to generate different design versions based on user portraits and preference information using a generative adversarial network, optimize design elements through a reinforcement learning algorithm based on user feedback data, and evaluate the design versions using an A / B test and a multi-armed bandit algorithm to select the best version for adjustment;

[0121] The marketing material generation module is used to combine user portraits and design elements, use natural language generation technology to generate marketing copy, optimize design drawings through image processing and generative adversarial networks, and automatically adjust the size, color tone and layout to optimize the display effect of marketing materials;

[0122] The delivery optimization module is used to deliver marketing materials and collect real-time market feedback to evaluate, adjust and optimize the design version.

[0123] The above formulas are all dimensionless and calculated numerically. Specific dimension removal can be achieved by various means such as standardization, which will not be elaborated here. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0124] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium may be a magnetic medium (e.g., a floppy disk, an ATA hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium may be a solid-state ATA hard disk.

[0125] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0126] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or 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. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0127] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0128] 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.

[0129] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0130] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present 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. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A new product material design method based on AI, characterized in that: The steps include: Collect and clean user behavior data, and identify user emotional states through sentiment analysis of user behavior data, segment user groups, extract interest preferences, and generate user portraits; Based on the generated user portraits, different design versions are generated using a generative adversarial network. Combined with user feedback data, the design elements are optimized using a reinforcement learning algorithm. The design versions are evaluated using an A / B test and a multi-armed bandit algorithm, and the best version is selected for adjustment. Combine user portraits and design elements, use natural language generation technology to generate marketing copy, optimize design images through image processing and generative adversarial networks, and automatically adjust the size, color tone and layout to optimize the display effect of marketing materials; Put marketing materials into use and collect real-time market feedback to evaluate, adjust and optimize the design version; Based on the generated user portraits, different design versions are generated using a generative adversarial network. Combined with user feedback data, the design elements are optimized using a reinforcement learning algorithm. The specific process is as follows: Generate user portraits as input to the generative adversarial network to generate different design versions; Generate a design graph through a generative adversarial network, which consists of two parts: a generator and a discriminator; The generator is used to generate designs that meet the preferences of the user group. The discriminator is used to evaluate the quality of the design and optimize the generator generation process. The objective function of the generator is: , where G represents the generator and the parameters of the generator are is the variable that needs to be learned during the training process, D represents the discriminator, which is used to determine the true probability of the generated samples; z is the input random noise, E is the expected value, is the probability distribution of noise; The user profile and preference information are passed as input to the generator, which generates different design solutions; Adjust design elements of the design plan based on click-through rate, number of interactions, purchase conversions, and preference information extracted from user portraits, including color, fonts, and layout; Combine A / B testing and the multi-armed bandit algorithm to evaluate the design versions and select the best version for adjustment. The specific process is as follows: A / B testing compares the click-through rate, conversion rate, and user interaction indicators of different designs to select the design that is most popular with the target group; The multi-armed bandit algorithm evaluates the performance of each design version and dynamically selects the design version; Set the design version filter target: , where R(a) is the cumulative reward of the selected design version, T is the number of experimental rounds, is the reward for round t; The best design version is selected according to the design version screening target, and preliminary design drawings are generated.

2. The AI-based new product material design method according to claim 1, characterized in that: Collect and clean user behavior data, and identify user emotional states through sentiment analysis of user behavior data, segment user groups, extract interest preferences, and generate user portraits. The specific process is as follows: Obtain user behavior data, including access records, click behaviors, purchase records, product browsing history, comment content, data timestamp, user ID, browsed product ID, purchased product ID, and review text; Detect abnormal fluctuations in user behavior data and correct or delete them using box plots or distribution-based anomaly detection algorithms; Segment the text in the user behavior data, remove stop words, and process special characters; Extract the characteristics of user behavior data, including user visit frequency, purchase frequency, preferred product category, interaction behavior intensity, and sentiment tendency, and use word frequency inverse document frequency to extract user interest topics; Use K-means clustering algorithm to cluster users according to their behavioral characteristics to obtain different user groups; Use the BERT model to perform sentiment analysis on user comments and generate each user’s sentiment score for the product design; After completing the user portrait construction and sentiment analysis, all user behavior characteristics, sentiment analysis results, and preference information are used as user portraits.

3. The AI-based new product material design method according to claim 2, characterized in that: Use the BERT model to perform sentiment analysis on user comments and generate each user's sentiment score for the product design. The specific process is as follows: Each user’s comment text is input into the BERT model after word segmentation. The BERT model generates context-aware word vector representation based on the input. The sentiment classification task is represented by minimizing the cross entropy loss: , where L is the loss function, It is a real emotional label. is the probability output by the BERT model; Based on the BERT model, a sentiment category and a probability value are generated for each comment. The sentiment categories include positive and negative. The BERT model is used to assign a sentiment label to each user comment and generate a sentiment score at the same time. The sentiment score is used to indicate the confidence level in the sentiment of the comment.

4. The AI-based new product material design method according to claim 3, characterized in that: Combining user portraits and design elements, we use natural language generation technology to generate marketing copy, optimize the design through image processing and generative adversarial networks, and automatically adjust the size, tone and layout to optimize the display effect of marketing materials. The specific process is as follows: Generate copy that matches user needs based on design elements and user sentiment analysis results combined with user portraits; Use natural language generation technology to automatically generate copy that meets emotional tendencies and market needs; Adjust the tone, vocabulary, and structure based on the generated copy; Automatically generate marketing materials based on the generated design drawings and copywriting, combined with the visual preferences of the target user group and the requirements of marketing channels, using image processing and generative adversarial networks to generate marketing materials; Optimize the color tone, layout, and size of marketing materials.

5. A new product material design system based on AI, used to implement a new product material design method based on AI as claimed in any one of claims 1 to 4, characterized in that: include: User portrait generation module, which is used to collect and clean user behavior data, identify user emotional states through sentiment analysis of user behavior data, segment user groups, extract interest preferences, and generate user portraits; The design analysis module is used to generate different design versions based on the preference information of user portraits using a generative adversarial network, optimize design elements through a reinforcement learning algorithm based on user feedback data, and evaluate the design versions using an A / B test and a multi-armed bandit algorithm to select the best version for adjustment; The marketing material generation module is used to combine user portraits and design elements, use natural language generation technology to generate marketing copy, optimize design drawings through image processing and generative adversarial networks, and automatically adjust the size, color tone and layout to optimize the display effect of marketing materials; The delivery optimization module is used to deliver marketing materials and collect real-time market feedback to evaluate, adjust and optimize the design version.

Images