A product design evaluation system based on consumer style preferences

Abstract

This invention discloses a product design evaluation system based on consumer style preferences. The system comprises a display and management subsystem, an analysis and modeling subsystem, and an evaluation and optimization subsystem. The display and management subsystem displays a product style knowledge graph and stores a database of consumer style preferences. The analysis and modeling subsystem automatically identifies the style of product proposals and then compares the identified product style with the consumer's preferred style in the consumer style preference database. The evaluation and optimization subsystem performs global optimization of product concept design proposals. This system provides businesses with targeted design guidance to improve product market acceptance.

Description

Technical Field

[0001] The present invention relates to the field of automated evaluation of product appearance designs, and in particular to a product appearance design evaluation system based on consumer style preferences. Background Art

[0002] Product style primarily refers to the characteristics exhibited by man-made objects. It is a designer's unique way of doing things. The common features created by the designer and appearing in their products are used to identify the designer's personal style. Style can be expanded by increasing the number of common features across products. Unlike works of art, products typically have practical functions and are produced for market sale. Therefore, the study of product style has expanded into fields such as ergonomics and marketing. Product designers must understand how consumers perceive product style attributes and effectively enhance these attributes to stimulate positive emotions and purchase desire. Products with strong style attributes can deepen the polarization of consumer evaluations. Product style highlights consumers' self-expression and is more pronounced when it expresses an individual rather than a group.

[0003] As consumption enters an era of differentiation, consumer aesthetics are becoming increasingly distinct. Chinese consumers are gradually becoming segmented, characterized by significant differences between groups and minimal differences within them. In this era of differentiation, consumers no longer simply follow the principle of copying or comparing with others, focusing instead on identifying their own interests and developing a lifestyle that suits them. Currently, a product's aesthetic is no longer determined by the tastes of designers and decision-makers, but rather by the aesthetics of its target consumers.

[0004] The following are some common examples of automated style assessment. For example, in architectural style classification, a convolutional neural network (CNN) is used as a predictive model to classify architectural styles into 17 categories, providing consumers with style recommendations and helping to improve the visual quality of buildings. In interior design style classification, targeted interior design style recommendations can be implemented within a 3D room virtual reality platform. In the automotive design field, machine learning is used to classify the style of car front faces. This aims to avoid duplication of car styles, effectively identify the consistency of brand styles across car front faces, and help infer and perceive brand characteristics in car styling, enabling designers to quickly classify them.

[0005] The design scheme evaluation method based on the popular style aesthetic preferences can effectively improve the attractiveness of the design, help product design companies more accurately capture the style preferences of the target consumer groups, so as to differentiate themselves from competitors, attract specific customer groups, and promote product portfolio, brand building and marketing strategies. It is crucial for designers and marketers to formulate effective strategies to meet the ever-changing customer needs.

[0006] The development of artificial intelligence has driven the automated generation of product design solutions, such as text-to-image and 2D-image-to-image, using stable diffusion models. Faced with tens of thousands of generated design options, it is crucial to identify the optimal solution that meets the style preferences of each consumer group.

[0007] However, there is currently no product appearance design evaluation method based on consumer style and aesthetic preferences in the market, which has led to the developed products not receiving a good response from the consumer market, resulting in losses for companies and serious economic losses. Summary of the Invention

[0008] In view of the above, the purpose of the present invention is to provide a product appearance design evaluation system based on consumer style preferences, which provides companies with targeted design guidance to improve the market acceptance of products.

[0009] The present invention includes a display and management subsystem, an analysis and modeling subsystem, and an evaluation and optimization subsystem.

[0010] The display and management subsystem is used to display the product style knowledge graph and store the consumer style preference database;

[0011] The analysis and modeling subsystem is used to automatically identify the style of product solutions, and then compare the identified product style with the consumer's preferred style in the consumer style preference database;

[0012] The evaluation and optimization subsystem is used for global optimization of product concept design solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0014] Figure 1 This is a schematic diagram of the structure of a product design scheme evaluation system based on popular style aesthetic preferences provided by an embodiment;

[0015] Figure 2 This is the framework diagram of the Resnet model;

[0016] Figure 3 It is a global optimization framework diagram of product concept design based on the stable diffusion model; DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

[0018] Figure 1 FIG. 1 is a schematic diagram of a product appearance design evaluation system based on consumer style preferences provided by an embodiment. Figure 1 As shown, the embodiment provides a product appearance design evaluation system based on consumer style preferences, including a display and management subsystem, an analysis and modeling subsystem, and an evaluation and optimization subsystem.

[0019] Among them, the display and management subsystem is used to display the product style knowledge graph and store the specific consumer style preference database.

[0020] Among them, the analysis and modeling subsystem is used to automatically identify the style of product solutions, and then compare the identified product style with the consumer's preferred style in the consumer style preference library.

[0021] Among them, the evaluation and optimization subsystem uses a stable diffusion model to perform global optimization of product concept design schemes.

[0022] The implementation process of the display and management subsystem provided in this embodiment is described as follows: It includes a product style knowledge graph display module, which contains the name and description of the product's parent style, the name and description of the product's child style, a vector relationship map of multiple historical product styles, heat map differences of the parent style across four dimensions: shape, color, material, context, and graphic analogy, and a lexical semantic map of the parent style. It also includes a specific consumer style preference database, which contains the preference percentages for various styles when combining various consumer labels (gender, age, education level, income).

[0023] The specific consumer style preference database module has the functions of entering, modifying, searching, deleting, and inferring consumer style preference data; indicators such as consumer labels (gender, age, education level, and income) influence consumer choices. Furthermore, this embodiment uses the logit function to calculate the probability of different consumers choosing different styles. By obtaining consumer label data f, collecting historical data, and fitting the data, different coefficient values ​​β for the four parameters of gender, age, education level, and income are calculated (i.e., the utility value of the popular style aesthetic of a specific group is calculated). The utility value calculation method used in this embodiment is described as follows:

[0024]

[0025] in represents the utility value of consumer group c choosing style s, Represents the proportion coefficients of gender, age, education level and income respectively. They represent the parameters of gender, age, education level, and income of different types of consumers. The calculated probability of the logit model is as follows:

[0026]

[0027] MS c,s represents the probability that customer c chooses style s, where the numerator represents the utility value of the passenger choosing style s, and the denominator represents the utility value of the passenger choosing all styles.

[0028] The machine learning style automatic identification module of the product solution described in the embodiment of this application adopts a style feature identification and classification algorithm based on supervised learning. Specifically, Resnet 50 is used to identify the product style of the product solution, and fine-grained style classification is performed to obtain the main style and secondary style labels corresponding to the product solution. Figure 2 The ResNet50 shown consists of five stages and a fully connected classification layer. The network input is a preprocessed product sample image. The features of the product sample image are extracted through convolution operations between the image input and the convolution kernel in the convolution layer. Convolution is the primary operation for feature extraction. The first stage is a common CONV+BN+ReLU+MAXPOOL convolution block, while stages 2 through 5 are residual convolution blocks. The output of the final fully connected layer, FC, is a multidimensional vector representing nine different styles. Each convolution layer is followed by a ReLU activation function.

[0029] f(x)=max(0,x) (3) The output formula of a neuron is:

[0030]

[0031] Where W i-1,k represents the weight k of i=1, X i-1,k Represents the input of neuron k in layer i=1. b i-1 Represents the output of neuron k in the i=1th layer, thereby realizing style classification of product sample images and obtaining the main style and secondary style labels corresponding to the product sample images.

[0032] The formula of the loss function is:

[0033]

[0034] The sample (x (i) ,y (i) ) represents the sample label of sample i, m represents the number of training samples, h w (x (i) ) indicates a pseudo function.

[0035] The product style is compared with the consumer's preferred style in the consumer style preference library to identify the style S identified by machine learning. m1 The probability that consumer c chooses style S in the same consumer style preference database C1 The comparison of the probability of is as follows:

[0036]

[0037] Consistency=1-Inconsistency N=2

[0038] Where N = the number of styles compared. Xjn = the X value calculated for style index n on design j. Xmax n is the style indicator n in the identified product solution S m1 and S in the consumer preference database C1 The maximum value on the style data, J is the product solution to be evaluated.

[0039] The global optimization module of the product solution described in the embodiment of this application includes a text encoder, an information image creator and an image decoder. Figure 3 shown.

[0040] The text decoder uses CLIPText for the text encoding module, inputs the style name and style probability information that need to be adjusted, and converts the text information into a matrix of digital expression.

[0041] The information image creator uses Attention U-Net to gradually diffuse the information converted into the latent space, including the ability to construct the underlying U-Net mainly from two-dimensional convolutional layers, and the ability to further focus the target on the most perceptually relevant bits. The formula is as follows:

[0042]

[0043] Stable diffusion model of neural agents ζ θ (z t ,t) is implemented in the form of a time-conditional UNet. During the training process, z can be efficiently obtained from ε t , thereby decoding the digital matrix into the image space as the model input.

[0044] The image decoder decodes the operation result of the latent space into the actual picture dimension, mainly using a variational encoder, and draws a new image based on the information passed from the image information creator.

[0045] The embodiment provides a product design evaluation system based on consumer style preferences, innovatively addressing the need for product evaluation and design methods that meet the aesthetic needs of different consumers in an era of differentiation. The embodiment addresses the significant challenges of existing technologies, such as their inability to comprehensively manage product styles and objectively evaluate the market potential of product design solutions.

[0046] The display and management subsystem provided in the embodiment realizes the product style knowledge and consumer style preference database, which solves the shortcoming of the existing technology that it cannot accurately capture the style preferences of specific consumer groups. It displays the style name and introduction of the product's parent, the style name and introduction of the product's child, the vector relationship map of multiple historical product styles, the heat map differences of the parent style in four dimensions of shape color elements, material elements, situational elements, and graphic analogy elements, the lexical semantic map of the parent style, and other content, thereby improving the product style knowledge.

[0047] The analysis and modeling subsystem provided in this embodiment implements automatic machine learning-based product style recognition. Using product style decoupling technology, it identifies both primary and secondary style characteristics within a product. This recognition of product characteristics is then compared with a database of consumer style preferences, enabling automated product scoring based on group consumer style preferences and identifying the most promising design solutions.

[0048] The evaluation and optimization subsystem provided in the embodiment realizes the generation of a two-dimensional concept design scheme with continuity, novelty and compliance with the evaluation index cognitive system based on the product concept design using the stable diffusion model method, and can further score the product based on the style preferences of group consumers, thereby realizing continuous iterative optimization of the product design scheme.

[0049] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A product appearance design evaluation system based on consumer style preferences, characterized by: Including display and management subsystem, analysis and modeling subsystem, evaluation and optimization subsystem; The display and management subsystem is used to display the product style knowledge graph and store the consumer style preference database; The analysis and modeling subsystem is used to automatically identify the style of product solutions, and then compare the identified product style with the consumer's preferred style in the consumer style preference database; The evaluation and optimization subsystem is used for global optimization of product concept design solutions; The evaluation and optimization subsystem includes a global optimization module for product solutions, which consists of a text encoder, an information image creator, and an image decoder; The text encoder uses CLIPText for the text encoding module, inputs the style name and style probability information to be adjusted, and converts the text information into a matrix of digital expression; The information image creator uses Attention U-Net to gradually diffuse the information converted into the latent space; The image decoder decodes the operation result of the latent space into the actual picture dimension, adopts the variational encoder, and draws a new image according to the information passed from the image information creator.

2. A product appearance design evaluation system based on consumer style preferences according to claim 1, characterized in that: The display and management subsystem includes a product style knowledge graph display module, which contains the product's parent style name and introduction, the product's child style name and introduction, vector relationship graphs of multiple historical product styles, and heat map differences of the parent style in four dimensions: shape color elements, material elements, situational elements, and graphic analogy elements, as well as a lexical semantic graph of the parent style.

3. The product appearance design evaluation system based on consumer style preference according to claim 1, characterized in that: The display and management subsystem includes the preference proportions of various styles corresponding to the combination of multiple consumer tags.

4. A product appearance design evaluation system based on consumer style preferences according to claim 3, characterized in that: The logit function is used to calculate the probability of different consumers choosing different types of styles.

5. The product appearance design evaluation system based on consumer style preference according to claim 1, characterized in that: The analysis and modeling subsystem includes a machine learning style automatic identification module for product solutions, which adopts a style feature recognition and classification algorithm based on supervised learning.

6. A product appearance design evaluation system based on consumer style preferences according to claim 1 or 5, characterized in that: The difference comparison between the product style and the consumer's preferred style in the consumer style preference library is a comparison of the probability of the style identified by machine learning and the probability of the consumer's selected style in the consumer style preference library.

Images