Generative AI Product Search via Style-Aware Knowledge Graph
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional e-commerce search systems fail to help customers find products in specific scenarios due to limitations in image representation and complex search query requirements, leading to increased computing resources usage and reduced conversion rates.
Innovation Solution
The implementation of generative AI to optimize product search queries by determining relationships between products using a knowledge graph, generating textual prompts for a text-to-image diffusion model, and ranking results based on color consistency and customer preferences, allowing for efficient visualization of products in desired scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional search query systems are used, then customers can perform simple product searches, but customers cannot find products in specific scenarios and search queries become too complex
Solution Approach 1:
The patent introduces an image generation system as an intermediary between the customer and product database. Instead of requiring customers to formulate complex text queries, they upload a reference image and the system generates similar product images automatically, mediating the search process between simple user input and accurate product retrieval
Solution Approach 2:
The patent replaces the traditional text-based mechanical search query system with an image-based generative system. Customers interact through image upload rather than text input, and the system uses image generation models (like Stable Diffusion) instead of text-based search algorithms, substituting the entire search mechanism paradigm
2Productivity
If conventional search systems are used, then computing resources are consumed for processing queries, but conversion rates decrease due to inability to show products in specific scenarios
Solution Approach 1:
The patent performs preliminary actions by pre-processing product images through style transfer and generating style embeddings before the actual search. The system pre-generates style-specific product representations and stores them in the database, so that during search, it only needs to match styles rather than process full images, reducing real-time computing resources
Solution Approach 2:
The patent creates style copies or style embeddings of product images rather than storing and processing the entire original image data. The generative model learns style representations and generates new images based on these compact style copies, significantly reducing the computational burden of image processing while maintaining visual fidelity
3Reliability
If product images are not available in specific scenarios, then storage requirements are reduced, but customer satisfaction and conversions decrease
Solution Approach 1:
The patent generates synthetic product images in various scenarios using style transfer and generative models based on existing product photos and style references. Instead of storing multiple physical product images in different settings, the system creates digital copies with different styles and backgrounds on-demand, reducing storage requirements while maintaining visualization accuracy
Solution Approach 2:
The patent changes the style parameters of existing product images through generative models. By adjusting style embeddings and generating variations with different color palettes, textures, and backgrounds, the system creates multiple scenario-specific visualizations from a single source image, eliminating the need to store multiple physical product photos
Data Source
AI summary
Methods and systems are provided for using generative AI to optimize product search queries. In embodiments described herein, product descriptions and product images for a plurality of products are obtained. A multi-modal style classification model classifies each product into a corresponding style of a plurality of styles based on the product's product description and product image. Relationships of each product to other products in the plurality of products are stored in a knowledge graph based on the corresponding style of each product and the corresponding product description of each product. An image is generated by a text-to-image diffusion model with a set of products of the plurality of products based on the relationships of each product of the plurality of products to other products in the plurality of products.