Online Product Recommendation via Image Color Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current online product recommendation technologies face challenges in accurately recommending similar products due to reliance on variable descriptive text, limited understanding of product content, and issues with data sparsity and cold starts, leading to inconsistent and inaccurate recommendations.
Innovation Solution
A process that utilizes image search technologies to extract and compare color features of product images, creating a high-dimensional vector representation of each image to establish an inverted index, allowing for efficient retrieval of similar products by measuring cosine similarity between query and candidate images, thereby reducing computation time and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional text-based recommendation technology is used, then the system can establish product inter-relationships through descriptive text, but the recommendation accuracy deteriorates due to variable quality of seller-written descriptions and fraudulent conduct
Solution Approach 1:
The patent replaces the mechanical text-based recommendation system with an image-based visual recognition system. Instead of relying on seller-written descriptive text to establish product relationships, the system uses image processing algorithms to extract visual features and compute similarity, thereby eliminating the impact of fraudulent or low-quality text descriptions on recommendation accuracy.
Solution Approach 2:
The patent introduces an image feature extraction and comparison mechanism as an intermediary between products and recommendation results. This intermediary process involves extracting visual features from product images, building an inverted index of these features, and using cosine similarity computation to determine product relationships, thereby mediating the recommendation process in a way that is independent of unreliable text descriptions.
2Productivity
If user behavior-based recommendation technology is used, then conversion rates can be increased through simulating consumer habitual behavior, but the system cannot understand product content leading to uncontrollable and inconsistent recommendation results
Solution Approach 1:
The patent changes the fundamental parameter used for recommendations from user behavior data to image visual features. By computing cosine similarity between image feature vectors, the system maintains consistent and controllable recommendation results based on actual product content rather than historical user behavior patterns, which vary by user type and time.
Solution Approach 2:
The patent substitutes the user behavior-based recommendation mechanism with an image-based visual similarity mechanism. This replacement allows the system to understand and compare actual product content through image features, providing consistent and explainable recommendations based on visual characteristics rather than opaque behavioral simulations.
3Quantity of substance
If user behavior-based recommendation technology is used, then the system can leverage historical data to find consumer inclinations, but it encounters data sparsity and cold start problems affecting recommendation quality
Solution Approach 1:
The patent performs preliminary action by pre-extracting visual features from all product images and building an inverted index before any user interaction occurs. This pre-processing step creates a ready-to-use recommendation framework that can immediately provide accurate recommendations without requiring historical user behavior data, thereby solving the cold start and data sparsity problems.
4Stability of the object's composition
If manual product selection for shopping advice columns is used, then products can be presented consistent with design and style themes, but the process expends large amounts of manual effort and does not ensure definite recall rate
Solution Approach 1:
The patent enables the system to automatically select and recommend products by computing visual similarity between seed products and the product catalog. This self-service mechanism eliminates manual effort in product selection while maintaining consistency with design and style themes, as the image-based similarity computation naturally captures visual characteristics such as color, pattern, and style.
Data Source
AI summary
Embodiments of the present application relate to a method for recommending online products, a system for recommending online products, and a computer program product for recommending online products. A method for recommending online products is provided. The method includes specifying a main product zone of a query product image, dividing the main product zone into a plurality of local zones, extracting color features from each local zone, looking up candidate recommended product images sharing common characteristics with a query product image based on the color features of each local zone, matching, among the found candidate recommended product images, product images that are similar in terms of color matching to the query product image, and regarding the matched product images as recommended product images.