Complementary Item Recommendation via Mixed-Query Latent Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation engines struggle to provide complementary item suggestions that cater to diverse user needs and preferences, particularly in e-commerce platforms like global fashion, due to annotation bias and lack of generalization, leading to inconsistent and biased recommendations.
Innovation Solution
A system and method utilizing a transformer encoder and decoder to generate latent representations of user queries, incorporating user preferences and compatibility criteria, enabling retrieval of complementary items from a gallery without requiring positive and negative annotations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing recommendation models use annotated compatible data for model fitting, then they can provide complementary item recommendations, but they become biased towards annotator preference and cannot be generalized
Solution Approach 1:
The patent extracts the compatibility modeling function from manual annotations and implements it through a Siamese network that learns compatibility patterns automatically from image data, removing the source of annotator bias while preserving the ability to provide personalized recommendations
Solution Approach 2:
The patent replaces the manual annotation mechanism with an automated computer vision system using Siamese networks and deep learning models that can process and understand image data independently of human annotator preferences, enabling unbiased and generalizable recommendations
2Adaptability or versatility
If existing methods use deterministic approaches for complementary item retrieval, then they can provide consistent recommendations, but they fail to cater to diverse user preferences and contradictions
Solution Approach 1:
The patent introduces dynamic user preference modeling where the recommendation system adapts to individual user contradictions and evolving tastes through iterative learning, allowing the same deterministic framework to produce diverse and personalized recommendations for different users
Solution Approach 2:
The patent changes the parameters of the recommendation model by incorporating user-specific contradiction data and preference patterns, allowing the deterministic algorithm to generate varied recommendations tailored to each user's unique taste while maintaining operational consistency
3Productivity
If existing recommendation engines rely on individual perception of compatibility, then they can provide recommendations, but it is difficult to obtain ground truth label for multiple user requests
Solution Approach 1:
The patent enables the system to self-evaluate compatibility through the Siamese network architecture that automatically computes compatibility scores between items and user profiles without requiring external ground truth labels, allowing rapid generation of personalized recommendations
Solution Approach 2:
The patent introduces compatibility scores as an intermediary metric that bridges the gap between individual perception and measurable accuracy, allowing the system to quantify and optimize recommendation quality without direct human validation for each request
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
This disclosure relates generally to method and system to recommend complementary items by generating candidate items. Complementary recommendation is an important problem in e-commerce platforms that gives compatible suggestions to the users based on recent purchase and pre-selected items. The method receives a mixed query as input from a user to obtain complementary target candidate image items. The mixed query includes a set of product category images along with product category label preselected by the user. Further, for the mixed query a target latent representation for the combined latent representation is generated. Then, a set of compatible complementary target candidate image items is retrieved for the one or more target candidate images from a retrieval gallery. Finally, the set of compatible complementary target candidate image items are displayed on electronic device of the user.