Triplet Network for Complementary Item Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current e-commerce systems fail to effectively recommend complementary items for users, especially for new products without past sales data, and struggle with low attribute coverage, particularly in categories like home decor, due to reliance on past purchase data and attribute matching which can be non-uniform or incorrect.
Innovation Solution
A system that generates multimodal embeddings for items based on their attributes, using a triplet network to predict and recommend complementary items by positioning anchor items, positive, and negative items in a shared embedding space, ensuring accurate recommendations even for new products.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current systems use past purchase data to generate recommendations, then recommendations are based on actual user behavior, but new products without past sales data cannot be associated with items in a user's cart
Solution Approach 1:
The patent introduces an attribute-based intermediary system that mediates between items and recommendations. Instead of directly relying on purchase data, the system uses attribute vectors and similarity calculations as intermediaries to connect items, enabling recommendations for new products through attribute matching rather than requiring historical transaction data.
Solution Approach 2:
The system performs preliminary action by pre-computing attribute vectors and similarity metrics for all items in the catalog before they are needed for recommendations. This allows new products to be immediately integrated into the recommendation system upon arrival, as the attribute-based framework is already in place and does not require historical data accumulation.
2Adaptability or versatility
If current systems use attribute matching for recommendations, then they can recommend items based on product features, but coverage of item attributes is generally low and attributes may be non-uniform and/or incorrect
Solution Approach 1:
The patent transforms attributes from discrete, potentially inconsistent categorical data into continuous attribute vectors with standardized dimensions. This parameter transformation allows for precise similarity calculations using mathematical distance metrics, converting qualitative attribute matching into quantitative measurements that improve both coverage and precision.
Solution Approach 2:
The system creates a universal attribute vector representation that can be applied across all product categories and items. This multi-functional framework handles diverse attribute types (color, size, material, style) through a unified vector space, allowing consistent and accurate comparisons across different product types while accommodating varying attribute coverage.
3Reliability
If systems rely on past sales data for recommendations, then they capture actual purchase patterns, but they fail to identify complementary items for new products and items with low attribute coverage
Solution Approach 1:
The patent replaces the mechanical system of counting and analyzing discrete purchase transactions with a field-based approach using attribute vectors and similarity fields. This substitution allows the system to infer complementary relationships through attribute proximity in vector space rather than requiring explicit co-purchase history, preserving information about potential complementary items even without sales data.
Data Source
AI summary
A system and method of generating complimentary items from a catalog of items is disclosed. A plurality of item attributes for each of a plurality of items is received and a multimodal embedding representative of the plurality of attributes is generated for each of the plurality of items. The multimodal embedding is configured to predict at least a subset of the received plurality of item attributes for each of the plurality of items. A triplet network including a node representative of each of the plurality of items is generated. The triplet network is generated based on the multimodal embedding for each of the plurality of items. A plurality of complimentary items is generated from the plurality of items. The plurality of complimentary items are selected by the triplet network based on an anchor item selection received from a user.