Item-to-item recommendations for cold start items
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional navigational models in e-commerce websites are ineffective in recommending items for about 80% of the catalog, known as cold start items, which are new or lack user interaction data, resulting in only 20% of items being recommended to customers.
Innovation Solution
A system and method that generate item-to-item recommendations by processing feature data, including similarities between items, using a processor to determine weights for feature vectors and apply logistic regression models to rank candidate items, thereby improving recommendations for cold start items.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional navigational models are used to recommend items, then items with sufficient user interaction data can be recommended, but cold start items (80% of catalog) cannot be recommended
Solution Approach 1:
The patent introduces feature vectors as an intermediary representation that bridges the gap between items without requiring direct user interaction data. By comparing feature vectors (images, titles, descriptions) of cold start items with user interaction data, the system can generate recommendations for items that would otherwise be invisible to conventional models, thereby improving recommendation coverage while maintaining reliability through the mediating feature comparison mechanism
Solution Approach 2:
The patent replaces the mechanical dependency on user interaction data (co-view, co-purchase) with a computational approach using feature vector comparison and logistic regression models. This substitution allows the system to generate recommendations for cold start items by computing similarities based on item features rather than relying on mechanical collection of user interaction data, thus improving coverage from 20% to include cold start items
2Reliability
If conventional navigational models rely on user interaction data, then recommendation accuracy is maintained, but only 20% of catalog items can be recommended
Solution Approach 1:
The patent performs preliminary action by pre-computing and storing feature vectors (images, titles, descriptions) for all catalog items before they are needed for recommendation. This preliminary preparation enables the system to quickly compare cold start items with items having user interaction data using logistic regression models, thereby improving catalog coverage without sacrificing recommendation accuracy through the use of pre-processed feature data
3Productivity
If feature-based comparison is used for cold start items, then recommendation coverage increases, but computational complexity increases
Solution Approach 1:
The patent segments the recommendation process into distinct components: feature extraction (images, titles, descriptions), feature vector comparison, and logistic regression modeling. By dividing the complex task of cold start item recommendation into these manageable segments, the system can process each aspect separately and efficiently, reducing overall computational complexity while improving coverage through systematic feature-based comparison
Data Source
AI summary
An approach is disclosed for providing item-to-item recommendations. The approach receives data for an item. The approach generates recommended candidate item data for the data of the received item. The approach generates feature data of the recommended candidate item data and the data of the received item. The feature data may include one or more similarities between the data of the received item and data corresponding to a respective recommended candidate item. The approach generates ranked recommended candidate item data based on the generated feature data and at least one weight determined for candidate test item data. The candidate test item data may include data corresponding to the recommended candidate item data.