Cold-Start Item Recommendation via Intermediary Static Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Online retailers face challenges in recommending cold-start items due to the lack of historical data, making it difficult to determine relevant products for users, and existing methods are inefficient and resource-intensive.
Innovation Solution
A system and method that utilize static features of items, such as titles, images, and seller information, to train models that recommend cold-start items by leveraging historical traffic patterns and existing recommendations, reducing the need for explicit traffic data and optimizing system resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If historical transaction data is used to determine recommended items, then recommendation accuracy is improved, but cold-start items cannot be recommended due to lack of historical data
Solution Approach 1:
The patent introduces an intermediary approach by using items with moderate historical data as a bridge. These intermediary items connect the well-established items (with abundant historical data) to cold-start items (with little or no historical data). The system learns from the intermediary items' performance and uses this knowledge to make recommendations for cold-start items, thus resolving the contradiction between needing historical data for accuracy and the inability to recommend items without historical data.
Solution Approach 2:
The system performs preliminary actions by pre-processing and analyzing historical transaction data to identify patterns and relationships among items before cold-start items are introduced. This preliminary analysis creates a knowledge base that can be applied to recommend cold-start items even when they lack their own historical data, allowing the system to prepare recommendation strategies in advance for new items.
2Productivity
If traditional recommendation methods are used for all items, then existing items can be recommended effectively, but cold-start items remain unrecommended
Solution Approach 1:
The patent implements a dynamic recommendation system that adapts its strategy based on item characteristics. For existing items with sufficient historical data, the system uses traditional recommendation methods. For cold-start items, it automatically switches to alternative strategies using intermediary items and static features. This dynamic adaptation allows the system to maintain high productivity for existing items while simultaneously achieving versatility for cold-start items.
Solution Approach 2:
The system changes key parameters in the recommendation process based on item status. For cold-start items, it modifies the weighting of different features (giving more weight to intermediary item relationships and seller quality) compared to existing items (where historical transaction patterns dominate). This parameter adjustment enables effective recommendations across both existing and new items without compromising the effectiveness for either category.
3Measurement precision
If comprehensive analysis of all items is performed, then recommendation quality is improved, but system resources are overwhelmed
Solution Approach 1:
The patent segments the item population into different categories based on historical data availability: well-established items, intermediary items, and cold-start items. Each segment is processed using appropriate analysis depth and computational methods. This segmentation allows comprehensive analysis for segments where it matters most while using lighter-weight methods for cold-start items, thereby maintaining recommendation quality without overwhelming system resources.
Solution Approach 2:
For cold-start items, the system applies partial analysis rather than comprehensive analysis. It focuses on specific critical features (intermediary item relationships, seller quality, static item features) rather than analyzing all possible dimensions. This partial action approach maintains adequate recommendation quality for cold-start items while significantly reducing computational resource requirements compared to exhaustive analysis.
Data Source
AI summary
Systems and methods including one or more processing modules and one or more non-transitory storage modules storing computing instructions configured to run on the one or more processing modules and perform acts of training one or more first models to recommend a first item after a user has had an interaction on the web site of the online retailer with a second item, determining static features common to both the first item and the second item, training a second model to determine whether to coordinate a display of any new item as one of one or more recommended items with any of a plurality of items, and coordinating the display of the new item as one of the one or more recommended items when the one or more of the plurality of items are displayed on the website of the online retailer based on the static features of the new item.