Core Concepts
CoHeat method outperforms existing cold-start methods by effectively leveraging user-item interactions and addressing skewed distribution in bundle interactions.
Abstract
Cold-start bundle recommendation is crucial due to the continuous creation of new bundles.
Existing methods for cold-start item recommendation are not suitable for bundles.
CoHeat proposes a novel approach, combining popularity-based coalescence and curriculum heating.
CoHeat demonstrates superior performance in cold-start bundle recommendation, achieving up to 193% higher nDCG@20 compared to competitors.
The method utilizes graph-based views, contrastive learning, and curriculum learning for accurate bundle recommendation.
Stats
CoHeat achieves 193% higher nDCG@20 compared to the best competitor.
The method dynamically adjusts weights based on bundle popularity.
Quotes
"CoHeat demonstrates superior performance in cold-start bundle recommendation, achieving up to 193% higher nDCG@20 compared to the best competitor."