Tricolore Framework for Multi-Behavior User Profiling

This technical report presents Tricolore, a multi-behavior user profiling framework designed to enhance candidate generation in recommender systems. Traditional recommender systems often optimize for a single target behavior, which limits their ability to capture the full spectrum of user interests and leads to issues such as cold-start and data sparsity. Tricolore addresses these limitations by utilizing a versatile multi-vector learning approach that uncovers connections between different behavior types. The framework incorporates a behavior-wise multi-view fusion module to manage variability in sparsity across behavior types and employs a popularity-balanced strategy to ensure a diverse recommendation list. Experimental results demonstrate Tricolore's effectiveness across various recommendation scenarios, including e-commerce and short video platforms. The report outlines the architecture of Tricolore, detailing its adaptive multi-task structure and the foundational layer that enhances learning through shared base embedding strategies. Overall, Tricolore aims to provide a comprehensive solution for improving user engagement through personalized recommendations.