Joint Modeling of Search and Recommendations Via an Unified Contextual Recommender (UniCoRn)
Moumita Bhattacharya, Vito Ostuni, Sudarshan Lamkhede
TL;DR
UniCoRn tackles the fragmentation of search and recommendation systems by unifying them into a single deep learning model that leverages a broad contextual input (user id, query, country, source entity id, and task). It uses embedding-based features, residual connections, and feature crossing, with context-aware imputation to handle missing inputs, and is trained using a binary cross-entropy objective. Personalization is introduced progressively—from semi-personalized clusters to fully personalized features derived from pre-trained representations—achieving 7% lift in search and 10% in recommendations. The results suggest that cross-task learning on a larger, shared dataset can outperform task-specific models and reduce maintenance, enabling scalable, low-latency serving across Netflix canvases.
Abstract
Search and recommendation systems are essential in many services, and they are often developed separately, leading to complex maintenance and technical debt. In this paper, we present a unified deep learning model that efficiently handles key aspects of both tasks.