X-Cross: Dynamic Integration of Language Models for Cross-Domain Sequential Recommendation
Guy Hadad, Haggai Roitman, Yotam Eshel, Bracha Shapira, Lior Rokach
2025-05-05
Summary
This paper talks about X-Cross, a new AI system that helps recommend things to people by learning from different areas, like music, movies, or shopping, and quickly adapting to new topics with very little extra training.
What's the problem?
Most recommendation systems struggle when they have to switch between different types of content or don't have much data for a new area, making their suggestions less accurate or useful.
What's the solution?
The researchers designed a model that uses special adapters to update and improve its understanding as it moves between different domains, so it can keep making good recommendations even with limited new information.
Why it matters?
This matters because it means recommendation systems can be smarter and more flexible, giving people better suggestions no matter what they're interested in, and helping companies serve their users more effectively.
Abstract
X-Cross is a novel cross-domain sequential-recommendation model using low-rank adapters to dynamically refine representations and adapt to new domains efficiently with minimal fine-tuning data and parameters.