On the Embedding Collapse When Scaling Up Recommendation Models [paper]
Recommendation models lack scalability due to the embedding collapse phenomenon. We propose multi-embedding to overcome embedding collapse while maintaining feature interactions modules.