Shuo Yang

Dramatically improve multimodal recommendation accuracy without any training by initializing user embeddings with item modality features and user cluster information.

General-purpose LLMs can extract some signal directly from raw DNA sequences, but still struggle with complex genomic inference, highlighting a gap between their capabilities and the demands of real-world genomic analysis.

Forget static fusion: CAMMSR adaptively weights multimodal signals in sequential recommendations based on item category and user preferences, unlocking synergistic effects between modalities.