1 paper
Fan Xia, Zhaoheng Zheng, Iman Setayesh +12
LLM-based recommendation has scaled along model capacity and sequence length, yet each position encodes only text, semantic IDs, or a few categorical features, discarding rich user…