arXiv 2509.02017
Empowering Large Language Model for Sequential Recommendation via Multimodal Embeddings and Semantic IDs
By Yuhao Wang, Junwei Pan, et al.
Published 2025-09-02
Mindmap
Browse the paper's core ideas, clusters, and relationships in a structured outline.
Sequential recommendation (SR) aims to capture users' dynamic interests and sequential patterns based on their historical interactions. Recently, the powerful capabilities of large language models (LLMs) have driven their adoption in SR. However, we identify two critical challenges in existing LLM-based SR methods: 1) embedding collapse when incorporating pre-trained collaborative embeddings and 2) catastrophic forg…