Researchers have developed GMTRouter, a novel system for personalized large language model (LLM) routing that addresses the scarcity and inconsistency of user preference data. This approach models multi-turn user-LLM interactions as a heterogeneous graph, capturing rich relational structures between users, LLMs, queries, responses, and turns. Utilizing a lightweight inductive graph learning framework with a tailored user-conditioned sampling mechanism, GMTRouter effectively learns user preferences from limited data, outperforming existing baselines with significant improvements in accuracy and AUC. The system demonstrates the ability to adapt to new users with minimal few-shot data, offering a more personalized and efficient LLM interaction experience. AI
IMPACT Enhances personalized LLM interactions by efficiently learning user preferences from limited data.
RANK_REASON The cluster is based on a research paper detailing a new system for LLM routing. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- GMTRouter
- Gotit.pub
- Hugging Face
- large language model
- ScienceCast
- Yihang Sun
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
