Cheap baselines often rival expensive LLM methods on graph node classification

Correct & Smooth on TF-IDF features trails the best reported LLM method by a median of 2.3 percentage points across 24 tasks, but some gains remain significant and evaluation flaws are common.

Industry
Michal Podstawski

NASK National Research Institute

Research Digest··3 min read
Podstawski tunes four cheap baseline models on splits from three recent studies of LLM-based node classification on text-attributed graphs, comparing them to the best reported LLM accuracy.

Podstawski selected three recent studies on LLM-based node classification for text-attributed graphs (TAGs): LLM4HeG, KCoT on the LLaGA split, and LLMNodeBed.

§