Factual Question Answering on PopQA (Accuracy, Recall)
72.38AccuracyHyGRAG +Inference
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HyGRAG +InferenceBackbone=Llama-3.1-8B-Instruct, Retrieval top-k=5, Embedding model=BGE-M32026.06 | 72.38 | 43.62 | |
| HyGRAGBackbone=Llama-3.1-8B-Instruct, Retrieval top-k=5, Embedding model=BGE-M32026.06 | 72.34 | 43.51 | |
| HippoRAG2Backbone=Llama-3.1-8B-Instruct2026.06 | 68.12 | 38.77 | |
| L-LightRAGBackbone=Llama-3.1-8B-Instruct, Mode=Local2026.06 | 67.05 | 39.97 | |
| VanillaRAGBackbone=Llama-3.1-8B-Instruct, Retrieval top-k=5, Embedding model=BGE-M3, Chunk size=1200 tokens2026.06 | 66.98 | 31.64 | |
| RAPTORBackbone=Llama-3.1-8B-Instruct, Retrieval top-k=5, Embedding model=BGE-M32026.06 | 66.33 | 30.98 | |
| RAPTOR-KBackbone=Llama-3.1-8B-Instruct, Retrieval top-k=5, Embedding model=BGE-M3, Clustering=K-means2026.06 | 66.26 | 30.86 | |
| EraRAGBackbone=Llama-3.1-8B-Instruct, Retrieval top-k=52026.06 | 61.4 | 29.94 | |
| HippoRAGBackbone=Llama-3.1-8B-Instruct2026.06 | 55.18 | 23.35 | |
| L-GraphRAGBackbone=Llama-3.1-8B-Instruct, Mode=Local2026.06 | 54.82 | 30.89 | |
| HiRAGBackbone=Llama-3.1-8B-Instruct2026.06 | 53.18 | 30.47 | |
| H-LightRAGBackbone=Llama-3.1-8B-Instruct, Mode=Hybrid2026.06 | 47.46 | 25.91 | |
| ArchRAGBackbone=Llama-3.1-8B-Instruct2026.06 | 30.81 | 14.04 | |
| G-LightRAGBackbone=Llama-3.1-8B-Instruct, Mode=Global2026.06 | 23.52 | 11.26 | |
| CoTBackbone=Llama-3.1-8B-Instruct, Decoding strategy=greedy decoding2026.06 | 16.03 | 5.58 | |
| Zero-shotBackbone=Llama-3.1-8B-Instruct, Decoding strategy=greedy decoding2026.06 | 15.37 | 5.62 |