Multi-hop Question Answering on HotpotQA distractor
90.08Support F1C2FM-F1
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| C2FM-F1Backbone=Electra large + DebertaV2 xx-large2023.07 | 90.08 | 84.65 | — | — | — | — | |
| FE2HBackbone=iterative Electra Large + Albert-xxlarge-v2, Iterative=true2023.07 | 89.14 | 84.44 | — | — | — | — | |
| SEQGRAPH-LargeBackbone=T5-large2023.07 | 88.28 | 81.62 | — | — | — | — | |
| SEQGRAPH-BaseBackbone=T5-base2023.07 | 87.72 | 77.6 | — | — | — | — | |
| SAE-LargeScale=Large2023.07 | 87.38 | 80.75 | — | — | — | — | |
| DFGN2023.07 | 81.62 | 69.69 | — | — | — | — | |
| Hybrid RAGMethod Type=Dense+Sparse Text Retrieval, Backbone LLM=GPT-4o-mini2026.05 | 39.42 | — | 46.05 | 65.45 | 27.62 | 61.92 | |
| Naive RAGMethod Type=Standard dense retrieval, Backbone LLM=GPT-4o-mini2026.05 | 38.56 | — | 45.01 | 62.48 | 26.99 | 59.41 | |
| TGS-RAGBackbone LLM=GPT-4o-mini2026.05 | 26.06 | — | 62 | 77.55 | 27.41 | 79.99 | |
| KG2RAGBackbone LLM=GPT-4o-mini2026.05 | 25.4 | — | 38.83 | 58.92 | 16.49 | 60.46 | |
| GraphRAGRetrieval Mode=Local, Backbone LLM=GPT-4o-mini2026.05 | 20.85 | — | 55.78 | 70.15 | 10.11 | 71.79 | |
| LightRAGRetrieval Mode=Max, Backbone LLM=GPT-4o-mini2026.05 | 15.58 | — | 60.55 | 72.04 | 10.23 | 72.09 |