List your company

Question Answering on HybridQA (dev)

68.4EM (Total)S3HQA
622.238.454.6Sep 9, 2021Feb 4, 2022Jul 2, 2022Nov 28, 2022Apr 25, 2023Sep 20, 2023Feb 16, 2024

Evaluation results

MethodLinks
S3HQAFine-tuning=true2024.0268.4——70.375.369.978.275.3
MAFIDFine-tuning=true2024.0266.2——69.475.266.575.574.1
MITQAFine-tuning=true2024.0265.5——68.173.366.775.672.7
POINTRFine-tuning=true2024.0263.4——68.674.262.871.971
POINTR + MATEBackbone=MATE2021.0963.4——68.674.262.871.971
POINTR + TAPASBackbone=TAPAS2021.0963.3——68.173.962.97270.8
POINTR + SATBackbone=SAT2021.0961.2——66.571.860.369.268.7
POINTR + LINFORMERBackbone=LINFORMER2021.0960.8——65.571.159.46968.4
MuGER2Fine-tuning=true2024.0257.1——60.969.256.968.967.3
HPROPROFine-tuning=false, Sampled cases=2002024.0248——51.455.946.854.454.6
HYBRIDERFine-tuning=true2024.0244——54.361.439.145.750.7
HYBRIDERT=0.82021.0944——54.361.439.145.750.7
POINTR + TABLEETCBackbone=TABLEETC2021.0936.1——3642.437.845.342.9
Unsupervised-QGFine-tuning=false2024.0225.7——————30.5
GPT-4 End-to-End QA w. RetrieverFine-tuning=false, Sampled cases=2002024.0224.5——5061.811.113.330
Passage-Only2021.0919.5——9.213.526.132.425.1
Table-Only2021.098.4——14.719.12.44.512.1
DOCHOPPERruntime=74.6/s2021.06—47.755—————
DOCHOPPERsparse_features=without2021.06—44.451.2—————
DOCHOPPERquery_update=without2021.06—44.250.9—————
DOCHOPPERscope=sentence-only2021.06—36.743.7—————
DOCHOPPERattention_steps=single-hop2021.06—27.834.1—————
DOCHOPPERinput_unit=cell2021.06—53.161.4—————
Hybriderruntime=5.1/s2021.06—4450.7—————
HYBRIDERBackbone=BERT-Base2021.09—43.750.9—————
HYBRIDERBackbone=BERT-Large2021.09—4450.7—————
MATEnote=failed to rerun2021.06—63.471—————
ReasonBERT2021.09—60.367.2—————
Retrieval + ETCretriever=BM25, runtime=8.3/s2021.06—3743.5—————
Sequential (ETC)strategy=sequential reading, runtime=0.6±0.1/s2021.06—39.444.8—————
TAPAS+ROBERTa2021.09—57.364—————