Extractive Summarization on NYT50 (test)
49.47ROUGE-1HIBERTM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| HIBERTMPre-training=Two stages (Open-Domain + In-Domain)2019.05 | 49.47 | 30.11 | 41.63 | |
| HIBERTSPre-training=Two stages (Open-Domain + In-Domain), Parameters=54.6M2019.05 | 49.25 | 29.92 | 41.43 | |
| HIBERTM (in-domain)Pre-training=One stage (NYT50 training set only)2019.05 | 49.06 | 29.7 | 41.23 | |
| HIBERTS (in-domain)Pre-training=One stage (NYT50 training set only), Parameters=54.6M2019.05 | 48.92 | 29.58 | 41.1 | |
| BERTParameters=110M2019.05 | 48.38 | 29.04 | 40.53 | |
| HeriTransformerPre-training=None2019.05 | 47.44 | 28.08 | 39.56 | |
| BERTSUM+Classifierbackbone=BERT2019.03 | 46.66 | 26.35 | 42.62 | |
| BERTSUMtype=Extractive2019.09 | 46.66 | 26.35 | 42.62 | |
| BERT-ext + abs + RLtype=Abstractive, training=RL2019.09 | 46.63 | 26.76 | 43.38 | |
| JECS2019.05 | 45.5 | 25.3 | 38.2 | |
| Two-Stage BERTtype=Abstractive2019.09 | 45.33 | 26.53 | — | |
| BERT-ext + abstype=Abstractive2019.09 | 44.41 | 24.61 | 41.4 | |
| EXTRACTIONArchitecture=Hierarchical LSTM2019.05 | 44.3 | 25.5 | 37.1 | |
| Deep Reinforcedtype=abstractive, source=Paulus et al. (2018)2019.03 | 42.94 | 26.02 | — | |
| Deep Reinforcedtype=Abstractive2019.09 | 42.94 | 26.02 | — | |
| Fulltype=extractive, source=Durrett et al. (2016)2019.03 | 42.2 | 24.9 | — | |
| Fulltype=Extractive2019.09 | 42.2 | 24.9 | — | |
| Lead2019.05 | 41.8 | 22.6 | 35 | |
| First-k wordsdescription=extracting first k words of the input article2019.03 | 39.58 | 20.11 | 35.78 | |
| First k wordstype=Extractive2019.09 | 35.7 | 21.6 | — | |
| First sentencestype=Extractive2019.09 | 28.6 | 17.3 | — |