Browse › Natural Language Processing › Relation Extraction › ACE 2005
ACE 2005 Benchmark (Relation Extraction)
Relation Extraction is the task of predicting attributes and relations for entities in a sentence. For example, given a sentence “Barack Obama was born in Honolulu, Hawaii.”, a relation classifier aims at predicting the relation of “bornInCity”. Relation Extraction is the key component for building relation knowledge graphs, and it is of crucial significance to natural language processing applications such as structured search, sentiment analysis, question answering, and summarization.
Source: Deep Residual Learning for Weakly-Supervised Relation Extraction
The archive carries no text for this table; the description above is the archive's text for the task Relation Extraction. archive 2025-07-28
Over time archive 2025-07-28
The chart needs JavaScript; the table below carries every value.
Direction inferred from the metric name, not from the archive: RE Micro F1 (higher is better), RE+ Micro F1 (higher is better), NER Micro F1 (higher is better), Relation classification F1 (higher is better), Relation F1 (higher is better). Not inferred (points only, no best-so-far line): Sentence Encoder, Cross Sentence. Points are placed at the row's paper date; 29 of 30 rows carry one.
Results archive 2025-07-28
Archive rows end at the archive snapshot, 2025-07-28: no result published after that date is in this table. Rank is the archive's row order at that snapshot; not re-ranked here. Metric values are the archive's strings. Column headers sort the table in your browser; each row keeps its archive rank.
| Paper | Code | Ran Syntology | Report | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | ITER | 75.1 ± 0.49 | 71.9 ± 0.56 | 91.6 ± 0.12 | FLAN T5 3B | Yes | – | Paper | Code | 2024 | linked, not harvested | report | ||
| 2 | PL-Marker | 73.0 | 71.1 | 91.1 | ALBERT | Yes | – | Paper | Code | 2021 | 3 of 9 ran · 6 unverified | report | ||
| 3 | ASP+T5-3B | 72.7 | 70.5 | 91.3 | T5-3B | Yes | – | Paper | Code | 2022 | linked, not harvested | report | ||
| 4 | GoLLIE | 70.1 | ✓ | Paper | Code | 2023 | 18 of 25 ran · 7 unverified | report | ||||||
| 5 | Ours: cross-sentence ALB | 69.4 | 90.9 | ALBERT | Yes | – | Paper | Code | 2020 | 3 of 4 ran · 1 unverified | report | |||
| 6 | Table-Sequence | 67.6 | 64.3 | 89.5 | ALBERT | No | – | Paper | Code | 2020 | 0 of 3 ran · 3 unverified | report | ||
| 7 | TriMF | 66.49 | 62.77 | 87.61 | BERT base | Yes | – | Paper | Code | 2021 | linked, not harvested | report | ||
| 8 | TablERT | 66.1 | 62.4 | 88.0 | BERT base | No | – | Paper | Code | 2020 | linked, not harvested | report | ||
| 9 | DYGIE++ | 63.4 | 88.6 | BERT base | Yes | ✓ | Paper | Code | 2019 | 0 of 7 ran · 7 unverified | report | |||
| 10 | DyGIE | 63.2 | 88.4 | ELMo | Yes | ✓ | Paper | Code | 2019 | linked, not harvested | report | |||
| 11 | Hierarchical Multi-task | 62.7 | 87.5 | ELMo | No | ✓ | Paper | Code | 2018 | 0 of 1 ran · 1 unverified | report | |||
| 12 | Attention | 55.9 | 53.6 | 82.6 | biLSTM | No | – | Paper | – | 2017 | no code linked | report | ||
| 13 | Joint w/ Global | 52.1 | 49.5 | 80.8 | - | No | – | Paper | – | 2014 | no code linked | report | ||
| 14 | MGE | 68.2 | 89.7 | ALBERT | No | – | Paper | – | no code linked | report | ||||
| 15 | PFN | 66.8 | 89.0 | ALBERT | No | – | Paper | Code | 2021 | 5 of 7 ran · 2 unverified | report | |||
| 16 | MRC4ERE++ | 62.1 | 85.5 | BERT base | No | – | Paper | Code | 2020 | linked, not harvested | report | |||
| 17 | Multi-turn QA | 60.2 | 84.8 | BERT base | No | – | Paper | Code | 2019 | linked, not harvested | report | |||
| 18 | MRT | 59.6 | 83.6 | biLSTM | No | – | Paper | – | 2018 | no code linked | report | |||
| 19 | GCN | 59.1 | 84.2 | biLSTM | No | – | Paper | – | 2019 | no code linked | report | |||
| 20 | Global | 57.5 | 83.6 | biLSTM | No | – | Paper | – | 2017 | no code linked | report | |||
| 21 | SPTree | 55.6 | 83.4 | biLSTM | No | – | Paper | Code | 2016 | linked, not harvested | report | |||
| 22 | Span-level | 85.98 | ELMo | No | – | Paper | – | 2019 | no code linked | report | ||||
| 23 | Dual Pointer Network(multi-head) | 80.8 | No | – | Paper | – | 2021 | no code linked | report | |||||
| 24 | Dual Pointer Network | 80.5 | No | – | Paper | – | 2019 | no code linked | report | |||||
| 25 | RNN+CNN | 67.7 | No | – | Paper | – | 2015 | no code linked | report | |||||
| 26 | Walk-based model | 64.2 | No | – | Paper | Code | 2019 | linked, not harvested | report | |||||
| 27 | CNN | 61.3 | No | – | Paper | – | 2015 | no code linked | report | |||||
| 28 | FCM | 58.2 | No | – | Paper | Code | 2015 | linked, not harvested | report | |||||
| 29 | HySPA (ours) w/ RoBERTa | 68.2 | – | Paper | Code | 2021 | 0 of 5 ran · 5 unverified | report | ||||||
| 30 | HySPA | ALBERT | No | – | Paper | Code | 2021 | 0 of 5 ran · 5 unverified | report |
All 30 rows shown. 30 link to a paper page on this site; 4 are marked as using additional training data in the archive. No GitHub stars are tracked; "Code" is the first repository the archive lists for the row. The archive carries no row tags, review links or community-submitted rows for this table; none are shown. archive 2025-07-28
Syntology Ran reads "N of M ran · U unverified": of the M code samples Syntology harvested from repositories linked to that row's paper (joined by arXiv id), N executed on a synthesized input and the other U = M−N are unverified (harvested, no recorded run). It counts code from repositories linked to that row's paper, not this result: the row's number was not reproduced and nothing here is a correctness claim. The other cell texts mean no graph line for the row: "linked, not harvested" (the archive links code, Syntology has not harvested it), "no code linked" (no code link in the archive), "not matched" (the row's paper URL matched no paper on this site). 9 rows have a graph line, from 8 distinct papers; 4 rows (4 papers) have at least one sample that ran. Counting each paper once: Syntology ran 29 of 61 samples; 32 unverified. Separately, 5 of those 61 are pointer-only (licence): the site points at that code rather than redistributing it, a licence property recorded for ran and unverified samples alike; each cell's tooltip carries the row's own pointer-only count. Read from the graph 2026-09-24. Per-sample status is on the paper page.
Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections