Packed Levitated Marker for Entity and Relation Extraction

ACL 2022  ·  Deming Ye, Yankai Lin, Peng Li, Maosong Sun ·

Recent entity and relation extraction works focus on investigating how to obtain a better span representation from the pre-trained encoder. However, a major limitation of existing works is that they ignore the interrelation between spans (pairs). In this work, we propose a novel span representation approach, named Packed Levitated Markers (PL-Marker), to consider the interrelation between the spans (pairs) by strategically packing the markers in the encoder. In particular, we propose a neighborhood-oriented packing strategy, which considers the neighbor spans integrally to better model the entity boundary information. Furthermore, for those more complicated span pair classification tasks, we design a subject-oriented packing strategy, which packs each subject and all its objects to model the interrelation between the same-subject span pairs. The experimental results show that, with the enhanced marker feature, our model advances baselines on six NER benchmarks, and obtains a 4.1%-4.3% strict relation F1 improvement with higher speed over previous state-of-the-art models on ACE04 and ACE05.

PDF Abstract ACL 2022 PDF ACL 2022 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Relation Extraction ACE 2004 PL-Marker RE Micro F1 69.7 # 1
NER Micro F1 90.4 # 1
RE+ Micro F1 66.5 # 1
Cross Sentence Yes # 1
Relation Extraction ACE 2005 PL-Marker RE Micro F1 73.0 # 1
NER Micro F1 91.1 # 1
RE+ Micro F1 71.1 # 1
Sentence Encoder ALBERT # 1
Cross Sentence Yes # 1
Named Entity Recognition CoNLL 2003 (English) PL-Marker F1 94.0 # 4
Named Entity Recognition Few-NERD (SUP) PL-Marker Precision 71.2 # 1
Recall 70.6 # 2
F1-Measure 70.9 # 1
Named Entity Recognition Ontonotes v5 (English) PL-Marker F1 91.9 # 2
Precision 92.0 # 1
Recall 91.7 # 1
Joint Entity and Relation Extraction SciERC PL-Marker Entity F1 69.9 # 4
Relation F1 53.2 # 1
RE+ Micro F1 41.6 # 3
Cross Sentence Yes # 1