Multi-Task Identification of Entities, Relations, and Coreference for Scientific Knowledge Graph Construction

By Yi Luan, Luheng He, Mari Ostendorf, Hannaneh Hajishirzi. 2018

789 citations recorded by the source corpus.

We introduce a multi-task setup of identifying entities, relations, and coreference clusters in scientific articles. We create SciERC, a dataset that includes annotations for all three tasks and develop a unified framework called SciIE with shared span representations. The multi-task setup reduces cascading errors between tasks and leverages cross-sentence relations through coreference links. Experiments show that our multi-task model outperforms previous models in scientific information extraction without using any domain-specific features. We further show that the framework supports construction of a scientific knowledge graph, which we use to analyze information in scientific literature.

Subfields: Artificial Intelligence, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

Citations in the corpus

SciBERT: A Pretrained Language Model for Scientific Text (2019, background): “In addition, SCIBERT achieves new SOTA results on ACLARC (Cohan et al., 2019), and the NER part of SciERC (Luan et al., 2018).”
Don’t Stop Pretraining: Adapt Language Models to Domains and Tasks (2020, neutral): “DAPT is more resource intensive (see Table 9 in §5.3), but TAPT manages to match its performance in some of the tasks, such as SCIERC.”
Continual Training of Language Models for Few-Shot Learning (2022, neutral): “The other three corresponding end task datasets are SemEval-res (Xu et al., 2019), ACL-ARC (Jurgens et al., 2018), and SCIERC (Luan et al., 2018).”
S2ORC: The Semantic Scholar Open Research Corpus (2020, neutral): “SciERC Luan et al. (2018) NER 67.”
SciBERT: Pretrained Contextualized Embeddings for Scientific Text (2019, neutral): “For relations in SciERC, our results are not comparable with those in Luan et al. (2018) because we are performing relation classification given gold entities, while they perform NER and relation extraction jointly.”
Unified Structure Generation for Universal Information Extraction (2022, neutral): “…ACE04 (Mitchell et al., 2005), ACE05 (Walker et al., 2006); CoNLL03 (Tjong Kim Sang and De Meul-der, 2003), CoNLL04 (Roth and Yih, 2004), Sci-ERC (Luan et al., 2018), NYT (Riedel et al., 2010), CASIE (Satyapanich et al., 2020), SemEval-14 (Pontiki et al., 2014), SemEval-15 (Pontiki et al.,…”
BERT: A Review of Applications in Natural Language Processing and Understanding (2021, methodology): “Datasets from subject areas were used: biomedicine (ChemProt [31], EBM-NLP [57] and others), computer science (SciERC [47], ACL-ARC [28]), and interdisciplinary (SciCite, described by A.”
Span-based Joint Entity and Relation Extraction with Transformer Pre-training (2019, methodology): “Recently, some span-based models towards joint entity and relation extraction have been proposed [20, 9], using span representations derived from a BiLSTM over concatenated ELMo, word and character embeddings.”
ScholarBERT: Bigger is Not Always Better (2022, methodology): “SciERC, introduced above, provides 4716 relations (Luan et al., 2018).”
A Frustratingly Easy Approach for Entity and Relation Extraction (2021, methodology): “4% on ACE05, ACE04, and SciERC respectively.”
Lifelong Pretraining: Continually Adapting Language Models to Emerging Corpora (2021, methodology): “…Lee, 2017) for the bio-medical domain; ACL-ARC citation intent classiﬁcation dataset (Jurgens et al., 2018) and SciERC relation extraction dataset (Luan et al., 2018) for the computer science domain; relation extraction over Synthesis procedures (Mysore et al., 2019) and named entity…”
A general framework for information extraction using dynamic span graphs (2019, methodology): “Luan et al. (2018a) previously achieved the state of the art on SciERC and use a span-based neural model like our D Y GIE.”
Time Waits for No One! Analysis and Challenges of Temporal Misalignment (2021, methodology): “For this domain, we study two tasks: Mention type classiﬁcation ( S CI ERC ) We use the SciERC dataset from Luan et al. (2018) which contains entity-relation annotations for computer science paper abstracts for a relatively wide range of years (1980s to 2019).”
RadGraph: Extracting Clinical Entities and Relations from Radiology Reports (2021, methodology): “Pipeline approaches [20, 21] decompose the task into separately trained subtasks (named entity recognition [22, 23] and relation extraction [24]), while joint extraction approaches [18, 25–27] model the two subtasks at the same time in order to capture interactions between entities and relations.”
Black-box Prompt Learning for Pre-trained Language Models (2022, background): “…et al., 2006; Giampiccolo et al., 2007; Bentivogli et al., 2009), and four domain-specific datasets: CITATIONINTENT (Jurgens et al., 2018), SCIERC (Luan et al., 2018), RCT (Dernoncourt and Lee, 2017), HYPERPARTISAN (Kiesel et al., 2019) from specific domains including computer science,…”
Text Generation from Knowledge Graphs with Graph Transformers (2019, methodology): “First, we apply the SciIE system of Luan et al. (2018), a state-of-the-art science-domain information extraction system.”
Data Selection for Language Models via Importance Resampling (2023, background): “Following Gururangan et al. [24], we consider 8 downstream datasets across 4 domains: Computer Science papers (ACL-ARC [31], Sci-ERC [51]), Biomedicine (ChemProt [41], RCT [16]) News (AGNews [92], HyperPartisan [34]), and Reviews (Helpfulness [53], IMDB [52]).”
OAG-BERT: Pre-train Heterogeneous Entity-augmented Academic Language Models (2021, background): “Besides pre-training for general purposes, more and more language models are targeting at specific domains, such as BioBERT [26] for biomedical field and SciBERT [2] for academic field, which establish new state-of-the-art on many domain-related benchmarks such as named entity recognition [10, 32], ”
Can We Automate Scientific Reviewing? (2021, background): “(4) External Knowledge: Besides the paper itself, review systems can also rely on external knowledge, such as a citation graphs constructed based on more scientific papers or a knowledge graph connecting concepts across different papers (Luan et al., 2018; Lo et al., 2020).”