A general framework for information extraction using dynamic span graphs

By Yi Luan, David Wadden, Luheng He, A. Shah, Mari Ostendorf, Hannaneh Hajishirzi. North American Chapter of the Association for Computational Linguistics, 2019

342 citations recorded by the source corpus.

We introduce a general framework for several information extraction tasks that share span representations using dynamically constructed span graphs. The graphs are dynamically constructed by selecting the most confident entity spans and linking these nodes with confidence-weighted relation types and coreferences. The dynamic span graph allow coreference and relation type confidences to propagate through the graph to iteratively refine the span representations. This is unlike previous multi-task frameworks for information extraction in which the only interaction between tasks is in the shared first-layer LSTM. Our framework significantly outperforms state-of-the-art on multiple information extraction tasks across multiple datasets reflecting different domains. We further observe that the span enumeration approach is good at detecting nested span entities, with significant F1 score improvement on the ACE dataset.

Subfields: Artificial Intelligence

View Paper: https://www.semanticscholar.org/paper/f262ef2f50dfcaf07dc6598f22fb9b2470b37cf1

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

Citations in the corpus

Span-based Joint Entity and Relation Extraction with Transformer Pre-training (2019, methodology): “Only for SciERC, entity type correctness is not considered when evaluating relation extraction, which is in line with prior work [20, 21, 34].”
DeepStruct: Pretraining of Language Models for Structure Prediction (2022, methodology): “We use the preprocessing scripts provided by (Luan et al., 2019).”
A Unified Generative Framework for Various NER Subtasks (2021, methodology): “, 2019; Shibuya and Hovy, 2020) and span-level classification (Luan et al., 2019; Li et al., 2020b; Wang et al., 2020a) models.”
A Frustratingly Easy Approach for Entity and Relation Extraction (2021, methodology): “Previous approaches (Luan et al., 2018, 2019; Wad-den et al., 2019) re-use the span representations h e ( s i ) , h e ( s j ) to predict the relationship between s i and s j .”
Entity, Relation, and Event Extraction with Contextualized Span Representations (2019, methodology): “Evaluation We follow the experimental setups of the respective state-of-the-art methods for each dataset: Luan et al. (2019) for entities and relations, and Zhang et al. (2019) for event extraction.”
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.”
Structured Prediction as Translation between Augmented Natural Languages (2021, methodology): “The second baseline is DyGIE++ (Wadden et al., 2019), a BERT based multi-task learning framework for the tasks of coreference resolution, relation extraction, named entity recognition, and event extraction.”
Unified Named Entity Recognition as Word-Word Relation Classification (2021, background): “ods (Luan et al. 2019; Li et al. 2021a) are another state-ofthe-art (SoTA) approaches for unified NER, enumerating all possible spans and conduct span-level classification.”
A Unified MRC Framework for Named Entity Recognition (2019, methodology): “• BiLSTM-CRF: Ma and Hovy (2016) uses a character CNN to extract features from characters, which are then output to CRFs.”
Named Entity Recognition as Dependency Parsing (2020, methodology): “Many other neural systems followed this approach but used instead LSTMs to encode the input and a CRF for the prediction (Lample et al., 2016; Ma and Hovy, 2016; Chiu and Nichols, 2016).”
Low Resource Multi-Task Sequence Tagging - Revisiting Dynamic Conditional Random Fields (2020, background): “In summary, existing work focuses on task specific solutions that either model a specific hierarchy in NNER (Alex et al., 2007; Finkel and Manning, 2009; Ju et al., 2018; Lin et al., 2019; Luan et al., 2019; Li et al., 2019), assume independence between tasks given the shared data representation (Sø”
A Comprehensive Survey on Automatic Knowledge Graph Construction (2023, methodology): “Dynamic Graph-based Model GP-GNNs [233] Bi-LSTM GCNs MLP + softmax Inter-node graph with generated parameters EoG [232] BiLSTM GCNN Node-feature aggregation + softmax Sentence-mention-entity pair graph GraphRel [220] BiLSTM BiGCN Threshold-based Relation-weighted graph DyGIE [238] ELmo + BiLSTM GCN ”
A Joint Neural Model for Information Extraction with Global Features (2020, methodology): “Recently, some efforts (Wadden et al., 2019; Luan et al., 2019) revisit global inference approaches by designing neural networks with embedding features to jointly model multiple subtasks.”