Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GraphERE: Jointly Multiple Event-Event Relation Extraction via Graph-Enhanced Event Embeddings

Haochen Li, Di Geng

TL;DR

GraphERE addresses the challenge of jointly extracting multiple event-event relations by enriching event representations with event arguments and structural information through static AMR and IE graphs. It then uses a Node Transformer to enable cross-event interactions and constructs four task-specific dynamic graphs for Coreference, Temporal, Causal, and Subevent relations, trained with a multi-task objective. On MAVEN-ERE, GraphERE achieves state-of-the-art F1 scores across all four relations and demonstrates the importance of both static graph enhancements and dynamic cross-task interactions, with ablations confirming their contributions. The work highlights the practical potential of integrated graph-based event representations for improving event-centric reasoning in document-level NLP tasks, while also noting limitations related to preprocessing costs and data requirements.

Abstract

Events describe the state changes of entities. In a document, multiple events are connected by various relations (e.g., Coreference, Temporal, Causal, and Subevent). Therefore, obtaining the connections between events through Event-Event Relation Extraction (ERE) is critical to understand natural language. There are two main problems in the current ERE works: a. Only embeddings of the event triggers are used for event feature representation, ignoring event arguments (e.g., time, place, person, etc.) and their structure within the event. b. The interconnection between relations (e.g., temporal and causal relations usually interact with each other ) is ignored. To solve the above problems, this paper proposes a jointly multiple ERE framework called GraphERE based on Graph-enhanced Event Embeddings. First, we enrich the event embeddings with event argument and structure features by using static AMR graphs and IE graphs; Then, to jointly extract multiple event relations, we use Node Transformer and construct Task-specific Dynamic Event Graphs for each type of relation. Finally, we used a multi-task learning strategy to train the whole framework. Experimental results on the latest MAVEN-ERE dataset validate that GraphERE significantly outperforms existing methods. Further analyses indicate the effectiveness of the graph-enhanced event embeddings and the joint extraction strategy.

GraphERE: Jointly Multiple Event-Event Relation Extraction via Graph-Enhanced Event Embeddings

TL;DR

GraphERE addresses the challenge of jointly extracting multiple event-event relations by enriching event representations with event arguments and structural information through static AMR and IE graphs. It then uses a Node Transformer to enable cross-event interactions and constructs four task-specific dynamic graphs for Coreference, Temporal, Causal, and Subevent relations, trained with a multi-task objective. On MAVEN-ERE, GraphERE achieves state-of-the-art F1 scores across all four relations and demonstrates the importance of both static graph enhancements and dynamic cross-task interactions, with ablations confirming their contributions. The work highlights the practical potential of integrated graph-based event representations for improving event-centric reasoning in document-level NLP tasks, while also noting limitations related to preprocessing costs and data requirements.

Abstract

Events describe the state changes of entities. In a document, multiple events are connected by various relations (e.g., Coreference, Temporal, Causal, and Subevent). Therefore, obtaining the connections between events through Event-Event Relation Extraction (ERE) is critical to understand natural language. There are two main problems in the current ERE works: a. Only embeddings of the event triggers are used for event feature representation, ignoring event arguments (e.g., time, place, person, etc.) and their structure within the event. b. The interconnection between relations (e.g., temporal and causal relations usually interact with each other ) is ignored. To solve the above problems, this paper proposes a jointly multiple ERE framework called GraphERE based on Graph-enhanced Event Embeddings. First, we enrich the event embeddings with event argument and structure features by using static AMR graphs and IE graphs; Then, to jointly extract multiple event relations, we use Node Transformer and construct Task-specific Dynamic Event Graphs for each type of relation. Finally, we used a multi-task learning strategy to train the whole framework. Experimental results on the latest MAVEN-ERE dataset validate that GraphERE significantly outperforms existing methods. Further analyses indicate the effectiveness of the graph-enhanced event embeddings and the joint extraction strategy.
Paper Structure (18 sections, 16 equations, 4 figures, 2 tables)

This paper contains 18 sections, 16 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Event-Event Relation Extraction
  4. Event Graph Construction
  5. Approach
  6. Task Formulation
  7. Sequence Encoder
  8. Graph-enhanced Event Embeddings by Static Event Graphs
  9. Joint Multiple Extraction by Task-specific Dynamic Event Graphs
  10. Training GraphERE
  11. Experiments
  12. Dataset and Setup
  13. Overall Results
  14. Ablation Analysis
  15. Analysis for IE Graph and AMR Graph
  16. ...and 3 more sections

Figures (4)

  • Figure 1: An example of the ERE task.
  • Figure 2: The overall architecture of GraphERE.
  • Figure 3: Procedures for the Node Transformer and Deep Graph Learning algorithms.
  • Figure 4: The performance along with the training data scale. We compare our GraphERE$_{joint}$ with the RoBERTa$_{joint}$ baseline, and the average F1 scores of 10 experiments are used.