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

Haochen Li,Di Geng

2024-03-19

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.

Computation and Language,Artificial Intelligence

What problem does this paper attempt to address?

The paper mainly proposes solutions to two major issues in the task of Event-Event Relation Extraction (ERE):

1. **Event representation issue**: Current methods only use the embeddings of event trigger words to represent event features, ignoring event arguments (such as time, place, people, etc.) and their structural information.
2. **Interaction between relations issue**: Existing methods often consider only one or two types of event relations, neglecting the interactions between different types of relations.

To address the above issues, the paper proposes a joint multi-event relation extraction framework based on graph-enhanced event embeddings (GraphERE). Specifically, the main contributions of this framework include:

- **Graph-enhanced event embeddings**: By constructing static event graphs (including AMR graphs and IE graphs), using graph encoders (such as graph attention networks) to obtain event embeddings that include event arguments and structural features.
- **Joint multi-event relation extraction framework**: Introducing node transformers and task-specific dynamic event graphs to model the feature interactions between four types of event relations (coreference, temporal, causal, sub-event), and adopting a multi-task learning strategy for joint training.

Experimental results show that GraphERE significantly outperforms baseline methods on the MAVEN-ERE dataset. Further analysis demonstrates the effectiveness of graph-enhanced event embeddings and the joint extraction strategy. Additionally, the authors conducted ablation experiments to verify the effectiveness of each component and explored the impact of different mixing ratios in static event graphs on model performance.