المفاهيم الأساسية
MulCo integrates knowledge from BERT and GNN to improve event temporal relation extraction performance.
الملخص
The content introduces MulCo, a model that combines knowledge from BERT and GNN to enhance event temporal relation extraction. It addresses the challenges of capturing cues for event pairs at different proximity bands. The paper discusses the importance of multi-scale contrastive knowledge co-distillation in improving performance on various datasets. Experimental results show MulCo achieves new state-of-the-art results on several benchmark datasets.
Directory:
Abstract
Event Temporal Relation Extraction (ETRE) is challenging due to varying proximity bands.
MulCo integrates linguistic cues across short and long proximity bands.
Introduction
ETRE predicts the order of events regardless of mention order in text.
TB-Dense dataset biases towards short-distance event pairs.
State-of-the-Art Models
BERT-based models excel in short-distance tasks but struggle with longer distances.
Graph Neural Networks capture long-distance structural cues effectively.
MulCo Model Formulation
Multi-Scale Distillation improves BERT's performance on both short and long distances.
Experiments and Results
MulCo outperforms baselines and achieves new SOTA on various datasets.
Limitations and Future Work
GNNs do not improve with multi-scale distillation from BERT, limiting potential applications.
الإحصائيات
"Our experimental results show that MulCo successfully integrates linguistic cues pertaining to temporal reasoning across both short and long proximity bands."
"MulCo achieves new state-of-the-art results on several ETRE benchmark datasets."