SPEECH: Structured Prediction with Energy-Based Event-Centric Hyperspheres

• DOI: 10.48550/arxiv.2305.13617
• Publication Date: 2023-01-01
• Authors: Shumin Deng; Shengyu Mao; Ningyu Zhang; Bryan Hooi
• Keywords: Structured prediction

Event-centric structured prediction involves predicting outputs of events. In most NLP cases, event structures are complex with manifold dependency, and it is challenging to effectively represent these complicated To address issues, we propose Structured Prediction Energy-based Event-Centric Hyperspheres (SPEECH). SPEECH models dependency among components energy-based modeling, represents classes simple but effective hyperspheres. Experiments on two unified-annotated datasets indicate that predominant in detection event-relation extraction tasks.