Abstract
This paper addresses automatic event prediction from unstructured text, specifically event chains. While current approaches employ LSTM for encoding full chains, learning long-range narrative orders, or learning partial orders and long-range narrative orders, none of them consider writer sentiment. To address this, we propose a deep learning-based approach that incorporates writer sentiment. We pre-process the text, extract events, compute sentiment scores using SentiWordNet, convert events to digital vectors, and feed them along with sentiment scores into a deep learning-based classifier. This classifier uses hidden states for event pair modeling, with each pair having its associated sentiment. Evaluation results show that our approach significantly surpasses state-of-the-art methods with 29.2% accuracy.