脚本事件预测:方法、评测与挑战

Script event prediction:methods, evaluation and challenges

相较于问答等经典的自然语言处理任务,脚本事件预测并未受到广泛关注。脚本事件预测旨在通过给定的上下文事件对未来事件进行准确预测,进而为后续可能发生的事件作出相应部署。为此,其研究具有重要意义,需要进行归纳与总结。针对脚本事件预测任务,首先给出脚本事件预测的基本概念;然后,梳理分析了事件表征中论元构成和事件表示学习两个子任务,其中事件表示学习从嵌入编码和语义增强两个方面叙述;接着,又以脚本建模方法为主线,从事件对、事件链、事件图以及结合型建模四个角度对其进行总结;并且,对现有模型进行测试,分析各类模型效果;最后,总结当前研究存在的问题,并展望未来可能的研究方向。

Compared to classic natural language processing tasks like Q&A,script event prediction has garnered less attention.In order to deploy subsequent events appropriately,script event prediction intends to accurately forecast future events with a given contextual event.This study is quite important and requires summarization.Based on the script event prediction task,this paper firstly proposed fundamental concept of script event prediction.Then it described argument composition and event representation learning in event characterization.Also,event representation learning was shown from embedded coding and semantic enhancement.Next,this paper summarized script modeling approaches from event pair,event chain,event graph,and integrated modeling in detail.In addition,it examined the impacts of different models,and horizontally evaluated the effects of the existing models.Finally,this paper provided a summary of the issues with the present study,along with an outlook on promising future research areas.