Extracting war incidents from news articles via deep Sequence Tagging

Abstract

An important natural language processing (NLP) task is to extract structured information from free text. In this thesis, we focus on the problem of extracting war incidents from news articles. A war incident is a tuple consisting of a location of the incident, the actor, the cause of death, and the number of casualties. We employ OpenTag [1], a deep sequence-tagging approach, followed by a series of flat classifiers to achieve this task. To train our sequence tagging model and the flat classifiers, we utilize a dataset of news articles surrounding the Syrian war. Our approach, which utilizes sequence tagging, outperforms baseline classifiers that rely solely on the text of the news articles.