Computational modelling of political discourse tasks has become an increasingly important area of research in natural language processing. Populist rhetoric has risen across the political sphere in recent years; however, computational approaches to it have been scarce due to its complex nature. In this paper, we present the new $\textit{Us vs. Them}$ dataset, consisting of 6861 Reddit comments annotated for populist attitudes and the first large-scale computational models of this phenomenon. We investigate the relationship between populist mindsets and social groups, as well as a range of emotions typically associated with these. We set a baseline for two tasks related to populist attitudes and present a set of multi-task learning models that leverage and demonstrate the importance of emotion and group identification as auxiliary tasks.
翻译：政治讨论任务的计算模型已成为自然语言处理研究的一个日益重要的领域。近年来,民粹主义言论在整个政治领域抬头;然而,由于它的复杂性,对它采取的计算方法很少。我们在本文件中介绍了一套新的“$\textit{Us vs.Them}$数据集”,其中包括6861条关于民粹主义态度的注释评论和这一现象的第一个大规模计算模型。我们研究了民粹主义思想和社会群体之间的关系,以及通常与此相关的一系列情感。我们为与民粹主义态度有关的两项任务确定了基准,并提出了一套多任务学习模式,这些模式利用并展示了情感和群体认同作为辅助任务的重要性。
相关内容
微信扫码咨询专知VIP会员