Global Agendas: Detection of Agenda Shifts in Cross-National Discussions Using Neural-Network Text Summarization for Twitter

摘要

Agendas in online media have become a scholarly focus nearly two decades ago, leading to shifting conceptualizations of what we see as agenda. Thus, agendas and agenda shifts inside online discussions have shown its potential to influence offline deliberation, aggregate support, fuel protest, passing through and/or bypassing traditional media's gatekeeping. Real-time (or nearly-real-time) learning about quick agenda movement inside globalized public debate might be particularly important for international organizations like UN or EU. However, we today lack both knowledge on how agendas move in such discussions and instruments on such analysis. In particular, we are next-to-unaware of to what extent globally relevant themes get contextualized within language-based discussion segments, as well as to what extent the latter depend on each other and lag behind each other in developing agendas and public opinion on quickly evolving issues or conflicts. In this paper, we propose a method of agenda detection based on neural-network text summarization and compare summaries of tweet packages across three languages within the Twitter hashtag #jesuischarlie. We show that sentiment detection may allow for quality assessment of the text summaries, as compared to aggregated sentiment to the original tweets. We show that, outside France, agendas were more interpretational, abstract, and non-contextualized. The pattern of news changing to 'issue outburt' was simultaneous in dense discussion segments and lagged behind in a sparser one. We also show that, globally, main issues of the discussion may be spotted within the first hour.