Exploration of summary informativeness and topic richness through mining multilingual tweets: a study on turkey earthquake 2023

Improved document richness requires analyzing both English and non-English tweets. This technique should yield a more simplified and enhanced view of the crisis. Even regional language tweets provided real-time disaster insights that may have been superior or equivalent to English tweets. In terms of information retrieval (IR), obtaining reliable ground truth topics at the earliest is necessary to develop an effective crisis response plan. Micro-blog summary strategies should curate such related content and deliver qualitative summary results at different points in the crisis timeline. Hence, the interpretation of multilingual and local language tweets is essential, along with English tweets, to improve the topic richness produced by the summary in the crisis timeline. Given these facts, the current work examines how to improve the informativeness of micro-blog summary techniques, such as ALBERT, SBERT, and DistilBERT, in terms of lexical and syntactical superiority, taking into account three corpora that include Turkish, Multilingual, and English tweets for a crisis event, namely the Turkey Earthquake 2023. An analytical study has been performed on summary results produced by ALBERT (found to be supreme in Lexical and Syntactic perspective), in context to Topic coverage and contrasts obtained concerning three corpora in crisis timeline. Finally, the diversity of crisis themes is also discussed by three corpora to understand the distribution of each theme in the crisis timeline. It is evident that the topics curated by regional and multilingual tweets extracted more ground truth and actionable insights. Whereas, English topics were having more lifetimes due to continuous reporting during crisis phases. Besides, regional and multilingual tweets have contributed around 61% of situational information throughout the crisis timeline i.e. 6 February–21 February, 2023.