Twitter as a Lifeline: Human-annotated Twitter Corpora for NLP of Crisis-related Messages

Twitter as a Lifeline: Human-annotated Twitter Corpora for NLP of Crisis-related Messages Muhammad Imran1, Prasenjit Mitra1, Carlos Castillo2 1Qatar Computing Research Institute (HBKU), Doha, Qatar 2Eurecat, Barcelona, Spain mimran@qf.org.qa, pmitra@qf.org.qa, chato@acm.org Abstract Microblogging platforms such as Twitter provide active communication channels during mass convergence and emergency events such as earthquakes, typhoons. During the sudden onset of a crisis situation, affected people post useful information on Twitter that can be used for situational awareness and other humanitarian disaster response efforts, if processed timely and effectively. Processing social media information pose multiple challenges such as parsing noisy, brief and informal messages, learning information categories from the incoming stream of messages and classifying them into different classes among others. One of the basic necessities of many of these tasks is the availability of data, in particular human-annotated data. In this paper, we present human-annotated Twitter corpora collected during 19 different crises that took place between 2013 and 2015. To demonstrate the utility of the annotations, we train machine learning classifiers. Moreover, we publish first largest word2vec word embeddings trained on 52 million crisis-related tweets. To deal with tweets language issues, we present human-annotated normalized lexical resources for different lexical variations. Keywords: Natural language processing, Twitter, Disaster response, Supervised classification 1. Introduction Twitter has been extensively used as an active communi- cation channel, especially during mass convergence events such as natural disasters like earthquakes, floods, ty- phoons (Imran et al., 2015; Hughes and Palen, 2009). Dur- ing the onset of a crisis, a variety of information is posted in real-time by affected people; by people who are in need of help (e.g., food, shelter, medical assistance, etc.) or by people who are willing to donate or offer volunteering ser- vices (Vieweg, 2012). Moreover, humanitarian and formal crisis response organizations such as government agencies, public health care NGOs, and military are tasked with re- sponsibilities to save lives, reach people who need help, etc. (Vieweg et al., 2014). Situation-sensitive requirements arise during such events and formal disaster response agen- cies look for actionable and tactical information in real- time to effectively estimate early damage assessment, and to launch relief efforts accordingly. Recent studies have shown the importance of social me- dia messages to enhance situational awareness and also in- dicate that these messages contain significant actionable and tactical information (Cameron et al., 2012; Imran et al., 2013; Purohit et al., 2013). Many Natural-Language- Processing (NLP) techniques such as automatic summa- rization, information classification, named-entity recogni- tion, information extraction can be used to process such social media messages (Bontcheva et al., 2013; Imran et al., 2015). However, many social media messages are very brief, informal, and often contain slangs, typograpi- cal errors, abbreviations, and incorrect grammar (Han et al., 2013). These issues degrade the performance of many NLP techniques when used down the processing pipeline (Ritter et al., 2010; Foster et al., 2011). We present Twitter corpora consisting of more than 52 million crisis-related messages collected during 19 differ- ent crises. We provide human annotations (volunteers …

Content truncated to 500 words