Abstract: Social media sources can be helpful in crisis situations, but discovering relevant messages is not trivial. Methods
have so far focused on universal detection models for all kinds of crises or for certain crisis types (e.g. floods).
Event-specific models could implement a more focused search area, but collecting data and training new models for
a crisis that is already in progress is costly and may take too much time for a prompt response. As a compromise,
manually collecting a small amount of example messages is feasible. Few-shot models can generalize to unseen
classes with such a small handful of examples, and do not need be trained anew for each event. We show how
these models can be used to detect crisis-relevant tweets during new events with just 10 to 100 examples and
counterexamples. We also propose a new type of few-shot model that does not require counterexamples.

Abstract: Radiation measurements are key information in post-nuclear accident situations. Automated Twitter accounts have been used to share the readings, but often in an incomplete way from the perspective of data sharing and risk communication between citizen and radiation experts. In this paper, we investigate the requirements for radiation measurements completeness, by analyzing the perceived usefulness of several metadata items that may go along the measurement itself. We carried out a benchmark of existing uses, and conducted a survey with both experts and lay citizens. We thus produced a set of guidelines regarding the metadata that should be used, and the way to publish it.

Abstract: Social Media and micro-blogging is being used during crisis events to provide live up-to-date information as events evolve (before, during and after). Messages are posted by citizens or public officials. To understand the effectiveness of these messages, we examined the content of geo-located Twitter messages (“tweets”) sent during the Moore, Oklahoma tornado of May 20th, 2013 (+/-1day) to explore the spatial and temporal relationships of real-time reactions of the general public. We found a clear transition of topics during each stage of the tornado event. Twitter was useful for posting and retrieving updates, reconstructing the sequence of events as well as capturing people's reactions leading up to, during and after the tornado. A long-term goal for the research reported here is to provide insights to forecasters and emergency response personnel concerning the impact of warnings and other advisory messages.

Abstract: There is evidence that people increasingly use social networking sites like Twitter in the aftermath of terrorist attacks to make sense of the events at the collective level. This work-in-progress paper focuses on the content of Twitter messages related to the 2016 terrorist attack on the Berlin Christmas market. We chose topic modeling to investigate the Twitter data and the terror management theory perspective to understand why people used Twitter in the aftermath of the attack. In particular, by connecting people and providing a real-time communication channel, Twitter helps its users collectively negotiate their worldviews and re-establish self-esteem. We provide first results and discuss next steps.

Abstract: This empirical analysis examines the use of the live-streaming application Periscope in three crises that occurred in 2015. Qualitative analyses of tweets relating to the Amtrak derailment in Philadelphia, Baltimore protests after Freddie Grey’s death, and Hurricane Joaquin flooding in South Carolina reveal that this recently deployed application is being used by both citizens and journalists for information sharing, crisis coverage and commentary. The accessibility and immediacy of live video directly from crisis situations, and the embedded chats which overlay on top of a video feed, extend the possibilities of real-time interaction between remote crowds and those on the ground in a crisis. These empirical findings suggest several potential challenges and opportunities for responders.

Abstract: Twitter is used for spreading information during crisis events. In this paper, we first retrieve event-related information
posted in English and Russian during six disasters and sports events that received wide media coverage in both
languages, using an adaptive information filtering method for automating the collection of about 100 000 messages.
We then compare the contents of these messages in terms of 17 informational and linguistic features using a
difference in differences approach. Our results suggest that posts in each language are focused on different types
of information. For instance, almost 50% of the popular people mentioned in these messages appear exclusively
in either the English messages or the Russian messages, but not both. Our results also suggest differences in the
adoption of platform mechanics during crises between Russian-speaking and English-speaking users. This has
important implications for data collection during crises, which is almost always focused on a single language.

Abstract: In this paper we present a new approach to enhance information extraction from social media that relies upon the geographical relations between twitter data and flood phenomena. We use specific geographical features like hydrological data and digital elevation models to analyze the spatiotemporal distribution of georeferenced twitter messages. This approach is applied to examine the River Elbe Flood in Germany in June 2013. Although recent research has shown that social media platforms like Twitter can be complementary information sources for achieving situation awareness, previous work is mostly concentrated on the classification and analysis of tweets without resorting to existing data related to the disaster, e.g. catchment borders or sensor data about river levels. Our results show that our approach based on geographical relations can help to manage the high volume and velocity of social media messages and thus can be valuable for both crisis response and preventive flood monitoring.

Abstract: In summer of 2014 the city of Münster experienced an urban flash flood not seen before with such intensity in Germany. This paper investigates the subsequent governmental and ad-hoc community response actions with a focus on the chronologies of Facebook and Twitter usage. Interviews identified drawbacks of coordinating volunteers in social media ecosystems. Possible solutions to overcome issues related to the interaction of community and official relief activities are identified.

Abstract: This paper offers a descriptive account of Twitter (a micro-blogging service) across four high profile, mass convergence events-two emergency and two national security. We statistically examine how Twitter is being used surrounding these events, and compare and contrast how that behavior is different from more general Twitter use. Our findings suggest that Twitter messages sent during these types of events contain more displays of information broadcasting and brokerage, and we observe that general Twitter use seems to have evolved over time to offer more of an information-sharing purpose. We also provide preliminary evidence that Twitter users who join during and in apparent relation to a mass convergence or emergency event are more likely to become long-term adopters of the technology.

Abstract: Microblogging sites such as Twitter can play a vital role in spreading information during “natural” or man-made disasters. But the volume and velocity of tweets posted during crises today tend to be extremely high, making it hard for disaster-affected communities and professional emergency responders to process the information in a timely manner. Furthermore, posts tend to vary highly in terms of their subjects and usefulness; from messages that are entirely off-topic or personal in nature, to messages containing critical information that augments situational awareness. Finding actionable information can accelerate disaster response and alleviate both property and human losses. In this paper, we describe automatic methods for extracting information from microblog posts. Specifically, we focus on extracting valuable “information nuggets”, brief, self-contained information items relevant to disaster response. Our methods leverage machine learning methods for classifying posts and information extraction. Our results, validated over one large disaster-related dataset, reveal that a careful design can yield an effective system, paving the way for more sophisticated data analysis and visualization systems.

Abstract: When Hurricane Sandy swept over the US eastern seaboard in October 2012, it was the most tweeted about event at the time. However, some of the most affected areas were underrepresented in the social media conversation about Sandy. Here, we examine the hurricane-related experiences and behaviors shared on Twitter by residents of Far Rockaway, a New York City neighborhood that is geographically and socioeconomically vulnerable to disasters, which was significantly affected by the storm. By carefully filtering the vast Twitter data, we focus on 41 Far Rockaway residents who offer rich personal accounts of their experience with Sandy. Analyzing their first-person narratives, we see risk perception and protective decision-making behavior in their data. We also find themes of invisibility and neglect when residents expressed feeling abandoned by the media, the city government, and the overall relief efforts in the aftermath of Sandy.

Abstract: We report on the development of a classifier to identify Twitter users contributing first-hand information during a disaster. Identifying such users helps social media monitoring teams identify critical information that might otherwise slip through the cracks. A parallel study (St. Denis et al., 2020) demonstrates that Twitter user filtering creates an information-rich stream of content, but the best way to approach this task is unexplored. A user's profile contains many different “modalities” of data, including numbers, text, and images. To integrate these different data types, we constructed a multimodal neural network that combines the loss function of all modalities, and we compared the results to many individual unimodal models and a decision-level fusion approach. Analysis of the results suggests that unimodal models acting on Twitter users' recent tweets are sufficient for accurate classification. We demonstrate promising classification of Twitter users for crisis response with methods that are (1) easy to implement and (2) quick to both optimize and infer.