When a disaster occurs, time is limited and safety is in question, so people need to act quickly with as much knowledge of the situation as possible. Knowing where to find information is a crucial step. However, affected populations, response agencies and other stakeholders may not know what to expect from different information sources. Previous research found that information shared on Twitter varies substantially from one crisis to another, but understanding the similarities that could ease the information overload wrought by social media during crisis situations.
To build this understanding, Olteanu et al. investigate several crises in a systemic manner and with a consistent methodology. Olteanu et al. used a retrospective sample of tweets from 2012 and 2013, related to 26 events that spawned significant activity on Twitter. The data was compiled by a crowdsourcing team and then categorized by crisis dimensions; such as the type of hazard and geographic spread ‚as well as by content dimensions, including the informativeness, type and source of information. The dataset (available at http://crisislex.org/) compiles on average 132 million tweets per month, amounting to roughly 38 GB of compressed data.
There are commonalities in the types of information people tend to be concerned with, given the particular dimensions of the crisis. Messages of caution and advice, and those containing information about infrastructure and utilities, were the most repeated. There are also subtle differences in Twitter responses that depend on the spread of the event over space and time. For example, when a crisis event is geographically diffused, the proportion of caution and advice tweets is higher, and when a crisis is localized the proportion of caution and advice tweets is lower. Despite these variations, the Twitter responses to human-induced crises are more similar to other human-induced crises than other types of hazards.
Eyewitness accounts as the source of information comprise, on average, only 9% of tweets for a crisis. Nearly 80% of all tweets relevant to a crisis are from outsiders reporting on the event with indirect and repeated sources of information. Furthermore, only an average of 47% of all crisis-related tweets contain the four main types of information (affected individuals, donations and volunteering, caution and advice, and infrastructure and utilities). Therefore, to use Twitter as a source of information during a crisis, a practitioner would have skip through the 80% of crisis-related tweets that are built on indirect and repeated information, then identify the 47% that concern the selected information types. This represents a relatively small number of tweets that could be of value for emergency management and response.
These findings will be of interest to members of the public, emergency managers and formal response agencies who are increasingly trying to understand how to effectively use social media during a crisis. Stakeholders who know what content and reliability to expect will not have to sift through masses of social media posts; instead they have a reasonable expectation of what they will find, and can then make more informed decisions regarding their situational assessment process. The results of this study show that there are a relatively small number of tweets that can help turn the information from Twitter into real-time information about the crisis.
There are a small number of tweets in a crisis that provide useful information for responding agencies. Focusing on these can enhance intelligence gathering and decision-making processes.