The European Physical Journal (EPJ) is a series of peer-reviewed journals covering the whole spectrum of physics and related interdisciplinary subjects. EPJ is committed to high scientific quality in publishing and is indexed in all main citation databases.
- Published on Monday, 04 December 2017 16:45
Nowadays, platforms like Twitter play a big role in the aftermath of disasters, such as natural disasters, mass shootings, or terror attacks, as people try to receive the latest information on what happened through social media channels. A new study published in EPJ Data Science shows how an analysis of social media responses to disasters might help us better understand the dynamic of the public’s attention during these events, what such an analysis shows about people’s attention spans and focus points in the aftermath of disasters, and how analyses like these could be performed in a cost-effective way.
(Guest post by Yu-Ru Lin, originally published on SpringerOpen blog)
It has become prevailing to leverage social and mobile communication technologies for better crisis response. In recent disasters, from Hurricanes Harvey, Irma and Maria to the Las Vegas mass shooting, from Paris and Brussels terror to most recent attacks in Europe, victims used social media to seek aid, connect with loved ones and helpers, and search for the latest updates. During these events, the flow and flood of information could lead to a drain on people’s attention, making them desensitized to more relevant and critical information.
How might we systematically understand the collective wandering and drift-away of people’s attention during disaster events? When measuring a person’s brain activity, doctors and scientists have used techniques like functional magnetic resonance imaging (fMRI), which identifies active brain areas that are being fed with oxygenated blood. However, at the collective level, could we produce something like fMRI scans that measure people’s attention shift?
Figure 1: fMRI scans for capturing brain activity. Courtesy of Dr. David Shin, UC San Diego ((http://fmri.ucsd.edu/Research/whatisfmri.html)
In our EPJ Data Science study, we proposed a novel method to capture the collective attention shift under exogenous shocks, such as mass shooting or terrorist events.
By using Twitter users’ communication streams, we analyze how individual users used different “hashtags” over time in order to capture the transitions between users’ focal topics at the collective level. These transitions, forming a sequence of “attention shift networks” (as shown in the figure below), reveal several properties of network structures and temporal dynamics that have been unseen before. Such properties are surprisingly consistent across multiple recent events.
Figure 2 the transitions between users’ focal topics at the collective level.
Using a set of network metrics, we characterized the most salient patterns in the attention shift networks. In the wake of a shocking event (e.g., the 2015 Paris attacks), we observed that users’ attention are highly concentrated to just few highly connected topics; however, meanwhile, users also became more liable to switch focal topics, compared with their pre-event attention span. This suggests that users at that very moment could adjust the way they pay attention to information; while having short attention span on all incoming information, more critical information may be quickly subsided by new or irrelevant topics.
Observing such a collective attention shift could be very expensive, as it requires tapping into many individuals’ social media streams. Therefore, we further proposed a sampling algorithm that can monitor collective attention in a cost-effective way. Our results show that the algorithm can retain about 75% of attended information from the entire community with only 25% sampled users, which can help guide the data collection strategy in practice.
Figure 3: Using a set of network metrics to characterize the dynamics of the attention shift networks, we observed that the temporal patterns are highly consistent across multiple recent events.