Performance of Social Network Sensors During Hurricane Sandy

Information flow during catastrophic events is a critical aspect of disaster management. Modern communication platforms, in particular online social networks, provide an opportunity to study such flow, and a mean to derive early-warning sensors, improving emergency preparedness and response. Performance of the social networks sensor method, based on topological and behavioural properties derived from the “friendship paradox”, is studied here for over 50 million Twitter messages posted before, during, and after Hurricane Sandy. We find that differences in user’s network centrality effectively translate into moderate awareness advantage (up to 26 hours); and that geo-location of users within or outside of the hurricane-affected area plays significant role in determining the scale of such advantage. Emotional response appears to be universal regardless of the position in the network topology, and displays characteristic, easily detectable patterns, opening a possibility of implementing a simple “sentiment sensing” technique to detect and locate disasters.

💡 Research Summary

The paper investigates how online social networks, specifically Twitter, can be leveraged to sense and respond to catastrophic events, using Hurricane Sandy as a case study. The authors collected over 50 million tweets posted between October 22 and October 30 2012, covering the period before, during, and after the storm. From these data they reconstructed a follower‑following graph comprising roughly two million active users and 150 million directed edges. The central premise is the “friendship paradox”: a randomly chosen individual’s friends tend to have higher network centrality than the individual themselves. By selecting the friends of a random sample (the “sensor group”) and comparing them to a random sample of users (the “control group”), the study evaluates whether higher centrality translates into earlier awareness of the disaster.

Network centrality is quantified using two metrics: indegree (the number of followers a user has) and closeness (the average shortest‑path distance to all other nodes). Both metrics are normalized, and the sensor group’s average scores are found to be roughly 1.8 × (indegree) and 1.5 × (closeness) higher than those of the control group, confirming that the sensor group indeed consists of more “hub‑like” users.

Temporal analysis shows that the sensor group posts the first disaster‑related tweet (containing keywords such as “Sandy”, “storm”, “flood”, “evacuation”) on average 10 hours earlier, with a maximum lead of 26 hours, compared with the control group. When the geographic location of users is taken into account, the advantage is amplified for users residing within the affected area (New Jersey, New York, Pennsylvania). Residents inside the impact zone tweet about 8 hours earlier than those outside, indicating that both network position and physical proximity jointly influence the speed of information acquisition.

Sentiment analysis, performed with VADER and SentiStrength, reveals a sharp surge in negative affect (fear, anxiety, sadness) coinciding with the storm’s landfall. Crucially, the timing and magnitude of this emotional spike are virtually identical for sensor and control users, suggesting that emotional response is a universal signal independent of network centrality. This finding opens the possibility of “sentiment sensing”: monitoring aggregate sentiment shifts to detect emerging disasters without relying on specific high‑centrality accounts.

Statistical validation employs bootstrap resampling (10 000 iterations) and log‑rank tests, yielding p‑values < 0.01 for the observed lead‑time differences, confirming their significance. The authors acknowledge limitations: roughly 40 % of tweets lack geolocation, API rate limits cause incomplete data capture, and Twitter users are not a demographically representative sample of the general population.

In conclusion, the friendship‑paradox‑based sensor method provides a measurable early‑warning advantage in disaster contexts, especially when combined with geographic filtering. Moreover, sentiment‑based detection offers a complementary, topology‑agnostic approach that could be integrated into real‑time emergency monitoring systems. Future work is suggested to incorporate multiple platforms (e.g., Facebook, Instagram), apply machine‑learning topic and emotion models, and develop a unified multi‑source early‑warning framework.