Better emergency responses by removing social bots

12 December, 2019
Fresh Science, Media releases

Image credit: Mehwish Nasim

Filtering out social bots can help critical response teams see what’s happening in real time

Network visualisation of genuine users (at the right side) and social bots (in purple on the left side). Bots form a densely connected graph by tweeting similar messages at the same time. Two users in this network are connected if they repeatedly tweeted similar messages about the same topics in a given time duration. Bots are very loosely connected to rest of the users.

Researchers have created an algorithm that distinguishes between misinformation and genuine conversations on Twitter, by detecting messages churned out by social bots.

Dr Mehwish Nasim and colleagues at the School of Mathematical Sciences at the University of Adelaide say the algorithm will make it easier for emergency services to detect major events such as civil unrest, natural disasters, and influenza epidemics in real time.

“When something really big is going on, people tweet a huge amount of useful information,” says Mehwish.

“Being able rapidly filter out the polluting inputs of social bots, which churn out multiple messages that distort the human information flow, will help law enforcement agencies, incident response teams and volunteers to act in a timely manner.”

The algorithm works by computing diversity in tweets and temporal coordination.

Bot accounts have less diverse tweets. For instance, they repeatedly use same URLs or hashtags and tweet around the same time.

Mehwish – who is now based at CSIRO’s Data61, the data and digital specialist arm of Australia’s national science agency– says that by filtering out automated misinformation, agencies will be able to easily access valuable first-hand information by people on the ground during large events.

“For instance,” she adds, “minute to minute tweets during the recent NSW and Queensland bushfire emergencies could be extremely valuable for guiding specific fire service responses.”

Mehwish Nasim, University of Adelaide

Further Reading:

  1. Social Bot detection paper in 2018: Real-time Detection of Content Polluters in Partially Observa ble Twitter Networks: https://doi.org/10.1145/3184558.3191574
  2. November 2019 – Note that Mehwish is not the lead author on this paper and is looking to do a media release on this one. This has an application of her work but the main concept of the paper is attributed to the first author: Pachinko Prediction: A Bayesian method for event prediction from social media data: https://doi.org/10.1016/j.ipm.2019.102147