THANOS: Election Prediction using Social Media Network

THANOS: Predicting Elections with Twitter and Polls 

Can we predict elections using Twitter data?

That’s the central question we asked in our latest paper—and we found a compelling answer: yes, but only if you do it right.

We developed a pair of models, THOS and THANOS, that combine Twitter data with traditional opinion polls to forecast election outcomes. While THOS uses hashtag frequency, THANOS takes it a step further, incorporating network centrality—a measure of how influential certain users are on the platform. Together, these models bridge the gap between online chatter and offline ballots.

Why Combine Twitter with Polls?

Opinion polls are slow. Twitter is fast, but noisy.

• Polls offer representative but lagging snapshots of voter sentiment.

• Twitter offers real-time insights, but it’s messy, biased, and often unrepresentative.

By combining both, we gain the depth of polling and the speed of social media.

Two Models: THOS and THANOS

1. THOS (Twitter Hashtag-based Opinion Survey)

   • Uses frequency of campaign-related hashtags over time.

   • Ideal for elections with clear front-runners and large victory margins.
     
     

2. THANOS (Twitter Hashtag and Network-based Opinion Survey)

   • Adds network features like:

     • Harmonic centrality (influence score of top users)

     • Retweet ratios of major influencers

   • Built for close races, where hashtag counts alone aren’t enough.

Tested on Real Campaigns

We applied both models to:

• Ireland’s 36th amendment referendum (2018)

  • THOS correctly predicted the landslide repeal of the abortion ban.

  • Actual vote: 66.4% Yes | THOS prediction: 67.3%

• 2018 US Senate Elections

  • THANOS correctly predicted the winner in 11 of the 12 closest races.

  • In Florida (decided by just 0.12%), THANOS gave a neutral forecast—exactly what you’d want in a toss-up.

Compared to other platforms (like FiveThirtyEight), our model performed just as well—or better—especially in tight contests.

Why Network Matters

It’s not just what people are tweeting—but who is driving the conversation.

• Influential users shape narratives, amplify messages, and sway opinion.

• Our model tracks this by building retweet networks and calculating centrality metrics.

• This helped us capture real momentum in dynamic races—something raw tweet counts can’t do.
  
  

Model Robustness

What if some tweets get deleted?
What if influencer accounts vanish?

No problem.

• We ran sensitivity analyses by randomly deleting up to 30% of the tweets.

• THANOS predictions barely changed, thanks to our focus on retweet structures and influential nodes.

Big Picture

THANOS shows that social media data—when used carefully—can enhance traditional election forecasting.

It's not about replacing polls. It's about enriching them with the real-time pulse of the electorate.

Whether it's a national referendum or a swing-state Senate race, this hybrid modeling approach offers a scalable, adaptive, and accurate tool for modern campaign analysis.

The paper is published in Data Science in Science.