'Senator, We Sell Ads': Analysis of the 2016 Russian Facebook Ads Campaign

TL;DR

This paper analyzes 3,500 Russian Facebook ads from the 2016 US election, revealing their characteristics, effectiveness, and strategies to sow division rather than influence the election outcome.

Contribution

It applies NLP techniques to categorize and evaluate the ads, providing insights into their content, effectiveness, and campaign strategies, which was not previously documented in detail.

Findings

Effective ads have less positive sentiment and focus on past events.

Effective campaigns are specific, personalized, and aim to sow division.

Ads' duration and promotion suggest intent to divide, not sway the election.

Abstract

One of the key aspects of the United States democracy is free and fair elections that allow for a peaceful transfer of power from one President to the next. The 2016 US presidential election stands out due to suspected foreign influence before, during, and after the election. A significant portion of that suspected influence was carried out via social media. In this paper, we look specifically at 3,500 Facebook ads allegedly purchased by the Russian government. These ads were released on May 10, 2018 by the US Congress House Intelligence Committee. We analyzed the ads using natural language processing techniques to determine textual and semantic features associated with the most effective ones. We clustered the ads over time into the various campaigns and the labeled parties associated with them. We also studied the effectiveness of Ads on an individual, campaign and party basis. The most effective ads tend to have less positive sentiment, focus on past events and are more specific and personalized in nature. The more effective campaigns also show such similar characteristics. The campaigns' duration and promotion of the Ads suggest a desire to sow division rather than sway the election.