Future Tech

Graph database shows Biden outspends Trump in social media ad war

Tan KW
Publish date: Sat, 18 May 2024, 07:58 AM
Tan KW
0 445,142
Future Tech

Although Joe Biden spends more on Facebook and Instagram ads than Donald Trump, ads attacking the US president outnumber those attacking his likely rival in this year's presidential election, according to data analysis.

With the backing of a $250,000 research grant from graph database vendor Neo4j, Syracuse University's Institute for Democracy, Journalism and Citizenship (IDJC) published its early findings from a research project to examine spending and sentiment trends across Meta's two popular social media platforms.

The research group, which employed Neo4j's graph database for the first time this year, found direct spending by Biden was greater than Trump's spending by about 7-to-1 on Facebook and Instagram ads. However, in attack ads from groups associated with either candidate, more mention Biden (47 percent) than Trump (37 percent).

The IDJC found $15.3 million spent by 1,802 groups that invested 24,000 ad buys for Trump or Biden. There were 5,545 unique ads between the beginning of September last year and February 29, 2024. These ads generated about 869 million impressions. In addition, Biden and Trump's campaigns spent another estimated $10 million, earning 303 million impressions.

The research project is led by Professor Jennifer Stromer-Galley, senior associate dean at Syracuse University's School of Information Studies, who has researched presidential campaigning in the internet age from 1996 to 2016.

The research group collects data from ads on Facebook and Instagram using Meta's ad library API. It tags the data using Google's natural language model BERT. Until a few months ago it had been storing the data in MongoDB - a document store database - for further analysis, but has now switched to graph.

Stromer-Galley told The Register she became interested in employing a knowledge graph for the project two years ago and even wrote an unsuccessful grant proposal. With funding and technical support from Neo4j, the team was able to explore the potential of using a graph database - which is designed to expose networks of relationships using nodes and edges - for the first time.

"It was surprising how quickly we could surface some of these networks," she said. "I'm not even sure we would have pursued these organizations in the past. We wanted to do some organizational analysis in 2020 but we were focused on the presidential candidates, and also governor and senator candidates. It's already a lot of candidates and it's hard to manage it all.

"There are a number of technical issues that always crop up. Dealing with that volume of data just seemed insurmountable. Coming into this election season, the knowledge graph structure makes it so much easier. I just don't have to worry about how do we really try to find and look at these relationships the way we used to."

The research found that conservative-leaning groups are the top ad spenders. For example, Liberty Defender Group was the highest spender at over $1.3 million. Although it is not a political organization, its ads indicate a pro-Trump alignment. The second highest spender at more than $1 million is AFP Action or Americans for Prosperity, another conservative organization.

Why focus only on Facebook and Instagram? It's because Meta makes the data available, even though it is not legally obliged to do so. Other social media platforms like TikTok, Google, YouTube (owned by Google), or Snapchat do not make their ad data available in the same way.

"We are notionally collecting Twitter data at the moment," Stromer-Galley said. "We were going to use the Twitter API, but it's incredibly expensive, so we're currently actually experimenting with Apify, which is a scraping tool. We're also exploring partnering with an organization to use a browser extension to collect other ads that people are being exposed to on Google, YouTube, and potentially Twitter. We're still working on that." ®



Be the first to like this. Showing 0 of 0 comments

Post a Comment