Posted by on Dec 13, 2011 in Research & Data

Hashtags have emerged as a way to mark keywords, topics, commentary or even add snark to content posted on Twitter. Adding the hash character (#) to the beginning of any word attaches the posted text to a wider conversation. For example, using the #immigration hashtag in the midst of a post on the topic of immigration exposes it to a wider audience, especially as it becomes hotter during the GOP primaries. Hashtags are being embraced by the political class as a way to both connect with their audiences but also battle with their opponents. Especially this year, we’re seeing a surge in hashtag usage leading to the 2012 US elections.

Looking at hashtag usage over time sheds light on how people’s perceptions of candidates are reflected through the language that they use. That’s exactly what we set out to track, analyzing over 8500 hashtags used over the past 3 weeks associated with each Republican candidate.

At SocialFlow we analyze the resonance of language within audiences by looking at historical and realtime language used in posts. We develop an understanding for people’s perception on any given topic, mapping out relationships between terms and topics amongst different audiences. We learn which topics are central, who the conversational outliers are and what topics act as bridges between different sides of the spectrum.

hashtag network of GOP candidates

Network graph of all hashtags used in reference with GOP candidates over the past 3 weeks.

From the data one learns that people’s perception of GOP candidates are certainly reflected through not only the language they use on Twitter, but also the frequency and combination of terms. In the graph above, each color represents connections between the candidate and associated hashtags. Every time a hashtag is published with a reference to a candidate in a tweet we added an edge to the graph. Every time hashtags appear with other hashtags, we also added an edge between them. The larger each node, the more times it appeared in tweets. Finally, we run a force-directed algorithm to organize the graph such that related hashtags appear closer.

Centrality of Hashtags

If we dive into the center of this graph, we note the dominant hashtags associated with all candidates – #tcot (Top Conservatives on Twitter), #gop, #teaparty, #fitn (First in the Nation – the New Hampshire primary), #IowaDebate, and so on. Their positioning in the graph represents how strongly associated they are in relation to each candidate. #teaparty is the largest, most central hashtag, evenly used in association with all candidates, while #tlot (Top Libertarians on Twitter) is situated closest to Ron Paul, being used over 3,500 times in association with Ron Paul, more than 7 times the number of times it was used with others such as Gingrich, Perry and Romney. #tlot is still a central hashtag in relation to the Republican primaries, though much more reflective of the conversation around Ron Paul, hence located closer to that candidate on the network graph.

zoom-in view of central hashtags

zoomed-in view of the central hashtags used in conversations referencing GOP candidates

Next, we can focus on the hashtag ecosystem surrounding each GOP candidate separately. By keeping the same graph structure but highlighting the neighborhoods of each candidate, we clearly see which hashtags are more strongly linked to each candidate. For example, in the middle graph below, all purple colored nodes represent the hashtags used in association with Mitt Romney. As we’d expect, the central hashtags all light up, as they were used multiple times in reference with the candidate.

Ron Paul Mitt Romney Newt Gingrich
above: Candidate Specific Hashtag Full Graphs

More interestingly, less central hashtags highlight unique aspects of each candidate, teaching us about the tone and topic of conversation. For example, in light of Romney claiming he is not a career politician, the DNC release the following youtube video. The buzz and reactions to this video are reflected in the #YoungerThanMittsPoliticalCareer hashtag (middle graph below). Hashtags like #immigracin, #primarias and #republicanas are highly correlated with Newt Gingrich, highlighting a Spanish speaking population referencing Gingrich on issues of immigration around the Republican primaries (right-most graph below).

Ron Paul - outliers Mitt Romney - outliers Newt Gingrich - outliers
above: Highly Correlated Hashtags for Referenced Candidates (zoomed-in)

Finally, it is interesting to compare the hashtag ecosystem ranging between two topics. In the example below, we compare hashtags used in reference with Romney versus those used with Gingrich. By eliminating the dominant central topics (#teaparty, #gop, …) we can now see the lay of the land – how the conversation differs between the candidates, and where it is similar.

Romney-Gingrich Tagcloud

Romney-Gingrich Hashtag Cloud

The hashtag #obama appears closer to Mitt Romney in the tagcloud, representing the stronger correlation between the appearance of those two terms – more people are referencing Romney with Obama (in comparison to Gingrich and Obama). Other noticeable correlations:

Mitt Romney: #gayrights, #lgbt, #jesus, #flipflop, #jobs, #economy

Newt Gingrich: #palestine, #OWS, #immigration, #abortion

Equally for Both: #republican, #dems, #economics, #amnesty

Hashtags offer us plenty of insight when trying to understand the public perception of any topic or event. Identifying abnormally spiking relationships in realtime help us cover the turn of events as they evolve, while keeping tabs on what is important for the general public. Relationships between hashtags used when referencing candidates give us important insight, while comparative language use helps us understand the differences between the conversations. Tracking these dynamics over time may unlock insight about candidates that would be hard to attain otherwise.

Now all this is *just* hashtags. Imagine how much we can learn across ALL topics! Questions? Thoughts? Feel free to comment below or ping me on Twitter: @gilgul