Given the heavy debate regarding our 2008 presidential election, I thought it would be fun to generate some semantic networks regarding the fore-runners as discussed on Twitter. The first (click for full view) focuses on “tweets” including John McCain.

Some quick context to help read the first graph:

• The branch to the right (red) focuses on the three candidates, who tend to be mentioned in conjunction with each other more than any other term. “i’d like to follow mccain on twitter but i can’t find him on here. oh well, obama and clinton it is.”
  
• The green area focuses around the controversy generated by Bill Cunningham’s remarks, and McCain’s subsequent disavowal. “mccain disavows bill cunningham’s ‘barack hussein obama’ rhetoric“
• The small-green branch (left) focuses on a recent New York times article questioning McCain’s relationship with a lobbyist.
• The purple branch (bottom left seems to focus on Howard Dean’s comments regarding McCain’s fundraising record: “‘we want john mccain to obey the law with his own name on it.’ - howard dean“
• … and typing in a few keywords into a twitter search engine like terraminds.com will provide you the context to understand some of the other clusters.

Next, I try something a little different; this semantic network only examines “twittered” adjectives and descriptives in relation to John McCain.

1. “scamp mccain takes on ‘angry liberal left’”
2. “getting smeared by the new york times was a diabolically clever move by the mccain campaign to court republicans who hate his guts, like me“
3. “texas poll: obama-clinton close mccain comfortable - wall street journal”

In the interest of parity, I also generated a similar twitter network for Barack Obama. This network is a little more complex, and I left in some outliers for comparison.

• “mccain, clinton are leading in missouri; clinton and obama are too close to even project a win this early. http//www.realclearpolitics.com “
• “obama seems to be focused on differentiating himself with mccain and not clinton. interesting.”
• “ap poll: number of republican voters satisfied with mccain as gop nominee 7/10. romney 6/10. for the democrats clinton 79%, obama 85%.”
• “morning thoughts: romney stays alive, sets dangerous precedent. mccain approaching must-win status. clinton, obama bury hatchet.”
• “rice is strong on foreign policy, considered a mccain weakness. in addition to giving voters an alternative to either obama or clinton.”
• “obama = ca shiraz; clinton = established expensive ca cab; mccain = northern rhone; romney = gin & tonic.”

I think these charts show a lot of the weaknesses and potential of natural language processing. Strengths, because associations are an interesting and underdeveloped part of our analytical toolkit, and even a simple NLP application can be a useful exploratory tool. Weaknesses, because it takes some pretty smart algorithms, bandwidth and processing power to adequately interpret human conversation - especially Twitter, which tends to contain more abbreviations, errors, and shorthand than most other Internet communication. In my examples, it’s clear that the associations are somewhat clouded because my co-occurrence algorithm can’t distinguish between descriptions about a target (mccain) and other sentence targets. However, a more sophisticated and domain-specific implementation might produce results more relevant for specific applications, be they political analysis, public relations, or market/brand research.

| Permalink | Trackback