I wrote an article in AllAnalytics.com on Don’t Count on Twitter for Presidential Predictions which was a Counterpoint to Pierre DeBois, Founder, Zimana article on All Hail Twitter as Presidential Predictor, and I suggest reading both in light of the second presidential debate that just happened on Tuesday.
From what I have heard on Cable/TV there are really 3 battleground states, Florida, Ohio and Virginia (assuming North Virginia). Figured I see if I can shed any light but the first thing I noticed (using Sysomos) is that I am picking up the sentiment of both candidates at one time in a single query, which isn’t really what I want to do. On the other hand, it’s really hard to find instances in social media around the debates where one candidate is mentioned, but not the other.
It seems to me that there were need to be an additional level of “entity analysis” that goes past what a boolean query could produce and you would almost have to read each verbatim (most of them are on Twitter, anyway) and then assign them to one or the other and decide if it’s really positive or negative in sentiment. Perhaps Crimson or DiscoverText can take this on, but then again, perhaps it’s something that needs to be set up well to run in the first place, as I’ve seen how easy it is to mis-configure searches like this. Since so much on Social is someone’s tweet, in this case, that’s repeated, it’s difficult to say just how representative Twitter really is (which was one of the points I made in my AllAnalytics.com article).
So i tried a different approach with Marketwire Sysomos, using the Popular Phases on Twitter for just Ohio, Florida and North Virginia, but honestly, when looking at the phrases themselves, they do not settle the matter.
So I decided to give Netbase a try though (they have a Presidential Mood Meter Infographic online) I wasn’t able to create a filter to the state level, just the country level, which would not work too well for Ohio, for example. Also using Ohio as term to filter on would show few verbatim since that’s not what people are writing about, I would think, instead they are taking positions on the candidates performance. I added an additional filter “lies”,”lied” and “lie”.
If I could but filter on locations or on specific people (entities) I might have a better shot at it (making the data tell the story I’m looking to find in it).
Of course, there’s no accounting for the filter bubble that is coloring all of this communication in the first place – it’s as if people see white as black and vice versa. As people see the same information, more or less differently, the tool are often helpless to tell the difference.
I’m told, so it’s thought, that I met the next Mayor of New York last night. There was also an excellent Gamification conference at the Metropolitan Pavilion created by Badgerville. More about both later.