trendpreneur

As someone interested in opinions and trends, the 2008 U.S. election coverage has been a goldmine of ideas and information.

The election result itself and the resulting discussion online has outlined a problem for anyone seeking to aggregate opinion. Two, in fact. The first is related to the disproportionate weight of conversation relating to the election’s winner and loser — there are many celebratory messages, but fewer expressing disappointment. (No hard numbers, yet, so this assessment may be incorrect, or skewed by the kind of person likely to comment on Twitter at 6am GMT.)

Why are there fewer negative than positive messages? Our system needs to answer that; it could be because there are more people supporting the winner than the loser (well, the vote result implies this!) or simply because those supporting the loser are not inclined to give comment. Instead, we need to look at the history of commenters, and fill in the gaps to some extent. Someone who’s a long-time Obama supporter expressing their joy is not news; someone who had endorsed McCain but is making some positive statement is interesting. How do we normalise the flood of content to gain a fair representation of the overall opinion when some of those who would comment are silent?

The other problem is related to semantics and entities. An English speaker with some world knowledge can tell when a negative comment about Obama is a positive one about McCain, or vice versa. Obama wins; cue outpouring of negative adjectives from McCain supporters, feeling “worried”, “annoyed”, “nervous”. If we’re ranking comments by the opinion contained therein, this is negative opinion about Obama, but implied positive about McCain (they wouldn’t feel this way if their candidate had won). How do we extract that information from a data and algorithmic point of view, and how do we display it?

Well, there’s nothing I like more on a Wednesday than a good problem, so I’ll get back to you on that one.

[Flickr image from raster.]