Natalie points out a basic error I made in charting the discussion about Royal and Hollande: BlogPulse doesn't do too well with normalizing for accented characters when it indexes. I had actually looked at this for Hollande (for whom I discovered that the form "François Hollande" dominated) but not for Royal. So take a look at the following:
This example is another good one for underlining the dangers of trivial keyword based analysis. One obviously couldn't search for the term 'royal' and claim that it tracked the discussion about the candidate. However, in the US the different form in which Hillary Clinton and Barack Obama appear in text suggest that using equivalent terms (e.g. "Hillary Clinton" versus "Barack Obama") is also misleading. This leads to the approach of using common constraining words (e.g. "Clinton +(2008 OR campaign)" versus "Obama +(2008 OR campaign)".
For what it's worth, I'll throw this one is as well:
Comments