4 ms·

> By looking for these kinds of partisan idioms in Wikipedia articles, the Harvard researchers could determine whether the text sounded more like the product of

> By looking for these kinds of partisan idioms in Wikipedia articles, the Harvard researchers could determine whether the text sounded more like the product of a Republican or a Democrat. They were also able to document how the articles evolved over time. Another explanation that fits the results of this study is that the popular party idioms changed over time. As the old idioms were replaced by new ones this analysis made the Wikipedia pages and editors appear to become less biased.

Bartweiss 10y ago
This seems like a serious risk, and I'd like to see how the "partisan idioms" were collected. "Illegal aliens" stood out to me as something I don't hear much from either party - the fight today is generally over "illegal immigrants" versus "undocumented immigrants". That looks like an obvious candidate for a marginalized phrase distorting the analysis. "Death tax" is still around, but I think there's a euphemistic treadmill thing happening - plenty of people on the right will now say "estate tax" and expect it to be understood as a thing they're strongly opposed to. Finally, the example given seems to be about maturation rather than moderation. "Turbans and terrorists" is rather un-encylopedic, and it would be totally possible to strip it from an article even while preserving a strong conservative bent. I wouldn't be shocked if these results were right, but it also feels like a sociology paper I wouldn't want to trust without a secondary analysis.
- smsm42 10y ago
  "Turbans and terrorists" gave me a pause too. Then again, that was in 2006, where Wikipedia was much less prominent than now, which means less editors, so I took a look. If we look at the article at the eve of 2006 it's not "turbans and terrorists" at all: https://en.wikipedia.org/w/index.php?title=Afghanistan&oldid=33409879 https://en.wikipedia.org/w/index.php?title=Afghanistan&oldid... I mean maybe it has some bias, it's quite large so I didn't analyze it, but it's not blatantly bad like WaPo seems to imply. I suspect they fallen the victim of what many journalists fall victim too - concentrating on rare exceptions like some troll putting there "turbans" for lulz and it surviving on the site for some short time - and not on the processes that govern the regular life of the site.
MichaelMoser123 10y ago
you are right, sentiment analysis algorithms look at a text as a bag of words (or n-grams where n is two or three), some of the tokens have been assigned some set score and the result score is a linear function based on the token counts; Changes in usage would significantly alter the result over longer time periods.

Bartweiss 10y ago

This seems like a serious risk, and I'd like to see how the "partisan idioms" were collected. "Illegal aliens" stood out to me as something I don't hear much from either party - the fight today is generally over "illegal immigrants" versus "undocumented immigrants". That looks like an obvious candidate for a marginalized phrase distorting the analysis. "Death tax" is still around, but I think there's a euphemistic treadmill thing happening - plenty of people on the right will now say "estate tax" and expect it to be understood as a thing they're strongly opposed to. Finally, the example given seems to be about maturation rather than moderation. "Turbans and terrorists" is rather un-encylopedic, and it would be totally possible to strip it from an article even while preserving a strong conservative bent. I wouldn't be shocked if these results were right, but it also feels like a sociology paper I wouldn't want to trust without a secondary analysis.

smsm42 10y ago
"Turbans and terrorists" gave me a pause too. Then again, that was in 2006, where Wikipedia was much less prominent than now, which means less editors, so I took a look. If we look at the article at the eve of 2006 it's not "turbans and terrorists" at all: https://en.wikipedia.org/w/index.php?title=Afghanistan&oldid=33409879 https://en.wikipedia.org/w/index.php?title=Afghanistan&oldid... I mean maybe it has some bias, it's quite large so I didn't analyze it, but it's not blatantly bad like WaPo seems to imply. I suspect they fallen the victim of what many journalists fall victim too - concentrating on rare exceptions like some troll putting there "turbans" for lulz and it surviving on the site for some short time - and not on the processes that govern the regular life of the site.

MichaelMoser123 10y ago

you are right, sentiment analysis algorithms look at a text as a bag of words (or n-grams where n is two or three), some of the tokens have been assigned some set score and the result score is a linear function based on the token counts; Changes in usage would significantly alter the result over longer time periods.