Home of the Muppet Imperial Moderator Corps

Bias

Is there any way of quantifying bias in the media? Using textual analysis or something like that.

Also, would it be possible to train a Bayesian filter to detect left-wing or right-wing writing?
Permalink Colm O'Connor 
August 25th, 2005
No. Shut the fuck up, you wrong-skinned hooten-anny.
Permalink muppet 
August 25th, 2005
I'd put Muppet at a fairly strong right-wing bias from that post...

I'm not aware of any way to categorise text automatically by content, since you would need to understand the meaning behind the words used to detect underlying political bias, which has got to be getting on for AI. You could probably use Bayesian filtering to give a fairly accurate account of the topic of the writing easily enough.

Having said that, wait until Chrissy shows up again and ask him, I'm sure he would be able to analyse the MindPixels contained in the message for you. :)
Permalink qwe 
August 25th, 2005
>I'm not aware of any way to categorise text automatically by
>content, since you would need to understand the meaning
>behind the words used to detect underlying political bias,
>which has got to be getting on for AI.

Well, supposing you had a corpus of left and right wing writings (make it self-declared left and right wing writings to make it easier) you could then train the bayesian filter using those writings.
Permalink Colm O'Connor 
August 25th, 2005
I wouldn't be surprised if you could detect things like talking points.
Permalink Tayssir John Gabbour 
August 25th, 2005
is this assuming that the democrats are left wing (which they are not)...?

real left wing rhetoric is filled with anger and talk of "brotherhood"... i'm sure you could pick up on that... you could use union speeches as a base...
Permalink Kenny 
August 25th, 2005
I think you're stereotyping the mannerisms of people with different political views too much here. Sure, you could probably identify individual writers in this way, but you woudln't be identifying their political bias, only that it was written by X who we are already aware is left/right/centre.

The major problem you would have is that the language used by people of differing bias will be similar when talking about different things - a liberal opinion on the rights of immigrants might very well use a lot of the same high ranking words as a more conservative piece on business. Then of course there is the problem that even the most one-sided reporting of events will undoubtably include quotes from opposing views, making it very difficult to identify which direction the article is coming from without being able to precisely understand the context in which each of the words is used.
Permalink qwe 
August 25th, 2005
I think you could do it with a large corpus of data.

For example, you could look for word proximity - "Bush" near "failed" or "opposed" or "Popularity" a lot may indicate liberal bias. Ditto "bush" near "Iraq" and "Oil." Obviously any significant occurrence of the words "neocon" or "impeach" indicates liberal bias.

OTOH, "Iraq" near "successful," "constitution," or "freedom" would indicate conservative bias; "death tax" is a significant indicator of conservatism, etc.

Best bet would be to run "known" biased sources and evaluate word pairs from the results, then weight them...

Philo
Permalink Philo 
August 25th, 2005
The big problem is that we are too diverse for national political terms to make a great deal of sense.
Permalink a cynic writes... 
August 25th, 2005
"The big problem is that we are too diverse for national political terms to make a great deal of sense."

should be

"The big problem is that we (?off) are too diverse for *anything* to make a great deal of sense."
Permalink Aaron F Stanton 
August 25th, 2005
The English to English translation can be interesting at times, yes.
Permalink a cynic writes... 
August 25th, 2005
I think Philo's examples (with the exception of Neocon) actually increase my opinion that this wouldn't work - 'Iraq' and 'constitution' together could reasonably be expected to appear in writing along the themes of 'Why the Iraqi constitution cannot work' as well as 'The Iraqi constitution will liberate the entire middle east region' simply because any major political idea will be discussed from both sides, and it is difficult to do without using the same language.

Having said that, if there are enough words like 'neocon', that will be almost exclusively used by one faction then perhaps this could work. It is likely, though, that these words will be used within a highly negative context (certainly 'neocon' seems to frequently be synonymous with 'slimy money obsessed b*****d') in which case they are less likely to show up in the more thought out pieces. This fact could be used to identify the more extreme examples, but I still believe you would have a hard time with more balanced writing without it coming down to identifying individual reporters' writing styles.
Permalink qwe 
August 25th, 2005
I only want to read posts written by people who think exactly like me and share my every opinion.
Permalink Winger 
August 25th, 2005
I agree with everything Winger just said, except the part about only wanting to read posts written by people who think exactly like me and share my every opinion. With that minor exception, I'm in total agreement.
Permalink Aaron F Stanton 
August 25th, 2005
Colm, I think you are looking for:

http://en.wikipedia.org/wiki/Fundamental_attribution_error
Permalink Peter 
August 25th, 2005
Peter, that's why I'd have to use self-attributed sources - i.e. sources that SAID that they were either left or right wing. Otherwise I'd get the error you just linked to because what looks left-wing to me might not to somebody else.
Permalink Colm O'Connor 
August 25th, 2005

This topic was orginally posted to the off-topic forum of the
Joel on Software discussion board.

Other topics: August, 2005 Other topics: August, 2005 Recent topics Recent topics