Interesting. I put more weight on the authors caveats about this method, than on the method itself, which is fine since it was I think just presented as an example of what could be done.
Also, if the person most likely (by this analysis) is actually the one who wrote the op-ed, it was either:
- the POTUS, which seems unlikely (although who knows)
- the secretary for the POTUS, who types his tweets as well
- the one person in the President's circle who he cannot fire
Awesome! I was actually thinking of doing something similar, but with some key differences:
- try and find transcripts and speeches each "suspect" has given.
- Probably use a classification algo to try and determine who the author could be (most likely using KNN)
- I hadn't determined feature building, but was thinking of a either a simple one-hot encoding, entity embedding, or tf-idf.
I then encountered a moral dilemma -- do I do this and it potentially becomes ammo for messing with someone else's career (when that is not my intention)?
I wonder if there could be a way to prevent this by e.g. paying a bunch of random people to rewrite parts of the text in their “own” style (with a review to ensure the meaning isn’t lost).
I wouldn't think so, it's not like doxxing someone. They author did something very public and very political and America at least doesn't have any rules against identifying that person.
Separately, I don't think any good will come from identifying them.
So the person interviewed is a linguist who's colleagues "helped crack the Unabomber case by analyzing Ted Kaczynski's 35,000 word manifesto."
So the Washington Post published the Unabomber's 35,000 word manifesto in it's daily paper in 1995 as a special section. The Unabomber's brother recognized and turned in his brother. There was no crack in the case by investigators. His brother received the $1 million dollar reward from the Department of Justice for turning him in.
This isn't fake news but it is incorrect. Linguistic experts were involved but it doesn't go much deeper than that.
This is definitely interesting. However, I have assumed from the beginning that, protests to the contrary notwithstanding, this op-ed was thoroughly edited by someone to prevent such an analytical outing.
10 comments
[ 1.9 ms ] story [ 34.7 ms ] threadAlso, if the person most likely (by this analysis) is actually the one who wrote the op-ed, it was either: - the POTUS, which seems unlikely (although who knows) - the secretary for the POTUS, who types his tweets as well - the one person in the President's circle who he cannot fire
- try and find transcripts and speeches each "suspect" has given.
- Probably use a classification algo to try and determine who the author could be (most likely using KNN)
- I hadn't determined feature building, but was thinking of a either a simple one-hot encoding, entity embedding, or tf-idf.
I then encountered a moral dilemma -- do I do this and it potentially becomes ammo for messing with someone else's career (when that is not my intention)?
Separately, I don't think any good will come from identifying them.
Obviously there are, I think.
So the Washington Post published the Unabomber's 35,000 word manifesto in it's daily paper in 1995 as a special section. The Unabomber's brother recognized and turned in his brother. There was no crack in the case by investigators. His brother received the $1 million dollar reward from the Department of Justice for turning him in.
This isn't fake news but it is incorrect. Linguistic experts were involved but it doesn't go much deeper than that.