Forumbled Options

[Simon Kirby]

Richard Garvie mentioned again the other day his idea that blackdog is actually Julian Swift Hook. I'm not interested in who blackdog is, but it did pique my interest.

I did this as a bit of fun, and I hope no one is offended, but I compared the text of the forumistas here with each other to measure how similar we are one from the other. It's called stylomety, and it's hardly any more scientific that reading the bumps on your head, but the theory is that you can isolate elements of a text's style such as its vocabulary, phraseology, and complexity, and then use those like a fingerprint as a comparison.

So for example you can compare the average sentence length, or some measure of clarity like the Gunning fog factor, or you can look for particular idioms and characteristic phrases, or even look for characteristic errors. I've done a bit of all of that - they all individually come up with different answers - and made a table of who's writing style resembles whose.

I'm not going to "out" anyone (well, not much anyway), but if you're interested in who your style is most like I'll post the top five comparisons for you.

To test Richard's idea I also threw in a bit of text attributed to JSH taken from a couple of Leader's reports from the town council web site. Bad news for Richard is that blackdog's style is only 20.8% like JSH. JSH's style did compare at 70% with one member here, but I'm not going to volunteer who that was.

Curiously enough my own style compared well with that of my friend Charlie Farrow at 72%, so that does get me wondering if outlook influences style.

More weird still is that Richard's own style compared best to Charlie Farrow's, and then my own!

This was me then:

Simon Kirby
stem vocabuary size: 969.0
words: 3350
complex word count: 385
mean word length: 4.5
mean sentence length: 22.185430463576157
******* Gunning fog index: 13.471187110803598
non-word metric: 0.9665671641791045

Brewmaster to Simon Kirby,75.01266010414504%
CharlieF to Simon Kirby,72.16331633621522%
Simon Kirby to badmummajamma,71.55895023816953%
Jack to Simon Kirby,71.0140122168921%
Sherlock to Simon Kirby,70.98092471873497%

If you'd like to see your own stats and comparisons just say.

[motormad]

Interesting.
I thought I was JSH to be honest.

What are my stats! And how do you find out such information

[Andy Capp]

Go-on: out me.

[MontyPython]

Go-on: out me.

You're northerner with a flat cap and a fag hanging out of his mouth! I am sure I saw you in Sainsbury's the other day!

[Andy Capp]

You're northerner with a flat cap and a fag hanging out of his mouth! I am sure I saw you in Sainsbury's the other day!

“'Infamy, infamy, they've all got it in for me:'

[Biker1]

This was me then:

Simon Kirby
stem vocabuary size: 969.0
words: 3350
complex word count: 385
mean word length: 4.5
mean sentence length: 22.185430463576157
******* Gunning fog index: 13.471187110803598
non-word metric: 0.9665671641791045

Brewmaster to Simon Kirby,75.01266010414504%
CharlieF to Simon Kirby,72.16331633621522%
Simon Kirby to badmummajamma,71.55895023816953%
Jack to Simon Kirby,71.0140122168921%
Sherlock to Simon Kirby,70.98092471873497%

It doesn't work then does it?
Because we know who you are and I think most of us know who Brewmaster is!

If you'd like to see your own stats and comparisons just say.

Yeah, go on then!!

[On the edge]

Lets have a look at mine too!

[motormad]

Who is Brewmaster?

[Andy Capp]

Because we know who you are and I think most of us know who Brewmaster is!

I think the idea is to see who's posting style we are most in line with. We would be able to some degree predict who we are, or who we are like on other forums.

Who is Brewmaster?

An out-of-town misery-guts of some repute.

[Andy Capp]

*Burp!*

[JeffG]

Perhaps Simon Kirby and Brewmaster went to the same school...

An out-of-town misery-guts of some repute.

That's not fair! His mulled cider is to die for.

[Andy Capp]

That's not fair! His mulled cider is to die for.

Surely it is a logical fallacy that a purveyor of fine cider and perries could not be a misery guts!

[JeffG]

So you are contradicting yourself?

[blackdog]

It doesn't work then does it?
Because we know who you are and I think most of us know who Brewmaster is!

I too know Brewmaster (I have the horrible feeling I owe him a pint), but perhaps the analysis does work and a 75% correlation is indicative of a non-match.

Simon - what level of correlation would the experts deem to be evidence that the two writers are one and the same?

[Cognosco]

I too know Brewmaster (I have the horrible feeling I owe him a pint), but perhaps the analysis does work and a 75% correlation is indicative of a non-match.

Simon - what level of correlation would the experts deem to be evidence that the two writers are one and the same?

Come on Julian your know Garvie is never wrong the games up?

[Andy Capp]

So you are contradicting yourself?

Err, no, I was describing your argument!

I said BM was a misery guts.
You said no he isn't: he makes good cider.
I said making good cider doesn't make him not a misery guts.

[Andy Capp]

I too know Brewmaster (I have the horrible feeling I owe him a pint), but perhaps the analysis does work and a 75% correlation is indicative of a non-match. Simon - what level of correlation would the experts deem to be evidence that the two writers are one and the same?

I doubt that can be established with certainty. perhaps 99%

It would be interesting to test the system by analysing ones self under the guise of being different people.

[MontyPython]

It would be interesting to test the system by analysing ones self under the guise of being different people.

Are you trying to make me a schizophrenic??


and me too!

[Nothing Much]

and me too!

You put it so well,
You put it so well.
ce
ce

[Simon Kirby]

Righto, I've been doing a little tinkering and tuning, and I've run the text with a slightly larger word-count limit which has excluded some of the less effusive forumistas. My own top-5 comparisons are now:

Simon Kirby and Exhausted - 71.2%
Simon Kirby and On the edge - 67.2%
Simon Kirby and Phil_D11102 - 67.2%
Simon Kirby and lordtup - 66.6%
Simon Kirby and NWNREADER - 65.9%

and I am least like - wait for it...
Simon Kirby and Swift Hook - 32.3% - ain't that the truth!

I did say that I wasn't going to out anyone, but as Richard asked the question I think it's only fair to say that Richard's second closest comparison was...
Richard Garvie and blackdog - 73.9% - so it turns out that blackdog is actually RIchard Garvie!

I will will just share one more - blackdog's least closest match is...
blackdog and Swift Hook - 29.8%

If the Leader's reports really were written by JSH (and I can't imagine that they'd have been written by anyone else) then that really puts the nails in the coffin of Richard's theory.


motormad's top five:
motormad and xjay1337 - 98.6%
motormad and Timbo - 98.0%
motormad and JeffG - 89.2%
motormad and x2lls - 85.0%
motormad and Strafin - 80.5%


Andy Capp's top five:
Andy Capp and Iommi - 90.0%
Andy Capp and Strafin - 70.2%
Andy Capp and JeffG - 69.5%
Andy Capp and Bloggo - 66.8%
Andy Capp and x2lls - 66.5%


Nothing Much's top five:
Nothing Much and xjay1337 - 60.2%
Nothing Much and motormad - 51.9%
Nothing Much and Bill1 - 49.2%
Nothing Much and Timbo - 48.8%
Nothing Much and JeffG - 48.6%


On the Edge's top five:
On the edge and Darren - 77.7%
On the edge and Bofem - 77.2%
On the edge and spartacus - 77.1%
On the edge and Jayjay - 75.5%
On the edge and lordtup - 74.3%


To answer blackdog's question about confidence levels, I think the answer is that the technique isn't really valid in that quantitive kind of way. If you have a text and you want to attribute it to an author and you have some samples from a set of candidate authors then you can use the technique to suggest the most likely author from that set of candidates but that's not the same thing. This was done with the Glabraith attribution to see if Galbraith was J K Rowling, but even comparing the text with the published text from just four candidate authors Rowling wasn't the best match for every test.

To give an idea though, I would suggest that numbers above 90% here are likely for texts from common authors, but I see at least one such correlation between two forumistas who I know to be different people so there really isn't any great confidence.

Can I interest anyone else in seeing their comparisons?