Waking Theta Brainwave states

Post Reply New Topic

I'll bet you guys are the same person

In fact I wonder whether I can use a Markov chain or something to detect similarity

Do share! I'm not too familiar with Markov Chains...

I pulled the idea out of my ass tbh

But apparently it is used commonly in these instances

http://www.perlmonks.org/index.pl?node_id=464805

I just downloaded Peter Millican's 'Signature Stylometric System' linguistic analysis software and cut and pasted 2000 to 3000 word writing samples for Alba and Prometheus - while I obviously am not an expert in that software, there are major statistical differences that suggest that they are in fact two different people.

At the top of the analysis, Alba has a much more egocentric writing style (not making normative judgement - simply far more I, me, my, I've, and I'm in those posts - perhaps the term 'personalized' would be a better description. And pairwise comparisons continue to suggest very different writing styles - one uses far more of certain words to establish conditions (if, would, for, both, probably, quite, etc.)

Also - Chi square tests comparing punctuation, letter composition of text, and word length all indicate highly significant differences between the two screen names. I must therefore reject your stated null hypothesis that Alba and PrometheusPan are the same.

http://www.philocomp.net/?pageref=human ... =signature

Just put together a sample of my writing for a negative control. On the chi-square tests, the differences between my writing and Prometheuspan's was not significant even at the .20 level!! Some of the curves were eerily similar. What's that strange feeling? Is my ego dissolving?? Have the pod-people snatched my brain? Is someone developing an artificial neural network and using my thought as training data?

Last edited by monty on 02 Sep 2008, 2:08 pm, edited 1 time in total.

Thanks Monty so generous of you. Why bother to dignify obnoxious posts with a response? These people aren't happy unless they're throwing insults. It should be obvious with even a pea size brain to work with, that we aren't the same individual.

Give it a rest man, and let them chatter. Some people may actually be interested in this thread. (To varying degrees.) This is rude and distracting.




See above admonishment only x2.

Sheesh.

I wasn't prepared for this

Rejecting hypotheses always seemed as easy as plugging and chugging for the right distribution then checking the limits

But I never knew that the implications could hit me so hard: someone actually entertains PrometheuSpam's inane drivel



Time to form a new hypothesis

Are you currently using any 'controlled substances'?

EDIT - maybe your software is ineffective ... that also seems very likely now

The last time I posted here was two days ago. I know you want to be noticed for saying this,
so this is your moment.

Sorry. I didn't look at your post date. It wasn't really my point though. I was trying to read the thread and found the repeated negative interjections distracting and annoying.

Perhaps I should have PM'd.

Thanks for noticing. I feel validated that you have acknowledged my meager existence, Mr. President.

No. The statute of limitations has long run out on any youthful indiscretions, and I believe I suffered no lasting effects banana manna ding dong! Watch out for the flying monkeys, duke.




Well, no, I think it is the nature of the test and the interpretation. The chi-square tests are relatively simple statistical tools that only give probablistic answers. Given enough comparisons with others (especially on a board like this), I would resemble some simply by chance. Sentence length and word length and punctuation frequency are simple measures of how we write, though they may be effective for some things. I suspect that my haikus (although mostly unwritten) would be conceptually different from prometheus's haikus, even though a statistical analysis would indicate some similarities. Markov chains, word bigrams and trigrams, principal components analysis, comparisons of sentence diagrams, and other techniques could give a more sophisticated comparison of longer passages.

Given two samples of unknown origin, it can never be completely certain if they were written by one person or two. The application of statistical analysis or pattern recognition cannot solve the mystery to the nth degree of certainty, although it can tell us that the styles are very different, and we can assume that something is likely or unlikely from that. The tests indicated that we can reject the hypothesis that Alba and Prometheus were the same with a 95%+ confidence level, but that it is not even 80% sure that Monty and Prometheus were 'different' - which means that we cannot assume the second hypothesis.

I'm not sure how your software applies chi square tests in this case, but can't you mash several posts together and then compare the results for better accuracy? I did something like that for ridiculing Dragonforce lyrics with a Markov generator ... the more songs I put in, the more realistic the output sounded.



I am familiar with both significance and goodness of fit testing.

Are you a statistician btw? I only started studying statistics about two weeks ago. It seems you really know a lot. I figured that having a good grasp of a field that everyone else seems to hate will give me a strong competitive edge in the future.

Yes, that's what I did - I simply cut and paste big chunks of what I thought were representative text from different posts by each person. I did not include most of Prometheus' lists, as those seemed to be a different form. I also tended to skip short, 1 sentence answers to others' posts, was looking for longer expository passages. My personal corpus was shorter as I was getting bored, so that might also affect the results - but it was actually from a different thread, so some of the tests was more likely to measure core structure, not the vocab of a particular conversation.



I don't consider myself a statistician per se - I have had a few years of graduate level statistics and research design. I also used statistical methods and neural nets in my dissertation - from a mathematician's standpoint, I didn't do anything new, but from my discipline's applied standpoint, it was considered novel - I showed that certain standard techniques could be used to mine good information from large datasets. (Actually, I came up with what I think is a new concept related to the structure and symmetry of neural nets, but haven't taken time to prove it).

Stats can be very useful - I think they are valuable. I haven't directly used them to get a job, though my programming skills did help me. But that is just the niche I live in - people in my discipline understand the idea that a programmer can make the computer crunch numbers and spit out answers. They are not always so quick to understand that a statistician can analyze whether those answers really mean anything.

Demons!! !

Oh yeah

Well, like I said earlier, if I can spin this topic into a thesis, I want to do mine on probabilistic methods for improving traffic lights, so that's in a similar vein.