Stylometry: Recognizing Authorship Through Writing Styles - Prof. Rachel A. Greenstadt, Study notes of Cryptography and System Security

Download Stylometry: Recognizing Authorship Through Writing Styles - Prof. Rachel A. Greenstadt and more Study notes Cryptography and System Security in PDF only on Docsity!

Prac%cal A)acks Against Authorship

Recogni%on Techniques

(…and a li)le bit of AI)

Mike Brennan

What is AI?

(Wait, I don’t think I need this slide! We’re all computer scien%sts here, right?)

What is a Classifier?

• Given some set of informa%on about a target,

determine it’s classifica%on.

• For example, if given the en%re course history of

a student at Drexel you would probably be able to

figure out what their major is.

• Feature set might include a comparison of the classes

they took to the required classes for each major.

• What if they switched majors?
  • We could add another feature that looks at how recently the comparison set of classes was taken.

Learning With Decision Trees

• Examples described by a)ribute values (Boolean, discrete, con%nuous)
• E.g., situa%ons where I will/won't wait for a table:
• Classifica%on of examples is posi%ve (T) or nega%ve (F)
• Goal: Learn the defini%on for the goal predicate WillWait (slide from Russell & Norvig)

Neural Networks

• A)empt to build a computa%on system based

on the parallel architecture of brains.

• Characteris%cs:

– Many simple processing elements

– Many connec%ons

– Simple messages

– Adap%ve interac%on

• Connec%on “weights” altered as data flows through. (slide from Rachel Greenstadt)

Neural Networks

(image from Russell & Norvig)

Naïve Bayesian Classifier

• Given a target, determine which class it belongs to, based on the prior probabili%es of features in the target. - Prior Probability: What is the probability of X given that we

know Y?

• Example: Movie Review Classifier
  • Features: frequency of each word in posi%ve and nega%ve

reviews.

• Mul%ply the probability of each word indica%ng a posi%ve

review and each word indica%ng a nega%ve review. Classify

the review as whichever has a greater probability.

Stylometry

• Stylometry is the study of a)ribu%ng

authorship to documents based on the

linguis%c styles they exhibit.

– Handwri%ng doesn’t count!

• In general the basic ques%on of Stylometry is

“who wrote this document?”

Why is Stylometry Important?

• Great, you can figure out who wrote a 200 year

old document, so what?

• From the Ins%tute for Linguis%c Evidence:

• “In some criminal, civil, and security ma)ers, language

can be evidence. A suicide note, a threatening le)er,

anonymous communica%ons business emails, blog

posts, trademarks – all of these can help inves%gators,

a)orneys, human resource execu%ves and private

individuals understand the heart of an incident. When

you are faced with [a threat or incident] you need

reliable, validated methods [of stylometry].”

The Problem

• An underlying assump%on in most Stylometry

research is that people are honest in their

wri%ng style. What if they’re not?

Related Work

• Some work has been done in looking into how

well methods of stylometry can dis%nguish

between true authors and imitators.

• Somers & Tweedie, 2003: Alice in Wonderland.
• Mixed results.

• Overall this area has been widely looked over.

Patrick Juola, an expert in Computer Linguis%cs

formerly of Oxford states in his 2008 book:

• “There is obviously great poten%al for further work

here.”

Study Format and Setup

• 15 Individual Authors. Par%cipa%on had three

parts:

• Submit 5000 words of pre‐exis%ng wri%ng from a

formal source.

• Formal means school essays, professional reports, etc. No slang, abbrevia%ons, casual conversa%on.
• Write a new 500 word passage as an obfusca%on

a)ack.

• Task: Describe your neighborhood.
• Write a new 500 word passage as an imita%on a)ack.
• Task: Imitate Cormac McCarthy using a passage from The Road, write a third person narra%ve about your day star%ng from when you wake up.

Some A)ack Examples…

• “Light sliced through the blinds, and construc%on

began in the adjacent apartment. The harsh

cacophony crashed through the wall.”

• “Hot water in the mug. Brush in the mug. The

blade read "Wilkinson Sword" on the layered wax

paper packaging.”

• “He fills the coffee pot with water, arer cleaning

out the putrid remains of yesterday's brew. The

beans are in the freezer, he remembers.”

Our Approach: Method 1

• Sta%s%cal Method using the Signature

Stylometric System.

– Assigns a Chi‐Square value for each comparison.

The higher the number, the less likely they are the

same author.

• Three features: word length, le)er usage,

punctua%on usage.

• Average Accuracy: 95%