doppelganger-finder

Doppelganger-finder finds multiple accounts (doppelgangers) of a user. It is useful to merge multiple aliases of a user. Given a dataset with n users user_1, user_2, ..., user_n, this code will compute what is the probability that user_i and user_j are the same person, for all i, j where i!=j.

Usage:

python doppelganger_finder.py -i INPUT -o OUTPUT -s MODEL_DIR

where, INPUT is the file with all features extracted (.arff or .svm),

OUTPUT is the .csv file which will contain the probabilities of two users being the same,

MODEL_DIR is the directory where the classifiers will be saved for future use.

Example:

python doppelganger_finder.py -i ../example/100-test-all.arff -o ../example/result.csv -s ../example/models/

Algorithm:

Let, A = set of users in a dataset

Calculate probability for every pair of users in A

2.a. For a user Ai in A:

   C <- classifier trained of all users in A except Ai
   R <- Test C using Ai and get the probability scores of Ai's document written by other users in A
   for an author Aj in R:
       p[Ai][Aj] = probability of Ai's document was written by Aj

Calculate combined pairwise probability

3.a For Ai, Aj in A:

  p[Ai][Aj] = probability of Ai's document was written by Aj
  p[Aj][Ai] = probability of Aj's document was written by Ai
  p[Ai==Aj] = Ai and Aj are the same author = p[Ai][Aj] * p[Aj][Ai]
  Ai, Aj are likely to be same person if p[Ai==Aj] is high.

Output p[Ai==Aj] for every pair of author Ai and Aj in A.