enron_attack

df_test = pd.DataFrame(test_ds, columns=['text', 'author'])
# make a unique list of authors
auth = sorted(set(df['author']))
# make a dict of possible signatures where key is the labeled entity and value is a list of possible signatures
auth_dict = {}
auth_dict[auth[0]] = ['ben', 'benjamin', 'rogers', 'benjamin rogers','ben rogers','br']
auth_dict[auth[1]] = ['chris', 'dorland','chris dorland','cd']
auth_dict[auth[2]] = ['drew','fossum','drew fossum','df']
auth_dict[auth[3]] = ['jeffrey','shankman','jeffrey shankman','js']
auth_dict[auth[4]] = ['kevin','presto','kevin presto','kp']
auth_dict[auth[5]] = ['kim','kimberly','watson','kimberly watson','kim watson','kw']
auth_dict[auth[6]] = ['lynn','blair','lynn blair','lb']
auth_dict[auth[7]] = ['mark','haedicke','mark haedicke','mh']
auth_dict[auth[8]] = ['mike','michelle','cash','michelle cash','mike cash','mc']
auth_dict[auth[9]] = ['phillip','allen','phillip allen']

def untargeted_signature_attack(auth_dict,text, target = ''):
  if target is not '':
    target = auth_dict[target][0]
  max_name_len = max([len(x) for x in auth_dict.keys()])
  body = text[-max_name_len:]
  body_l = body.lower()
  for author in auth:
    found = False
    for sig in sorted(auth_dict[author], key=lambda k: len(k), reverse=True):
      if sig in body_l:
        found = True
        startidx = body_l.find(sig)
        endidx = startidx + len(sig)
        text = text[:-max_name_len] + body.replace(body[startidx:endidx], target)
        break
    if found:
      break
  # return pd.Series([text, found])
  return text
  
df_test[['perturbed_text', 'modified']] = df_test['text'].apply(lambda text:untargeted_signature_attack(auth_dict,text))
print('Percentage of perturbation: {:.2f}%\n'.format((len(df_test[df_test['modified'] == True])/len(df_test))*100))
for a in auth:
  print('Modified data percentage for ' + a + ' = {:.2f}%\n'.format((len(df_test[(df_test['author']==a) & (df_test['modified']==True)])/len(df_test[df_test['author']==a]))*100))