def run():
rows = csv_util.query_csv_for_rows('labeled_data/entities.csv')
labeled_entities = set()
resolved_entities = set()
for row in rows:
entity_id = row[0]+'_'+row[3] # (i.e., "surfaceform_shorttext")
label = row[2]
if label == 'Y':
labeled_entities.add(entity_id)
resolved_entities.add(entity_id)
elif label=='N':
labeled_entities.add(entity_id)
print str(len(labeled_entities))+" annotated entities."
print str(len(resolved_entities))+" unanimously annotated entities."
def run():
rows = csv_util.query_csv_for_rows('labeled_data/entities.csv')
labeled_entities = set()
resolved_entities = set()
for row in rows:
entity_id = row[0] + '_' + row[3] # (i.e., "surfaceform_shorttext")
label = row[2]
if label == 'Y':
labeled_entities.add(entity_id)
resolved_entities.add(entity_id)
elif label == 'N':
labeled_entities.add(entity_id)
print str(len(labeled_entities)) + " annotated entities."
print str(len(resolved_entities)) + " unanimously annotated entities."
def get_resolved_ambiguous_entities():
""" Returns the ambiguous entities for which the intended
meaning has been unanimously resolved by human annotators. """
all_entities = defaultdict(list)
correct_meaning_label = "Y"
row_count = -1
labeled_entities_dataset = csv_util.query_csv_for_rows("labeled_data/entities.csv", False)
for candidate_row in labeled_entities_dataset:
row_count = row_count + 1
if row_count == 0:
# header row
surfaceform_col = candidate_row.index("surface_form")
shorttext_col = candidate_row.index("short_text")
candidate_meaning_col = candidate_row.index("candidate_meaning")
candidate_label_col = candidate_row.index("candidate_is_relevant")
userkey_col = candidate_row.index("user_key")
continue
# use "surfaceform_shorttext" as ID for entity
surfaceform = candidate_row[surfaceform_col]
shorttext = candidate_row[shorttext_col]
entity_id = surfaceform + "_" + shorttext
meaning = candidate_row[candidate_meaning_col]
label = candidate_row[candidate_label_col]
userkey = candidate_row[userkey_col]
all_entities[entity_id].append((meaning, label, surfaceform, shorttext, userkey))
# test if entity is ambiguous (i.e., has more than one candidate meaning) and
# if so if entity has been resolved (i.e., has at least one candidate labeled
# as the intended meaning)
resolved_entities = {}
for entity in all_entities:
entity_tuple_list = all_entities[entity]
if len(entity_tuple_list) < 2:
continue
candidate_meanings = []
intended_meanings = []
user = None
for (meaning, label, surfaceform, shorttext, userkey) in entity_tuple_list:
# title of a potential meaning of the ambiguous entity
if not meaning in candidate_meanings:
candidate_meanings.append(meaning)
# annotated label indicating whether this candidate
# meaning is the intended meaning of the entity
if label == correct_meaning_label and not meaning in intended_meanings:
intended_meanings.append(meaning)
if user is None:
user = userkey
if len(intended_meanings) > 1 and len(intended_meanings) > 0 and user != None:
# this entity is ambiguous, has been manually resolved,
# and we know the user who wrote it
entity_obj = ResolvedEntity(candidate_meanings, intended_meanings, surfaceform, shorttext, user)
entity_id = entity_obj.get_id()
resolved_entities[entity_id] = entity_obj
return resolved_entities
def get_bridged_usernames():
usernames = {}
userhashes = csv_util.query_csv_for_rows("labeled_data/user_privacy/anonymized_userhash.csv")
for (userkey, username) in userhashes:
usernames[userkey] = username
def get_resolved_ambiguous_entities():
''' Returns the ambiguous entities for which the intended
meaning has been unanimously resolved by human annotators. '''
all_entities = defaultdict(list)
correct_meaning_label = 'Y'
row_count = -1
labeled_entities_dataset = csv_util.query_csv_for_rows('labeled_data/entities.csv', False)
for candidate_row in labeled_entities_dataset:
row_count = row_count+1
if row_count==0:
# header row
surfaceform_col = candidate_row.index('surface_form')
shorttext_col = candidate_row.index('short_text')
candidate_meaning_col = candidate_row.index('candidate_meaning')
candidate_label_col = candidate_row.index('candidate_is_relevant')
userkey_col = candidate_row.index('user_key')
continue
# use "surfaceform_shorttext" as ID for entity
surfaceform = candidate_row[surfaceform_col]
shorttext = candidate_row[shorttext_col]
entity_id = surfaceform+'_'+shorttext
meaning = candidate_row[candidate_meaning_col]
label = candidate_row[candidate_label_col]
userkey = candidate_row[userkey_col]
all_entities[entity_id].append((meaning, label, surfaceform, shorttext, userkey))
# test if entity is ambiguous (i.e., has more than one candidate meaning) and
# if so if entity has been resolved (i.e., has at least one candidate labeled
# as the intended meaning)
resolved_entities = {}
for entity in all_entities:
entity_tuple_list = all_entities[entity]
if len(entity_tuple_list) < 2:
continue
candidate_meanings = []
intended_meanings = []
user = None
for (meaning, label, surfaceform, shorttext, userkey) in entity_tuple_list:
# title of a potential meaning of the ambiguous entity
if not meaning in candidate_meanings:
candidate_meanings.append(meaning)
# annotated label indicating whether this candidate
# meaning is the intended meaning of the entity
if label==correct_meaning_label and not meaning in intended_meanings:
intended_meanings.append(meaning)
if user is None:
user = userkey
if len(intended_meanings)>1 and len(intended_meanings)>0 and user!=None:
# this entity is ambiguous, has been manually resolved,
# and we know the user who wrote it
entity_obj = ResolvedEntity(candidate_meanings, intended_meanings, surfaceform, shorttext, user)
entity_id = entity_obj.get_id()
resolved_entities[entity_id] = entity_obj
return resolved_entities
def get_bridged_usernames():
usernames = {}
userhashes = csv_util.query_csv_for_rows('labeled_data/user_privacy/anonymized_userhash.csv')
for (userkey, username) in userhashes:
usernames[userkey] = username
def run():
username_rows = csv_util.query_csv_for_rows('labeled_data/user_identity.csv')
total_flickr = 0
total_twitter = 0
total_youtube = 0
exists_flickrs = 0
exists_twitters = 0
exists_youtubes = 0
FP_Flickr = 0
FP_Twitter = 0
FP_Youtube = 0
TP_Flickr = 0
TP_Twitter = 0
TP_Youtube = 0
for row in username_rows:
#username = row[0]
flickr_label = row[1]
twitter_label = row[2]
youtube_label = row[3]
exists_wikipedia = row[4]
exists_flickr = row[5]
exists_twitter = row[6]
exists_youtube = row[7]
if exists_flickr=='TRUE' and exists_wikipedia=='TRUE':
exists_flickrs+=1
if exists_twitter=='TRUE' and exists_wikipedia=='TRUE':
exists_twitters+=1
if exists_youtube=='TRUE' and exists_wikipedia=='TRUE':
exists_youtubes+=1
if flickr_label!='':
total_flickr+=1
if twitter_label!='':
total_twitter+=1
if youtube_label!='':
total_youtube+=1
if flickr_label=='TP':
TP_Flickr+=1
if twitter_label=='TP':
TP_Twitter+=1
if youtube_label=='TP':
TP_Youtube+=1
if flickr_label=='FP':
FP_Flickr+=1
if twitter_label=='FP':
FP_Twitter+=1
if youtube_label=='FP':
FP_Youtube+=1
print "Table 3"
print "Initial Sample Twitter: "+str(total_twitter)
print "Reused: "+str(exists_twitters)+" "+str(100*float(exists_twitters)/total_twitter)
print " "
print "Initial Sample YouTube: "+str(total_youtube)
print "Reused: "+str(exists_youtubes)+" "+str(100*float(exists_youtubes)/total_youtube)
print " "
print "Initial Sample Flickr: "+str(total_flickr)
print "Reused: "+str(exists_flickrs)+" "+str(100*float(exists_flickrs)/total_flickr)
print " "
print "Twitter Bridged: "+str(TP_Twitter)+" "+str(100*float(TP_Twitter)/(TP_Twitter+FP_Twitter))
print "Youtube Bridged: "+str(TP_Youtube)+" "+str(100*float(TP_Youtube)/(TP_Youtube+FP_Youtube))
print "Flickr Bridged: "+str(TP_Flickr)+" "+str(100*float(TP_Flickr)/(TP_Flickr+FP_Flickr))
