def extract_entities(shorttext_rows, site):
# { short text id -> (noun entities, named entities) }
shorttext_entities = {}
# nltk entity classes
nltk_entity_types = __get_nltk_entity_types__()
for shorttext_row in shorttext_rows:
shorttext_id = shorttext_row[0]
shorttext_str = shorttext_row[1]
noun_entities = []
named_entities = []
sentences = nltk.sent_tokenize(shorttext_str)
tokenized_sentences = [nltk.word_tokenize(sentence) for sentence in sentences]
tagged_sentences = [nltk.pos_tag(sentence) for sentence in tokenized_sentences]
chunked_sentences = nltk.batch_ne_chunk(tagged_sentences)
for tree in chunked_sentences:
__extract_valid_entities__(tree, (noun_entities, named_entities), nltk_entity_types)
shorttext_entities[shorttext_id] = (noun_entities, named_entities)
# Cache extracted entities
pkl_util.write_pickle(__output_str__, shorttext_entities, __get_nltk_entities_cache_path__(site))
def build_wikipedia_editor_username_cache():
''' Fetches large numbers of active Wikipedia editors who have made edits
recently and stores them in a cache, which we can access while attempting
to find usernames that exist on both Wikipedia and a short text source site.
(The occurrence of such cross-site username matces may be low, so we want
to have a large cache of Wikipedia editors to draw upon).
The cache is a mapping of { username -> { edited page -> number of edits on page } } '''
# Load the Wikipedia usernames+edits cache
output_str = "Wikipedia editor usernames and their edited pages..."
editor_usernames = pkl_util.load_pickle(output_str,
__wikipedia_editors_cache_path__)
if editor_usernames is None:
editor_usernames = []
# Prompt how many Wikipedia usernames to fetch and query Wikipedia until retrieved that many
desired_num_editors = prompt_and_print.prompt_num_entries_to_build("active Wikipedia editors",
editor_usernames)
pre_fetch_len = len(editor_usernames)
wikipedia_api_util.query_editors_of_recentchanges(desired_num_editors, editor_usernames)
print "Fetched "+str(len(editor_usernames)-pre_fetch_len)+" more recent and active Wikipedia editors"
# make sure all usernames are lowercase
editor_usernames = [u.lower() for u in editor_usernames]
# Update cache
print "Cached a total of "+str(len(editor_usernames))+" Wikipedia editor usernames"
pkl_util.write_pickle(output_str, editor_usernames, __wikipedia_editors_cache_path__)
def build_crosssite_username_dataset(site):
''' Searches the given site for the Wikipedia editor usernames we have previously cached.
Does so until have a sufficient set of unique users who have both active Wikipedia accounts
and active accounts on the given site. Saves those users in a csv file and a pkl cache
and also writes to cache the usernames that are determined to NOT exist on the given
site so we don't bother searching for them again in the future.
@param site: Should be a Site object
'''
siteNameStr = str(site.siteName)
# Load or create/initialize the spreadsheet of usernames
usernames_csv_path = __get_usernames_csv_path__(site)
csv_string = 'usernames that exist on both Wikipedia and '+siteNameStr
headers = [COLUMN_USERNAME, __COLUMN_SAME_INDIVIDUAL__]
usernames_in_csv = csv_util.load_or_initialize_csv(usernames_csv_path, csv_string, headers, COLUMN_USERNAME)
# Load the caches of Wikipedia usernames:
editor_names_cache = pkl_util.load_pickle("Wikipedia editor usernames",
__wikipedia_editors_cache_path__)
editor_usernames = [] if (editor_names_cache is None) else editor_names_cache
# editor usernames that do NOT exist on the given site
nonexistent_usernames_path = __get_nonexistent_usernames_cache_path__(site)
nonexistent_usernames_cache = pkl_util.load_pickle("Wikipedia usernames that do NOT exist on "+siteNameStr+"...",
nonexistent_usernames_path)
if nonexistent_usernames_cache==None:
nonexistent_usernames_cache = []
# only need to analyze those usernames that we haven't
# already determined do or do not exist on given site
usernames_todo = __get_remaining_todo__(editor_usernames,
[usernames_in_csv, nonexistent_usernames_cache])
# Prompt how many matching usernames to fetch from the given site
desired_num_usernames = prompt_and_print.prompt_num_entries_to_build(csv_string, usernames_in_csv)
num_to_append = desired_num_usernames - len(usernames_in_csv)
if len(usernames_todo) < num_to_append:
print "Only "+str(len(usernames_todo))+" unanalyzed Wikipedia usernames in cache. If you "+\
"want "+str(desired_num_usernames)+" total in the cross-site usernames csv, you'll have to "+\
"re-run script and choose to first fetch more Wikipedia editor usernames."
prompt_count = 0
while(len(usernames_in_csv)<desired_num_usernames and len(usernames_todo)>0):
# Intermittently prompt user whether to continue fetching matching usernames or exit script
if prompt_count >= __PROMPT_COUNT__:
continue_searching = prompt_and_print.prompt_continue_building(csv_string, usernames_in_csv, desired_num_usernames)
if not continue_searching:
break
prompt_count = 0 # reset count
prompt_count = prompt_count + 1
# get lists of usernames that do or do not also exist on the given site
match_response = site.fetching_existence_status(usernames_todo, desired_num_usernames)
existing = match_response[site.get_existing_response_key()]
nonexisting = match_response[site.get_nonexisting_response_key()]
print "Found "+str(len(existing))+" existing and active usernames on "+siteNameStr
# update the spreadsheet with any new usernames that have been fetched
existing_rows = [[username, __VALUE_UNCONFIRMED__] for username in existing]
csv_util.append_to_spreadsheet(csv_string, usernames_csv_path,
usernames_in_csv, existing_rows)
# and update the list of usernames in the csv so we know how
# many more we still need to fetch to reach the desired num
usernames_in_csv.extend(existing)
# Also update the cache of Wikipedia usernames that do NOT exist on the given site
nonexistent_usernames_cache.extend(nonexisting)
nonexistent_write_str = "usernames that DO NOT exist on both Wikipedia and "+siteNameStr+"..."
pkl_util.write_pickle(nonexistent_write_str, nonexistent_usernames_cache, nonexistent_usernames_path)
# remove any usernames that we now determined do not exist on given site
usernames_todo = __get_remaining_todo__(usernames_todo, [existing, nonexistent_usernames_cache])
rate_limited = match_response[site.get_rate_limit_key()]
if rate_limited:
break # reached rate limit, so break
def build_wikipedia_edits_dataset(crosssite_usernames, prompt=True):
# Load or create/initialize the spreadsheet of users' wikipedia edits
csv_string = 'Wikipedia edits made by usernames that also exist on a site that is a source of short texts'
headers = [COLUMN_USERNAME, __COLUMN_ARTICLE_ID__, __COLUMN_NUM_EDITS__]
usernames_in_csv = csv_util.load_or_initialize_csv(__edits_csv_path__, csv_string, headers, COLUMN_USERNAME)
# Load the cache of edits, a dict: { username -> {edited page -> num edits } }
editor_names_to_edits_cache = pkl_util.load_pickle("Wikipedia editor usernames to their edited pages+counts",
__edits_cache_path__)
if editor_names_to_edits_cache is None:
editor_names_to_edits_cache = {}
# only need to fetch the edits for usernames that we haven't already done
editors_todo = [u for u in crosssite_usernames if u not in usernames_in_csv]
# Exit if all available names are done
if len(editors_todo)==0:
print "Wikipedia edit data fetched and stored for all "+\
str(len(crosssite_usernames))+" confirmed cross-site editors. Exiting."
return
print str(len(crosssite_usernames))+" cross-site editors total, and "+\
str(len(editors_todo))+" editors not yet in spreadsheet of edits "
# Prompt how many users to fetch edits for
if prompt:
desired_num_editors = prompt_and_print.prompt_num_entries_to_build(csv_string, usernames_in_csv)
num_to_append = desired_num_editors - len(usernames_in_csv)
if len(editors_todo) < num_to_append:
print "Only "+str(len(editors_todo))+" cross-site usernames available. If you want "+\
"want "+str(desired_num_editors)+" total editors' edits in the edits csv, you'll have to "+\
"re-run script and choose to first fetch more cross-site usernames."
else:
desired_num_editors = 1
edits_rows = []
#prompt_count = 0
progress_count = 1
for username in editors_todo:
if len(usernames_in_csv) >= desired_num_editors:
# have enough so exit
break
'''
# Intermittently prompt user whether to continue fetching matching usernames or exit script
if prompt and prompt_count >= __PROMPT_COUNT__:
continue_searching = prompt_and_print.prompt_continue_building(csv_string, usernames_in_csv, desired_num_editors)
if not continue_searching:
break
prompt_count = 0 # reset count
prompt_count = prompt_count + 1
'''
if progress_count%10==0:
print "Querying for pages edited by cross site usernames... Number "+\
"usernames whose edits have been fetched so far: "+str(progress_count)
progress_count = progress_count+1
user_edits = wikipedia_api_util.query_usercontribs(username, False)
for article_id in user_edits:
num_times_edited = user_edits[article_id]
edits_row = [username, article_id, num_times_edited]
edits_rows.append(edits_row)
# keep track that we'll be adding this username to the csv
usernames_in_csv.append(username)
editor_names_to_edits_cache[username] = user_edits # add that user+edits to cache
# update the spreadsheet with any new editors' edits that have been fetched
csv_util.append_to_spreadsheet(csv_string, __edits_csv_path__, usernames_in_csv, edits_rows, False)
# update the edit mapping cache
pkl_util.write_pickle("user edits to file...",
editor_names_to_edits_cache, __edits_cache_path__)
def __save_resolved_entities__(resolved_entities, site, use_cache):
if use_cache:
pkl_util.write_pickle(__resolved_entities_output_str__, resolved_entities, __get_resolved_entities_cache_path__(site))
def save_annotator_decisions(annotator_decisions, site):
pkl_util.write_pickle(__annotator_output_str__, annotator_decisions, __get_annotator_cache_path__(site))
def save_entity_judgements(judgments, site):
pkl_util.write_pickle(__candidate_judgments_output_str__, judgments, __get_candidate_judgments_cache_path__(site))
def build_entities_dataset(shorttext_rows, site):
siteNameStr = str(site.siteName)
# Load or create/initialize the spreadsheet of users' short texts
entity_csv_path = __get_entities_csv_path__(site)
output_str = __get_detected_entities_output_str__(site)
headers = [COLUMN_ENTITY_ID, __COLUMN_ENTITY_STRING__, COLUMN_SHORTTEXT_ID, COLUMN_SHORTTEXT_STRING, COLUMN_USERNAME]
entities_in_csv = csv_util.load_or_initialize_csv(entity_csv_path, output_str, headers, COLUMN_ENTITY_ID)
shorttexts_in_csv = csv_util.get_all_column_values(entity_csv_path, COLUMN_SHORTTEXT_ID)
print "A total of "+str(len(shorttext_rows))+" short texts available to detect and resolve entities in..."
# Load the cache of ambiguous entity objects
ne_objs = pkl_util.load_pickle(output_str, __get_ne_cache_path__(site))
if ne_objs is None:
ne_objs = []
# Load the cache of short texts that contain no entities
# and that we don't need to keep querying services with
entityless_shorttexts = get_entityless_shorttexts(site)
# Load the cache of problematic short texts that we can
# go back and look at later..
problematic_shorttexts = get_problematic_shorttexts(site)
# Prompt how many users to fetch short texts for
desired_num_entities = prompt_and_print.prompt_num_entries_to_build(output_str, shorttexts_in_csv)
entities_rows = []
progress_count = 1
all_shorttexts_done = True
for shorttext_row in shorttext_rows:
shorttext_id = shorttext_row[0]
if shorttext_id in shorttexts_in_csv or shorttext_id in entityless_shorttexts or shorttext_id in problematic_shorttexts:
# already did entities for this shorttext (and either successfully
# detected some, successfully detected none, or encountered an error)
continue
all_shorttexts_done = False
try:
if len(entities_in_csv) >= desired_num_entities:
# have enough so exit
break
if progress_count%10==0:
print "Detecting named entities in short texts posted on "+siteNameStr+\
" by cross-site usernames... Number of short texts whose entities have been fetched so far: \n"+\
str(len(entities_in_csv))
progress_count = progress_count+1
original_shorttext = shorttext_row[1]
username = shorttext_row[2]
# get the entities contained in each short text
# clean the short text before attempting to detect entities in it
clean_shorttext = text_util.format_text_for_NER(original_shorttext, site)
if clean_shorttext=='':
# whole string was invalid, perhaps a URL or
# some other content that gets totally filtered
problematic_shorttexts.append(shorttext_id)
continue
detected_entities = named_entity_finder.find_and_construct_named_entities(shorttext_id, original_shorttext, username, site)
if len(detected_entities)==0:
entityless_shorttexts.append(shorttext_id)
for ne_obj in detected_entities:
# cache this entity object
ne_objs.append(ne_obj)
# make a row in the spreadsheet for this entity
ne_id = ne_obj.get_entity_id()
entity_row = [ne_id, ne_obj.surface_form,
shorttext_id, original_shorttext,
username]
entities_rows.append(entity_row)
# keep track that we'll be adding this entity to the csv
entities_in_csv.append(ne_id)
except Exception as st_e:
print "Problematic short text "+str(shorttext_row[1]), st_e
if 'referenced before assignment' in str(st_e):
raise # it's a server error so we need to stop
problematic_shorttexts.append(shorttext_id)
continue
# update the spreadsheet with any new users' short texts that have been fetched
csv_util.append_to_spreadsheet(output_str, entity_csv_path, entities_in_csv, entities_rows, False)
# update the cache of ambiguous surface form objects
pkl_util.write_pickle(output_str, ne_objs, __get_ne_cache_path__(site))
pkl_util.write_pickle(__entityless_output_str__, entityless_shorttexts, __get_entityless_cache_path__(site))
pkl_util.write_pickle(__problematic_output_str__, problematic_shorttexts, __get_problematic_cache_path__(site))
print "Cached a total of "+str(len(ne_objs))+" ambiguous named entities"
if all_shorttexts_done:
print "Completed detecting and resolving entities in all short texts available."
else:
print "More short texts available to detect and resolve entities for."
