def batch_extract_content(websiteElementsPath, urlData):
## 1) Extract webpage data
print "[INFO] ==== Extracting webpage data ===="
data_extractor = WebsiteDataExtractor(websiteElementsPath)
out = pd.DataFrame(urlData["URL"])
keyterms = []
for url in urlData["URL"]:
print url
data_dict = data_extractor.crawlPage(url)
## 2) Extract candidate keyterms
print "[INFO] ==== Extracting candidate keyterms ===="
keyterm_extractor = KeyTermExtractor(data_dict)
keyterm_extractor.execute()
#print keyterm_extractor.result_dict
## 3) Compute candidate keyterm features
print "[INFO] ==== Computing candidate keyterm features ===="
keyterm_feat = KeyTermFeatures(url, data_dict, keyterm_extractor.result_dict, lang=utils.LANG_FR)
candidate_keyterm_df = keyterm_feat.compute_features()
selected_keyterms = []
if not candidate_keyterm_df.empty:
## 4) Filter for relevancy and output top 10 keyterms
print "[INFO] ==== Selecting relevant keyterms ===="
relevance_filter = RelevanceFilter(candidate_keyterm_df, "dataset/keyterm-classifier-model-v2.pickle", topk=10)
selected_keyterms = relevance_filter.select_relevant()
keyterms.append(",".join(selected_keyterms))
out["keyterms"] = keyterms
return out
def create_raw_dataset(output_filename):
df_raw = read_raw_data("dataset/preProc2_lower.json")
page_scraper = WebsiteDataExtractor("dataset/WebsiteElementsPathDef.xml")
# training dataset is made up only of pages published in 2015
df_dataset = df_raw.loc[df_raw['dateTime'].map(lambda x: x.year >= 2015)]
# get all URLs
all_urls = [x[1] for x in df_dataset['link'].iteritems()]
# get urls common with grapeshot - these will be our test set
test_urls = None
with open("dataset/extracted_terms_grapeshot_common_v3.json") as fp:
d = json.load(fp, encoding="utf-8")
test_urls = d.keys()
train_urls = [x for x in all_urls if x not in test_urls]
train_urls = random.sample(train_urls, 4 * len(test_urls))
# dataset urls are test + train
dataset_urls = test_urls + train_urls
dataset_dict = {}
idx = 1
for url in dataset_urls:
print "[INFO] " + str(idx) + " :: Parsing URL: " + url
page_data = page_scraper.crawlPage(url)
dataset_dict[url] = page_data
idx += 1
with open(output_filename, mode="w") as fp:
json.dump(dataset_dict, fp, encoding="utf-8")
with open("dataset/test_url_list.json", mode="w") as fp:
json.dump(test_urls, fp, encoding="utf-8")
with open("dataset/train_url_list.json", mode="w") as fp:
json.dump(train_urls, fp, encoding="utf-8")
print "[INFO] Page scraping dataset created."
def get_candidate_keyterms_dataset(output_file, url_list):
from website_data_extractor import WebsiteDataExtractor
from keyterm_extractor import KeyTermExtractor2
data_scraper = WebsiteDataExtractor("dataset/WebsiteElementsPathDef.xml")
candidate_extractor = KeyTermExtractor2(tagger, lang="french")
candidate_extractor.initialize()
dataset_dict = {}
for link in url_list:
print "Processing URL: " + link
data_dict = data_scraper.crawlPage(link)
candidate_extractor.execute(data_dict)
dataset_dict[link] = candidate_extractor.candidates
candidate_extractor.cleanup()
with open(output_file, "w") as fp:
json.dump(dataset_dict, fp)
def _initialize(self):
self.tagger = ttw.TreeTagger(TAGLANG=self.lang_abrev, TAGDIR=KeyTermExtractor2.TREETAGGER_DIR)
self.data_scraper = WebsiteDataExtractor("dataset/WebsiteElementsPathDef.xml")
self.candidate_extractor = KeyTermExtractor2(self.tagger, lang = self.lang)
self.candidate_extractor.initialize()
self.feature_extractor = KeyTermFeatures2(self.tagger, lang = self.lang)
#self.relevance_filter = RelevanceFilter("dataset/keyterm-classifier-model-v3.pickle", topk = self.topk)
#self.relevance_filter = RelevanceFilter("dataset/keyterm-classifier-model-updated.pickle", topk = self.topk)
self.relevance_filter = RelevanceFilter("dataset/keyterm-classifier-model-general.pickle", topk = self.topk)
self.keytermClassifier = KeytermClassification(
classesFile="dataset/top10-keywords-ecommerce-filtered.txt",
classesClusterPath="dataset/keyterm_clustering/top_adv_keyterm_clusters.dump")
__author__ = "alex"
import pprint
import utils.functions as utils
from website_data_extractor import WebsiteDataExtractor
from keyterm_extractor import KeyTermExtractor, KeyTermExtractor2
from keyterm_features import KeyTermFeatures
from keyterm_classifier import RelevanceFilter
if __name__ == "__main__":
url = 'http://www.generation-nt.com/blackview-a8-smartphone-petit-budget-pas-cher-mwc-2016-actualite-1925283.html'
## 1) Extract webpage data
print "[INFO] ==== Extracting webpage data ===="
data_extractor = WebsiteDataExtractor("dataset/WebsiteElementsPathDef.xml")
data_dict = data_extractor.crawlPage(url)
## 2) Extract candidate keyterms
print "[INFO] ==== Extracting candidate keyterms ===="
keyterm_extractor = KeyTermExtractor(data_dict)
keyterm_extractor.execute()
keyterm_extractor2 = KeyTermExtractor2(data_dict, lang="french")
keyterm_extractor2.execute()
print "======== Results from Extractor 1 ========"
pprint.pprint(keyterm_extractor.result_dict)
# print "Nr t1grams: " + str(len(keyterm_extractor.result_dict['t1gram']['term']))
# print "Nr t2grams: " + str(len(keyterm_extractor.result_dict['t2gram']['term']))
# print "Nr t3grams: " + str(len(keyterm_extractor.result_dict['t3gram']['term']))
class KeytermServerExtractor(object):
def __init__(self, port = 8080, lang = utils.LANG_FR, topk = 10):
print "Initializing Term Extractor Server"
## setup server port
self.port = port
self.topk = topk
## setup keyterm service extraction language
self.lang = lang
self.lang_abrev = utils.LANG_ABREV[lang]
## setup http request handling classes
self.server_class = HTTPServer
self.handler_class = makeServerHandlerClass(self)
## setup logging
## setup logging
root_log = logging.getLogger()
root_log.setLevel(logging.ERROR)
stdout_log = logging.StreamHandler(sys.stdout)
stdout_log.setLevel(logging.ERROR)
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
stdout_log.setFormatter(formatter)
root_log.addHandler(stdout_log)
## initialize keyterm extraction service modules
self._initialize()
def _initialize(self):
self.tagger = ttw.TreeTagger(TAGLANG=self.lang_abrev, TAGDIR=KeyTermExtractor2.TREETAGGER_DIR)
self.data_scraper = WebsiteDataExtractor("dataset/WebsiteElementsPathDef.xml")
self.candidate_extractor = KeyTermExtractor2(self.tagger, lang = self.lang)
self.candidate_extractor.initialize()
self.feature_extractor = KeyTermFeatures2(self.tagger, lang = self.lang)
#self.relevance_filter = RelevanceFilter("dataset/keyterm-classifier-model-v3.pickle", topk = self.topk)
#self.relevance_filter = RelevanceFilter("dataset/keyterm-classifier-model-updated.pickle", topk = self.topk)
self.relevance_filter = RelevanceFilter("dataset/keyterm-classifier-model-general.pickle", topk = self.topk)
self.keytermClassifier = KeytermClassification(
classesFile="dataset/top10-keywords-ecommerce-filtered.txt",
classesClusterPath="dataset/keyterm_clustering/top_adv_keyterm_clusters.dump")
def _cleanup(self):
self.tagger = None
self.data_scraper.cleanup()
self.candidate_extractor.cleanup()
self.feature_extractor.cleanup()
self.relevance_filter.cleanup()
def runServer(self):
server_address = ('', self.port)
httpd = self.server_class(server_address, self.handler_class)
print 'Starting httpd...'
try:
httpd.serve_forever()
except KeyboardInterrupt:
self._cleanup()
sys.exit(0)
except Exception as ex:
logging.getLogger().exception("Error in keyterm extraction!")
sys.exit(0)
def extracTermsFromLink(self, link):
default_return = {
"available_domains": ["http://www.generation-nt.com/", "http://www.maison.com/",
"http://www.journaldugeek.com/", "http://www.journaldugamer.com/",
"http://www.jdubuzz.com/", "http://news.pixelistes.com/",
"http://www.societe.com/", "http://www.pausecafein.fr/",
"http://worldofwarcraft.judgehype.com/news/",
"http://hearthstone.judgehype.com/news/", "http://diablo3.judgehype.com/news/",
"http://www.judgehype.com/news/",
"http://www.jeuxonline.info", "http://heroes.judgehype.com/news/",
"http://overwatch.judgehype.com/news/",
"http://film-warcraft.judgehype.com/news/", "http://judgehype.com/",
"http://portail.free.fr/", "http://www.planet.fr/",
"http://aliceadsl.closermag.fr/", "http://aliceadsl.lemonde.fr/",
"http://aliceadsl.gqmagazine.fr/"],
"defaultPath": False, "dataIntegrity":False, "keyTerms":[]}
try:
## 1) Extract webpage data
print "[INFO] ==== Extracting webpage data ===="
data_dict = self.data_scraper.crawlPage(link)
default_return["defaultPath"] = data_dict["defaultPath"]
default_return["dataIntegrity"] = data_dict["dataIntegrity"]
if data_dict["defaultPath"] or not data_dict["dataIntegrity"]:
return default_return
#pprint.pprint(data_dict)
## 2) Extract candidate keyterms
print "[INFO] ==== Extracting candidate keyterms ===="
self.candidate_extractor.execute(data_dict)
# print keyterm_extractor.result_dict
## 3) Compute candidate keyterm features
print "[INFO] ==== Computing candidate keyterm features ===="
candidate_keyterm_df = self.feature_extractor.compute_features(link, data_dict, self.candidate_extractor.candidates)
## 4) Filter for relevancy and output top 10 keyterms
print "[INFO] ==== Selecting relevant keyterms ===="
selected_keyterms = self.relevance_filter.select_relevant(candidate_keyterm_df, self.candidate_extractor.candidates)
# print "[INFO] ==== FINAL SELECTION ====="
default_return["keyTerms"] = selected_keyterms
return default_return
except:
return default_return
def extractTermsFromText(self, text):
default_return = {
"available_domains": ["http://www.generation-nt.com/", "http://www.maison.com/",
"http://www.journaldugeek.com/", "http://www.journaldugamer.com/",
"http://www.jdubuzz.com/", "http://news.pixelistes.com/",
"http://www.societe.com/", "http://www.pausecafein.fr/",
"http://worldofwarcraft.judgehype.com/news/",
"http://hearthstone.judgehype.com/news/", "http://diablo3.judgehype.com/news/",
"http://www.judgehype.com/news/",
"http://www.jeuxonline.info", "http://heroes.judgehype.com/news/",
"http://overwatch.judgehype.com/news/",
"http://film-warcraft.judgehype.com/news/", "http://judgehype.com/",
"http://portail.free.fr/", "http://www.planet.fr/",
"http://aliceadsl.closermag.fr/", "http://aliceadsl.lemonde.fr/",
"http://aliceadsl.gqmagazine.fr/"],
"defaultPath": False, "dataIntegrity": False, "keyTerms": []}
try:
candidate_keyterms = self.candidate_extractor.execute_with_snippet(text)
keyterms = self.filter_candidates_from_snippet(candidate_keyterms)
default_return["keyTerms"] = keyterms
return default_return
except:
return default_return
def recommendKeytermsForBase(self, link):
default_return = {
"type": "Recommendations based on clustering.",
"text_used_from_link": ["title", "description", "keywords", "urlTokens"],
"keyTerms_recommandations": []}
try:
## 1) Extract webpage data
print "[INFO] ==== Extracting webpage data USING Specific PathDef===="
data_dict = self.data_scraper.crawlPage(link, elementsPathDef="baseCluster")
#Check integrity of list
if len(data_dict) <= 0:
return default_return
# #TEST
# default_return["crawled_data"] = data_dict
#Simple extraction of possible terms (not using trained model)
#Concatanate into text all components with sentence separation
text_for_analysis = u''
for key, value in data_dict.iteritems():
if isinstance(value, basestring):
text_for_analysis = text_for_analysis + ". " + value + ". "
elif isinstance(value, list):
text_for_analysis = text_for_analysis + ". ".join(value)
# #TEST
# default_return["text_for_analysis"] = text_for_analysis
#pprint.pprint(data_dict)
## 2) Extract candidate keyterms
print "[INFO] ==== Extracting candidate keyterms ===="
candidates = self.candidate_extractor.execute_with_snippet(text_for_analysis)
#SHOW CANDIDATES
# default_return["keyTerms_candidates"] = candidates
## 3) Compute keyterm recommendations comparing cluster centroids
print "[INFO] ==== Computing keyterm recommendations ===="
orig_list, keyterm_recommendations = self.keytermClassifier.match_adv_keyterm_clusters_base(candidates,
min_similarity_threshold=0.5)
# print "[INFO] ==== FINAL SELECTION ====="
default_return["keyTerms_recommandations"] = keyterm_recommendations
return default_return
except:
return default_return
def recommendKeytermsSimple(self, link):
default_return = {
"type": "Simple recommendation based only on individual extracted keyterms.",
"defaultPath": False, "dataIntegrity": False,
"keyTerms_recommandations": []}
try:
## 1) Extract webpage data
print "[INFO] ==== Extracting Terms From Link ===="
keyterms = self.extracTermsFromLink(link)["keyTerms"]
if len(keyterms) <= 0:
return default_return
##2) Compute keyterm recommendations comparing cluster centroids
print "[INFO] ==== Computing keyterm recommendations ===="
orig_list, keyterm_recommendations = self.keytermClassifier.match_adv_keyterm_website(keyterms,
min_similarity_threshold=0.5, min_diff_distance=0.90, top=5)
print "[INFO] ==== FINAL SELECTION ====="
default_return["keyTerms_recommandations"] = list(keyterm_recommendations)
return default_return
except:
return default_return
def filter_candidates_from_snippet(self, candidate_keyterms):
from functools import cmp_to_key
ordered_keyterms = sorted(candidate_keyterms.itervalues(), key = lambda item: item['cvalue'], reverse = True)
selected_keyterms = [item for item in ordered_keyterms if item['cvalue'] > 0]
def pos_cmp(keyterm1, keyterm2):
if not "NAM" in keyterm1['pos'] and "NAM" in keyterm2['pos']:
return -1
elif "NAM" in keyterm1['pos'] and "NAM" not in keyterm2['pos']:
return 1
else:
return 0
filtered_keyterms = sorted(selected_keyterms, key=cmp_to_key(pos_cmp), reverse=True)
keyterms = [{'term' : " ".join(t['words']), 'cvalue': t['cvalue'], 'lemma': t['lemma_string'], 'pos_tag': t['pos']} for t in filtered_keyterms]
return keyterms