def compute_sentiment(tweets_df):
tweets_df["polarity"] = tweets_df["clean_tweets"].apply(
lambda x: sentiment(x)[0])
tweets_df["subjectivity"] = tweets_df["clean_tweets"].apply(
lambda x: sentiment(x)[1])
return tweets_df
def calculate_sentiment(row):
if pypath not in sys.path:
sys.path.insert(1, pypath)
from pattern.text.nl import sentiment
entry = row.asDict()
entry['description_score'], entry['text_score'] = sentiment(
entry['description']), sentiment(entry['text'])
return entry
def get_analysis(self, sSent):
"""Return the sentiment-analysis of one sentence"""
try:
score = sentiment(sSent)
return score
except:
# act upon error
self.errHandle.DoError("SentiAna/get_analysis")
return None
def getEntities(cleanSentence, tfidf, max):
relevantWords = {}
counter = 0
x = pl.sentiment(cleanSentence).assessments
listSent = []
for x1 in x:
w, a, b, c = x1
for p in w:
listSent.append(p)
for w in sorted(tfidf, key=tfidf.get, reverse=True):
if w not in listSent and 2 < len(w) < 25:
relevantWords[w] = ([], [], tfidf[w])
counter += 1
return relevantWords
def getSentiments(max_val_tf, relevantWords, text):
sentiment_count = 0
for sentence in text: #passing each sentence the number of times a sentiment has occurred. do this before.
sentiments_in_sentence = {}
sentiments_in_sentence_list = []
for sentiment in pl.sentiment(sentence).assessments:
sentiment_words, pol, obj, x = sentiment
# only add sentiments with a bit of flavor
if abs(pol) + abs(obj) > 0.1: # interestingness
sentiments_in_sentence[" ".join(
sentiment_words)] = sentiment # appending a tuple
# assumption: if the sentiment is longer than 1 item,
# the head of the first item will be the second item, etc.
# This assumption is not always correct (case of conjunction)
# TODO: fix conjunction case
sentiments_in_sentence_list.append(sentiment_words[0])
sentiment_count += 1
if sentiments_in_sentence: # if there are interesting sentiments in the sentence, we want to do a recursion to match them with an entity/subject
i, flag = 0, 0
for chunk in nlp(
sentence
): # we are going through the nlp (spacy) tagged sentence
if chunk.text == 'niet': # we found a negation
flag = 1
if chunk.text in sentiments_in_sentence_list and chunk.dep_ not in [
'case', 'mark', 'advcl'
] and chunk.pos_ not in ["VERB"]:
sentiment_chunk = get_sentiments_from_text_chunk(
sentiments_in_sentence, chunk.text)
recursion_path_list = recurse_to_find_entity(
chunk, 0, 0, 0, 0, [])
missed_negation = find_missed_negation(
i, sentiment_chunk[0])
context_fragment = missed_negation + sentiment_chunk[0]
if 'naar' not in sentiment_chunk[
0]: # hand-filtering out "naar", which keeps being classified as a sentiment
word = ""
for item in recursion_path_list:
if item.isupper():
word = item.lower()
if word not in context_fragment: # the word can't be in "context_fragment" already, that's how you get "ramp ramp chernobyl"
context_fragment.append(word)
if word != "": # if we found at least one potential noun to be the entity/subject
relevantWords = attach_direct_list(
relevantWords, recursion_path_list,
context_fragment, word, sentiment_chunk)
if flag == 1: # for the missed_negation
i += 1
#print("SENTIMENT COUNT: ", sentiment_count)
return relevantWords
def transform_text(text):
from pattern.text.nl import sentiment
polarity, subjectivity = sentiment(text)
return numpy.array([polarity, subjectivity])