from nest_backend.app import app from nest_backend.database import * import tweepy as tw import nltk from nltk.corpus import stopwords ext.init_app(app=app) def authenticate(): c_k = "GEhtSyP9e98mzFeiOCSW0lvQX" c_s = "438cmYrl5xqaX2W7I2Bf5A9nF1pN5VtM9f77WYQnAXg1BwKJ27" a_t = "1380217745732689921-8gCfr8Zx9YHKvo4OVP3HAr3kfMRkgz" a_t_s = "jGOlgTs1i1itGMxDxAqFEDnv7QAui772n9hGxeSIKcwzS" auth = tw.OAuthHandler(c_k, c_s) auth.set_access_token(a_t, a_t_s) api = tw.API(auth, wait_on_rate_limit=True) return api def start_exploring(): api = authenticate() most_popular_hashtags = dict() most_popular_words = dict() geocode="34.0999996,-118.333332,1000km" for tweet in tw.Cursor(method = api.search, q="Draghi OR Oscar", lang="it").items(50): print(tweet.user.name + ' : ' + tweet.text) for hashtag in tweet.entities['hashtags']: if hashtag['text'] in most_popular_hashtags.keys(): most_popular_hashtags[hashtag['text']] += 1 else: most_popular_hashtags[hashtag['text']] = 1 stop_words = set(stopwords.words('italian')) stop_words.add("RT") word_tokens = nltk.word_tokenize(tweet.text) filtered_sentence = [w for w in word_tokens if not w.lower() in stop_words and w.isalpha()] for word in filtered_sentence: if word in most_popular_words.keys(): most_popular_words[word] += 1 else: most_popular_words[word] = 1 print(dict(sorted(most_popular_hashtags.items(), key=lambda item: item[1]))) print(dict(sorted(most_popular_words.items(), key=lambda item: item[1]))) def search_repo_conditions(repository_id): repo = Repository.query.filter_by(id=repository_id).first() conditions = [use.condition for use in repo.uses] conditions_type = dict() for condition in conditions: # print(condition.id) if condition.type not in conditions_type.keys(): conditions_type[condition.type]=[condition.content] else: conditions_type[condition.type].append(condition.content) for types in conditions_type.keys(): print(types, ":", conditions_type[types]) if __name__ == "__main__": search_repo_conditions() #print(stopwords.words('italian')) with app.app_context(): ext.create_all(app=app) #start_exploring()