Recon / sentiment.py
Dragneel's picture
Update sentiment.py
317535b
raw
history blame
5.37 kB
import pickle
import os
import praw
import torch
from transformers import RobertaTokenizer, RobertaForSequenceClassification
import nltk
from nltk.stem.porter import PorterStemmer
from nltk.corpus import stopwords
import spacy
import string
import matplotlib.pyplot as plt
from wordcloud import WordCloud
def save_data(data, filename):
with open(filename, 'wb') as file:
pickle.dump(data, file)
def load_data(filename):
if os.path.exists(filename):
with open(filename, 'rb') as file:
return pickle.load(file)
else:
return None
# PRAW configs
REDDIT_CLIENT_ID = os.environ['client_id']
REDDIT_CLIENT_SECRET = os.environ['secret_key']
REDDIT_USERNAME = os.environ['username']
reddit = praw.Reddit(
client_id=REDDIT_CLIENT_ID,
client_secret=REDDIT_CLIENT_SECRET,
user_agent=f"script:sentiment-analysis:v0.0.1 (by {REDDIT_USERNAME})"
)
# NLP configs
stemmer = PorterStemmer()
nlp = spacy.load("en_core_web_sm")
nltk.download('punkt')
# Model configs
tokenizer = RobertaTokenizer.from_pretrained('aychang/roberta-base-imdb')
model = RobertaForSequenceClassification.from_pretrained(
'aychang/roberta-base-imdb', num_labels=2)
model.classifier = torch.nn.Linear(768, 2)
def get_sentiment(query):
print('inside get sentiment')
filename = f"data/sentiment_analysis/{query}_results.pkl"
saved_data = load_data(filename)
if saved_data:
print('inside saved_data')
positive, negative, _ = saved_data
wordcloud = f'static/images/wordcloud/{query}_cloud.png'
return positive, negative, wordcloud
else:
print(' inside else not saved data')
results = get_reddit_results(query)
if not results:
print('no results')
error = "No results found for query"
return error
positive, negative, wordcloud = analyze_comments(
results, query=query)
print(f'positive:{positive}')
save_data((positive, negative, wordcloud), filename)
return positive, negative, f'static/images/wordcloud/{query}_cloud.png'
def get_reddit_results(query):
print('inside get reddit result')
try:
sub = reddit.subreddit('noveltranslations+progressionfantasy')
results = sub.search(query, limit=1)
results_list = list(results)
if results_list:
print(f'result from reddit: {results_list[0]}')
return results_list
else:
print("No results found for query.")
return []
except Exception as e:
print(f"Error occurred: {e}")
return []
def transform_text(text):
print('inside transformtext0')
text = text.lower()
text = nltk.word_tokenize(text)
print('inside transformtext1')
text = [i for i in text if i.isalnum()]
print('inside transformtext2')
text = [i for i in text if i not in stopwords.words(
'english') and i not in string.punctuation]
print('inside transformtext3')
text = [stemmer.stem(i) for i in text]
print('inside transformtext4')
return ' '.join(text)
def tokenize(text):
print('inside tokenize')
doc = nlp(text)
return [token.text for token in doc]
def analyze_comments(results, query):
total_positive = 0
total_negative = 0
total_comments = 0
comments_for_cloud = []
for submission in results:
print('inside submission')
submission.comments.replace_more(limit=None)
all_comments = submission.comments.list()
for comment in all_comments:
print('inside comment')
comment_body = comment.body
print(comment_body)
text = transform_text(comment_body)
print(text)
comments_for_cloud.append(comment_body)
if text:
print('inside text')
tokens = tokenize(text)
tokenized_input = tokenizer(
tokens, return_tensors='pt', truncation=True, padding=True)
outputs = model(**tokenized_input)
probabilities = torch.softmax(outputs.logits, dim=-1)
mean_probabilities = probabilities.mean(dim=1)
positive_pct = mean_probabilities[0][1].item() * 100
negative_pct = mean_probabilities[0][0].item() * 100
total_positive += positive_pct
total_negative += negative_pct
total_comments += 1
if total_comments > 0:
avg_positive = total_positive / total_comments
avg_negative = total_negative / total_comments
else:
avg_positive = 0
avg_negative = 0
if total_comments > 0:
all_comments_string = ' '.join(comments_for_cloud)
wordcloud = WordCloud(width=400, height=400,
background_color='white',
max_words=30,
stopwords=stopwords.words('english'),
min_font_size=10).generate(all_comments_string)
# Save the WordCloud image as a static file
wordcloud.to_file(
f'static/images/wordcloud/{query}_cloud.png')
else:
wordcloud = None
print(f'positive:{avg_positive}')
return round(avg_positive), round(avg_negative), wordcloud