NewsSearch / tools /sentiment_analysis_util.py
KatGaw's picture
adding files
1a4d79b
import os
from dotenv import load_dotenv
from transformers import pipeline
import pandas as pd
from GoogleNews import GoogleNews
from langchain_openai import ChatOpenAI
import praw
from datetime import datetime
import numpy as np
from tavily import TavilyClient
load_dotenv()
TAVILY_API_KEY = os.environ["TAVILY_API_KEY"]
def fetch_news(topic):
""" Fetches news articles within a specified date range.
Args:
- topic (str): Topic of interest
Returns:
- list: A list of dictionaries containing news. """
load_dotenv()
days_to_fetch_news = os.environ["DAYS_TO_FETCH_NEWS"]
googlenews = GoogleNews()
googlenews.set_period(days_to_fetch_news)
googlenews.get_news(topic)
news_json=googlenews.get_texts()
urls=googlenews.get_links()
no_of_news_articles_to_fetch = os.environ["NO_OF_NEWS_ARTICLES_TO_FETCH"]
news_article_list = []
counter = 0
for article in news_json:
if(counter >= int(no_of_news_articles_to_fetch)):
break
relevant_info = {
'News_Article': article,
'URL': urls[counter]
}
news_article_list.append(relevant_info)
counter+=1
return news_article_list
# def fetch_tavily_news(topic):
# """ Fetches news articles.
# Args:
# - topic (str): Topic of interest
# Returns:
# - list: A list of dictionaries containing news. """
# # Step 1. Instantiating your TavilyClient
# tavily_client = TavilyClient(api_key=TAVILY_API_KEY)
# #response = tavily_client.search(topic)
# # Step 2.1. Executing a context search query
# answer = tavily_client.get_search_context(query=f"Give me news on {topic}")
# line=[]
# tavily_news=[]
# for i in range(len(answer.split("url")))[1:]:
# https_link=(answer.split("url")[i].split("\\\\\\")[2]).split('"')[1]
# topic_answer=answer.split("url")[i].split("\\\\\\")[-3]
# tavily_news=np.append(tavily_news,{'https':https_link,'topic_answer':topic_answer})
# return tavily_news
def fetch_tavily_news(prompt):
try:
# Assuming answer contains the Tavily API response
# First, let's make the URL extraction more robust
urls = []
# Method 1: Using string manipulation with error handling
try:
parts = answer.split("url")
for part in parts[1:]: # Skip the first part before 'url'
try:
# Try different splitting patterns
if '\\\\' in part:
url = part.split('\\\\')[2].split('"')[1]
elif '"' in part:
url = part.split('"')[1]
else:
continue
if url.startswith('http'): # Validate URL
urls.append(url)
except (IndexError, AttributeError):
continue
except Exception as e:
print(f"Error extracting URLs: {e}")
# If no URLs found, try alternative parsing
if not urls:
# Method 2: Try JSON parsing if the response is JSON formatted
try:
import json
data = json.loads(answer)
if isinstance(data, list):
for item in data:
if isinstance(item, dict) and 'url' in item:
urls.append(item['url'])
except json.JSONDecodeError:
pass
# If still no URLs found, try regex
if not urls:
import re
url_pattern = r'https?://[^\s<>"]+|www\.[^\s<>"]+|http?://[^\s<>"]+'
urls = re.findall(url_pattern, answer)
# Remove duplicates while preserving order
urls = list(dict.fromkeys(urls))
return urls
except Exception as e:
print(f"Error in fetch_tavily_news: {e}")
return [] # Return empty list on error
def fetch_reddit_news(topic):
load_dotenv()
REDDIT_USER_AGENT= os.environ["REDDIT_USER_AGENT"]
REDDIT_CLIENT_ID= os.environ["REDDIT_CLIENT_ID"]
REDDIT_CLIENT_SECRET= os.environ["REDDIT_CLIENT_SECRET"]
#https://medium.com/geekculture/a-complete-guide-to-web-scraping-reddit-with-python-16e292317a52
user_agent = REDDIT_USER_AGENT
reddit = praw.Reddit (
client_id= REDDIT_CLIENT_ID,
client_secret= REDDIT_CLIENT_SECRET,
user_agent=user_agent
)
headlines = set ( )
for submission in reddit.subreddit('nova').search(topic,time_filter='day'):
headlines.add(submission.title + ', Date: ' +datetime.utcfromtimestamp(int(submission.created_utc)).strftime('%Y-%m-%d %H:%M:%S') + ', URL:' +submission.url)
for submission in reddit.subreddit('fednews').search(topic,time_filter='day'):
headlines.add(submission.title + ', Date: ' +datetime.utcfromtimestamp(int(submission.created_utc)).strftime('%Y-%m-%d %H:%M:%S') + ', URL:' +submission.url)
for submission in reddit.subreddit('washingtondc').search(topic,time_filter='day'):
headlines.add(submission.title + ', Date: ' +datetime.utcfromtimestamp(int(submission.created_utc)).strftime('%Y-%m-%d %H:%M:%S') + ', URL:' +submission.url)
if len(headlines)<10:
for submission in reddit.subreddit('washingtondc').search(topic,time_filter='year'):
headlines.add(submission.title + ', Date: ' +datetime.utcfromtimestamp(int(submission.created_utc)).strftime('%Y-%m-%d %H:%M:%S') + ', URL:' +submission.url)
if len(headlines)<10:
for submission in reddit.subreddit('washingtondc').search(topic): #,time_filter='week'):
headlines.add(submission.title + ', Date: ' +datetime.utcfromtimestamp(int(submission.created_utc)).strftime('%Y-%m-%d %H:%M:%S') + ', URL:' +submission.url)
return headlines
def analyze_sentiment(article):
"""
Analyzes the sentiment of a given news article.
Args:
- news_article (dict): Dictionary containing 'summary', 'headline', and 'created_at' keys.
Returns:
- dict: A dictionary containing sentiment analysis results.
"""
#Analyze sentiment using default model
#classifier = pipeline('sentiment-analysis')
#Analyze sentiment using specific model
classifier = pipeline(model='tabularisai/robust-sentiment-analysis') #mrm8488/distilroberta-finetuned-financial-news-sentiment-analysis')
sentiment_result = classifier(str(article))
analysis_result = {
'News_Article': article,
'Sentiment': sentiment_result
}
return analysis_result
def generate_summary_of_sentiment(sentiment_analysis_results): #, dominant_sentiment):
news_article_sentiment = str(sentiment_analysis_results)
print("News article sentiment : " + news_article_sentiment)
OPENAI_API_KEY = os.environ["OPENAI_API_KEY"]
model = ChatOpenAI(
model="gpt-4o",
temperature=0,
max_tokens=None,
timeout=None,
max_retries=2,
api_key=OPENAI_API_KEY, # if you prefer to pass api key in directly instaed of using env vars
# base_url="...",
# organization="...",
# other params...
)
messages=[
{"role": "system", "content": "You are a helpful assistant that looks at all news articles with their sentiment, hyperlink and date in front of the article text, the articles MUST be ordered by date!, and generate a summary rationalizing dominant sentiment. At the end of the summary, add URL links for all the articles in the markdown format for streamlit. Make sure the articles as well as the links are ordered descending by Date!!!!!!! Example of adding the URLs: The Check out the links: [link](%s) % url. "},
{"role": "user", "content": f"News articles and their sentiments: {news_article_sentiment}"} #, and dominant sentiment is: {dominant_sentiment}"}
]
response = model.invoke(messages)
summary = response.content
print ("+++++++++++++++++++++++++++++++++++++++++++++++")
print(summary)
print ("+++++++++++++++++++++++++++++++++++++++++++++++")
return summary
def plot_sentiment_graph(sentiment_analysis_results):
"""
Plots a sentiment analysis graph
Args:
- sentiment_analysis_result): (dict): Dictionary containing 'Review Title : Summary', 'Rating', and 'Sentiment' keys.
Returns:
- dict: A dictionary containing sentiment analysis results.
"""
df = pd.DataFrame(sentiment_analysis_results)
#print(df)
#Group by Rating, sentiment value count
grouped = df['Sentiment'].value_counts()
sentiment_counts = df['Sentiment'].value_counts()
# Plotting pie chart
# fig = plt.figure(figsize=(5, 3))
# plt.pie(sentiment_counts, labels=sentiment_counts.index, autopct='%1.1f%%', startangle=140)
# plt.axis('equal') # Equal aspect ratio ensures that pie is drawn as a circle.
#Open below when u running this program locally and c
#plt.show()
return sentiment_counts
def get_dominant_sentiment (sentiment_analysis_results):
"""
Returns overall sentiment, negative or positive or neutral depending on the count of negative sentiment vs positive sentiment
Args:
- sentiment_analysis_result): (dict): Dictionary containing 'summary', 'headline', and 'created_at' keys.
Returns:
- dict: A dictionary containing sentiment analysis results.
"""
df = pd.DataFrame(sentiment_analysis_results)
# Group by the 'sentiment' column and count the occurrences of each sentiment value
#print(df)
#print(df['Sentiment'])
sentiment_counts = df['Sentiment'].value_counts().reset_index()
sentiment_counts.columns = ['sentiment', 'count']
print(sentiment_counts)
# Find the sentiment with the highest count
dominant_sentiment = sentiment_counts.loc[sentiment_counts['count'].idxmax()]
return dominant_sentiment['sentiment']
#starting point of the program
if __name__ == '__main__':
#fetch news
news_articles = fetch_news('AAPL')
analysis_results = []
#Perform sentiment analysis for each product review
for article in news_articles:
sentiment_analysis_result = analyze_sentiment(article['News_Article'])
# Display sentiment analysis results
print(f'News Article: {sentiment_analysis_result["News_Article"]} : Sentiment: {sentiment_analysis_result["Sentiment"]}', '\n')
result = {
'News_Article': sentiment_analysis_result["News_Article"],
'Sentiment': sentiment_analysis_result["Sentiment"][0]['label']
}
analysis_results.append(result)
#Graph dominant sentiment based on sentiment analysis data of reviews
dominant_sentiment = get_dominant_sentiment(analysis_results)
print(dominant_sentiment)
#Plot graph
plot_sentiment_graph(analysis_results)