# import streamlit as st
# import pandas as pd
# import numpy as np
# import os
# import ast
# import openai
# from openai import OpenAI
# import json
# from getpass import getpass
# from scipy.spatial.distance import cosine
# from tqdm import tqdm
# import matplotlib.pyplot as plt
# import financial_analysis as fa
# from financial_analysis import alphalens_analysis, alphalens_analysis_by_sector, calculate_information_ratio, process_sentiment_data
# def get_sentiment_gpt(company, SASB, news, max_retries=5, model = 'gpt-4-turbo-2024-04-09'):
# system_prompt = """
# As a specialist in ESG analytics,
# You possess a deep understanding of evaluating environmental, social, and governance factors in the context of corporate news.
# Your expertise lies in discerning the underlying sentiment of news segments that pertain to a company's ESG practices,
# determining whether the coverage reflects a positive, negative, or neutral stance.
# """
# allowed_sentiments = ['Negative', 'Positive', 'Neutral']
# attempt = 0
# while attempt < max_retries:
# main_prompt = f"""
# Classify the sentiment (Only options: Positive, Negative, Neutral) of the following news: {news} |
# The sentiment classification should be about the sections of the news talking about the company {company}. |
# The ESG part of the news should be around topics within the following SASB topics {SASB}
# The output should be a structured JSON object with the key: "sentiment".
# Here is the format I expect for the JSON object:
# {{
# "sentiment": "Enter 'Positive', 'Neutral', or 'Negative'",
# }}
# Do not return any additional text or information outside of this JSON structure.
# """
# messages = [
# {"role": "system", "content": system_prompt},
# {"role": "user", "content": main_prompt}
# ]
# response = openai.chat.completions.create(
# model=model,
# messages=messages,
# response_format={"type": "json_object"} # Enable JSON mode
# )
# response_json = json.loads(response.choices[0].message.content)
# json_sentiment = response_json.get('sentiment')
# if json_sentiment in allowed_sentiments:
# return json_sentiment
# attempt += 1
# # After max retries, if no valid sentiment is found, handle as needed (e.g., return a default sentiment)
# print("Failed to obtain a valid sentiment after maximum retries. Defaulting to 'Neutral'.")
# return 'Neutral' # Default return value if no valid sentiment is obtained
# def update_dataset_with_gpt_sentiment(df, model, column_name='GPT_based_sentiment'):
# # Initialize the new column to store GPT-based sentiment
# df['GPT_based_sentiment'] = None
# # Use tqdm to show a progress bar for the operation
# for index, row in tqdm(df.iterrows(), total=len(df), desc="Processing rows"):
# # Extract necessary information for each row
# company = row['Company'] # Make sure this matches your DataFrame's column name
# SASB = row['SASB'] # Make sure this matches your DataFrame's column name
# news = row['title & content'] # Make sure this matches your DataFrame's column name
# # Call the function to get the sentiment
# sentiment = get_sentiment_gpt(company, SASB, news, model=model)
# # Update the DataFrame with the obtained sentiment
# df.at[index, column_name] = sentiment # Now correctly assigns the sentiment
# return df
# # Function to get embeddings, provided by you
# def get_embedding(text, model="text-embedding-3-small"):
# client = OpenAI()
# text = text.replace("\n", " ")
# return client.embeddings.create(input=[text], model=model).data[0].embedding
# # Function to calculate cosine similarity
# def cosine_similarity(v1, v2):
# return 1 - cosine(v1, v2)
# def calculate_sasb_embeddings(sasb_str):
# # Safely convert the string representation of a dictionary into an actual dictionary
# try:
# sasb_dict = ast.literal_eval(sasb_str)
# if not isinstance(sasb_dict, dict):
# raise ValueError("SASB column does not contain a valid dictionary.")
# except ValueError as e:
# print(f"Error converting SASB column to dictionary: {e}")
# return {}
# sasb_embeddings = {}
# for topic, content in sasb_dict.items():
# # Join the list of keywords into a single string
# combined_content = ' '.join(content)
# sasb_embeddings[topic] = get_embedding(combined_content)
# return sasb_embeddings
# # Function to process ESG classification
# def classify_esg(data):
# # Calculate embeddings for the news
# data['news_embeddings'] = data['title & content'].apply(get_embedding)
# # Calculate embeddings for SASB topics (you need to have your SASB topics defined)
# data['sasb_embeddings'] = data['SASB'].apply(calculate_sasb_embeddings)
# # Compute cosine similarities
# data['cosine_similarities'] = data.apply(
# lambda row: {topic: cosine_similarity(row['news_embeddings'], emb)
# for topic, emb in row['sasb_embeddings'].items()},
# axis=1
# )
# # Extract max cosine similarity
# data['max_cosine_similarity'] = data['cosine_similarities'].apply(lambda x: max(x.values()))
# # Mark the top 10% of news by max_cosine_similarity within each 'Sector' as 'Yes'
# sector_thresholds = data.groupby('Sector')['max_cosine_similarity'].quantile(0.9).to_dict()
# data['ESG_relevance'] = data.apply(
# lambda row: 'Yes' if row['max_cosine_similarity'] >= sector_thresholds[row['Sector']] else 'No',
# axis=1
# )
# return data
# def main():
# st.set_page_config(page_title="NLP ESG Project", page_icon="📈")
# # Custom styles
# st.markdown(
# """
#
# """,
# unsafe_allow_html=True,
# )
# # Header section
# st.write("# NLP Project: ESG News Analysis and Financial Impact")
# st.sidebar.write("## Configuration")
# # API Key input
# openai_api_key = st.sidebar.text_input("Enter your OpenAI API key", type="password")
# openai_api_key = os.getenv('OPENAI_API_KEY')
# os.environ["OPENAI_API_KEY"] = openai_api_key
# openai.api_key = openai_api_key
# # File Upload
# st.sidebar.write("## Upload Data")
# uploaded_file = st.sidebar.file_uploader("", type="csv")
# # Investment Strategy Slider
# st.sidebar.markdown("### Investment Strategy")
# investment_strategy = st.sidebar.slider(
# "Investment Strategy",
# min_value=0.0, max_value=1.0, value=0.5, step=0.01,
# format="",
# help="0 is Conservative, 1 is Aggressive",
# label_visibility="collapsed"
# )
# st.sidebar.text(f"Current Strategy: {'Conservative' if investment_strategy <= 0.5 else 'Aggressive'}")
# # Main container
# st.sidebar.write("## Upload Data")
# uploaded_file = st.sidebar.file_uploader("Please upload a CSV file", type="csv", label_visibility="collapsed")
# if uploaded_file:
# # Displaying the file
# data = pd.read_csv(uploaded_file)
# st.session_state.classified_data = classify_esg(data)
# st.write("### Uploaded News Data:")
# st.dataframe(data, use_container_width=True)
# if st.button("🔍 Classify ESG"):
# st.write("Classifying ESG-related news...")
# try:
# with st.spinner("Calculating embeddings and similarities..."):
# st.session_state.classified_data = classify_esg(st.session_state.classified_data)
# st.write("Classified News Data:")
# st.dataframe(st.session_state.classified_data, use_container_width=True)
# except Exception as e:
# st.error(f"An error occurred: {e}")
# if st.button("😊 Determine Sentiment"):
# st.write("Determining sentiment using GPT...")
# # Run sentiment analysis with GPT
# try:
# with st.spinner("Analyzing sentiment..."):
# # Assume you have your API key set and a function defined to handle sentiment analysis
# st.session_state.updated_data = update_dataset_with_gpt_sentiment(st.session_state.classified_data, model='gpt-4-turbo-2024-04-09')
# st.write("News with GPT-based Sentiment Analysis:")
# st.dataframe(st.session_state.updated_data, use_container_width=True)
# except Exception as e:
# st.error(f"An error occurred: {e}")
# if st.button("📊 Alphalens Analysis"):
# # process_sentiment_data(sentiment_data = 'finbert_sentiment.csv', sector_ticker = 'sector_ticker.csv', prices = 'prices.csv')
# prices = pd.read_csv('prices.csv')
# factor_data = pd.read_csv('factor_data.csv')
# merged_data = pd.read_csv('merged_data.csv')
# alphalens_analysis(merged_data, prices)
# # Expander for advanced settings
# with st.expander("Advanced Settings"):
# st.write("Any advanced settings and configurations will go here.")
# if __name__ == "__main__":
# main()
import streamlit as st
import pandas as pd
import numpy as np
import os
import ast
import openai
from openai import OpenAI
import json
from getpass import getpass
from scipy.spatial.distance import cosine
from tqdm import tqdm
import matplotlib.pyplot as plt
def get_sentiment_gpt(company, SASB, news, max_retries=5, model = 'gpt-4-turbo-2024-04-09'):
system_prompt = """
As a specialist in ESG analytics,
You possess a deep understanding of evaluating environmental, social, and governance factors in the context of corporate news.
Your expertise lies in discerning the underlying sentiment of news segments that pertain to a company's ESG practices,
determining whether the coverage reflects a positive, negative, or neutral stance.
"""
allowed_sentiments = ['Negative', 'Positive', 'Neutral']
attempt = 0
while attempt < max_retries:
main_prompt = f"""
Classify the sentiment (Only options: Positive, Negative, Neutral) of the following news: {news} |
The sentiment classification should be about the sections of the news talking about the company {company}. |
The ESG part of the news should be around topics within the following SASB topics {SASB}
The output should be a structured JSON object with the key: "sentiment".
Here is the format I expect for the JSON object:
{{
"sentiment": "Enter 'Positive', 'Neutral', or 'Negative'",
}}
Do not return any additional text or information outside of this JSON structure.
"""
messages = [
{"role": "system", "content": system_prompt},
{"role": "user", "content": main_prompt}
]
response = openai.chat.completions.create(
model=model,
messages=messages,
response_format={"type": "json_object"} # Enable JSON mode
)
response_json = json.loads(response.choices[0].message.content)
json_sentiment = response_json.get('sentiment')
if json_sentiment in allowed_sentiments:
return json_sentiment
attempt += 1
# After max retries, if no valid sentiment is found, handle as needed (e.g., return a default sentiment)
print("Failed to obtain a valid sentiment after maximum retries. Defaulting to 'Neutral'.")
return 'Neutral' # Default return value if no valid sentiment is obtained
def update_dataset_with_gpt_sentiment(df, model, column_name='GPT_based_sentiment'):
# Initialize the new column to store GPT-based sentiment
df['GPT_based_sentiment'] = None
# Use tqdm to show a progress bar for the operation
for index, row in tqdm(df.iterrows(), total=len(df), desc="Processing rows"):
# Extract necessary information for each row
company = row['Company'] # Make sure this matches your DataFrame's column name
SASB = row['SASB'] # Make sure this matches your DataFrame's column name
news = row['title & content'] # Make sure this matches your DataFrame's column name
# Call the function to get the sentiment
sentiment = get_sentiment_gpt(company, SASB, news, model=model)
# Update the DataFrame with the obtained sentiment
df.at[index, column_name] = sentiment # Now correctly assigns the sentiment
return df
# Function to get embeddings, provided by you
def get_embedding(text, model="text-embedding-3-small"):
client = OpenAI()
text = text.replace("\n", " ")
return client.embeddings.create(input=[text], model=model).data[0].embedding
# Function to calculate cosine similarity
def cosine_similarity(v1, v2):
return 1 - cosine(v1, v2)
def calculate_sasb_embeddings(sasb_str):
# Safely convert the string representation of a dictionary into an actual dictionary
try:
sasb_dict = ast.literal_eval(sasb_str)
if not isinstance(sasb_dict, dict):
raise ValueError("SASB column does not contain a valid dictionary.")
except ValueError as e:
print(f"Error converting SASB column to dictionary: {e}")
return {}
sasb_embeddings = {}
for topic, content in sasb_dict.items():
# Join the list of keywords into a single string
combined_content = ' '.join(content)
sasb_embeddings[topic] = get_embedding(combined_content)
return sasb_embeddings
# Function to process ESG classification
def classify_esg(data):
# Calculate embeddings for the news
data['news_embeddings'] = data['title & content'].apply(get_embedding)
# Calculate embeddings for SASB topics (you need to have your SASB topics defined)
data['sasb_embeddings'] = data['SASB'].apply(calculate_sasb_embeddings)
# Compute cosine similarities
data['cosine_similarities'] = data.apply(
lambda row: {topic: cosine_similarity(row['news_embeddings'], emb)
for topic, emb in row['sasb_embeddings'].items()},
axis=1
)
# Extract max cosine similarity
data['max_cosine_similarity'] = data['cosine_similarities'].apply(lambda x: max(x.values()))
# Mark the top 10% of news by max_cosine_similarity within each 'Sector' as 'Yes'
sector_thresholds = data.groupby('Sector')['max_cosine_similarity'].quantile(0.9).to_dict()
data['ESG_relevance'] = data.apply(
lambda row: 'Yes' if row['max_cosine_similarity'] >= sector_thresholds[row['Sector']] else 'No',
axis=1
)
return data
def main():
st.set_page_config(page_title="NLP ESG Project", page_icon="📈")
# Custom styles
st.markdown(
"""
""",
unsafe_allow_html=True,
)
# Header section
st.write("# NLP Project: ESG News Analysis and Financial Impact")
st.sidebar.write("## Configuration")
# API Key input
openai_api_key = st.sidebar.text_input("Enter your OpenAI API key", type="password")
os.environ["OPENAI_API_KEY"] = openai_api_key
openai.api_key = openai_api_key
# File Upload
st.sidebar.write("## Upload Data")
uploaded_file = st.sidebar.file_uploader("", type="csv")
# Investment Strategy Slider
st.sidebar.markdown("### Investment Strategy")
investment_strategy = st.sidebar.slider(
"Investment Strategy",
min_value=0.0, max_value=1.0, value=0.5, step=0.01,
format="",
help="0 is Conservative, 1 is Aggressive",
label_visibility="collapsed"
)
st.sidebar.text(f"Current Strategy: {'Conservative' if investment_strategy <= 0.5 else 'Aggressive'}")
# Main container
if uploaded_file:
# Displaying the file
data = pd.read_csv(uploaded_file)
st.session_state.classified_data = classify_esg(data)
st.write("### Uploaded News Data:")
st.dataframe(data, use_container_width=True)
if st.button("🔍 Classify ESG"):
st.write("Classifying ESG-related news...")
try:
with st.spinner("Calculating embeddings and similarities..."):
st.session_state.classified_data = classify_esg(st.session_state.classified_data)
st.write("Classified News Data:")
st.dataframe(st.session_state.classified_data, use_container_width=True)
except Exception as e:
st.error(f"An error occurred: {e}")
if st.button("😊 Determine Sentiment"):
st.write("Determining sentiment using GPT...")
# Run sentiment analysis with GPT
try:
with st.spinner("Analyzing sentiment..."):
# Assume you have your API key set and a function defined to handle sentiment analysis
st.session_state.updated_data = update_dataset_with_gpt_sentiment(st.session_state.classified_data, model='gpt-4-turbo-2024-04-09')
st.write("News with GPT-based Sentiment Analysis:")
st.dataframe(st.session_state.updated_data, use_container_width=True)
except Exception as e:
st.error(f"An error occurred: {e}")
if st.button("📊 Alphalens Analysis"):
st.write("Alphalens analysis will be here") # placeholder
# Expander for advanced settings
with st.expander("Advanced Settings"):
st.write("Any advanced settings and configurations will go here.")
if __name__ == "__main__":
main()