pages/Comparision.py

import streamlit as st
import requests
import nltk
from transformers import pipeline
from rake_nltk import Rake
from nltk.corpus import stopwords
from fuzzywuzzy import fuzz
import openai
import os
from dotenv import load_dotenv

# Load environment variables for Llama 3
load_dotenv()

# Title of the app
st.title("Sentiment Analysis Comparison: Transformers vs Llama 3")

# Define the options for the dropdown menu, selecting a remote txt file already created to analyze the text
options = ['None', 'Appreciation Letter', 'Regret Letter', 'Kindness Tale', 'Lost Melody Tale', 'Twitter Example 1', 'Twitter Example 2']

# Create a dropdown menu to select options
selected_option = st.selectbox("Select a preset option", options)

# Define URLs for different options
urls = {
    'Appreciation Letter': "https://raw.githubusercontent.com/peteciank/public_files/main/Transformers/Appreciation_Letter.txt",
    'Regret Letter': "https://raw.githubusercontent.com/peteciank/public_files/main/Transformers/Regret_Letter.txt",
    'Kindness Tale': "https://raw.githubusercontent.com/peteciank/public_files/main/Transformers/Kindness_Tale.txt",
    'Lost Melody Tale': "https://raw.githubusercontent.com/peteciank/public_files/main/Transformers/Lost_Melody_Tale.txt",
    'Twitter Example 1': "https://raw.githubusercontent.com/peteciank/public_files/main/Transformers/Twitter_Example_1.txt",
    'Twitter Example 2': "https://raw.githubusercontent.com/peteciank/public_files/main/Transformers/Twitter_Example_2.txt"
}

# Function to fetch text content based on selected option
def fetch_text_content(selected_option):
    return requests.get(urls[selected_option]).text if selected_option in urls else ""

# Fetch text content based on selected option
text = fetch_text_content(selected_option)

# Display text content in a text area
text = st.text_area('Enter the text to analyze', text)

# Download NLTK resources
nltk.download('punkt')
nltk.download('stopwords')

# Initialize sentiment, summarization, and keyword extraction pipelines for Transformers
pipe_sent = pipeline('sentiment-analysis')
pipe_summ = pipeline("summarization", model="facebook/bart-large-cnn")

# Llama 3 initialization
llama_api_key = os.getenv('HFSecret')
llama_base_url = "https://api-inference.huggingface.co/v1"
llama_repo_id = "meta-llama/Meta-Llama-3-8B-Instruct"

# Function to use Llama 3 for sentiment analysis, summarization, and keyword extraction
def analyze_with_llama(text):
    headers = {
        "Authorization": f"Bearer {llama_api_key}"
    }
    data = {
        "inputs": text,
        "parameters": {
            "max_new_tokens": 200
        }
    }
    # Perform the request
    response = requests.post(f"{llama_base_url}/models/{llama_repo_id}", headers=headers, json=data)
    return response.json()

# Function to extract keywords using RAKE and remove duplicates
def extract_keywords(text):
    r = Rake()
    r.extract_keywords_from_text(text)
    phrases_with_scores = r.get_ranked_phrases_with_scores()
    stop_words = set(stopwords.words('english'))
    keywords = [(score, phrase) for score, phrase in phrases_with_scores if phrase.lower() not in stop_words]
    keywords.sort(key=lambda x: x[0], reverse=True)
    unique_keywords = []
    seen_phrases = set()
    for score, phrase in keywords:
        if phrase not in seen_phrases:
            similar_phrases = [seen_phrase for seen_phrase in seen_phrases if fuzz.ratio(phrase, seen_phrase) > 70]
            if similar_phrases:
                merged_phrase = max([phrase] + similar_phrases, key=len)
                unique_keywords.append((score, merged_phrase))
            else:
                unique_keywords.append((score, phrase))
            seen_phrases.add(phrase)
    return unique_keywords[:10]

# Create two columns
col1, col2 = st.columns(2)

# Transformer-based analysis in the first column
with col1:
    st.header("Transformer-based Analysis")
    if st.button("Analyze with Transformers"):
        with st.spinner("Analyzing with Transformers..."):
            # Sentiment analysis
            out_sentiment = pipe_sent(text)
            sentiment_score = out_sentiment[0]['score']
            sentiment_label = out_sentiment[0]['label']
            sentiment_emoji = '😊' if sentiment_label == 'POSITIVE' else '😞'
            sentiment_text = f"Sentiment Score: {sentiment_score}, Sentiment Label: {sentiment_label.capitalize()} {sentiment_emoji}"
            
            with st.expander("Sentiment Analysis (Transformers)"):
                st.write(sentiment_text)
            
            # Summarization
            out_summ = pipe_summ(text)
            summarized_text = out_summ[0]['summary_text']
            
            with st.expander("Summarization (Transformers)"):
                st.write(summarized_text)
            
            # Keyword extraction
            keywords = extract_keywords(text)
            keyword_list = [keyword[1] for keyword in keywords]
            
            with st.expander("Keywords (Transformers)"):
                st.write(keyword_list)

# Llama 3-based analysis in the second column
with col2:
    st.header("Llama 3-based Analysis")
    if st.button("Analyze with Llama 3"):
        with st.spinner("Analyzing with Llama 3..."):
            llama_response = analyze_with_llama(text)
            
            if llama_response:
                # Assuming the response returns in the same format, adjust if needed
                sentiment_text = llama_response.get('sentiment_analysis', 'No sentiment detected')
                summarized_text = llama_response.get('summarization', 'No summary available')
                keywords = llama_response.get('keywords', 'No keywords available')
                
                with st.expander("Sentiment Analysis (Llama 3)"):
                    st.write(sentiment_text)
                
                with st.expander("Summarization (Llama 3)"):
                    st.write(summarized_text)
                
                with st.expander("Keywords (Llama 3)"):
                    st.write(keywords)