Update app.py

app.py

@@ -2,10 +2,9 @@ import streamlit as st
 import nltk
 import spacy
 import matplotlib.pyplot as plt
-from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification
+from transformers import pipeline
 import random
 import subprocess
-import torch
 
 # Ensure spaCy model is installed
 try:
@@ -19,14 +18,12 @@ nltk.download("vader_lexicon")
 from nltk.sentiment import SentimentIntensityAnalyzer
 sia = SentimentIntensityAnalyzer()
 
-# Load multilingual emotion detection model
-model_name = "joeddav/xlm-roberta-large-xnli"
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-emotion_model = AutoModelForSequenceClassification.from_pretrained(model_name)
-
 # Load English emotion detection pipeline
 emotion_pipeline = pipeline("text-classification", model="j-hartmann/emotion-english-distilroberta-base", return_all_scores=True)
 
+# Load a better multilingual sentiment analysis model
+multilingual_pipeline = pipeline("sentiment-analysis", model="cardiffnlp/twitter-xlm-roberta-base-sentiment")
+
 # Sample texts
 sample_texts = [
     "The digital world is transforming the way we read and engage with text.",
@@ -96,21 +93,16 @@ if st.button("Analyze Engagement"):
         # Sentiment Analysis
         sentiment_score = sia.polarity_scores(text)
 
-        # English Emotion Detection
+        # Emotion Detection
         if selected_language == "English":
             emotion_results = emotion_pipeline(text)
             top_emotion = max(emotion_results[0], key=lambda x: x['score'])['label']
             emotion_scores = {e['label']: e['score'] for e in emotion_results[0]}
         else:
             # Multilingual Sentiment Analysis
-            inputs = tokenizer(text, return_tensors="pt", truncation=True)
-            with torch.no_grad():
-                outputs = emotion_model(**inputs)
-            probabilities = torch.nn.functional.softmax(outputs.logits, dim=-1)
-            top_emotion_idx = torch.argmax(probabilities).item()
-            emotion_labels = ["neutral", "entailment", "contradiction"]
-            top_emotion = emotion_labels[top_emotion_idx]
-            emotion_scores = {top_emotion: probabilities[0][top_emotion_idx].item()}
+            multilingual_results = multilingual_pipeline(text)
+            top_emotion = multilingual_results[0]["label"]
+            emotion_scores = {top_emotion: multilingual_results[0]["score"]}
 
         # Display Sentiment
         st.subheader("📊 Sentiment Analysis")