app.py

import streamlit as st
import pickle
import pandas as pd
import os
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.metrics import accuracy_score, f1_score
from sklearn.model_selection import train_test_split

# Custom Styling (Changes text area label color)
st.markdown(
    """
    <style>
        html, body, [class*="st-"] {
            background-color: white !important;
            color: black !important;
        }
        label[data-testid="stTextAreaLabel"] {
            color: blue !important;
            font-size: 18px;
            font-weight: bold;
        }
    </style>
    """,
    unsafe_allow_html=True,
)

# Display an Image (Check if File Exists)
image_path = "Innomatics-Logo1.png"
if os.path.exists(image_path):
    st.image(image_path, width=700)
else:
    st.warning("Image not found! Check the file path.")

# Streamlit UI
st.title("😊 Emotion Detection from Text")

# 1. Business and Data Understanding
st.header("📌 1. Business and Data Understanding")

# a) Business Problem
st.subheader("📍 a) Business Problem of Emotion Detection Using ML")
st.write("""
Emotion detection using machine learning aims to classify human emotions from text data. 
The goal is to develop an AI-driven system that can accurately recognize emotions from written sentences.  
This has applications in:
- **Customer Support:** Detecting customer sentiment in reviews and chat conversations.
- **Mental Health Monitoring:** Identifying distress or emotional patterns in patients.
- **Marketing & Brand Monitoring:** Understanding customer sentiment towards products and services.
- **Social Media & Content Analysis:** Identifying trends and reactions from social media posts.
""")

# b) Business Objective
st.subheader("📍 b) Business Objective")
st.write("""
The primary objective is to build a machine learning model that can classify emotions accurately based on textual input.  
The model should:
- Provide **high accuracy** in emotion classification.
- Be **efficient** for real-time or batch processing.
- Offer **interpretability** for business use cases.
""")

# c) Business Constraints
st.subheader("📍 c) Business Constraints")
st.write("""
- **Accuracy vs. Speed:** The model must balance accuracy and computational efficiency.
- **Data Privacy & Compliance:** Sensitive emotional data should be handled with care.
- **Scalability:** The system should be scalable for real-time analysis.
- **Bias & Fairness:** The model should minimize biases in emotion detection.
""")

# d) Data Understanding
st.subheader("📍 d) Data Understanding")
st.write("""
- **Dataset Size:** 422,746 text samples.
- **Columns:**
  1. **sentence (Text Feature)** – Contains textual data representing different emotional expressions.
  2. **emotion (Target Label)** – Represents the classified emotion associated with each sentence (e.g., happy, sad, fear, etc.).
""")

# Load the trained model
with open("emotion_model.pkl", "rb") as file:
    model = pickle.load(file)

# Load CountVectorizer (required for text processing)
vectorizer = model.named_steps["BOW"]

# Project Description
st.header("🎭 About This Project")
st.write("""
This **Emotion Detection Model** analyzes text and predicts the **emotion behind the sentence**.  
It uses **Natural Language Processing (NLP)** and a **Naïve Bayes classifier** trained on labeled emotions.  
Simply enter a sentence, and the model will detect whether it expresses happiness, sadness, anger, surprise, or other emotions!  
""")

# Load dataset again to calculate accuracy & F1-score
file_path = "combined_emotion.csv"  # Ensure this file is available
if os.path.exists(file_path):
    df = pd.read_csv(file_path)

    # Extract features and labels
    X = df["sentence"]
    y = df["emotion"]

    # Split dataset to evaluate model
    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

    # Transform test data
    X_test_transformed = vectorizer.transform(X_test)

    # Predict and calculate metrics
    y_pred = model.named_steps["Algorithm"].predict(X_test_transformed)
    accuracy = accuracy_score(y_test, y_pred)
    f1 = f1_score(y_test, y_pred, average="weighted")

    # Display model performance **AFTER** project description
    st.subheader("📊 Model Performance")
    st.write(f"✅ **Accuracy:** {accuracy:.2f}")
    st.write(f"✅ **F1-Score:** {f1:.2f}")
else:
    st.warning("Dataset not found! Accuracy and F1-score cannot be displayed.")

# Define emotion emojis
emotion_emojis = {
    "joy": "😊",
    "sad": "😢",
    "anger": "😠",
    "fear": "😨",
    "surprise": "😲",
    "neutral": "😐",
    "love": "❤️",
    "disgust": "🤢"
}

# User input with custom color
user_input = st.text_area("Enter text here:")

if st.button("Predict Emotion"):
    if user_input.strip() != "":
        # Transform input using the same vectorizer
        transformed_input = vectorizer.transform([user_input])

        # Predict emotion
        prediction = model.named_steps["Algorithm"].predict(transformed_input)[0]

        # Get corresponding emoji (default to 😶 if emotion not in dictionary)
        emoji = emotion_emojis.get(prediction.lower(), "😶")

        # Display result with emoji
        st.success(f"Predicted Emotion: **{prediction}** {emoji}")
    else:
        st.warning("Please enter some text before predicting.")