nlp_app.py

import streamlit as st
import re
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import LogisticRegression

# 1. Page Layout & UI Configuration
st.set_page_config(page_title="NLP Pipeline Tool", layout="centered")
st.title("📝 Natural Language Processing Pipeline")
st.write("This app demonstrates a complete text classification pipeline using Tokenisation, TF-IDF, and Machine Learning.")

# 2. The Internal Training Dataset
# (The model trains itself instantly on this data every time the app starts!)
training_data = [
    "This is an amazing and fantastic tool, I love it!",
    "Excellent quality, highly recommend this to everyone.",
    "Brilliant customer care service and great experience.",
    "The product works perfectly and smoothly.",
    "Completely useless, a total waste of money and time.",
    "Horrible experience, it broke immediately on day one.",
    "I hate this application, it crashes constantly.",
    "Worst service ever, deeply disappointed with the quality."
]
# Target Labels: 1 = Positive Sentiment, 0 = Negative Sentiment
training_labels = [1, 1, 1, 1, 0, 0, 0, 0]

# --- THE NLP PIPELINE IMPLEMENTATION ---

# Pipeline Steps 1 & 2: Tokenisation & TF-IDF Vectorisation
# The TfidfVectorizer handles tokenisation automatically by cleaning and splitting text
vectorizer = TfidfVectorizer(lowercase=True, stop_words='english')
X_train_tfidf = vectorizer.fit_transform(training_data)

# Pipeline Step 3: The Trained Classifier Model
# Training a Logistic Regression model brain on our numeric numeric dataset
classifier = LogisticRegression()
classifier.fit(X_train_tfidf, training_labels)

st.success("🤖 NLP Pipeline Model Trained Successfully!")

# --- USER INTERACTION INTERFACE ---
st.markdown("### 🔍 Test the Pipeline Live")
user_input = st.text_input("Enter a custom sentence or product review to classify:", "The build quality is absolutely horrible and broken.")

if user_input:
    st.markdown("---")
    st.subheader("🛠️ Pipeline Execution Breakdown")
    
    # Showcase Pipeline Step 1: Tokenisation
    # Emulating text cleanup to show raw tokens explicitly on screen
    clean_text = re.sub(r'[^\w\s]', '', user_input.lower())
    tokens = clean_text.split()
    st.write("#### 1️⃣ Tokenisation Stage")
    st.info(f"**Extracted Text Tokens (Individual Words):** `{tokens}`")
    
    # Showcase Pipeline Step 2: TF-IDF Vectorisation
    user_tfidf = vectorizer.transform([user_input])
    feature_names = vectorizer.get_feature_names_out()
    matched_features = [word for word in tokens if word in feature_names]
    
    st.write("#### 2️⃣ TF-IDF Vectorisation Stage")
    st.info(f"**Key Evaluated Features Found (Math Weights Applied):** `{matched_features}`")
    
    # Showcase Pipeline Step 3: Trained Classifier Prediction
    prediction = classifier.predict(user_tfidf)[0]
    probabilities = classifier.predict_proba(user_tfidf)[0]
    
    st.write("#### 3️⃣ Trained Classifier Prediction")
    if prediction == 1:
        st.markdown("### Final Sentiment Class: **🟢 POSITIVE**")
        st.progress(float(probabilities[1]))
        st.write(f"📊 Model Confidence Score: **{probabilities[1]*100:.2f}%**")
    else:
        st.markdown("### Final Sentiment Class: **🔴 NEGATIVE**")
        st.progress(float(probabilities[0]))
        st.write(f"📊 Model Confidence Score: **{probabilities[0]*100:.2f}%**")