app.py

# gradio_sentiment_app.py

import gradio as gr
import pandas as pd
from sklearn.feature_extraction.text import CountVectorizer, TfidfVectorizer
from sklearn.svm import SVC
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split
from transformers import pipeline

# ------------------------------
# Prepare Dataset
# ------------------------------
data = [
    ("I love this product, it's amazing!", "positive"),
    ("This is the worst experience ever.", "negative"),
    ("Absolutely fantastic! Highly recommend.", "positive"),
    ("Not good, really disappointing.", "negative"),
    ("I am extremely happy with this.", "positive"),
    ("Would never buy this again.", "negative"),
    ("It's okay, nothing special.", "negative"),
    ("Could be better but not bad.", "negative"),
    ("Oh wow, this product is just *so* amazing. Totally worth every penny... NOT!", "negative"),
    ("Yeah, sure, this was the best purchase of my life... said no one ever.", "negative"),
    ("If disappointment had a face, it would look like this product.", "negative"),
    ("Wow! I didn’t expect it to be this great. Oh wait, I did. It’s terrible!", "negative"),
    ("Oh sure, because everyone loves wasting money on garbage, right?", "negative"),
    ("Well, it works, I guess... if you're into that sort of thing.", "negative"),
    ("I can't believe how much I love this! Just kidding, it's awful.", "negative"),
    ("This product is life-changing! I can't imagine living without it!", "positive"),
    ("Absolutely horrible, I wouldn't even gift this to my worst enemy.", "negative"),
    ("Surprisingly good! Didn’t expect to love it this much.", "positive"),
    ("I regret every penny I spent on this.", "negative"),
    ("If you enjoy wasting money, this is perfect for you!", "negative"),
    ("It does the job. Nothing more, nothing less.", "positive"),
    ("Fantastic! Way beyond my expectations!", "positive"),
    ("I wouldn’t recommend this to my worst enemy.", "negative"),
    ("Superb quality, really impressed!", "positive"),
    ("How do I return this?", "negative"),
    ("Game-changer! I absolutely love it!", "positive"),
    ("Wow, this exceeded all my expectations!", "positive"),
    ("Disappointed but not surprised.", "negative"),
    ("They really outdid themselves. This is amazing!", "positive"),
    ("This is what I call a waste of money.", "negative"),
    ("Couldn’t be happier with my purchase!", "positive"),
    ("Just another product that failed to impress.", "negative"),
    ("Best decision I ever made!", "positive"),
    ("Literally the worst thing I’ve ever bought.", "negative"),
    ("Incredible value for money!", "positive"),
    ("One word: trash.", "negative"),
    ("I didn’t expect much, and I still got disappointed.", "negative")
]

df = pd.DataFrame(data, columns=['text', 'sentiment'])

# Split data
X_train, X_test, y_train, y_test = train_test_split(df['text'], df['sentiment'], test_size=0.2, random_state=42)

# ------------------------------
# Train Traditional ML Models
# ------------------------------
# BoW + SVM
vectorizer_bow = CountVectorizer()
X_train_bow = vectorizer_bow.fit_transform(X_train)
X_test_bow = vectorizer_bow.transform(X_test)
svm_model = SVC(kernel='linear')
svm_model.fit(X_train_bow, y_train)

# TF-IDF + Logistic Regression
vectorizer_tfidf = TfidfVectorizer()
X_train_tfidf = vectorizer_tfidf.fit_transform(X_train)
X_test_tfidf = vectorizer_tfidf.transform(X_test)
lr_model = LogisticRegression()
lr_model.fit(X_train_tfidf, y_train)

# Transformer Model
transformer_model = pipeline(
    "sentiment-analysis",
    model="distilbert-base-uncased-finetuned-sst-2-english",
    device=-1  # use CPU
)

# ------------------------------
# Gradio Prediction Function
# ------------------------------
def predict_sentiment(text, model_type):
    model_type = model_type.lower()
    
    if model_type == "svm":
        X_input = vectorizer_bow.transform([text])
        pred = svm_model.predict(X_input)[0]
    elif model_type == "tfidf":
        X_input = vectorizer_tfidf.transform([text])
        pred = lr_model.predict(X_input)[0]
    elif model_type == "transformer":
        result = transformer_model(text)[0]
        pred = 'positive' if result['label'] == 'POSITIVE' else 'negative'
    elif model_type == "prompt":
        prompt = f"Classify the sentiment of the following review as positive or negative: '{text}'"
        result = transformer_model(prompt)[0]
        pred = 'positive' if result['label'] == 'POSITIVE' else 'negative'
    else:
        pred = "Invalid model type"
    
    return pred

# ------------------------------
# Gradio Interface
# ------------------------------
model_choices = ["SVM", "TFIDF", "Transformer", "Prompt"]

demo = gr.Interface(
    fn=predict_sentiment,
    inputs=[
        gr.Textbox(label="Enter Product Review", placeholder="Type your review here..."),
        gr.Dropdown(label="Select Model", choices=model_choices)
    ],
    outputs=gr.Label(label="Predicted Sentiment"),
    title="Product Sentiment Analysis",
    description="A Gradio app to classify product reviews as positive or negative using different ML and Transformer models."
)

if __name__ == "__main__":
    demo.launch(share=True)