app.py

import gradio as gr
import joblib
import re
import numpy as np
from sastrawi.stemmer.stemmer_factory import StemmerFactory
from sastrawi.stopwords.stopwords_factory import StopWordRemoverFactory
import nltk

# --- Download NLTK data (only needs to run once) ---
try:
    nltk.data.find('tokenizers/punkt')
except nltk.downloader.DownloadError:
    nltk.download('punkt')

# --- 1. Load Pre-trained Model and Vectorizer ---
# These files should be in the same directory as your app.py file.
model = joblib.load('best_svm_model.pkl')
vectorizer = joblib.load('tfidf_vectorizer.pkl')

# --- 2. Recreate the Preprocessing Functions ---
# Initialize Sastrawi components
stemmer = StemmerFactory().create_stemmer()
stopword_remover = StopWordRemoverFactory().create_stop_word_remover()

# Slang dictionary from your notebook
slang_dict = {
    'yg': 'yang', 'ga': 'tidak', 'gak': 'tidak', 'udh': 'sudah', 'tdk': 'tidak',
    'bgt': 'banget', 'dg': 'dengan', 'klo': 'kalau', 'kalo': 'kalau', 'mksh': 'terima kasih',
    'terimakasih': 'terima kasih', 'bgs': 'bagus', 'ok': 'oke', 'blm': 'belum', 'sy': 'saya',
    'sya': 'saya', 'ak': 'aku', 'utk': 'untuk', 'tpi': 'tapi', 'tp': 'tapi', 'jd': 'jadi',
    'jg': 'juga', 'trs': 'terus', 'skrg': 'sekarang', 'bkin': 'bikin', 'dr': 'dari',
    'dn': 'dan', 'pke': 'pakai', 'gausah': 'tidak usah', 'ngga': 'tidak', 'bkn': 'bukan',
    'sdh': 'sudah', 'aja': 'saja', 'lg': 'lagi', 'mls': 'malas', 'gk': 'tidak',
    'knp': 'kenapa', 'krn': 'karena', 'gmn': 'bagaimana', 'gimana': 'bagaimana',
    'udah': 'sudah', 'sm': 'sama', 'gbs': 'tidak bisa', 'nggak': 'tidak', 'mantap': 'bagus',
    'cek': 'periksa', 'bansos': 'bantuan sosial'
}

def preprocess_text(text):
    # 1. Cleaning: numbers, punctuation, extra spaces
    text = re.sub(r'\d+', '', text)
    text = re.sub(r'[^\w\s]', '', text)
    text = re.sub(r'\s+', ' ', text).strip()
    
    # 2. Case folding
    text = text.lower()
    
    # 3. Slang normalization
    words = text.split()
    normalized_words = [slang_dict.get(word, word) for word in words]
    text = ' '.join(normalized_words)
    
    # 4. Stopword removal
    text = stopword_remover.remove(text)
    
    # 5. Stemming
    text = stemmer.stem(text)
    
    return text

# --- 3. Prediction Function ---
def predict_sentiment(sentence):
    # Preprocess the input sentence
    processed_text = preprocess_text(sentence)
    
    # Vectorize the text using the loaded TF-IDF vectorizer
    text_vector = vectorizer.transform([processed_text])
    
    # The model was trained with an additional 'thumbs_up_log_scaled' feature.
    # Since we only have a sentence, we'll assume a neutral value (0) for this feature.
    thumbs_up_feature = np.array([[0]])
    
    # Combine the TF-IDF vector with the thumbs_up feature
    # Note: hstack is used for sparse matrices
    final_vector = np.hstack([text_vector.toarray(), thumbs_up_feature])
    
    # Predict using the loaded model
    prediction = model.predict(final_vector)
    
    # Return the result
    return prediction[0].capitalize()

# --- 4. Create Gradio Interface ---
iface = gr.Interface(
    fn=predict_sentiment,
    inputs=gr.Textbox(lines=3, placeholder="Masukkan kalimat ulasan dalam Bahasa Indonesia..."),
    outputs="text",
    title="Analisis Sentimen Ulasan Aplikasi",
    description="Analisis sentimen untuk ulasan aplikasi 'Cek Bansos' menggunakan model SVM. Masukkan sebuah kalimat untuk memprediksi sentimennya (Positif, Negatif, atau Netral).",
    examples=[
        ["aplikasinya bagus sekali dan sangat membantu"],
        ["tidak bisa daftar, gagal terus padahal sinyal bagus"],
        ["aplikasi ini biasa saja, tidak ada yang spesial"]
    ]
)

# --- 5. Launch the App ---
iface.launch()