app.py

from hazm import stopwords_list, Lemmatizer, Normalizer
import gradio as gr
import numpy as np
import pickle
from huggingface_hub import hf_hub_download

# Load model
model_path = hf_hub_download(
    repo_id="sirunchained/bad-comment",
    filename="offensive_text_classifier.pkl"
)
with open(model_path, "rb") as f:
    pickle_model = pickle.load(f)

stopwords = stopwords_list()
lemmatizer = Lemmatizer()
normalizer = Normalizer()

def clean_sentences(sentence: str) -> str:
    normalized = normalizer.normalize(sentence)
    return " ".join(
        lemmatizer.lemmatize(word)
        for word in normalized.split()
        if word not in stopwords
    )

def predict_with_model(comment: str) -> str:
    cleaned_comment = clean_sentences(comment)

    output_pred = pickle_model.predict([cleaned_comment])
    output_proba = pickle_model.predict_proba([cleaned_comment])

    label = "Offensive" if output_pred[0] == 1 else "Neutral"
    confidence = output_proba.max()

    return f"{confidence:.2f} : {label}"

iface = gr.Interface(
    fn=predict_with_model,
    inputs=gr.Textbox(lines=2, placeholder="Enter a comment here..."),
    outputs="text",
    title="Persian Offensive Comment Classifier",
    description="Enter a Persian comment to classify it."
)

if __name__ == "__main__":
    iface.launch()