import os import json import re import csv import io from flask import Flask, request, jsonify, render_template_string from transformers import AutoTokenizer, AutoModelForSequenceClassification import torch app = Flask(__name__) MODEL_PATH = "./hate_speech_model" tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH) model = AutoModelForSequenceClassification.from_pretrained(MODEL_PATH) threshold = 0.5 if os.path.exists(os.path.join(MODEL_PATH, "training_config.json")): try: with open(os.path.join(MODEL_PATH, "training_config.json"), "r") as f: config = json.load(f) threshold = config.get("optimal_threshold", 0.5) print(f"Loaded optimal threshold from config: {threshold:.4f}") except Exception as e: print(f"Gagal memuat training_config.json: {e}. Menggunakan threshold default 0.5") dataset_cache = {"data": None, "loaded": False} def preprocess_text(text): text = re.sub(r'@[^\s]+', '@USER', text) text = re.sub(r'https?://[^\s]+', 'HTTPURL', text) text = re.sub(r'\s+', ' ', text).strip() return text WEB_UI_HTML = r""" Hate Speech Detector

Hate Speech Detector

Klasifikasi teks ujaran kebencian

0 karakter
Non-Hate 0% Hate 

                

            

        

    


    
    

        

            

                
                
Memuat dataset IndoDiscourse dari HuggingFace...

            


            
        

    


    
Powered by Hate Speech Shield · Threshold: {{ threshold }}







"""

@app.route("/")
def home():
    return render_template_string(WEB_UI_HTML, threshold=threshold)

@app.route("/api/dataset")
def get_dataset():
    if dataset_cache["loaded"]:
        return jsonify({"data": dataset_cache["data"]})

    try:
        import urllib.request

        url = "https://huggingface.co/datasets/Exqrch/IndoDiscourse/resolve/main/indotoxic2024_annotated_data_v2_final.csv"
        req = urllib.request.Request(url, headers={"User-Agent": "HateSpeechDetector/1.0"})
        with urllib.request.urlopen(req, timeout=30) as response:
            raw = response.read().decode("utf-8")

        reader = csv.DictReader(io.StringIO(raw))
        rows = []
        for row in reader:
            text = row.get("text", "").strip()
            if not text:
                continue

            rows.append({"text": text})

        dataset_cache["data"] = rows
        dataset_cache["loaded"] = True

        return jsonify({"data": rows})

    except Exception as e:
        return jsonify({"error": f"Gagal memuat dataset: {str(e)}"}), 500

@app.route("/predict", methods=["POST"])
def predict():
    api_key = request.headers.get("X-API-Key")
    if api_key != "sh1eld-hate-speech-key-2026":
        return jsonify({"error": "Akses ditolak: API Key tidak valid atau tidak disediakan"}), 401

    data = request.get_json()

    if not data or "text" not in data:
        return jsonify({"error": "Silakan kirimkan parameter 'text'"}), 400

    text = data["text"]
    
    text_clean = preprocess_text(text)

    inputs = tokenizer(text_clean, return_tensors="pt", truncation=True, padding=True)
    with torch.no_grad():
        outputs = model(**inputs)
        probs = torch.nn.functional.softmax(outputs.logits, dim=1)
        hate_prob = probs[0][1].item()
        
        clean_text_for_len = text.strip()
        if len(clean_text_for_len) > 500:
            clean_text_for_len = clean_text_for_len[:500]
        text_len = len(clean_text_for_len)

        if text_len < 80:
            effective_threshold = 0.92
        elif text_len < 200:
            effective_threshold = 0.82
        else:
            effective_threshold = threshold

        pred_label = 1 if hate_prob >= effective_threshold else 0
        label = "Hate" if pred_label == 1 else "Non-Hate"
        confidence = hate_prob if pred_label == 1 else (1 - hate_prob)

    return jsonify({
        "label": label,
        "confidence": confidence,
        "hate_probability": float(hate_prob),
        "text_length": len(text),
        "threshold_used": effective_threshold,
        "probabilities": {
            "Non-Hate": float(probs[0][0]),
            "Hate": float(probs[0][1])
        }
    })

if __name__ == "__main__":
    app.run(host="0.0.0.0", port=7860)