body_analyzer.py

import requests
import os
import re

HF_API_KEY = os.getenv("HF_API_KEY")
HF_HEADERS = {"Authorization": f"Bearer {HF_API_KEY}"} if HF_API_KEY else {}

MODELS = {
    "ai_detector": "roberta-base-openai-detector",
    "sentiment": "finiteautomata/bertweet-base-sentiment-analysis",
    "spam": "mrm8488/bert-tiny-finetuned-sms-spam-detection",
}

SUSPICIOUS_PATTERNS = [
    r"verify your account",
    r"urgent action",
    r"click here",
    r"reset (your )?password",
    r"confirm (your )?identity",
    r"bank account",
    r"invoice",
    r"payment (required|overdue|failed|method expired)",
    r"unauthorized login",
    r"compromised",
    r"final reminder",
    r"account (suspended|deactivated|locked)",
    r"update your (information|details|billing)",
    r"legal action",
]

def query_hf(model, text):
    if not HF_API_KEY:
        return None
    try:
        res = requests.post(
            f"https://api-inference.huggingface.co/models/{model}",
            headers=HF_HEADERS,
            json={"inputs": text[:1000]},
        )
        return res.json()
    except Exception:
        return None

def analyze_body(text):
    findings = []
    score = 0
    body_lower = text.lower()
    highlighted_body = text

    # --- 1. Suspicious keyword detection ---
    for pattern in SUSPICIOUS_PATTERNS:
        matches = re.findall(pattern, body_lower)
        for match in matches:
            findings.append(f"Suspicious phrase detected: \"{match}\"")
            score += 20
            highlighted_body = re.sub(
                match, f"<mark>{match}</mark>", highlighted_body, flags=re.IGNORECASE
            )

    # --- 2. AI-generated text detection ---
    result = query_hf(MODELS["ai_detector"], text)
    if result and isinstance(result, list) and len(result) > 0:
        label = result[0]["label"]
        confidence = result[0]["score"]
        findings.append(f"Body: AI Detector → {label} (confidence {confidence:.2f})")

    # --- 3. Sentiment analysis ---
    result = query_hf(MODELS["sentiment"], text)
    if result and isinstance(result, list) and len(result) > 0:
        label = result[0]["label"]
        confidence = result[0]["score"]
        findings.append(f"Body: Sentiment → {label} (confidence {confidence:.2f})")
        if label.lower() == "negative":
            score += 10

    # --- 4. Spam vs Ham detection ---
    result = query_hf(MODELS["spam"], text)
    if result and isinstance(result, list) and len(result) > 0:
        label = result[0]["label"]
        confidence = result[0]["score"]
        findings.append(f"Body: Spam Detector → {label} (confidence {confidence:.2f})")
        if label.lower() == "spam":
            score += 20

    if not findings:
        return ["No suspicious content detected in body."], 0, text

    return findings, score, highlighted_body