Spaces:
Sleeping
Sleeping
File size: 19,544 Bytes
63d7edb |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 |
import os
import io
import json
import requests
from typing import Dict, List, Any, Optional
from fastapi import FastAPI, HTTPException, Body
from pydantic import BaseModel
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification
from newspaper import Article
from bs4 import BeautifulSoup
import easyocr
from PIL import Image
import google.generativeai as genai
from datetime import datetime
import logging
from fastapi.middleware.cors import CORSMiddleware
# Setup logging
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
)
logger = logging.getLogger("TruthLens")
# Initialize FastAPI app
app = FastAPI(title="TruthLens Backend")
# Add CORS middleware
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
# Load Hugging Face model (RoBERTa-based fake news detector)
MODEL_NAME = "Pulk17/Fake-News-Detection"
tokenizer = None
model = None
def load_model():
"""Lazy load the Hugging Face model"""
global tokenizer, model
if tokenizer is None or model is None:
print("Loading Hugging Face model...")
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME)
model.eval()
return tokenizer, model
# Environment variables
# Ensure you set these in your environment or .env file
GOOGLE_FACT_CHECK_API_KEY = os.environ.get('GOOGLE_FACT_CHECK_API_KEY', '')
HIVE_API_KEY = os.environ.get('HIVE_API_KEY', '')
GEMINI_API_KEY = os.environ.get('GEMINI_API_KEY', '')
# Initialize Gemini
if GEMINI_API_KEY:
genai.configure(api_key=GEMINI_API_KEY)
gemini_model = genai.GenerativeModel('gemini-pro')
else:
gemini_model = None
# Request Models
class TextRequest(BaseModel):
text: str
class UrlRequest(BaseModel):
url: str
class ImageRequest(BaseModel):
image_url: str
# Source Credibility Database
CREDIBLE_SOURCES = {
"apnews.com": "Associated Press",
"reuters.com": "Reuters",
"bbc.com": "BBC News",
"bbc.co.uk": "BBC News",
"nytimes.com": "The New York Times",
"npr.org": "NPR",
"pbs.org": "PBS NewsHour",
"wsj.com": "The Wall Street Journal",
"bloomberg.com": "Bloomberg",
"theguardian.com": "The Guardian",
"washingtonpost.com": "The Washington Post",
"propublica.org": "ProPublica",
"aljazeera.com": "Al Jazeera",
"economist.com": "The Economist",
"forbes.com": "Forbes"
}
SATIRE_SOURCES = {
"theonion.com": "The Onion",
"babylonbee.com": "The Babylon Bee",
"clickhole.com": "ClickHole",
"newyorker.com/humor/borowitz-report": "The Borowitz Report",
"thebeaverton.com": "The Beaverton",
"cracked.com": "Cracked",
"dailymash.co.uk": "The Daily Mash",
"waterfordwhispersnews.com": "Waterford Whispers News"
}
# Helper functions
def get_fact_checks(text: str) -> List[Dict[str, str]]:
"""Get fact checks from Google Fact Check Tools API"""
if not GOOGLE_FACT_CHECK_API_KEY:
return []
try:
# Extract key claims (first 100 chars as query)
query = text[:100]
url = "https://factchecktools.googleapis.com/v1alpha1/claims:search"
params = {
"query": query,
"key": GOOGLE_FACT_CHECK_API_KEY,
"languageCode": "en"
}
response = requests.get(url, params=params, timeout=10)
if response.status_code == 200:
data = response.json()
claims = data.get('claims', [])
fact_checks = []
for claim in claims[:3]: # Top 3 fact checks
fact_check = {
"claim": claim.get('text', ''),
"claimant": claim.get('claimant', ''),
"rating": claim.get('claimReview', [{}])[0].get('textualRating', 'Unknown'),
"url": claim.get('claimReview', [{}])[0].get('url', '')
}
fact_checks.append(fact_check)
return fact_checks
else:
print(f"Fact check API error: {response.status_code}")
return []
except Exception as e:
print(f"Error getting fact checks: {e}")
return []
def extract_claims_with_gemini(text: str) -> List[str]:
"""Use Gemini to extract key factual claims for building a search query"""
if not gemini_model:
return [text[:100]]
try:
prompt = f"""
Extract the single most important factual claim from the following text that can be used to search in a fact-check database.
Output ONLY the extracted claim string, nothing else.
Text: {text[:1000]}
"""
response = gemini_model.generate_content(prompt)
if response and hasattr(response, 'text'):
claim = response.text.strip()
return [claim] if claim else [text[:100]]
return [text[:100]]
except Exception as e:
print(f"Gemini claim extraction error: {e}")
return [text[:100]]
def generate_explanation_with_gemini(text: str, label: str, confidence: float, fact_checks: List[Dict]) -> str:
"""Use Gemini to explain the reasoning behind the detection result"""
if not gemini_model:
return f"The news has been classified as {label} with {confidence:.2%} confidence."
try:
fact_check_context = ""
if fact_checks:
fact_check_context = "Relevant fact checks found:\n" + "\n".join([f"- {fc['claim']} (Rating: {fc['rating']})" for fc in fact_checks])
prompt = f"""
Act as a professional fact-checker for an app called TruthLens.
Analyze the following news text and the AI detection result.
News Text: {text[:1000]}
AI Classification: {label}
Confidence: {confidence:.2%}
{fact_check_context}
Provide a concise, human-readable explanation (2-3 sentences) explaining why this news is likely {label}.
Focus on style, source (if present), or specific fact-check evidence.
"""
response = gemini_model.generate_content(prompt)
if response and hasattr(response, 'text'):
return response.text.strip()
return f"The model identified this content as {label} with {confidence:.2%} confidence."
except Exception as e:
print(f"Gemini explanation error: {e}")
return f"Analysis complete: The model identified this content as {label}."
def detect_ai_image(image_bytes: bytes) -> Dict[str, Any]:
"""Detect AI-generated content using Hive Moderation API"""
if not HIVE_API_KEY:
return {"probability": 0.0, "generator": None}
try:
url = "https://api.hivemoderation.com/v2/task/sync"
headers = {
"Authorization": f"Token {HIVE_API_KEY}",
"Content-Type": "application/json"
}
# Convert image to base64
import base64
image_b64 = base64.b64encode(image_bytes).decode('utf-8')
payload = {
"image": image_b64,
"models": ["ai_generated"]
}
response = requests.post(url, headers=headers, json=payload, timeout=30)
if response.status_code == 200:
data = response.json()
ai_generated = data.get('status', [{}])[0].get('response', {}).get('output', [{}])[0]
return {
"probability": ai_generated.get('score', 0.0),
"generator": ai_generated.get('class', None)
}
else:
print(f"Hive API error: {response.status_code}")
return {"probability": 0.0, "generator": None}
except Exception as e:
print(f"Error detecting AI image: {e}")
return {"probability": 0.0, "generator": None}
def calculate_risk_level(
label: str,
confidence: float,
fact_checks: List[Dict],
image_ai_result: Optional[Dict] = None
) -> str:
"""Calculate overall risk level"""
# Base risk on label and confidence
if label == "FAKE" and confidence > 0.8:
base_risk = "high"
elif label == "FAKE" and confidence > 0.5:
base_risk = "medium"
elif label == "REAL" and confidence > 0.8:
base_risk = "low"
else:
base_risk = "medium"
# Adjust based on fact checks
if fact_checks:
fake_ratings = sum(1 for fc in fact_checks if 'false' in fc['rating'].lower() or 'fake' in fc['rating'].lower() or 'satire' in fc['rating'].lower())
if fake_ratings >= 2:
base_risk = "high"
# Adjust based on AI image detection
if image_ai_result and image_ai_result['probability'] > 0.7:
if base_risk == "low":
base_risk = "medium"
elif base_risk == "medium":
base_risk = "high"
return base_risk
def check_source_credibility(url: str) -> Dict[str, Any]:
"""Check if the URL belongs to a known credible or satire source"""
from urllib.parse import urlparse
try:
domain = urlparse(url).netloc.lower()
if domain.startswith("www."):
domain = domain[4:]
# Check Satire first
for satire_domain, name in SATIRE_SOURCES.items():
if satire_domain in url.lower():
return {"status": "satire", "name": name, "label": "FAKE", "confidence": 1.0}
# Check Credible
if domain in CREDIBLE_SOURCES:
return {"status": "credible", "name": CREDIBLE_SOURCES[domain], "label": "REAL", "confidence": 0.95}
return {"status": "unknown", "name": None, "label": None, "confidence": 0.0}
except Exception as e:
print(f"Error checking credibility: {e}")
return {"status": "unknown", "name": None, "label": None, "confidence": 0.0}
def extract_article_text(url: str) -> str:
"""Extract article text from URL using newspaper3k"""
try:
article = Article(url)
article.download()
article.parse()
return article.text
except Exception as e:
print(f"Error extracting article with newspaper3k: {e}")
# Fallback to BeautifulSoup
try:
response = requests.get(url, timeout=30)
soup = BeautifulSoup(response.content, 'html.parser')
# Remove script and style elements
for script in soup(["script", "style"]):
script.decompose()
# Get text
text = soup.get_text()
# Clean up whitespace
lines = (line.strip() for line in text.splitlines())
chunks = (phrase.strip() for line in lines for phrase in line.split(" "))
text = ' '.join(chunk for chunk in chunks if chunk)
return text
except Exception as e2:
logger.error(f"Error with BeautifulSoup fallback: {e2}")
return ""
@app.get("/")
async def root():
return {"status": "healthy", "service": "TruthLens API"}
# Endpoints
@app.post("/detect-text")
async def detect_text(request: TextRequest, skip_extras: bool = False):
"""Detect fake news in text with optional fact-check and explanation"""
try:
text = request.text
if not text:
raise HTTPException(status_code=400, detail="Text is required")
# Load model
tok, mdl = load_model()
# Tokenize and predict
inputs = tok(text, return_tensors="pt", truncation=True, max_length=512, padding=True)
with torch.no_grad():
outputs = mdl(**inputs)
logits = outputs.logits
probabilities = torch.softmax(logits, dim=1)
prediction = torch.argmax(probabilities, dim=1).item()
confidence = probabilities[0][prediction].item()
label = "REAL" if prediction == 1 else "FAKE"
if skip_extras:
return {
"input_type": "text",
"label": label,
"confidence": confidence,
"timestamp": datetime.now().isoformat()
}
# Enhanced Fact Checking with Gemini
extracted_claims = extract_claims_with_gemini(text)
fact_checks = get_fact_checks(extracted_claims[0])
# Gemini Explanation
explanation = generate_explanation_with_gemini(text, label, confidence, fact_checks)
risk_level = calculate_risk_level(label, confidence, fact_checks)
return {
"input_type": "text",
"text": text,
"label": label,
"confidence": confidence,
"explanation": explanation,
"fact_checks": fact_checks,
"risk_level": risk_level,
"timestamp": datetime.now().isoformat()
}
except Exception as e:
print(f"Error in detect_text: {str(e)}")
raise HTTPException(status_code=500, detail=str(e))
@app.post("/detect-url")
async def detect_url(request: UrlRequest, skip_extras: bool = False):
"""Detect fake news in URL with optional fact-check and explanation"""
try:
url = request.url
if not url:
raise HTTPException(status_code=400, detail="URL is required")
# 1. Check Source Credibility First
source_info = check_source_credibility(url)
article_text = extract_article_text(url)
if not article_text:
raise HTTPException(status_code=400, detail="Failed to extract article text from URL")
tok, mdl = load_model()
inputs = tok(article_text, return_tensors="pt", truncation=True, max_length=512, padding=True)
with torch.no_grad():
outputs = mdl(**inputs)
logits = outputs.logits
probabilities = torch.softmax(logits, dim=1)
prediction = torch.argmax(probabilities, dim=1).item()
confidence = probabilities[0][prediction].item()
label = "REAL" if prediction == 1 else "FAKE"
# Override with source credibility if it's definitive
if source_info["status"] == "satire":
label = "FAKE"
confidence = 1.0
elif source_info["status"] == "credible" and label == "FAKE":
# If a credible source is flagged as fake, we lower risk but keep label
# or we could trust the source more. Let's provide it in metadata.
pass
if skip_extras:
return {
"input_type": "url",
"label": label,
"confidence": confidence,
"source_metadata": source_info,
"timestamp": datetime.now().isoformat()
}
# Enhanced Fact Checking with Gemini
extracted_claims = extract_claims_with_gemini(article_text)
fact_checks = get_fact_checks(extracted_claims[0])
# Gemini Explanation
explanation = generate_explanation_with_gemini(
f"Source: {source_info['name'] if source_info['name'] else 'Unknown'}. Content: {article_text}",
label,
confidence,
fact_checks
)
risk_level = calculate_risk_level(label, confidence, fact_checks)
return {
"input_type": "url",
"url": url,
"source_metadata": source_info,
"text": article_text[:500],
"label": label,
"confidence": confidence,
"explanation": explanation,
"fact_checks": fact_checks,
"risk_level": risk_level,
"timestamp": datetime.now().isoformat()
}
except Exception as e:
print(f"Error in detect_url: {str(e)}")
raise HTTPException(status_code=500, detail=str(e))
@app.post("/detect-image")
async def detect_image(request: ImageRequest, skip_extras: bool = False):
try:
image_url = request.image_url
logger.info(f"Processing image: {image_url}")
if not image_url:
raise HTTPException(status_code=400, detail="Image URL is required")
response = requests.get(image_url, timeout=30)
response.raise_for_status()
image_bytes = response.content
reader = easyocr.Reader(['en'])
# EasyOCR can read from bytes directly
ocr_results = reader.readtext(image_bytes)
extracted_text = ' '.join([result[1] for result in ocr_results])
image_ai_result = detect_ai_image(image_bytes)
if extracted_text.strip():
tok, mdl = load_model()
inputs = tok(extracted_text, return_tensors="pt", truncation=True, max_length=512, padding=True)
with torch.no_grad():
outputs = mdl(**inputs)
logits = outputs.logits
probabilities = torch.softmax(logits, dim=1)
prediction = torch.argmax(probabilities, dim=1).item()
confidence = probabilities[0][prediction].item()
label = "REAL" if prediction == 1 else "FAKE"
if skip_extras:
return {
"input_type": "image",
"label": label,
"confidence": confidence,
"image_ai_result": image_ai_result,
"timestamp": datetime.now().isoformat()
}
# Enhanced Fact Checking with Gemini
extracted_claims = extract_claims_with_gemini(extracted_text)
fact_checks = get_fact_checks(extracted_claims[0])
else:
label = "FAKE" if image_ai_result['probability'] > 0.7 else "REAL"
confidence = image_ai_result['probability'] if label == "FAKE" else (1 - image_ai_result['probability'])
fact_checks = []
if skip_extras:
return {
"input_type": "image",
"label": label,
"confidence": confidence,
"image_ai_result": image_ai_result,
"timestamp": datetime.now().isoformat()
}
# Gemini Explanation
explanation = generate_explanation_with_gemini(extracted_text if extracted_text else "No text found in image", label, confidence, fact_checks)
risk_level = calculate_risk_level(label, confidence, fact_checks, image_ai_result)
return {
"input_type": "image",
"image_url": image_url,
"text": extracted_text[:500] if extracted_text else None,
"label": label,
"confidence": confidence,
"explanation": explanation,
"fact_checks": fact_checks,
"image_ai_result": image_ai_result,
"risk_level": risk_level,
"timestamp": datetime.now().isoformat()
}
except Exception as e:
print(f"Error in detect_image: {str(e)}")
raise HTTPException(status_code=500, detail=str(e))
if __name__ == "__main__":
import uvicorn
uvicorn.run(app, host="0.0.0.0", port=8000)
|