Spaces:

tomalex04
/

Credible

Sleeping

App Files Files Community

tomalex04 commited on 22 days ago

Commit

5906cc6

1 Parent(s): 32244bd

Sync latest code from Hugging Face

Browse files

Files changed (5) hide show

misinformationui/static/front.html +25 -0
misinformationui/static/main.py +327 -0
misinformationui/static/script.js +88 -0
misinformationui/static/style.css +188 -0
misinformationui/static/test_server.py +0 -0

misinformationui/static/front.html ADDED Viewed

	@@ -0,0 +1,25 @@

+<!DOCTYPE html>
+<html lang="en">
+<head>
+  <meta charset="UTF-8">
+  <meta name="viewport" content="width=device-width, initial-scale=1.0">
+  <title>Fake News Detection Chatbot</title>
+  <link rel="stylesheet" href="style.css">
+</head>
+<body>
+  <div class="chat-container" id="chat"></div>
+  <div class="input-area">
+    <input type="text" id="query" placeholder="Type news to verify..." />
+    <button id="sendBtn">Send</button>
+  </div>
+  <div class="chat-background" id="chat-background">
+    <p id="p1">Hey,</p>
+    <p id="p2">Discover misinformations around you,</p>
+  </div>
+  <script src="script.js"></script>
+</body>
+</html>

misinformationui/static/main.py ADDED Viewed

	@@ -0,0 +1,327 @@

+#!/home/tom/miniconda3/envs/fake_news_detection/bin/python
+"""
+main.py - Server for the Fake News Detection system
+This script creates a Flask server that exposes API endpoints to:
+1. Take user input (news query) from the UI
+2. Process the request through the fake news detection pipeline
+3. Return the results to the UI for display
+"""
+import os
+import json
+import time
+from dotenv import load_dotenv
+from flask import Flask, request, jsonify
+from flask_cors import CORS
+# Import required functions from modules
+from gdelt_api import (
+    fetch_articles_from_gdelt,
+    filter_by_whitelisted_domains,
+    normalize_gdelt_articles
+)
+from ranker import ArticleRanker
+from gdelt_query_builder import generate_query, GEMINI_MODEL
+import bias_analyzer
+# Global variable for embedding model caching across requests
+print("Preloading embedding model for faster request processing...")
+# Preload the embedding model at server startup
+global_ranker = ArticleRanker()
+# The function has been removed since bias category descriptions are provided directly by the Gemini model
+# and stored in the bias_analysis["descriptions"] dictionary
+def format_results(query, ranked_articles):
+    """
+    Format the ranked results in a structured way for the UI.
+    Args:
+        query (str): The original query
+        ranked_articles (list): List of ranked article dictionaries
+    Returns:
+        dict: Dictionary with formatted results
+    """
+    result = {}
+    if not ranked_articles:
+        result = {
+            "status": "no_results",
+            "message": "⚠️ No news found. Possibly Fake.",
+            "details": "No reliable sources could verify this information.",
+            "articles": []
+        }
+    else:
+        # Get display configuration from environment variables
+        show_scores = os.getenv('SHOW_SIMILARITY_SCORES', 'true').lower() == 'true'
+        show_date = os.getenv('SHOW_PUBLISH_DATE', 'true').lower() == 'true'
+        show_url = os.getenv('SHOW_URL', 'true').lower() == 'true'
+        formatted_articles = []
+        for article in ranked_articles:
+            formatted_article = {
+                "rank": article['rank'],
+                "title": article['title'],
+                "source": article['source']
+            }
+            if show_scores:
+                formatted_article["similarity_score"] = round(article['similarity_score'], 4)
+            if show_url:
+                formatted_article["url"] = article['url']
+            if show_date:
+                formatted_article["published_at"] = article['published_at']
+            formatted_articles.append(formatted_article)
+        result = {
+            "status": "success",
+            "message": f"✅ Found {len(ranked_articles)} relevant articles for: '{query}'",
+            "articles": formatted_articles,
+            "footer": "If the news matches these reliable sources, it's likely true. If it contradicts them or no sources are found, it might be fake."
+        }
+    return result
+def remove_duplicates(articles):
+    """
+    Remove duplicate articles based on URL.
+    Args:
+        articles (list): List of article dictionaries
+    Returns:
+        list: List with duplicate articles removed
+    """
+    unique_urls = set()
+    unique_articles = []
+    for article in articles:
+        if article['url'] not in unique_urls:
+            unique_urls.add(article['url'])
+            unique_articles.append(article)
+    return unique_articles
+# This function has been removed since Gemini is a cloud API service
+# that does not require local caching - models are instantiated as needed
+def main():
+    """Main function to run the fake news detection pipeline as a server."""
+    # Load environment variables
+    load_dotenv()
+    # Create Flask app
+    app = Flask(__name__, static_folder='static')
+    CORS(app)  # Enable CORS for all routes
+    @app.route('/static/')
+    def index():
+        """Serve the main page."""
+        return app.send_static_file('front.html')
+    @app.route('/api/detect', methods=['POST'])
+    def detect_fake_news():
+        """API endpoint to check if news is potentially fake."""
+        # Start timing the request processing
+        start_time = time.time()
+        data = request.json
+        query = data.get('query', '')
+        if not query:
+            return jsonify({
+                "status": "error",
+                "message": "Please provide a news statement to verify."
+            })
+        # =====================================================
+        # 1. Input Handling
+        # =====================================================
+        # Generate three variations of the query using Gemini
+        query_variations = generate_query(query)
+        # Check if the query was flagged as inappropriate
+        if query_variations == ["INAPPROPRIATE_QUERY"]:
+            return jsonify({
+                "status": "error",
+                "message": "I cannot provide information on this topic as it appears to contain sensitive or inappropriate content."
+            })
+        # =====================================================
+        # 2. Data Fetching
+        # =====================================================
+        # Fetch articles from GDELT API for each query variation
+        all_articles = []
+        for query_var in query_variations:
+            articles = fetch_articles_from_gdelt(query_var)
+            if articles:
+                all_articles.extend(articles)
+        # Store unique articles in a set to ensure uniqueness
+        unique_articles = remove_duplicates(all_articles)
+        # Apply domain whitelist filtering if enabled in .env
+        use_whitelist_only = os.getenv('USE_WHITELIST_ONLY', 'false').lower() == 'true'
+        if use_whitelist_only:
+            print(f"Filtering articles to only include whitelisted domains...")
+            unique_articles = filter_by_whitelisted_domains(unique_articles)
+            print(f"After whitelist filtering: {len(unique_articles)} articles remain")
+        # Normalize the articles to a standard format
+        normalized_articles = normalize_gdelt_articles(unique_articles)
+        if not normalized_articles:
+            return jsonify(format_results(query, []))
+        # =====================================================
+        # 3. Embedding & Ranking
+        # =====================================================
+        # Initialize the ranker with model from environment variable
+        model_name = os.getenv('SIMILARITY_MODEL', 'intfloat/multilingual-e5-base')
+        # Use global ranker if it matches the requested model, otherwise create a new instance
+        if global_ranker.model_name == model_name:
+            ranker = global_ranker
+        else:
+            ranker = ArticleRanker(model_name)
+        # Get TOP_K_ARTICLES from .env file
+        TOP_K_ARTICLES = int(os.getenv('TOP_K_ARTICLES', 250))
+        min_threshold = float(os.getenv('MIN_SIMILARITY_THRESHOLD', 0.1))
+        # Prepare article texts for embedding
+        article_texts = [f"{article['title']} {article['description'] or ''}" for article in normalized_articles]
+        # Create embeddings and calculate similarities
+        query_embedding, article_embeddings = ranker.create_embeddings(query, article_texts)
+        similarities = ranker.calculate_similarities(query_embedding, article_embeddings)
+        # Get top articles based on similarity
+        top_indices = ranker.get_top_articles(similarities, normalized_articles, TOP_K_ARTICLES, min_threshold)
+        top_articles = ranker.format_results(top_indices, similarities, normalized_articles)
+        # =====================================================
+        # 4. Bias Categorization
+        # =====================================================
+        # Extract outlet names from the TOP_K_ARTICLES
+        # In top_articles, the source is already extracted as a string
+        outlet_names = [article['source'] for article in top_articles]
+        unique_outlets = list(set(outlet_names))
+        print(f"Analyzing {len(unique_outlets)} unique news outlets for bias...")
+        # Analyze bias using Gemini - send just the outlet names, not the whole articles
+        bias_analysis = bias_analyzer.analyze_bias(query, unique_outlets, GEMINI_MODEL)
+        # =====================================================
+        # 5. Category Embeddings
+        # =====================================================
+        print("\n" + "=" * 80)
+        print("EMBEDDING VECTORS BY BIAS CATEGORY")
+        print("=" * 80)
+        # Create embedding vectors for each bias category
+        # 1. Group articles based on their outlet's bias category
+        # 2. Create an embedding vector for each category using ONLY article titles
+        # 3. Rank articles within each category by similarity to query
+        category_rankings = bias_analyzer.categorize_and_rank_by_bias(
+            query, normalized_articles, bias_analysis, ranker, min_threshold
+        )
+        # =====================================================
+        # 6. Top N Selection per Category
+        # =====================================================
+        # Get TOP_N_PER_CATEGORY from .env file (default: 5)
+        TOP_N_PER_CATEGORY = int(os.getenv('TOP_N_PER_CATEGORY', 5))
+        # Get total counts of articles per category before filtering
+        category_article_counts = {
+            category: len(articles)
+            for category, articles in category_rankings.items()
+            if category not in ["descriptions", "reasoning"]
+        }
+        # For each bias category, select the top N articles
+        # These are the most relevant articles within each bias perspective
+        filtered_category_rankings = {}
+        for category, articles in category_rankings.items():
+            # Skip non-category keys like "descriptions" or "reasoning"
+            if category in ["descriptions", "reasoning"]:
+                continue
+            filtered_category_rankings[category] = articles[:TOP_N_PER_CATEGORY]
+            # Only print if there are articles in this category
+            if len(filtered_category_rankings[category]) > 0:
+                print(f"\n===== Top {len(filtered_category_rankings[category])} articles from {category} category =====")
+                # Print detailed information about each selected article
+                for i, article in enumerate(filtered_category_rankings[category], 1):
+                    print(f"Article #{i}:")
+                    print(f"  Title: {article['title']}")
+                    print(f"  Source: {article['source']}")
+                    print(f"  Similarity Score: {article['similarity_score']:.4f}")
+                    print(f"  Rank: {article['rank']}")
+                    print(f"  URL: {article['url']}")
+                    print(f"  Published: {article['published_at']}")
+                    print("-" * 50)
+        # =====================================================
+        # 7. Summarization
+        # =====================================================
+        # Generate summary from articles in all categories
+        print("\nGenerating factual summary using top articles from all categories...")
+        # Pass the original bias_analysis to include the reasoning in the summary
+        # We need to add the reasoning to filtered_category_rankings since that's what gets passed to generate_summary
+        filtered_category_rankings["reasoning"] = bias_analysis.get("reasoning", "No reasoning provided")
+        # Call the bias_analyzer's generate_summary function with articles from all categories
+        summary = bias_analyzer.generate_summary(
+            query,
+            normalized_articles,
+            filtered_category_rankings,
+            GEMINI_MODEL
+        )
+        # Print the summary to terminal (already includes its own formatting)
+        print(summary)
+        # Prepare response with only the summary and reasoning
+        result = {
+            "query": query,
+            "summary": summary,
+            "reasoning": bias_analysis.get("reasoning", "No reasoning provided")
+        }
+        return jsonify(result)
+    @app.route('/api/health', methods=['GET'])
+    def health_check():
+        """API endpoint to check if the server is running."""
+        return jsonify({
+            "status": "ok",
+            "message": "Fake News Detection API is running"
+        })
+    # Get port from environment variable or use default 5000
+    port = int(os.getenv('PORT', 5000))
+    debug = os.getenv('DEBUG', 'false').lower() == 'true'
+    print(f"Starting Fake News Detection API server on port {port}...")
+    # Start the Flask server
+    app.run(host='0.0.0.0', port=port, debug=debug)
+if __name__ == "__main__":
+    main()

misinformationui/static/script.js ADDED Viewed

	@@ -0,0 +1,88 @@

+const chat = document.getElementById('chat');
+const input = document.getElementById('query');
+const sendBtn = document.getElementById('sendBtn');
+function addMessage(text, sender, isPre = false) {
+  const msg = document.createElement('div');
+  msg.classList.add('message', sender);
+  // Add special class for pre-formatted messages to style them properly
+  if (isPre) {
+    msg.classList.add('pre-formatted');
+    // For pre-formatted text (terminal output)
+    const pre = document.createElement('pre');
+    pre.textContent = text;
+    // No inline styles - all styling comes from CSS
+    msg.appendChild(pre);
+  } else {
+    msg.textContent = text;
+  }
+  chat.appendChild(msg);
+  // Smooth scroll to new message
+  setTimeout(() => {
+    msg.scrollIntoView({ behavior: 'smooth', block: 'end' });
+  }, 100);
+  return msg;
+}
+async function sendMessage() {
+  const query = input.value.trim();
+  if (!query) return;
+  const bg = document.getElementById('chat-background');
+  if (bg && !bg.classList.contains('blurred')) {
+    bg.classList.add('blurred');
+  }
+  addMessage(query, 'user');
+  input.value = '';
+  const loader = addMessage('Processing...', 'bot');
+  try {
+    const response = await fetch('/api/detect', {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({ query: query })
+    });
+    const data = await response.json();
+    loader.remove();
+    if (data && data.summary) {
+      // Display summary exactly as it comes from the backend
+      addMessage(data.summary, 'bot', true); // scrollable <pre> block
+    } else {
+      addMessage("Could not generate a summary.", 'bot');
+    }
+  } catch (e) {
+    loader.remove();
+    addMessage("Error checking news.", 'bot');
+  }
+}
+function formatBackendData(data) {
+  // If we have a summary, only display that
+  if (data && data.summary) {
+    if (typeof data.summary === 'string') {
+      return data.summary;
+    } else if (typeof data.summary === 'object' && data.summary.text) {
+      return data.summary.text;
+    } else {
+      return JSON.stringify(data.summary, null, 2);
+    }
+  }
+  // If no summary is available, return null so we can fall back to showing basic results
+  return null;
+}
+sendBtn.addEventListener('click', sendMessage);
+input.addEventListener('keypress', (e) => {
+  if (e.key === 'Enter') sendMessage();
+});

misinformationui/static/style.css ADDED Viewed

	@@ -0,0 +1,188 @@

+body {
+  font-family: Arial, sans-serif;
+  margin: 0;
+  padding: 0;
+  height: 100vh;
+  display: flex;
+  flex-direction: column;
+  overflow: hidden;
+}
+/* Blurred background image */
+body::before {
+  content: "";
+  position: fixed;
+  top: 0;
+  left: 0;
+  width: 100vw;
+  height: 100vh;
+  z-index: -1;
+  background: rgb(23, 23, 23);
+}
+.chat-container {
+  flex: 1;
+  display: flex;
+  flex-direction: column;
+  padding: 15px;
+  overflow-y: auto;
+  overflow-x: hidden; /* prevent horizontal scrolling */
+  margin-bottom: 70px;
+  background: transparent;
+  width: 100%;
+  max-width: 95%;     /* wider to accommodate terminal output */
+  margin-left: auto;   /* center align */
+  margin-right: auto;  /* center align */
+  scroll-behavior: smooth; /* Smooth scrolling */
+  height: calc(100vh - 70px); /* Full height minus input area */
+}
+.message {
+  width: fit-content;   /* shrink to text */
+  max-width: 100%;      /* allow full width for terminal output */
+  margin-bottom: 12px;
+  padding: 12px 15px;
+  border-radius: 15px;
+  line-height: 1.4;
+}
+/* Special styling for bot messages with pre-formatted text */
+.message.bot.pre-formatted {
+  width: 100%;          /* full width for terminal output */
+  max-width: 100%;      /* no width restriction */
+  white-space: pre-wrap; /* wrap text to prevent horizontal scroll */
+  overflow-wrap: break-word; /* break long words if needed */
+}
+.user {
+  align-self: flex-end;   /* right side */
+  background: #414141;
+  color: #fff;
+  border-bottom-right-radius: 5px;
+}
+.bot {
+  align-self: flex-start; /* left side */
+  background: transparent;
+  color: #ffffff;
+  border-bottom-left-radius: 5px;
+}
+/* Terminal output style - matching exactly what appears in the terminal */
+.message.bot pre {
+  font-family: monospace;
+  background-color: transparent; /* No background color */
+  color: inherit; /* Use the same text color as the parent */
+  padding: 0;
+  border: none;
+  width: 100%;
+  max-height: none; /* No height limit */
+  overflow-x: visible; /* No horizontal scrolling */
+  white-space: pre-wrap; /* Wrap text to prevent horizontal scrolling */
+  font-size: inherit;
+  line-height: 1.4;
+}
+.input-area {
+  position: fixed;
+  bottom: 1rem;
+  left: 50%;
+  transform: translateX(-50%);
+  width: 100%;
+  max-width: 95%; /* Match the width of chat container */
+  display: flex;
+  padding: 10px;
+  background: rgba(54, 54, 54, 0.7); /* make input area semi-transparent */
+  box-shadow: 0 -2px 5px rgba(0,0,0,0.05);
+  border-radius: 30px;
+  z-index: 10; /* Ensure input stays on top */
+}
+.input-area input {
+  flex: 1;
+  padding: 14px 18px;
+  border: 1px solid #1b1b1b;
+  box-shadow: #414141;
+  border-radius: 25px;
+  outline: none;
+  font-size: 1rem;
+}
+.input-area button {
+  margin-left: 10px;
+  padding: 0 20px;
+  border: none;
+  background: #1f1f1f;
+  color: white;
+  border-radius: 25px;
+  cursor: pointer;
+  font-size: 1rem;
+  transition: background 0.2s ease;
+}
+.input-area button:hover {
+  background: #6a6a6a;
+}
+.loader {
+  font-size: 0.9rem;
+  color: gray;
+  margin: 5px 0;
+}
+.chat-background{
+  position: fixed;
+  font-family: 'Courier New', Courier, monospace;
+  top: 40%;
+  left: 50%;
+  transform: translate(-50%, -50%);
+  font-weight: bold;
+  color: rgba(255, 255, 255, 0.8);  /* semi-transparent */
+  text-align: center;
+  z-index: 0;  /* below chat messages */
+  pointer-events: none; /* makes it "untouchable" */
+  transition: all 0.4s ease;
+}
+.chat-background{
+  display: inline-block;
+  text-align: left;
+}
+.chat-background #p1 {
+  font-size: 4rem;
+}
+.chat-background #p2 {
+  font-size: 3rem;
+}
+/* When blurred */
+.chat-background.blurred {
+  filter: blur(12px); /* strong blur */
+  opacity: 0.4; /* fade slightly for readability */
+  transition: all 0.4s ease;
+}
+@media (max-width: 600px) {
+  .message {
+    max-width: 85%;
+  }
+  .chat-container {
+    padding: 10px;
+    max-width: 100%;
+  }
+  .input-area {
+    max-width: 95%;
+    bottom: 0.5rem;
+  }
+  .chat-background #p1 {
+    font-size: 3rem;
+  }
+  .chat-background #p2 {
+    font-size: 2rem;
+  }
+}

misinformationui/static/test_server.py ADDED Viewed

File without changes