New 4 features

.gitignore

@@ -0,0 +1 @@
+tests.ipynb

frontend/src/components/ClusterCard.jsx

@@ -57,10 +57,24 @@ export default function ClusterCard({ cluster }) {
                                     className="block p-3 rounded-lg bg-slate-800/40 hover:bg-slate-800 transition-colors group"
                                 >
                                     <div className="flex justify-between items-start gap-3">
-                                        <h4 className="text-sm text-slate-300 font-medium line-clamp-2 group-hover:text-blue-300">
-                                            {article.title}
-                                        </h4>
-                                        <ExternalLink className="w-4 h-4 text-slate-500 shrink-0 group-hover:text-blue-400" />
+                                        <div className="flex-1">
+                                            <h4 className="text-sm text-slate-300 font-medium line-clamp-2 group-hover:text-blue-300 mb-1">
+                                                {article.title}
+                                            </h4>
+                                            <div className="flex flex-wrap gap-2">
+                                                {article.is_clickbait && (
+                                                    <span className="text-[10px] uppercase font-bold text-red-400 bg-red-900/30 border border-red-900/50 px-1.5 py-0.5 rounded">
+                                                        Clickbait
+                                                    </span>
+                                                )}
+                                                {article.is_sensationalist && (
+                                                    <span className="text-[10px] uppercase font-bold text-amber-400 bg-amber-900/30 border border-amber-900/50 px-1.5 py-0.5 rounded">
+                                                        Sensationalist
+                                                    </span>
+                                                )}
+                                            </div>
+                                        </div>
+                                        <ExternalLink className="w-4 h-4 text-slate-500 shrink-0 group-hover:text-blue-400 mt-1" />
                                     </div>
                                     <div className="mt-2 flex items-center gap-2 text-xs text-slate-500">
                                         <span className="font-semibold text-slate-400">{article.newspaper}</span>

main.py

@@ -149,7 +149,7 @@ def get_clusters():
     # Usually newest first is better for news.
     # The user asked for "3 days consecutive". Just the set.
     # Let's order them chronologically descending.
-    target_dates = sorted(target_dates, reverse=True)
+    target_dates = sorted(target_dates, reverse=False)
 
     for day_dt in target_dates:
         # Filter by datetime
@@ -171,7 +171,9 @@ def get_clusters():
                     "title": row.get('title', 'No Title'),
                     "newspaper": row.get('newspaper', 'Unknown'),
                     "url": row.get('article_url', '#'),
-                    "newspaper_url": row.get('newspaper_url', '#')
+                    "newspaper_url": row.get('newspaper_url', '#'),
+                    "is_clickbait": bool(row.get('is_clickbait', False)) if pd.notna(row.get('is_clickbait')) else None,
+                    "is_sensationalist": bool(row.get('is_sensationalist', False)) if pd.notna(row.get('is_sensationalist')) else None
                 })
             
             # Size calc
@@ -219,12 +221,15 @@ def analyze_article(request: AnalyzeRequest):
         with torch.no_grad():
             cb_outputs = models["cb_model"](**cb_inputs)
         cb_probs = torch.softmax(cb_outputs.logits, dim=1)
-        # Assuming label 0 is NO, 1 is YES (need to verify mapping usually, usually 1 is positive class)
-        # Checking model config locally or on HF would be ideal. Let's assume standard mapping:
-        # For clickbait: likely labels are "no_clickbait", "clickbait".
-        # We can check specific model details if needed. For now assume:
-        # Label 1 is clickbait.
-        is_clickbait = torch.argmax(cb_probs).item() == 1
+        
+        # Feature 3 & 4: Inverted Clickbait Logic & Thresholds
+        # Clickbait Model: Label 0 is CLICKBAIT (based on user info)
+        # Check for decision_threshold in config, default to 0.5
+        cb_threshold = getattr(models["cb_model"].config, 'decision_threshold', 0.5)
+        
+        # Prob of class 0 (Clickbait)
+        prob_clickbait = cb_probs[0][0].item()
+        is_clickbait = prob_clickbait >= cb_threshold
         
         # Sensationalism: using title + text
         sens_input_text = f"TITULAR: {title}\nCUERPO: {text}"
@@ -232,15 +237,22 @@ def analyze_article(request: AnalyzeRequest):
         with torch.no_grad():
             sens_outputs = models["sens_model"](**sens_inputs)
         sens_probs = torch.softmax(sens_outputs.logits, dim=1)
-        is_sensationalist = torch.argmax(sens_probs).item() == 1
+        
+        # Sensationalism Model: Standard (Label 1 is Sensationalist)
+        # Check for decision_threshold in config, default to 0.5
+        sens_threshold = getattr(models["sens_model"].config, 'decision_threshold', 0.5)
+        
+        # Prob of class 1 (Sensationalist)
+        prob_sensationalist = sens_probs[0][1].item()
+        is_sensationalist = prob_sensationalist >= sens_threshold
         
         return {
             "title": title,
             "text_snippet": text[:200] + "..." if len(text) > 200 else text,
             "is_clickbait": is_clickbait,
             "is_sensationalist": is_sensationalist,
-            "clickbait_conf": float(cb_probs[0][1]),
-            "sensationalist_conf": float(sens_probs[0][1])
+            "clickbait_conf": float(prob_clickbait), # Sending confidence of being clickbait
+            "sensationalist_conf": float(prob_sensationalist)
         }
         
     except Exception as e: