feat: implement aggressive temporal weighting and 14-day default window for news relevance

src/core/use_cases/rag_chat_use_case.py

@@ -271,13 +271,18 @@ JSON:"""
         # ── Regex fallback for temporal + source ─────────────────────────────
         query_lower = query.lower()
 
-        # Temporal fallback
-        if any(w in query_lower for w in ("today", "tonight", "now", "breaking", "just")):
+        # Temporal fallback - more aggressive keywords
+        if any(w in query_lower for w in ("today", "tonight", "now", "breaking", "latest", "current", "update")):
             result["days_back"] = 1
         elif "yesterday" in query_lower:
             result["days_back"] = 2
-        elif any(w in query_lower for w in ("this week", "recently", "past few days")):
-            result["days_back"] = 7
+        elif any(w in query_lower for w in ("this week", "recently", "past few days", "days ago")):
+            # Check for "X days ago"
+            days_match = re.search(r'(\d+)\s+days?\s+ago', query_lower)
+            if days_match:
+                result["days_back"] = int(days_match.group(1)) + 1
+            else:
+                result["days_back"] = 7
         elif "this month" in query_lower:
             result["days_back"] = 30
 
@@ -330,7 +335,12 @@ JSON:"""
                     try:
                         pub_date = datetime.fromisoformat(pub_at.replace("Z", "+00:00"))
                         days_old = (now.replace(tzinfo=None) - pub_date.replace(tzinfo=None)).days
-                        score_multiplier = max(0.5, 1.0 - (days_old / 60))
+                        # Aggressive news freshness multiplier
+                        if days_old <= 0: score_multiplier = 1.05   # Boost today
+                        elif days_old == 1: score_multiplier = 1.0  # Normal yesterday
+                        elif days_old < 7: score_multiplier = 0.85  # Slight penalty this week
+                        elif days_old < 30: score_multiplier = 0.6  # Heavy penalty this month
+                        else: score_multiplier = 0.3                 # Archive data
                     except:
                         pass
 
@@ -408,6 +418,11 @@ JSON:"""
         # Sanitize placeholder filter values from API
         actual_source_filter = None if source_filter == "string" else source_filter
         actual_language_filter = None if (not language_filter or language_filter == "string") else language_filter
+        
+        # Default temporal window (last 14 days) if not specified — avoids ancient data
+        if days_back is None:
+            days_back = 14
+            logger.info(f"[RAG] No temporal filter provided — defaulting to 14 days for relevance")
 
         # ── HYBRID SEARCH INTEGRATION ──────────────────────────────────────────
         # Check if hybrid search is enabled and orchestrator is available