update

src/recommendation/recommender.py

@@ -49,22 +49,29 @@ class RecommendationService:
     # Step 2: keywords from latest note per category
     # ──────────────────────────────────────────────
     def _latest_note_per_category(
-        self, notes: List[Dict], categories: List[str]
-    ) -> Dict[str, Dict]:
+        self, notes: List[Dict], categories: List[str], top_k: int = 3
+    ) -> Dict[str, List[Dict]]:
         """
-        For each target category return the note with the largest createdAt value.
-        Notes are pre-fetched so no extra DB calls needed.
+        Return top K latest notes per category instead of a single note.
         """
-        latest: Dict[str, Dict] = {}
+
+        grouped: Dict[str, List[Dict]] = {c: [] for c in categories}
+
         for note in notes:
             cat = note.get("category")
             cats = cat if isinstance(cat, list) else [cat] if cat else []
             ts = note.get("createdAt", 0)
+
             for c in cats:
                 if c in categories:
-                    if c not in latest or ts > latest[c].get("createdAt", 0):
-                        latest[c] = note
-        return latest
+                    grouped[c].append(note)
+
+        # sort each category by latest first
+        for c in grouped:
+            grouped[c].sort(key=lambda x: x.get("createdAt", 0), reverse=True)
+            grouped[c] = grouped[c][:top_k]
+
+        return grouped
 
     async def _extract_keywords_with_claude(
         self, note: Dict, category: str
@@ -213,23 +220,39 @@ class RecommendationService:
         # ── Step 2: keywords via Claude ──────────
         latest_notes = self._latest_note_per_category(notes, top_categories)
 
-        keyword_tasks = [
-            self._extract_keywords_with_claude(latest_notes[cat], cat)
-            for cat in top_categories
-            if cat in latest_notes
-        ]
+        keyword_tasks = []
+
+        for cat in top_categories:
+            if cat not in latest_notes:
+                continue
+
+            notes_list = latest_notes[cat]
+
+            # نجمع كل النوتس في نص واحد
+            combined_content = "\n".join(
+                n.get("content", "")[:500]
+                for n in notes_list
+            )
+
+            fake_note = {
+                "content": combined_content
+            }
+
+            keyword_tasks.append(
+                self._extract_keywords_with_claude(fake_note, cat)
+            )
         keyword_results = await asyncio.gather(*keyword_tasks)
 
-        category_keywords: Dict[str, List[str]] = {
-            cat: kws
-            for cat, kws in zip(top_categories, keyword_results)
-        }
+        category_keywords = {}
+
+        for cat, kws in zip(latest_notes.keys(), keyword_results):
+            category_keywords[cat] = kws
 
         # ── Step 3: build query & recommend ──────
         all_videos = []
 
         for category, keywords in category_keywords.items():
-            query = " ".join(keywords[:3])
+            query = " ".join(keywords[:3]) +" tutorial explanation"
 
             logger.info(f"🎯 Searching category: {category} | Query: {query}")