Feat: Claude-Keywords für RSS-Suche, Jahreszahlen-Filter, strikteres Matching

- rss_researcher liefert jetzt Keywords zurück, die direkt für RSS-Suche genutzt werden
- Neue _clean_search_words() filtert rein-numerische Begriffe (Jahreszahlen etc.)
- Matching-Schwelle aufgerundet: bei 3 Keywords müssen mindestens 2 matchen

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/agents/orchestrator.py

@@ -579,14 +579,15 @@ class AgentOrchestrator:
                 all_feeds = await get_feeds_with_metadata(tenant_id=tenant_id)
 
                 feed_usage = None
+                keywords = None
                 if len(all_feeds) > 20:
-                    selected_feeds, feed_usage = await rss_researcher.select_relevant_feeds(
+                    selected_feeds, keywords, feed_usage = await rss_researcher.select_relevant_feeds(
                         title, description, international, all_feeds
                     )
                     logger.info(f"Feed-Selektion: {len(selected_feeds)} von {len(all_feeds)} Feeds ausgewählt")
-                    articles = await rss_parser.search_feeds_selective(title, selected_feeds)
+                    articles = await rss_parser.search_feeds_selective(title, selected_feeds, keywords=keywords)
                 else:
-                    articles = await rss_parser.search_feeds(title, international=international, tenant_id=tenant_id)
+                    articles = await rss_parser.search_feeds(title, international=international, tenant_id=tenant_id, keywords=keywords)
 
                 logger.info(f"RSS: {len(articles)} relevante Artikel gefunden (international={international})")
                 return articles, feed_usage

src/feeds/rss_parser.py

@@ -20,21 +20,33 @@ class RSSParser:
         "über", "unter", "ist", "sind", "hat", "the", "and", "for", "with", "from",
     }
 
-    async def search_feeds(self, search_term: str, international: bool = True, tenant_id: int = None) -> list[dict]:
+    @staticmethod
+    def _clean_search_words(words: list[str]) -> list[str]:
+        """Entfernt rein-numerische Wörter (Jahreszahlen etc.) aus Suchbegriffen."""
+        cleaned = [w for w in words if not w.isdigit()]
+        return cleaned if cleaned else words
+
+    async def search_feeds(self, search_term: str, international: bool = True, tenant_id: int = None, keywords: list[str] | None = None) -> list[dict]:
         """Durchsucht RSS-Feeds nach einem Suchbegriff.
 
         Args:
             search_term: Suchbegriff
             international: Wenn False, nur deutsche Feeds + Behoerden (keine internationalen)
             tenant_id: Optionale Org-ID fuer tenant-spezifische Quellen
+            keywords: Optionale Claude-generierte Keywords (bevorzugt gegenüber Title-Split)
         """
         all_articles = []
-        search_words = [
-            w for w in search_term.lower().split()
-            if w not in self.STOP_WORDS and len(w) >= 3
-        ]
-        if not search_words:
-            search_words = search_term.lower().split()[:2]
+        if keywords:
+            search_words = [w.lower().strip() for w in keywords if w.strip()]
+            logger.info(f"RSS-Suche mit Claude-Keywords: {search_words}")
+        else:
+            search_words = [
+                w for w in search_term.lower().split()
+                if w not in self.STOP_WORDS and len(w) >= 3
+            ]
+            if not search_words:
+                search_words = search_term.lower().split()[:2]
+            search_words = self._clean_search_words(search_words)
 
         rss_feeds = await self._get_rss_feeds(tenant_id=tenant_id)
 
@@ -62,20 +74,26 @@ class RSSParser:
         all_articles = self._apply_domain_cap(all_articles)
         return all_articles
 
-    async def search_feeds_selective(self, search_term: str, selected_feeds: list[dict]) -> list[dict]:
+    async def search_feeds_selective(self, search_term: str, selected_feeds: list[dict], keywords: list[str] | None = None) -> list[dict]:
         """Durchsucht nur die übergebenen Feeds (vorselektiert durch Claude).
 
         Args:
             search_term: Suchbegriff
             selected_feeds: Liste von Feed-Dicts mit mindestens {"name", "url"}
+            keywords: Optionale Claude-generierte Keywords (bevorzugt gegenüber Title-Split)
         """
         all_articles = []
-        search_words = [
-            w for w in search_term.lower().split()
-            if w not in self.STOP_WORDS and len(w) >= 3
-        ]
-        if not search_words:
-            search_words = search_term.lower().split()[:2]
+        if keywords:
+            search_words = [w.lower().strip() for w in keywords if w.strip()]
+            logger.info(f"RSS-Selektiv mit Claude-Keywords: {search_words}")
+        else:
+            search_words = [
+                w for w in search_term.lower().split()
+                if w not in self.STOP_WORDS and len(w) >= 3
+            ]
+            if not search_words:
+                search_words = search_term.lower().split()[:2]
+            search_words = self._clean_search_words(search_words)
 
         tasks = []
         for feed_config in selected_feeds:
@@ -124,8 +142,8 @@ class RSSParser:
                 summary = entry.get("summary", "")
                 text = f"{title} {summary}".lower()
 
-                # Flexibles Keyword-Matching: mindestens die Hälfte der Suchworte muss vorkommen
-                min_matches = max(1, len(search_words) // 2)
+                # Flexibles Keyword-Matching: mindestens die Hälfte der Suchworte muss vorkommen (aufgerundet)
+                min_matches = max(1, (len(search_words) + 1) // 2)
                 match_count = sum(1 for word in search_words if word in text)
 
                 if match_count >= min_matches: