neu graph explorer service

This commit is contained in:
Lars 2025-12-14 08:27:15 +01:00
parent dc16bbf8a4
commit 772a202d6e

View File

@ -100,6 +100,7 @@ EDGE_COLORS = {
"derived_from": "#ff9ff3",# Pink
"references": "#bdc3c7", # Grey
"belongs_to": "#2e86de" # Dark Blue
"contributes_to": "#1dd1a1"
}
# --- HELPER FUNCTIONS ---
@ -245,27 +246,19 @@ class GraphExplorerService:
self.notes_col = f"{prefix}_notes"
self.chunks_col = f"{prefix}_chunks"
self.edges_col = f"{prefix}_edges"
self._note_cache = {} # Simple in-memory cache for the session
self._note_cache = {}
def get_ego_graph(self, center_note_id: str):
"""
Bidirektionaler Ego-Graph:
1. Lädt Center Node.
2. Findet OUTGOING Edges (Source = Chunk von Center).
3. Findet INCOMING Edges (Target = Chunk von Center ODER Target = Titel von Center).
4. Dedupliziert auf Notiz-Ebene.
"""
nodes_dict = {} # note_id -> Node Object
unique_edges = {} # (source_note_id, target_note_id) -> Edge Data
nodes_dict = {}
unique_edges = {}
# 1. Zentrale Note laden
center_note = self._fetch_note_cached(center_note_id)
if not center_note: return [], []
self._add_node_to_dict(nodes_dict, center_note, is_center=True)
center_title = center_note.get("title")
# 2. Chunks der Note finden (für Edge-Suche)
# Chunks laden
scroll_filter = models.Filter(
must=[models.FieldCondition(key="note_id", match=models.MatchValue(value=center_note_id))]
)
@ -276,7 +269,7 @@ class GraphExplorerService:
raw_edges = []
# 3. OUTGOING EDGES Suche
# 1. OUTGOING: Source ist einer unserer Chunks
if center_chunk_ids:
out_filter = models.Filter(
must=[models.FieldCondition(key="source_id", match=models.MatchAny(any=center_chunk_ids))]
@ -286,28 +279,26 @@ class GraphExplorerService:
)
raw_edges.extend(res_out)
# 4. INCOMING EDGES Suche
# Case A: Target ist einer unserer Chunks
# 2. INCOMING: Target ist Chunk, Titel oder exakte Note-ID
# Hinweis: Target mit #Section (z.B. 'note#header') kann via Keyword-Index schwer gefunden werden,
# wenn wir den Header-Teil nicht kennen.
must_conditions = []
if center_chunk_ids:
in_chunk_filter = models.Filter(
must=[models.FieldCondition(key="target_id", match=models.MatchAny(any=center_chunk_ids))]
)
res_in_c, _ = self.client.scroll(
collection_name=self.edges_col, scroll_filter=in_chunk_filter, limit=100, with_payload=True
)
raw_edges.extend(res_in_c)
# Case B: Target ist unser Titel (Wikilinks)
must_conditions.append(models.FieldCondition(key="target_id", match=models.MatchAny(any=center_chunk_ids)))
if center_title:
in_title_filter = models.Filter(
must=[models.FieldCondition(key="target_id", match=models.MatchValue(value=center_title))]
)
res_in_t, _ = self.client.scroll(
collection_name=self.edges_col, scroll_filter=in_title_filter, limit=50, with_payload=True
)
raw_edges.extend(res_in_t)
must_conditions.append(models.FieldCondition(key="target_id", match=models.MatchValue(value=center_title)))
# 5. Kanten verarbeiten und auflösen
# NEU: Auch exakte Note-ID als Target prüfen
must_conditions.append(models.FieldCondition(key="target_id", match=models.MatchValue(value=center_note_id)))
if must_conditions:
in_filter = models.Filter(should=must_conditions) # 'should' wirkt wie OR
res_in, _ = self.client.scroll(
collection_name=self.edges_col, scroll_filter=in_filter, limit=100, with_payload=True
)
raw_edges.extend(res_in)
# Verarbeitung
for record in raw_edges:
payload = record.payload
@ -316,29 +307,23 @@ class GraphExplorerService:
kind = payload.get("kind", "related_to")
provenance = payload.get("provenance", "explicit")
# Resolve Source Note
src_note = self._resolve_note_from_ref(src_ref)
# Resolve Target Note
tgt_note = self._resolve_note_from_ref(tgt_ref)
if src_note and tgt_note:
src_id = src_note['note_id']
tgt_id = tgt_note['note_id']
# Keine Self-Loops und valide Verbindung
if src_id != tgt_id:
# Nodes hinzufügen (falls noch nicht da)
self._add_node_to_dict(nodes_dict, src_note)
self._add_node_to_dict(nodes_dict, tgt_note)
# Deduplizierung: Wir behalten die "stärkste" Kante
# Wenn bereits eine explizite Kante existiert, überschreiben wir sie nicht mit einer AI-Kante
key = (src_id, tgt_id)
existing = unique_edges.get(key)
is_current_explicit = (provenance == "explicit" or provenance == "rule")
should_update = True
if existing:
is_existing_explicit = (existing['provenance'] == "explicit" or existing['provenance'] == "rule")
if is_existing_explicit and not is_current_explicit:
@ -346,39 +331,25 @@ class GraphExplorerService:
if should_update:
unique_edges[key] = {
"source": src_id,
"target": tgt_id,
"kind": kind,
"provenance": provenance
"source": src_id, "target": tgt_id, "kind": kind, "provenance": provenance
}
# 6. Agraph Edge Objekte erstellen
final_edges = []
for (src, tgt), data in unique_edges.items():
kind = data['kind']
prov = data['provenance']
color = EDGE_COLORS.get(kind, "#bdc3c7")
is_smart = (prov != "explicit" and prov != "rule")
label = f"{kind}"
# AI Edges gestrichelt
dashes = is_smart
final_edges.append(Edge(
source=src,
target=tgt,
label=label,
color=color,
dashes=dashes,
title=f"Provenance: {prov}, Type: {kind}"
source=src, target=tgt, label=kind, color=color, dashes=is_smart,
title=f"Provenance: {prov}\nType: {kind}"
))
return list(nodes_dict.values()), final_edges
def _fetch_note_cached(self, note_id):
if note_id in self._note_cache: return self._note_cache[note_id]
res, _ = self.client.scroll(
collection_name=self.notes_col,
scroll_filter=models.Filter(must=[models.FieldCondition(key="note_id", match=models.MatchValue(value=note_id))]),
@ -390,54 +361,50 @@ class GraphExplorerService:
return None
def _resolve_note_from_ref(self, ref_str):
"""Löst eine ID (Chunk) oder einen String (Titel) zu einer Note Payload auf."""
if not ref_str: return None
# Fall 1: Chunk ID (enthält '#')
# Fall A: Chunk ID (Format: note_id#cXX)
if "#" in ref_str:
# Wir könnten den Chunk holen, aber effizienter ist es, die note_id aus dem Chunk-String zu parsen,
# WENN das Format strikt 'note_id#cXX' ist. Um sicher zu gehen, fragen wir Qdrant.
# 1. Versuch: Echte Chunk ID in DB suchen
try:
res = self.client.retrieve(collection_name=self.chunks_col, ids=[ref_str], with_payload=True)
if res:
parent_id = res[0].payload.get("note_id")
return self._fetch_note_cached(parent_id)
except: pass # Falls ID nicht existiert
except: pass
# Fall 2: Vermutlich ein Titel (Wikilink) oder Note ID
# Versuch als Note ID
# 2. Versuch (NEU): Es ist ein Link auf eine Section (z.B. "note-id#Header")
# Wir entfernen den Hash-Teil und suchen die Basis-Notiz
possible_note_id = ref_str.split("#")[0]
note_by_id = self._fetch_note_cached(possible_note_id)
if note_by_id: return note_by_id
# Fall B: Es ist direkt die Note ID
note_by_id = self._fetch_note_cached(ref_str)
if note_by_id: return note_by_id
# Versuch als Titel
# Fall C: Es ist der Titel (Wikilink)
res, _ = self.client.scroll(
collection_name=self.notes_col,
scroll_filter=models.Filter(must=[models.FieldCondition(key="title", match=models.MatchValue(value=ref_str))]),
limit=1, with_payload=True
)
if res:
payload = res[0].payload
self._note_cache[payload['note_id']] = payload
return payload
p = res[0].payload
self._note_cache[p['note_id']] = p
return p
return None
def _add_node_to_dict(self, node_dict, note_payload, is_center=False):
nid = note_payload.get("note_id")
if nid in node_dict: return
ntype = note_payload.get("type", "default")
color = GRAPH_COLORS.get(ntype, GRAPH_COLORS["default"])
size = 40 if is_center else 20
size = 35 if is_center else 20
node_dict[nid] = Node(
id=nid,
label=note_payload.get("title", nid),
size=size,
color=color,
shape="dot" if not is_center else "diamond",
title=f"Type: {ntype}\nTags: {note_payload.get('tags')}",
font={'color': 'black', 'face': 'arial'}
id=nid, label=note_payload.get("title", nid), size=size, color=color, shape="dot",
title=f"Type: {ntype}\nTags: {note_payload.get('tags')}", font={'color': 'black'}
)
# Init Graph Service