yujunjun/LightRAG
1
0
Fork 0
mirror of https://github.com/HKUDS/LightRAG.git synced 2025-11-09 06:13:47 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix: Deduplicate entities and relationships in a single chunk with multiple gleaning results during KG rebuild

Browse Source
This commit is contained in:
yangdx 2025-07-03 13:47:52 +08:00
parent 70e154b0aa
commit 6b6d14bc3a
1 changed files with 15 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
lightrag/operate.py
View File

@ -284,6 +284,7 @@ async def _rebuild_knowledge_from_chunks(
pipeline_status["history_messages"].append(status_message)
# Get cached extraction results for these chunks using storage
# cached_results： chunk_id -> [list of extraction result from LLM cache sorted by created_at]
cached_results = await _get_cached_extraction_results(
llm_response_cache,
all_referenced_chunk_ids,
@ -309,6 +310,7 @@ async def _rebuild_knowledge_from_chunks(
chunk_entities[chunk_id] = defaultdict(list)
chunk_relationships[chunk_id] = defaultdict(list)
# process multiple LLM extraction results for a single chunk_id
for extraction_result in extraction_results:
entities, relationships = await _parse_extraction_result(
text_chunks_storage=text_chunks_storage,
@ -317,9 +319,20 @@ async def _rebuild_knowledge_from_chunks(
)
# Merge entities and relationships from this extraction result
# Only keep the first occurrence of each entity_name in the same chunk_id
for entity_name, entity_list in entities.items():
if (
entity_name not in chunk_entities[chunk_id]
or len(chunk_entities[chunk_id][entity_name]) == 0
):
chunk_entities[chunk_id][entity_name].extend(entity_list)
# Only keep the first occurrence of each rel_key in the same chunk_id
for rel_key, rel_list in relationships.items():
if (
rel_key not in chunk_relationships[chunk_id]
or len(chunk_relationships[chunk_id][rel_key]) == 0
):
chunk_relationships[chunk_id][rel_key].extend(rel_list)
except Exception as e: