Open LevickCG opened 2 days ago
After further investigation, I identified that the root cause of this issue is due to the mismatch of uuid to human readable id in the search process.
I plan to work on a fix and submit a pull request. I’d appreciate any feedback or guidance from the maintainers to ensure my approach aligns with the project’s design principles.
Do you need to file an issue?
Describe the bug
I’ve been exploring the codebase for GraphRAG and recently noticed that the community reports used for query augmentation in local search appear empty.
In consequnce, we see the
community_context_data
is empty.For the final response we see the lack of reports info.
This leads to missing community structure information in local search, which seems to degrade GraphRAG's performance and creates a discrepancy between the code implementation and the paper.
Steps to reproduce
You can specify the raw text on your own.
your_path_to_graphrag/graphrag/graphrag/graphrag/cli/
add the codes below to debug_query.py, it will launch a local search according to your query.
Add
import pdb;pdb.set_trace()
tographrag/query/structured_search/local_search/mixed_context.py:254
run the code and print debug info
run the code and it will stop at
mixed_context.py:254
, print the community information and you'll see it's empty.Expected Behavior
1.The selected_communities should not be empty.
2.Accordingly, community context should not be empty.
3.For the local search response, it should show the data source with
entity
,relationship
,report
(nowreport
is missing).GraphRAG Config Used
Logs and screenshots
See images provided.
Additional Information