InvalidArgument: 400 when using 'keyword_similarity_score' in Discovery Engine custom ranking #14550
Description
Determine this is the right repository
- I determined this is the correct repository in which to report this bug.
Summary of the issue
I'm trying to use the SearchServiceClient within Discovery Engine to perform a search with a custom ranking expression. According to the documentation, I should be able to use the keyword_similarity_score signal, which is described as implementing the Best Match 25 (BM25) ranking function for keyword matching. However, when I execute a search with this signal, the API returns an InvalidArgument: 400 error.
I tryed discoveryengine, discoveryengine_v1, discoveryengine_v1alpha and discoveryengine_v1beta
Is this an issue with the documentation or a bug in the API?
- Is the keyword_similarity_score signal no longer supported?
- Has it been officially replaced by topicality_rank * -1.0?
If so, the documentation should be updated to reflect this change and clarify the correct way to implement BM25-based ranking. Any guidance on how topicality_rank * -1.0 achieves the same effect as BM25 would also be highly appreciated.
API client name and version
google-cloud-discoveryengine v0.13.12
Reproduction steps: code
from google.cloud import discoveryengine_v1 as discoveryengine
serving_config =f"projects/{project_id}/locations/{datastore_location}/collections/default_collection/engines/{app_id}/servingConfigs/default_search"
search_request = discoveryengine.SearchRequest(
ranking_expression_backend=discoveryengine.SearchRequest.RankingExpressionBackend.RANK_BY_FORMULA,
ranking_expression="keyword_similarity_score",
query=query,
serving_config=serving_config,
)
client = discoveryengine.SearchServiceClient(
client_options = ClientOptions()
)
response = client.search(search_request)
Reproduction steps: supporting files
Reproduction steps: actual results
The API call fails with the following error:
InvalidArgument: 400 Invalid signal: keyword_similarity_score. Please use topicality_rank * -1.0 instead.
Reproduction steps: expected results
I expected the search to execute successfully, with the results ranked based on the keyword_similarity_score. The documentation implies that this signal is a valid and supported option for custom ranking expressions.
OS & version + platform
No response
Python environment
3.12
Python dependencies
No response
Additional context
No response