v5

latestOpenAPI 3.1.02026-08-025631,1012.8 MB

Retriever Evaluations

Generate evaluation dataset from interactions

Auto-generate a ground truth evaluation dataset by mining user interactions for this retriever. Queries with sufficient positive interactions (click, purchase, bookmark, etc.) are used to infer relevance labels, creating a dataset suitable for offline evaluation.

post/v1/retrievers/{retriever_id}/evaluations/generate-from-interactions

Path parameters

retriever_idstring required

Request body

dataset_namestring nullable

Name for the generated dataset. If omitted, defaults to 'auto_{retriever_id}_{date}'.

min_interactionsinteger

Minimum number of positive interactions a query must have to be included in the dataset. Queries with fewer interactions are excluded as low-confidence ground truth.

lookback_daysinteger

How many days of interaction history to consider.

positive_interaction_typesstring[] nullable

Interaction types considered positive signals. Defaults to: click, purchase, positive_feedback, bookmark, share.

Response

Successful Response

dataset_idstring required

Unique dataset identifier

dataset_namestring required

Human-readable dataset name

descriptionstring nullable

Dataset description

created_atstring date-time required

When dataset was created

updated_atstring date-time required

Last update timestamp

namespace_idstring required

Namespace this dataset belongs to

internal_idstring required

Internal organization ID

query_countinteger required

Number of queries in dataset

metadataobject nullable

Additional metadata (e.g., labeling instructions, version info)