Continuous Relevancy Training

Discovery Continuous Relevancy Training can learn from user behavior automatically, significantly reducing the effort required to improve the relevancy ranking of results. It uses interactions from users to learn how to surface the most relevant results. It can learn from interactions like clicks to determine which results are most valuable for users and uncover the most important signals for future queries. Continuous Relevancy Training reranks documents to surface the most relevant information at the top of the results. You can also use it for the following purposes:

• Help improve the relevance of results for support agents, based on the results they click on
• Display more relevant results in a customer chatbot based on the long-tail results selected
• Provide experts with more relevant responses, based on the results they explore

To setup Continuous Relevancy Training:

• Continuous Relevancy Training can only be enabled at the environment level. Queries must use either the /api/v1/environment/{environment_id}/query and /api/v1/environments/{environment_id}/query endpoints to query one or more collections.
• Because Discovery events (clicks) are used to create the training data needed, first integrate event tracking. See Usage monitoring for details.
• A minimum of 1000 natural language queries with an associated click event are required for Continuous Relevancy Training to start. Events and logs are retained for 30 days across your instance so the 1000 clicks must be collected during that time period.
• Continuous Relevancy Training is available for Advanced plans size Small or larger, and Premium plans. It is not available for Lite plans.
• The number of collections for Continuous Relevancy Training is limited to 5. Because Continuous Relevancy Training works across multiple collections, both query and training times might be extended.
• Continuous Relevancy Training only occurs on top-level fields, and there are limits to how many fields can be used in the training process. In cases where there are more than 10 top level fields, training is more likely to encounter errors.

To use Continuous Relevancy Training:

Once trained, Continuous Relevancy Training is used to influence the results of a natural_language_query when using an environment-level query.

Continuous Relevancy Training can be used at query time by running a multi-collection natural_language_query across all collections in your environment. See Querying multiple collections for instructions.

Continuous Relevancy Training does not affect queries made to a trained or untrained collection-level (/v1/environments/{environment_id}/collections/{collection_id}/query) query call.

Tracking status:

The status of Continuous Relevancy Training can be viewed from the /api/v1/environments endpoint. See the API reference. The status provides information about the state of the training operations and informs you if Continuous Relevancy Training is active:

"search_status" : [
        {
            "scope" : "environment",
            "status" : "NO_DATA",
            "status_description" : "Discovery is employing a default strategy for document search natural_language_query. Enable query and event logging so we can initiate relevancy training to improve search accuracy.”
        }
    ]