Adjusting the relevancy score filter level

You can either apply restrictive filtering or turn off relevancy score filtering for document chunks that the vector database returns.

About this task

When a natural language query is submitted to the IBM® Content Assistant, the Content Platform Engine queries a vector database to find document chunks that are relevant to a user question.

The results of this query are ordered by a relevancy score, which reflects how closely the document chunks match the input query. You can specify a numerical threshold for this relevancy score to filter out unrelated document chunks. You can adjust this numerical threshold by editing the GenAI configuration object and specifying a relevancy score filter level as an advanced configuration option.

Procedure

  1. In the Administration Console for Content Platform Engine (ACCE), go to the object store where you want to adjust the relevancy score filter level for the Content Assistant.
  2. In the left-pane, go to Search.
  3. Click New Object Store Search.
  4. In Simple View, specify the Class as Gen AI Configuration and click Run.
    The search results display a single Gen AI Configuration object.
  5. From the search results, click the object ID to access the properties for the Gen AI Configuration object.
  6. In the Properties tab, specify the relevancy score filter level in the Gen AI Advanced Options field.

    Specify the relevancy score filter level as a JSON value. The value of relevancy score must be a value between 0.0 and 1.0. The default value is 0.5. Enter a value of 0.0 to turn off relevancy score filtering. To allow less relevant document chunks to be used in your GenAI query prompts, enter a value less than 0.5. To apply more restrictive filtering of document chunks by relevancy score, enter a value greater than 0.5. A value of 1.0 provides the most restrictive filtering.

    For example, if you want to set a relevancy score of 0.3, enter the value as {"RelevanceFilterLevel":"0.3"}.
  7. Save your changes.