Configuring Coveo Machine Learning Models

All Coveo™ for ServiceNow components automatically log Coveo Usage Analytics (Coveo UA) events. Once sufficient usage analytics data has been gathered, this data can be leveraged by Coveo Machine Learning (Coveo ML) models to provide highly relevant AI-powered recommendations.

This article explains how a ServiceNow instance administrator or developer who has access to the Coveo Cloud organization linked to their instance can configure their Coveo ML models.

To configure your Coveo ML models

In your Coveo Cloud organization:

  1. Create a new query pipeline without a condition (see Adding and Managing Query Pipelines - Create a Query Pipeline).

    All Coveo ML models must be associated with a query pipeline. While nothing prevents you from associating your Coveo ML models to your default query pipeline, it is a good practice to rather associate them to a distinct query pipeline so that you can measure the impact of Coveo ML through A/B testing before allowing all queries to be processed by your Coveo ML models.

  2. Configure an Automatic Relevance Tuning (ART) model (see Add a Model).

    An ART model uses Intelligent Term Detection (ITD) to automatically refine queries with additional contextual information (e.g., important keywords from a large textual case description). It also ensures that items deemed highly relevant are included in the query result set and have a fairly high ranking score value, even if those items do not actually match the original query.

  3. Associate the ART model with the query pipeline you created at Step 1 (see Associate a Model With a Query Pipeline).

  4. Configure a Query Suggestions (QS) model (see Add a Model).

    A Query Suggestions model provides a list of relevant query completion suggestions as the end user is typing in a search box. It can also make the search-as-you-type feature more powerful and reliable.

  5. Associate the QS model with the query pipeline you created at Step 1 (see Associate a Model With a Query Pipeline).

  6. Optionally, if you want to evaluate how well your Coveo ML models are performing before routing a larger part of queries to the query pipeline in which they are defined, setup an A/B test between the default empty query pipeline and your new query pipeline, which now contains your Coveo ML models (see Adding and Managing A/B Tests).

What’s Next?