Achieve relevance

The next step in implementing Coveo is to maximize the relevance of your search results and recommendations. Coveo Usage Analytics (Coveo UA), query pipelines, and Coveo Machine Learning (Coveo ML) models enable you to do just that.

Usage analytics

Coveo UA is a cloud service that captures usage data from your Coveo-powered search interfaces. This data can not only be used to analyze user behavior, but it also feeds machine learning models.

To leverage usage analytics

  1. Get familiar with the usage analytics concepts.

  2. Ensure that your search interfaces send usage analytics data including a search hub value.

    Note

    When using the Atomic library, you set the search hub value using the atomic-search-interface component searchHub attribute.

  3. Review analytics reports to identify areas of improvement in your search interfaces.

  4. Take advantage of Coveo machine learning features that feed on usage analytics.

Query pipelines

When a user performs a search, the query undergoes a series of processing steps before the results are returned to the user. These steps make up the query pipeline.

Some steps in the query pipeline allow you to define rules to alter the user query, boost or bury items based on specific criteria. These query pipeline rules are:

  • Thesaurus entries

    Thesaurus entries let you add equivalent keywords or phrases to the query entered by a user.

  • Stop words

    Stop words are words that are filtered out from a query.

  • Featured results

    Featured result rules make specific items appear at the top of the result list whenever a query satisfies a given condition.

  • Ranking expressions

    Ranking expressions let you increase or decrease item ranking scores based on specific criteria.

  • Filters

    Filters let you define the scope of the search results displayed to your users.

  • Query parameters

    Query parameter rules may be defined to override search parameter values when a query matches a condition.

  • Ranking weights

    Ranking weights let you increase or decrease the relative importance of ranking factors in the computed relevance score of an item. Examples of ranking factors include:

    • the presence of a query term in the title of an item

    • the number of times a query term appears in the item body

  • Triggers

    Triggers let you define an action to execute in a search interface when a query meets a given condition.

To leverage query pipelines

  1. Create a query pipeline.

    Tip

    You may want to create several query pipelines in your Coveo organization. Though there are different mechanisms to route your search queries to a query pipeline, we recommend using condition-based routing.

  2. Add query pipeline rules (or components) to the query pipeline.

    Tip

    Though query pipeline rules allow significant control over the search query and results search interfaces yield, leveraging Coveo Machine Learning models in a query pipeline is a better time and relevance investment.

  3. Repeat these steps to create other query pipelines as needed.

  4. Use the Relevance Inspector to assess and troubleshoot the relevance of search results.

Machine learning models

The real power of the Coveo Platform lies in its machine learning models. Models learn from user behavior to automatically and continuously improve the relevance of your search results or recommendations.

Coveo provides the following machine learning model types:

  • Relevance Generative Answering (RGA)

    RGA models are used to generate on-the-fly answers to user queries by analyzing the content of indexed items.

  • Query Suggestions (QS)

    Query suggestions are suggested queries that appear in a drop-down list as soon as a user starts typing in the search box.

  • Automatic Relevance Tuning (ART)

    ART models analyze user behavior patterns from many usage analytics search visit actions to understand and deliver the content that users seek.

  • Content Recommendations (CR)

    Content recommendations are personalized recommendations based on the repeated occurrence of clicks and view events in past user visits. Think of content recommendations as a people who viewed this also viewed feature.

  • Dynamic Navigation Experience (DNE)

    DNE models are used to make the most relevant facets and facet values appear first in your search interfaces. The model uses the most popular facet values for a given query to promote the most relevant results.

  • Smart Snippets

    Smart Snippet models compare the user’s search query to the headings in items returned by the index. When there’s a strong similarity between the two, the model extracts the item content that’s most likely to answer the query. This content is displayed as a snippet directly on the results page.

To leverage machine learning

  1. Ensure that your search interfaces send usage analytics data including a search hub value.

    Note

    When using the Atomic library, you set the search hub value using the atomic-search-interface component searchHub attribute.

  2. If you’re using the Coveo for Sitecore integration to index content and you want Smart Snippets, create an indexing pipeline extension to index items with the proper language field value.

  3. Create a model for the machine learning feature you want to leverage. Prerequisites and instructions are specific to the type of model (that is, RGA, QS, ART, CR, DNE, or Smart Snippets).

  4. Some machine learning features require that you add a front-end component to your search interface. If you’re using Coveo Atomic for your search interfaces, these components are:

  5. Use the Relevance Inspector to assess and troubleshoot the relevance of search results.