Asset recommender

The asset recommender suggests relevant assets based on the assets that were visited in the past.
The provided recommendations can be based on the overall asset popularity across the organization or targeted for a specific user. Depending on where the recommendations are shown, they can also be categorized and filtered by asset type, domain or community.
To ensure the recommendations are based on up-to-date information, the asset recommender is retrained every day. For more information, go to Asset recommender strategy.

Asset recommendations are shown in Collibra for Desktop, Home page, and Data Marketplace. In the following example, the asset recommender is used to display popular data and recommended data in Data Marketplace.

Important 
  • The asset recommender is a Cloud-only feature. The asset recommender also requires that application usage data is collected for the environment.
  • Recommendations are not available in environments with instance names that contain one of the following strings: -qa, beta, -poc, ie-, infra-, docker, training, sandbox, or :.
  • Recommendations in new environments will be empty or less meaningful for some time until the asset recommender collected enough user behavior data.
  • If you just started using Collibra as a user, you won't receive any recommendations. You'll receive recommendations after using Collibra and after the daily retraining of the asset recommender.

Asset recommender strategy

The asset recommender is trained on asset visit data. It roughly follows these steps:

  1. Data on asset visits, like which users visited which assets, when, and how often, is collected and processed to assign a score to all asset and visitor combinations. The score reflects the interest of a user in the asset.
    Note 

    Asset visits in Collibra platform, Collibra for Desktop, Collibra for Mobile, and Data Marketplace are taken into account.
    Visits to the following instance names are not taken into account:

    • Instance names that contain the following strings: -qa, beta, -poc, ie-, infra-, docker, training, sandbox, and :.
    • Instance names that do not contain .collibra.com
  2. A model is trained to predict all user-asset interest scores. The model learns to generate informative recommendations for assets and users based on the collected data.

    Note Certain asset visits are filtered out when the asset recommender is retrained, for example, assets visited by only one user, assets that are only visited once, and asset visits that take less than six seconds.

  3. The data collection and model training are triggered daily to ensure that asset recommendations are based on up-to-date information.

Troubleshooting asset recommender

Issue: You do not receive any recommendations where you expect them.

Possible reasons:

  • If you just started using Collibra as a user, you won't receive any recommendations. You'll receive recommendations after using Collibra and after the daily retraining of the asset recommender.
  • You visited assets, but the visits were filtered out during retraining. In this case, the asset recommender cannot provide recommendations.
  • It is possible that no relevant assets can be suggested because of applied asset type, community or domain filters where the recommendations are shown. For example, assets outside the Data Marketplace scope will not be recommended in the context of Data Marketplace.
  • It is possible that no relevant assets can be suggested in the instance.