Feature Selection

The Feature Selection algorithm enables Coveo™ Machine Learning (Coveo ML) models to automatically filter and refine which context keys to use or ignore when building ML models. The Feature Selection algorithm uses a wide variety of factors to determine whether to use a dimension to personalize suggestions or recommendations.

Examples of tests which the Feature Selection model carries out to determine whether to blacklist a context key:

  • Check whether the data contains many long strings of text which might not be intended/suitable for classification
  • Check whether the data is empty
  • Check whether the data contains too many unique values
  • Check whether the data contains only a single value
  • Check whether the data is considered irrelevant according to some ML/Statistical tests

It is possible to override the Feature Selection algorithm to forcefully use or ignore custom context keys by using the whitelist and blacklist parameters (see Custom Model Parameters).