- Crossed feature is a very good way to combine two features, which doesn't have proper meaning by themselves
- for example, in uber car pick up, only the latitude or longitude is not enough to properly identify a location, but multiplication of them would specifically identify a location.
- In some cases, both of the features can be high cardinal, which makes the Crossed feature Very High cardinal. In those cases, Feature Hashing is used on each of the category before crossed feature to reduce the dimension
Use Cases:
- LinkedIn uses Cross Feature on user location vs. user job title in their recommendation model
References