In content-based filtering (CBF) recommenders, when there are is no user profile, similar items are recommended an item that a user is currently inspecting. For instance, if you are looking for a comedy movie, other comedies might be recommended.
In the literature I found that often such recommenders use linear combination (i.e., weighted sums) of distances between items features to measure similarity between items.
I wonder how common the approach with weighted sum is for CBF?
Is industry using rather linear classifiers to measure similarity between items?