How the model calculates the bias?

Nobody has claimed this yet.

I'm having a hard time understanding how the model learns the users and items biases and I'd loooove to have some guidance! haha

I do understand the fact that **the bias term for user u is given by the sum of the features' biases**, as Kula explains in his paper, however, I can't seem to understand **how the model calculates the features' biases**.

In other words, I know that when creating the FM, the model represents users (and items) as its' features latent vectors, and then users' biases originate from the sum of the features biases of users features. But how the model calculates these features biases? What are those?

love, Bec.

Start with the issue's description and its reference to Kula's paper, then trace how feature biases are defined and learned in LightFM. Done means providing a clear explanation of what the feature biases represent and how they produce user and item bias terms.

Written by the indexing model from the issue text.