Module sbr::models::ewma

Model based on exponentially-weighted average (EWMA) of past embeddings.

The model estimates three sets of parameters:

• n-dimensional item embeddings
• item biases (capturing item popularity), and
• an n-dimensional alpha parameter, capturing the rate at which past interactions should be decayed.

The representation of a user at time t is given by an n-dimensional vector u:

u_t = sigmoid(alpha) * i_{t-1} + (1.0 - sigmoid(alpha)) + i_t

where i_t is the embedding of the item the user interacted with at time t.

Structs

Hyperparameters	Hyperparameters describing the EWMA model.
ImplicitEWMAModel	Implicit EWMA model.