The current version of Novelty relies on assigning a discrete popularity score to items depending on their "position" in the distribution of ratings/interactions (long tail/short head).
I suggest implementing another version, in which the value of popularity is continuous, and defined as the percentage of users who rated/interacted with item i.
Hi!
The current version of Novelty relies on assigning a discrete popularity score to items depending on their "position" in the distribution of ratings/interactions (long tail/short head).
I suggest implementing another version, in which the value of popularity is continuous, and defined as the percentage of users who rated/interacted with item i.
See, e.g. Current challenges and visions in music recommender systems research, Equation 12
Cheers! Marta