Paper Authors and Title
Jorge Jimenez Garcia (Technological University Dublin), Rajesh Jaiswal, “Recommender models for items in the game ‘Destiny 2′”
Abstract
With the goal of exploring the field of recommendation models, a system for the game ‘Destiny 2’ is developed as a case study.
Alongside analysis and study of data for this particular domain, a novel approach, making use of a heuristic objective function that features discounted reward based on the gravity of the error is explored. This approach is proposed to be generalizable and useful for all systems that deal with recommendations in popularity-based domains, rivaling or surpassing accuracy-based metrics and often resulting in smaller or/and quicker to converge models.
For the case study, hyperparameter tuning, model evaluation and selection is carried out. A model card discussing potential biases and behaviour specific to the model and domain is provided.
Future improvements to the objective function approach are proposed, as well as for the case study’s field.