Weighing Dynamic Availability and Consumption for Twitch Recommendations

Accepted workshop paper co-authored by Edgar Chen, Mark Ally, Eder Santana and Saad Ali, all of Twitch, San Francisco (his address appears in the paper as [contact redacted]). It proposes two loss-weighting methods for Twitch's implicit-feedback recommender: one correcting for channels not being online simultaneously with users, and one adjusting for minutes-watched as a preference signal. The OARS CFP states that accepted submissions are presented at the workshop; which co-author delivered the presentation is not stated. Also indexed at https://www.amazon.science/publications/weighing-dynamic-availability-and-consumption-for-twitch-recommendations.