Multitask Ranking System for Immersive Feed and No More Clicks: A Case Study of Short-Form Video Recommendation

What is it about?

We focus on Short-Form Video (SFV) recommendation with the challenges rooted from the recent innovation in UI: immersive feed and no clicks. Instead of providing choices for users to click, SFV platforms actively recommend content to users to watch one at a time. We highlight unique challenges rooted from such UI changes for SFV recommendation system design: a) new and more challenging implicit user biases appear without the common click-based position biases and b) user behaviors can be extremely skewed and sparse and raise severe task conflicts when training multiple types of user activities. To tackle these challenges, we introduce a unified multi-task ranking framework which puts two novel components all together into an overall system for SFV recommendation. We demonstrate the effectiveness and efficiency of the framework on one of today's largest SFV platforms.

Featured Image

Why is it important?

Short-Form Videos, or SFVs have generated huge success during the past several years and become the newest social media stars. It is important to investigate and understand the unique challenges for recommendation system design. Two key findings regarding the innovative UI introduced by SFVs and highly contributed to their success are: a) we identify that there are position biases of SFVs in the recommendation sequence, namely “watch trail biases”, and introduce biases correction using trail-related information, and b) to get the most benefits from multi-task learning, especially co-training tasks with extremely skewed and sparse labels, we adapt a disentangle regularization to mitigate task conflicts, introduce loss upweighting for sparse task co-training and adopt a meta-learning algorithm for efficient weight selection.

Perspectives