Toward User Engagement Optimization in 2D Presentation

Given a collection of items to display, such as news, videos, or products, how can we optimize their presentation order to maximize user engagements, such as clicks and purchases? The problem is often regarded as a ranking task and items are usually displayed in a vertical order. However, widescreens and grid-based presentations pose novel challenges for online content presentation. For example, many platforms such as E-Commerce may display results in a 2-dimensional, grid-like format, which is different from the 1-dimensional presentation of classic ranking applications. In this work, we propose to investigate the influence of 2-dimensional presentation on user engagements.

In a nutshell, the study presents how we find a fair way of presentation in grid-based environment - more relevant content should be placed in a more noticeable position. The position noticeability is further quantified to help ranking models better understand the signal of relevance manifested in user feedbacks. Our investigation paves the way for an end-to-end item presentation framework that learns the optimal layout for optimizing user engagements. Experimental results based on large-scale and real-world data show the superiority over state-of-the-art approaches. The proposed method has been deployed in an E-Commerce platform where online A/B testing results generated by millions of user sessions show significant growth of key user feedback metrics.