Slate-Aware Ranking for Recommendation

• DOI: 10.48550/arxiv.2302.12427
• Publication Date: 2023-01-01
• Authors: Yi Ren; Xiao Han; Zhao Xu; Shenzheng Zhang; Yan Zhang
• Keywords: Ranking SVM; Relevance; Rank (graph theory); Learning to Rank

We see widespread adoption of slate recommender systems, where an ordered item list is fed to the user based on interests and items' content. For each recommendation, can select one or several items from for further interaction. In this setting, significant impact behaviors mutual influence among well understood. The existing methods add another step re-ranking after ranking stage which considers recommended re-rank generate recommendation results so as maximize expected overall utility. However, model complex interaction multiple items, usually just handle dozens candidates because constraint limited hardware resource system latency. Therefore, still essential most applications provide high-quality candidate set stage. paper, we propose a solution named Slate-Aware (SAR) By implicitly considering relations it significantly enhances quality stage's boosts relevance diversity systems. Both experiments with public datasets internal online A/B testing are conducted verify its effectiveness.