Plackett-Luce model for learning-to-rank task
This addresses the gap in real-world ranking systems for search and recommendation tasks, though it is incremental as it builds on existing list-wise frameworks.
The authors tackled the underperformance of list-wise learning-to-rank methods in real-world applications by proposing ListMLE, a non-linear algorithm using the Plackett-Luce loss, which matched or surpassed state-of-the-art systems on Yahoo challenge 2010 and Microsoft 30K datasets.
List-wise based learning to rank methods are generally supposed to have better performance than point- and pair-wise based. However, in real-world applications, state-of-the-art systems are not from list-wise based camp. In this paper, we propose a new non-linear algorithm in the list-wise based framework called ListMLE, which uses the Plackett-Luce (PL) loss. Our experiments are conducted on the two largest publicly available real-world datasets, Yahoo challenge 2010 and Microsoft 30K. This is the first time in the single model level for a list-wise based system to match or overpass state-of-the-art systems in real-world datasets.