Bi-Rank: A New Bi-Directional Ranking Method for Goods Selection

Citations Over Time

Abstract

Goods selection is a typical daily routine faced by e-commerce platforms, such as choosing the right goods for on-shelf and off-shelf . In this paper, we turned goods selection problem into a learning-to-rank task (LTR). Instead of ranking the head part and the tail part separately, we proposed a Bi-directional Ranking model, abbreviated as Bi-Rank, to solve this task. Bi-Rank relies on a customized loss function/metric named NDCG PLUS, which is an improved version of NDCG. NDCG PLUS incorporates the ranking loss of tail part in the total ranking loss. In addition, Bi-Rank model can choose different size and weight to balance head and tail. It can also downgrade from double-end to single-end by turn off a designed switch, avoiding the trivial process of manually changing items' label. The experiment shows that Bi-Rank model can achieve a good enough ranking result on collections' head part, while output a similar ranking result on tail part compared to the model that specifically optimizing the tail. In addition, this Bi-Rank model is also very flexible, efficient, and easy to use.

Related Papers

• → QoRank: A query-dependent ranking model using LSE-based weighted multiple hyperplanes aggregation for information retrieval(2010)7 cited
• → Querywise Fair Learning to Rank through Multi-Objective Optimization(2023)5 cited
• → Bi-Rank: A New Bi-Directional Ranking Method for Goods Selection(2020)1 cited
• → OrdRank: Learning to Rank with Ordered Multiple Hyperplanes(2009)
• Survey of information retrieval ranking algorithm(2009)