Learning to Rank for Search Results Re-ranking in Learning Experience Platforms

Abstract

The ability to search and retrieve the right resources in a Learning Experience Platform (LXP) is critical in helping the workforce of an enterprise to upskill and deepen their expertise effectively. To ensure the best resources are shown as high in the result set as possible to catch learners’ attention, a supervised learning approach of training and deploying a Learning to Rank (LTR) model for re-ranking is proposed. This work specifically focuses on judgement list preparation taking advantage of the learning progress data available in LXPs, as well as on defining and measuring model performance through metrics in both test and production setups. In particular, it highlights the positive impact of the deployed LTR model in production using the defined metrics like average search result click position and percentage top N clicks.