Multi-funnel Recommender System for Cold Item Boosting

In order to achieve a healthy two sided marketplace, a recommender system balancing the user and the item interests must be deployed. One of the challenges for such a system is to give new items a fair chance to compete against an existing supply. In the context of an online travel agency, this challenge arises not only from the classical cold start problem in recommender systems but also from the customer’s reluctance to book new properties due to high uncertainty about their quality, which is often alleviated by past customer reviews. In this paper, we share how a multi-funnel recommender system was developed to address this specific challenge and how the effect of such intervention can be measured on both the customers and the supply. Along with the main recommender delivering a personalized ranking maximizing customers conversion, a second funnel focusing on only cold properties can be used to generate an alternative optimistic ranking. Then these two rankings are merged in a way to balance between short term conversion rate and long term metrics assessing the supply health and properties performance. We show this intervention has different effects when segmenting both customers and properties and suggest future directions given these observations.