|
Abstract:
|
Customer Life Time Value estimation is central to long term business planning. A ride-sharing platform business is a non-contractual business with continuous purchase opportunity, thus facing the most difficult scenario in LTV estimation. In this work, we propose an LTV model based on flexible growth curve models that can accommodate highly heterogenous population using Embedding techniques. Ride-sharing usage is highly personal and it is necessary to provide a model that captures unique potential growth experience of each user to maximize future user utility. The model is trained and evaluated on a large-scale deep learning system using real world ride-sharing consumption data across highly heterogenous markets and user segments. Our contributions include: 1. Our work is one of initial studies on personalized life time value estimation using deep learning techniques within a growth curve framework; 2. Our work developed an embedding technique that jointly learns the latent representation of user’s experience. Using both internal and publicly available transaction data, we benchmark our personalized LTV estimation against existing models.
|
