Neural ODE for Multi-channel Attribution

Multi-Touch Attribution (MTA) emerges as a pivotal tool in both marketing and advertising landscapes, shedding light on the intricate web of interactions within customer journeys during transactions or impressions. This comprehensive methodology empowers marketers with strategic allocation of attribution credits across diverse channels, not only optimizing campaigns but also enhancing overall marketplace strategies.In this study, we recognize the inherent irregularity in customer journey data and present a pioneering exploration into the effectiveness and constraints of neural ordinary differential equations (ODE) in estimating attributions and predicting conversions. We introduce an innovative application of ODE-LSTM to tackle the MTA challenge, integrating an attention mechanism into the original model.Our research finds that ODE-LSTM surpasses other methods, particularly in scenarios where time intervals maintain a moderate irregularity. Nevertheless, its performance experiences a decline with increasing irregularity. However, it distinguishes itself in attribution estimation compared to alternative approaches, thus marking a significant advancement in this field.