A Dynamic Structural Model for Heterogeneous Mobile Data Consumption and Promotion Design

Inverse Reinforcement Learning for Marketing

“…al. [4] develops a structural SOC-based model for estimation of mobile phone users' preferences using their observed data daily consumption. On the side of Inverse Reinforcement Learning, our framework is rooted in The Maximum Entropy IRL (MaxEnt-IRL) [5,6] method.…”

“…As will be clear below, our approach has a number of advantages over the method of Ref. [4]. Most importantly, it does not need Monte Carlo to estimate parameters of the user utility, and instead relies on a straightforward Maximum Likelihood Estimation (MLE) with a convex negative loglikelihood function with 5 variables, which can be done very efficiently using the standard off-the-shelf convex optimization software.…”

“…Note that Ref. [4] uses a two-step structural approach to estimating the consumption model which first estimates the empirical policy and then finds structural parameters of the utility function that are consistent with this empirically "observed" optimal policy. A similar approach that tries to simultaneously estimate both the reward and policy function compatible with this reward is sometimes employed in the IRL literature.…”

“…al. [4], a framework presented here focuses on consumption data. While our method can be applied to a number of different business settings as outlined in the introduction (such as cloud plans, data plans, utility plans etc.…”

“…We note that our specification of no-consumption behavior is more flexible than what suggested by Xu et. al [4],. who interpret zero-consumption events as observations of a Gaussian process censored at zero.…”