A model for reasoning about persistence and causation
Computational Intelligence
| Hi-index | 0.00 |
Sequence analysis has been employed for the analysis of longitudinal consumer behavior with the aim to support marketing decision making. One of the most popular applications involves Acquisition Pattern Analysis exploiting the existence of typical acquisition patterns to predict customer's most likely next purchase. Typically, these cross-sell models are restricted to the prediction of acquisitions for a limited number of products or within product categories. After all, most authors represent the acquisition process by an extensional, unidimensional sequence taking values from a symbolic alphabet. This sequential information is then modeled by (hidden) Markov models suffering from the state-space explosion problem. This paper advocates the use of intensional state representations exploiting structure and consequently allowing to model complex sequential phenomena like acquisition behavior. Dynamic Bayesian Networks (DBNs) represent the state of the environment (e.g. customer) by a set of variables and model the probabilistic dependencies of the variables within and between time steps. The advantages of this intensional state space representation are demonstrated on a cross-sell application for a financial-services provider. The DBN models multidimensional customer behavior as represented by acquisition, product ownership and covariate sequences. In addition to the ability to model structured multidimensional, potentially coupled, sequences, the DBN exhibits adequate predictive performance to support the financial-services provider's cross-sell strategy.