Temporal Modelling for Customer Behaviour

The customer behaviour analysis is a critical component of business intelligence and marketing. Understanding the customer behaviour can support businesses to develop appropriate strategies and communicate with the right customers at the right time. In this thesis, we propose advanced temporal modelling techniques for customer behaviour analysis, in order to track customer behaviour changes, identify different types of customers and compare customer responses to marketing strategies and programs. We design temporal modelling techniques based on the temporal collaborative filtering and stochastic processes for customer behaviour analysis. To handle the sparse and noisy purchase records, our methods utilise latent variable modelling to segment customers into groups and discover shared purchase patterns. Our methods can detect the long-term and short-term behaviour patterns due to various factors such as seasonal effects and preference drifts. Our methods can also monitor the evolution of customer groups and track how individual customers shift across groups. We conduct four case studies on a real-world supermarket health program, which encourages participants to adopt a healthier lifestyle. The participants can access a supporting website with personalised tools for the health program. They can also get discount on fruits and vegetables when purchasing in the supermarket. Through the case studies, we show the advantages of our techniques to model the customer behaviour. We also evaluate the impact of promotions and the health program on different types of customers. Our models and results can be used by business stakeholders to increase the customer engagement and optimise their promotional campaigns. Importantly, the proposed methods contribute to the body of knowledge in data mining and user behaviour analytics. These methods can be applied to model other types of behaviour such as activities on social networks and educational systems