Abstract:
Transaction data that are obtained from financial institutes contain a description string from which the merchant identity and an encoded store identifier may be parsed. Personal Financial Management (PFM) services aggregate these data in order to provide value to individuals in the form of insights and targeted suggestions. To do so, it is crucial for PFMs to have a good characterization both of individuals (in terms of demographic attributes), and of transactions (especially what they are and where they happened). This information, while not directly present in the transaction data, can be obtained using a range of data science approaches. In the first part of the talk, I will present a deep learning approach for inferring demographic attributes from purchase history and discuss the implications of estimated demographics on representative sample weighing. In the second part, I will present a statistical model for predicting location of points of sale, operating on the graph of shared customers.