Understanding ecommerce clickstreams: a tale of two states
Humphrey Sheil, OMER F. RANA, Richard B. Reilly · ORCA Online Research @Cardiff (Cardiff University) · 2018
We present an analysis of Ecommerce clickstream data using Recurrent Neural Networks (RNN), Gated Recurrent Units (GRU) and Long-Short Term Memory (LSTM). Our analysis highlights the substantial difference in the predictive power of LSTM models depending on whether or not hidden state is shared across batches and also assesses the ability of RNNs to learn and use both session-local and dataset-global information under different sampling strategies. We propose random sampling combined with stateless LSTM for optimal performance of LSTM in an Ecommerce domain.