Concept to Code: Neural Networks for Sequence Learning

Abstract

Deep Learning has shown significant results. Advances in deep learning are applied to many aspects of modern IR systems. In this tutorial we provide conceptual understanding of Embedding methods and Recurrent Neural Networks (RNNs) which are currently applied in IR. As RNNs are effective in modelling sequential data (e.g. clicks, add to cart, purchase data) that is generated by users in a session and across sessions. We provide a hands-on case study for sequence aware Recommender System. Proposed tutorial begins with learning of Embedding (e.g. user, product, product features), traditional sequence-based and session-aware recommendation systems and their evaluation methods. It then covers various models based on hierarchical representation and RNNs: attention and attribute aware, memory based models, multi-layer LSTMs, and combining RNNs with CNNs. We walk-through the code for techniques covered in each section and for a sequence-aware recommender system on e-commerce dataset, summarize these models, parameters and understand what is going on behind the scene with various visualizations. We will use Jupyter notebook with already executed code for walk through. We believe that a self contained tutorial giving good conceptual understanding of deep learning techniques with sufficient mathematical background along with actual code will be of immense help to participants.