As deep learning techniques and applications continue to spread across business and society, this technology is now making an impact on retail and advertising. Data mining and deep learning algorithms are applied to understand customer insights in greater detail than ever before; allowing retailers to evolve successfully in an increasingly competitive market.
Customer Lifetime Value (CLTV) prediction is an important problem in e-commerce. An accurate estimate of CLTV allows retailers to correctly: allocate marketing spend, identify and nurture high value customers, minimise exposure to unprofitable customers and attribute value to indirect marketing such as content production.
At the inaugural Deep Learning in Retail & Advertising Summit in London, Ben Chamberlain, Senior Data Scientist at ASOS, will share expertise on combined automatic feature learning through deep neural models with hand-crafted features to produce CLTV estimates that outperform either paradigm used in isolation. RE•WORK asked him a few questions ahead of the summit to learn more.
What do you feel are the leading factors enabling recent advancements of deep learning?
I think most people would say that it is a combination of three factors:
- Really good public, labelled datasets appearing, such as ImageNet;
- Massive GPUs, which make things like deep reinforcement learning work in practice;
- Open source frameworks like TensorFlow and Theano, which allow people to build and share components for reuse by others.
A lot of the modelling improvements were made decades ago, but the level of interest in deep learning is so intense now, it seems likely that major algorithmic advances will crystallise in industry soon too.
What was your motivation to get involved in deep learning at ASOS?
Deep learning has been massively successful in computer vision and natural language processing, but has not had the same success in highly stochastic problems like predicting if a customer will churn. Despite this, it does provide a very elegant way to represent heterogenous data like images and text and combine then into predictive models. This is where my primary interest in deep learning at ASOS lies, though we do have people who are working on text, video and images as well.
What present or potential future applications of deep learning excite you most?
There was certainly a lot of talk about Generative Adversarial Networks at this year’s NIPS conference. These are networks that in a sense can exhibit a form of creativity. It’s early days however and personally I’m more interested in how recurrent networks can be used to incorporate temporal patterns into retail models and do things like diagnose a customer context switch in real-time.
Which industries do you feel will be most disrupted by deep learning, and artificial intelligence in the future?
Most large technology and automotive companies are investing billions in self driving cars. This will have a profound effect, not just on transportation industries, but on the structure of our cities and how we plan and build. Much longer commutes become viable in an age of affordable, productive, point to point transportation.
There is also a massive potential in healthcare, where the major blocker at the moment is not the technology, but the availability of data.
What developments in deep learning can we expect to see in the next 5 years?
Deep learning is an incredibly dynamic research area that has attracted the attention of many brilliant minds. Trying to predict what people much smarter than you are will think of over the next five years is difficult. The current trends however, appear to be improving generative models (like GANs) and to learn more efficiently from smaller datasets.
