How many times have you been shopping online, filled up your basket and abandoned your purchase? This could be due to countless factors - you’re over budget and can’t decide what to get rid of, you were just browsing, you’re out of time, or you’re just overwhelmed by the sheer amount of choice on the website. E-commerce platforms are desperately trying to overcome this problem by serving personalised and targeted adverts based not only on your previous browsing history but on countless factors and data sets they’ve collected whilst you’ve been on their website. Both retailers and search engines are facing the challenge of an abundance of choice and to break through this and encourage purchases requires adverts to be deployed.

Which platforms do you use? Who should you target? How do you target?

We've gone from being exposed to about 500 ads a day back in the 1970's to as many as 5,000 a day today, so for companies who are spending huge amounts on their advertising, how do they make them stand out from the crowd? Yahoo Labs are one of many leveraging AI to try and cut through the noise and are using applied research such as search and display ads and prediction. Their goal is to deploy models and algorithms into production that appeal directly to each specific user. Miao Lu Research Scientist at Yahoo Labs said that ‘in online advertising, big data with structured or unstructured format of text, image or video impose big challenges in recommendation and forecasting. For instance, an important goal of ad is to reach the potential audiences who are interested in the ads, and finally obtain valuable actions (e.g., click, sign up, purchase). Heterogeneous information of ads description such as the text, image and video, together with historical information, make it harder to do recommendation (such as bid / price) and forecasting (like Click-through rate).’

Miao will be joining RE•WORK at the Deep Learning Summit in San Francisco this January 25 & 26 where he will share his work in ‘Deep Learning in Advertisement’ and discuss the work Yahoo Labs are doing in  training deep learning models in click-through rate (CTR) forecasting in online advertising. Specifically, he will describe a context-aware convolutional neural network (CNN), to capture the highly non-linear and local information of the historical time series, as well as the underlying association between the time series of CTR and the context information. Yahoo is a platform with huge amounts of data, and Miao explained that when implementing AI software, they ‘encounter challenges like how to deal with massive data efficiency. Fortunately, Yahoo have created TensorFlowOnSpark which easily solved these problems.’

AI, and specifically deep learning is a huge breakthrough with dealing with rich format big data - there are word embeddings, CNNs and recurrent neural networks which can handle image, video and sequential data, and Miao explains that these have huge potential when employed in advertising. Of course there are ethical implications of using AI in advertising, and this is something that every platform has to consider when building their models and using customer information. It’s no surprise to hear that the most successful adverts are those that evoke an emotional response as opposed to value-driven advertising - think the John Lewis Christmas advert - especially prevalent in video advertising! This is currently produces to appeal to the masses rather than based on customer profiles and activity, however the next step is for emotionally intelligent AIs to be built into advertising models, although the ethics of this is still questionable.

To learn more about deep learning in advertising from Miao register now for the Deep Learning Summit in San Francisco this January.

Can’t make it to San Fran?

Next March 15 & 16 the Deep Learning in Retail & Advertising Summit will be taking place in London where discussions will focus around the impact of deep learning and AI in the retail and advertising sector. Applications include computer vision for sizing; image analysis for shopping efficiency; and natural language processing for personalised shopping experiences. Confirmed speakers include Kostas Perifanos, Lead Machine Learning Engineer, Argos, George Platon, CTO & Founder, BuddyGuard, Yuriy Mukhin, Co-Founder, Lalafo, Honglei Li, Senior Lecturer, Northumbria University and many more leading names more to be confirmed.