Sat | Mar 23, 2019

Artificial Intelligence and the Economy | Andre Smart applies Machine Learning to dynamically engage customers via smartphones

Published:Tuesday | July 17, 2018 | 12:20 AMJordan Micah Bennett/Contributor
Andre Smart

The following article was published in The Gleaner on Monday, July 16, 2018.

 

Artificial Intelligence and the Economy features machine-learning computer models in Jamaica. These models are computer algorithms, or smart apps, that seek to give computers the ability to learn like children for a variety of tasks.

Here, we highlight how an author’s work may solve a particular set of real-world tasks or problems. By doing this, we aim to encourage more local research and development in artificial intelligence.

Today, we will highlight machine learning applied to realtime customer shopping behaviour. This is work being done by André Smart, the chief executive officer of Contexual.io. Wow! I am constantly surprised by, and proud of, the existence of artificial intelligence being leveraged by Jamaicans, and the feeling is the same when it comes to this fellow Jamaican.

 

Artificial intelligence solves difficult problems

Jordan Micah Bennett: Briefly explain what you do at your company Contexual.io as it relates to artificial intelligence/machine learning.”

André Smart: We are a technology company that assists businesses in understanding more about their customers. We do that by creating dynamic personalised user experiences for customers at stores, events and other locations through their smartphones. A typical example would be in a retail store where the store owner wants to know which products in the store customers are interested in, which items customers physically interact with or to do a targeted sales promotion and measure how well customers are responding to it. Our technology allows us to collect this information and provide it in an intelligent way to businesses while maximising the on-site experience of customers.

For customers, we use machine learning to create dynamic and personalised experiences that are presented to those customers. The information that is collected is used to tailor their experience from the language that is used to the suggestions that are made [in relation to] products the customer may be interested in. On the business side, we analyse the data to empower businesses to make informed decisions. In the example of the retail store, the data could inform product placement decisions.

So, why do you love artificial intelligence?

Generally, I am excited about solving challenging problems. As a computer scientist, I am very interested in how technology can be applied. Artificial intelligence has many applications in many industries and we are already seeing it used from systems that work with doctors to inform patient diagnosis, to systems for farmers to determine the health of their crops. Both are difficult problems for which artificial intelligence is providing a solution. So for me, artificial intelligence represents myriad possibilities for solving difficult problems.

 

Solving problems by creating persons user experiences

Let’s talk about a large problem you helped to solve at Contexual.io, with the help artificial intelligence. Briefly describe the issue faced, as well as tell us briefly about the solution you employed.

When we started, we immediately had a problem with user interactions in places with high customer traffic and multiple businesses, for example, a mall or shopping centre. In this setting, we have multiple businesses for whom we are pushing customer interactions in order to promote and measure user engagement. This presented two challenges: first, users were overwhelmed with the interactions that they were getting, which negatively impacted the utility that they would normally receive. Second, businesses needed a way to cut through the noise to reach both existing and potential customers effectively and efficiently.

We realised that with the data we were collecting, we could solve both problems by personalising the user experience using machine learning. Our algorithms take into account previous interactions that the user had, such as items they’ve liked, shared, or physically interacted with and most importantly for us, their context. That is where the customer is and what they are doing at a given moment. So, for example, if a user is physically closer to ‘Store A’ than to ‘Store B’ our algorithm takes that into account when personalising the experience for that customer. Our algorithms also take into account new and existing customers when creating these interactions.
 

Excellent. Let’s get a bit technical. What type of learning algorithm(s) did you decide to utilise to solve the problem prior described?

Our system uses neural networks for candidate generation and ranking. The neural network that does candidate generation takes into account the customer’s previous interactions on our platform and finds suitable interactions based on the user’s context. The network uses collaborative filtering, which is a heuristics based model to find items that would generally be of interest to the user. Our second network ranks those interactions by assigning a score to each interaction. Essentially, the task of the candidate generator is to learn embeddings that map users to interactions as a function of the user’s history and their context which is then passed to a softmax classifier.

Additionally, we use decision trees to overcome the cold start problem. This is where the system does not have historical data on a user and is thus unable to make effective recommendations. The platform attempts to seed the recommender with information by posing questions to the user. The decision trees algorithm is used to learn a model for maximising the information garnered while minimising the time spent by the user.

 

A solution for businesses

Can you talk about some shortfalls of your solution?

The number of training examples we used when training the neural network was not bound. That is to say the number of training examples per user varied. Consequently, during the loss calculation, the loss was dominated by users with more training examples. Research suggests that we should be able to improve the performance of the network by ensuring that there is a fixed number of examples per user.

How does society benefit from your solution?

Our platform benefits both businesses and customers. Businesses need a model for continuously testing and engaging customers with their brand and products in order to make intelligent decisions about what your customers want. For customers, it is an experience that gives them the most out of where they are by personalising it to them and their needs through a medium that they already use to interact with the world around them.

Can you describe some of your larger machine learning related personal projects too?

I have been dedicating a lot of my time to computer vision research. That is teaching a computer to infer useful information from some visual input, like videos or images. My team is currently working on a robust solution that utilises this to address several challenges that we have identified in Jamaica. I look forward to talking with you more about it in the future.

I enjoyed your time/answers. Looking forward to interviewing you again sometime soon.

Thank you Jordan. It was my pleasure.

Send feedback to editorial@gleanerjm.com, or jordanmicahbennett@gmail.com.