Machine Learning Specialist
Month Dev Time
For this project our client was a marketing agency working with the Korean Tourism Office. They wanted a dynamic website with AI controlled and optimized content. Their goal was to attract web users using pay-per-click (PPC) advertising campaigns and then monitor their activity. Our advanced Machine Learning (ML) technology would then analyze user interactions and make content adjustments as necessary, aiming to retain web users for longer.
In order to satisfy the requirements of our client, we split the project into three parts: microsite, admin portal, and an Artificial Intelligence (AI)/ML engine.
The microsite acted as a landing page for our client to direct users through their PPC ad campaigns. It needed to be capable of hosting multiple pages dedicated to different aspects of Korean life, such as food, cultural heritage, and entertainment news. What’s more, each of these additional pages also needed to support dynamic content compatible with our ML technology.
Our clients requested an admin portal that would allow them to manually create, modify, and publish assets to their various landing pages. Assets needed to have different types that would determine their location on the site and deployed on any combination of pages. They also required the ability to create landing pages populated by sets of assets and content.
The Machine Learning (ML) featured on our client’s project needed to be able to capture user data and information, then make predictions about their future interactions and behavior. This would allow our custom algorithms to determine what content is most likely to attract specific user interest and increase site dwell time. Ultimately, the goal of the AI/ML engine was to improve Key Performance Indicators (KPIs), such as bounce rate and traffic, while refining the platform's User Experience (UX).
One of the biggest challenges with this project was developing an AI/ML engine capable of collecting large volumes of precise user interactions and data. Once collected, the data needed to be analyzed by AI and transformed into tangible algorithmic steps able to appropriately update site content.
In order to create an AI/ML engine that predicted user behavior and interactions with a minimal margin of error, we elected to deploy a bandits algorithm and a supervised ML algorithm.
Bandit algorithms are a set of algorithmic steps designed to perform under uncertain conditions. This made them the obvious choice for our clients project, as web users can be unpredictable, making forecasting their behavior and content preferences difficult. The bandit algorithm allowed us to analyze user interactions by averages, using both new and existing online data. This algorithm is used during the exploration phase, primarily deployed to collect and understand user data.
Supervised ML algorithms are used to define and label datasets while training other algorithms to accurately classify data. Once the data is classified it can be used to help predict outcomes, in this case, user interactions and behavior. Together with the bandits algorithm, the supervised ML algorithm provided our client’s platform with a way of exactly projecting future user actions. As a result, we were able to solve the issues associated with creating an effective AI/ML engine.
After working with the team at Idea Maker, our client now has a platform that satisfies all their requirements. They are now able to direct web users to their dynamically updated and AI driven website. Here are just a handful of features we implemented on their platform.
Get in touch today to learn more about our development services.