Shopping Lies in the AI of the Beholder
Associate Professor Hady Lauw from the Singapore Management University explains how the ThriftCity framework makes sense of the multitude of products across different ecommerce platforms, providing consumers with the best offer.
A smart shopper would look at multiple e-commerce providers to compare prices, but the diversity of how products are listed with different model numbers, names, images, and specifications makes it challenging to find them at scale.
This is where an Artificial Intelligence (AI) framework of data mining and machine learning technology can help to make sense of the myriad of products.
Meet ThriftCity – a framework developed by Associate Professor Hady Lauw and his team from Singapore Management University (SMU) ’s School of Computing and Information Systems.
One key aspect of this technology is about collecting data all over the Internet, but products are often not identified by a single unique code, but instead have descriptors such as product names, brands and images.
Prof Lauw said, “There are billions of pages on the web and we can’t possibly go through all of it, so we have to build a crawling framework that is intelligent. This is where the AI comes in... where we train the framework to collect information efficiently.”
Training AI with Human Signals
Another aspect of the ThriftCity framework is determining when two products are the same product, and this is done by training the AI algorithms to make use of “signals that humans would use”- such as looking at the products’ appearance or its descriptions.
For example, to compare how the two products look, the AI system will require computer vision technologies to do image processing, or use Natural Language Processing technologies to identify the different descriptors in the product names, before comparing the products’ similarities through the use of machine learning.
Assoc Prof Hady Lauw (L) and Entrepreneur Lead of ThriftCity, Darryl Ong, will be looking at working with commercial partners in the South East Asia region in the coming year.
The team also designed the framework to specify the degree of confidence when determining when two products are similar.
“For instances where we are less confident, we can flag it and surface it to human labellers, who can then lend some supervisory judgment. In that case, this model will get better over time,” said Prof Lauw.
He also emphasised that such AI technologies are not developed to keep the human element completely kept out of the loop, but rather to reduce the effort required by the human, so that they can work on more products with the assistance of AI models.
Focus on Electronic Products
Currently, the ThriftCity platform has a focus on electronic products, specifically print cartridges, headsets and smart watches, as they have the biggest price differences across platforms.
Prof Lauw explained that electronic products tend to have more unique datasets in terms of textual information such as model number or make, which makes it a good starting point to train the AI model. On the other hand, product domains like fashion may be more focused on the visual descriptors, and would require more vision and image resources.
ThriftCity was conceptualised as a research prototype around two years ago, and when Darryl Ong, a graduate from SMU’s School of Computing and Information Systems came on board last year.
He took up the role of Entrepreneur Lead and the team then started to consider commercialising the system. This was achieved with the support from the SMU Institute of Innovation and and Entrepreneurship.
Darryl said the team sees themselves as a technology player, so that ThriftCity will not be tied to a specific product domain. Although the technology is capable of sourcing products from anywhere reachable by the Internet, the team will be looking to work with commercial partners in the South East Asia region in the coming year, where they are able to contribute to a growing e-commerce market.
Instead of providing a direct service to consumers, he said that businesses that tap on their technology could find out how their competitor products are priced, and in turn price their products more competitively, to which consumers will still benefit.
Prof Lauw added that, “If things are progressing well in the next two years, we can revisit the idea of proving more consumer-facing services or products on ThriftCity.”