Data Scientist to Present Findings at E-commerce Workshop During Prestigious Conference
San Francisco, July 10, 2018 — Michael Skinner, the resident “Data Whisperer” at Duetto, hospitality’s only Revenue Strategy Platform, placed first in the 2018 Rakuten Data Challenge and has been invited to present at the 2018 SIGIR Workshop on eCommerce, July 12 in Ann Arbor, Mich.
The competition, organized as part of the 41st International ACM SIGIR Conference on Research and Development in Information Retrieval, tasked data scientists with developing a large-scale taxonomy classification to accurately predict the category of hundreds of thousands of individual products in a sample e-commerce catalog. As the winner of the Data Challenge, Skinner will showcase his findings — including the “Balanced Pooling Views” method he developed — at a poster session during SIGIR 2018.
Skinner won the Data Challenge by developing and applying Balanced Pooling Views, a strategy that enhances the ability of a statistical model to “read” the product titles in the data set and accurately sort them into their appropriate categories.
“The big takeaway from this experiment is that solving a massive challenge, like classifying a giant e-commerce catalogue, doesn’t necessarily need major changes to the standard architecture of common recurrent neural networks,” Skinner said. “By focusing on the learning rate — or how fast the machine ‘learns’ and makes adjustments — and experimenting with new ideas that let scientists ‘train’ the networks with fewer iterations, we can tackle complex problems in a more methodical fashion.”
Skinner, a graduate of the University of Washington and a former software engineer for Google, said his new method built upon other recent findings in the field of machine learning. Skinner’s system description paper for his winning solution builds on several previously published ideas, most notably a recent paper by Jeremy Howard and Sebastian Ruder, which applied similar techniques to traditional natural-language processing tasks.
“As a tight-knit team of software developers and data scientists, we at Duetto couldn’t be prouder of Michael’s achievement,” said Craig Weissman, Duetto’s Co-Founder and Chief Technology Officer, and the former CTO of Salesforce.com. “His contribution to a greater understanding of machine learning’s application to e-commerce will move many industries forward, especially our own in hospitality technology.”
Skinner added that Balanced Pooling Views’ unique method, which “reads” any giant data set — be it millions of product titles in a catalogue or an occupancy forecast for a hotel — already has influenced new ideas and new approaches to projects underway at Duetto.