Mandy Korpusik, assistant professor of computer science at Loyola Marymount University, was recently awarded $50,000 in funding by eBay to conduct research on purchase intent prediction using deep learning.
The collaboration was born when a member of eBay’s university research division reached out to see if Korpusik might be interested in conducting specific research on eBay’s behalf. Kishore Paul, a senior member of eBay’s technical staff, had read a paper Korpusik published outlining research applying deep learning for recommender systems. “eBay invited me to conduct this research by offering my expertise in designing a novel neural network architecture,” Korpusik said.
Korpusik, who teaches “Computer Programming and Laboratory,” “Machine Learning,” and “Natural Language Processing” courses, focuses a portion of her research portfolio on deep learning (a subset of machine learning), which is one reason she was selected by eBay to lead this research project. “Deep learning is a specific type of machine learning model called a neural network,” said Korpusik. “Inspired by the neurons in the human body, the model processes different layers of neurons that are all connected together similar to the human nervous system.”
The questions eBay is seeking to address through this research include: What is the probability that a customer will proceed to check out during the online purchase process? Can we predict this with a very high rate of accuracy at greater than 95 percent? Will adjusted pricing positively affect purchase decisions for different customer groups?
Korpusik will partner with one of her master’s students, Brandon Golshirazian, who will implement the research for his thesis. It is estimated to be a six-month research project that will conclude in spring 2023. The research will begin by running new customer activity data that eBay has collected through the company’s existing long short-term memory model to determine if the output accuracy is the same as previous results. Then the research team will add new customer activity features to analyze and expand on eBay’s previous research. The final step will involve the design of a different neural network model to run eBay’s new set of customer data through with the goal of eBay using the results to enhance the online purchasing process for its customers.
The model’s mathematical algorithm is trained to come to the same prediction as a human expert when provided the same information or data. These neural networks are built from interconnected layers of neurons with activation functions designed to learn to recognize patterns in the same manner as the human brain. Many recent advances in artificial intelligence were made possible by deep learning’s ability to identify patterns and classify different types of information which is crucial for processing vast amounts of data with minimal human input.
Korpusik, who joined LMU’s faculty in 2019, earned a bachelor’s degree in electrical and computer engineering at Franklin W. Olin College of Engineering in Massachusetts, and a master’s and Ph.D. in electrical engineering and computer science at Massachusetts Institute of Technology. “I am grateful for eBay’s initiative to collaborate with LMU’s Frank R. Seaver College of Science and Engineering through this research funding gift,” said Korpusik. “Often, professors may wait years to have research opportunities such as this one that will also provide valuable research experience for an LMU graduate student. It will be exciting if we can design a novel neural network architecture that improves outcomes over eBay’s current model, while also advancing what is capable with deep learning.”