eBay Research Engineers Visit CDS
Aditi Nair and Daniel Galron discussed ranking recommendations and search results with CDS students
CDS students at last week’s Company Information Session heard about life at eBay from Aditi Nair, Research Engineer and NYU CDS graduate, and Daniel Galron, Technical Team Lead and NYU CS graduate.
Nair explained the research engineer role, which she said intersects data science, data analytics, and data engineering. She emphasized that research engineers are involved with the full end-to-end process of prototyping, developing, and productionizing algorithms. Her role at eBay focuses particularly on ranking item recommendations. For banners like “Frequently Bought Together” on the website, many algorithms are needed to populate results. Nair has to address the problem of how to rank the results of different algorithms and determine which data science or machine learning solution best suits the problem.
Her workflow requires a decision regarding these questions which leads to the development of data science, machine learning, or deep learning models. Once a model is developed, she is responsible for writing production modeling code, designing live traffic experiments to facilitate launch decisions, and eventually developing applications for real-time traffic. A fundamental skill to succeed in this process, according to Nair, is the ability to frame business problems as data science problems.
Galron works on machine learning solutions for search systems — on eBay, most purchases are made through search. His team engineers search features which consider personalization, price references, and product aggregation. Some of the main challenges he faces include aggregating similar items based on unreliable labels (vendors label their own products on eBay) and addressing presentation bias, the paradox that new models are trained on user behavior, but user behavior reacts to prior models, a pervasive search problem not unique to eBay.
One of the most interesting problems Galron works on is query understanding, which refers to interpretations of user intent for a search query. Models attuned to user intent can increase the probability someone will buy a product, which serves the ultimate goal of building a search platform — maximizing revenue.
Galron and Nair concluded with a Q&A and offered earnest advice to prospective hirees for interviews and resume-building.
By Paul Oliver
