photobank.kiev.ua - Fotolia
Among those in the forefront of the invention of e-commerce is eBay Inc. While sometimes overshadowed by others, the company has helped forge a new generation of distributed data processing tools, at the same time become an important proponent of open source software. Its engineers have participated in Apache Hadoop and Spark projects, and the company has contributed its own projects to open source, including the Apache Kylin multidimensional analytics engine.
To find out more, SearchDataManagement spoke recently with Debashis Saha, vice president of commerce platform infrastructure at eBay, who leads a talented cadre dedicated to data engineering. He said Kylin is undergoing some changes, as eBay adds Spark to an arsenal once based more around the MapReduce data framework. He also tells us that e-commerce sites are undergoing a big shift from search-orientation to recommendation-driven designs. Below are edited excerpts of Saha's conversation.
How does use of open source software for data engineering relate to the overall philosophy of eBay?
Debashis Saha: EBay was founded to create more opportunity for everyone. We wanted to take that spirit to the way we did everything. The technology team here has a culture that is focused on how technology can be used to create more openness and opportunity. So we place a priority on open source for its importance to both the tech community and our own business.
We feel we are empowering our engineers by allowing them to be the decision makers in how eBay.com is built, and in turn, their contributions to open source are helping to accelerate and grow the tech ecosystem. Open source has become an important part of our DNA.
What is the purpose of Kylin? Can you place it in a business context for our readers?
Saha: Using Kylin, the team has been able to use real-time data to capture trends and insights about what shoppers are looking for on eBay. We have over 800 million items listed on eBay, and we're able to dig into the data and ensure we're marketing and surfacing the right products for our buyers. In addition, our insights also help our sellers make better decisions about what inventory the eBay shopper is interested in.
For example, with the Super Bowl, we looked at transactions and saw how merchandise for the Carolina Panthers and Denver Broncos compared and which items were hot sellers. More recently, we developed an [NCAA basketball] March Madness themed bracket using eBay data. Using Kylin, we were able to quickly determine the most-popular teams in the country from the opening of the college hoops season through March 6th.
How do you view your use of MapReduce in relation to Kylin -- is it fair to say there is a migration underway to more use of Spark with Kylin?
Debashis Sahavice president of commerce platform infrastructure, eBay
Saha: Yes, we have already implemented a Spark-based cube building engine in our 2.0 code base and this will be the default engine in Kylin 2.0. Another important area where Spark has an impact would be streaming integration. We are actively looking at Spark as the stream processing engine for near-real-time updates of the Kylin cubes.
Can you describe the shift you have seen from a search-oriented e-commerce platform to one that is more "real time," less batch-oriented, incorporating predictive and cognitive analytics?
Saha: There is a deluge of data, and e-commerce is leveraging it in a way that provides more relevant and personalized options for customers. At eBay, we're reinventing our data platform efforts to make it even easier to find things, compare items and surface exactly what customers need. We've embarked on a mission to structure our unstructured data that will allow eBay to be an open-sourced catalog with unique identifiers. Using state-of-the-art machine learning tools, we're organizing, categorizing and surfacing more searchable product pages in real time.
Tools like Kylin are making it easier for businesses to bridge the gap between the data and its business implications. This year, eBay started a business war room to look at the incoming metrics on its platform in real time so [analysts] may react to any shifts immediately. We have a team that can see what shoppers are looking to buy and determine where to quickly source and integrate this inventory in commerce ecosystem. Results can be surprising. During Valentine's Day, our data found that more shoppers were looking to antique and vintage styles of engagement rings, so we were able to quickly surface this merchandise to our buyers.
Learn how one CTO used open source data tools to serve ads in real time.
Check out an analytics experts view on open source.
Be there as Hadoop's creator celebrates its 10th birthday.