By: eBay News Team
A minute before conferences were delayed or moved online, Gilad Fuchs from our research team has presented one of our hottest projects, in the 6th Data Science Summit.
Hi Gilad, tell us what you do in eBay.
I am a research scientist on the selling group. Our job is to help sellers improve sales on the platform and boost sellers experience. We develop tools for improving seller abilities, like product pricing or product promotions in search results. Our tools enable sellers to make better decisions.
What interesting project did you talk about in the conference?
I talked about Ad Rates. We wanted to find a way to help sellers set the correct price for their ads, compared to the items sold by competitors. In order to do that, we needed to understand what similar items are going to appear when you do a certain search. It’s challenging, because there are huge amounts of data, and most data is unstructured, so sellers and buyers use different keywords for item descriptions.
We based our work on the Word2Vec model, in order to identify items that are similar. It wasn’t enough, because the “weight” of the word in a sentence is very important. Other than words that describe the product, sellers use marketing words, such as “free shipping”. If these words have the same weight as the product’s name, for example, then “matching” items that come up when you search, won’t really match.
Our solution was scoring the words in the product’s title, according to search words that are being used. We’ve developed a new model, called Listing2Query, which can score any word in the product’s title according to the chance that it will appear in the search terms. This score is the “weight” of the word, which enables us to distinguish between important words and “noises” and identify similar products. After the model was tested and validated, we started using Listing2Query in other places on eBay, like the Search group.
Gilad, this sounds fascinating. Thanks for sharing!