A Distributed Application Architecture for a Search Platform
EBay sponsored this post.
EBay’s global marketplace connects people with items they love. With so many choices in the marketplace, one challenge for our sellers is how to have their listings stand out and inspire buyers to make a purchase. In this post, we present a distributed application architecture built using event-driven systems, microservices, recommendation systems and data aggregation layer.
A good quality listing helps buyers get to know the product. It includes information about the item — ranging from the condition to shipping — through a descriptive title, item specifics, photos and a detailed product description. But creating an excellent listing is only half the equation for a successful sale — sellers must also surface their items to their most interested buyers.
Buyers search in two different ways: by keywords and/or by applying left-hand navigation filters to their query. Even if a seller creates a good-quality listing that would be a perfect match for a buyer, the listing won’t surface in the search and perform to its full potential if it doesn’t map to buyer tokens (or keyword terms) and filters. Sellers often try to improve the search performance of their listings by tweaking the wording to optimize their position in search results, but they may not be aware of the important keywords or filters buyers are using that might impact their item’s visibility.
Now, eBay sellers can access Buyer Demand Data in a simple and elegant way right in the listing flow, to help surface items that buyers will love. This insight into what buyers want helps drive seller engagement and increases the completion of item specifics, which in turn further enhances listing quality, visibility and sales.
Buyer Demand Data
Buyer Demand Data is the number of times buyers have searched for a particular item specific over a defined period of time. Buyers tend to use consistent search terms for used or commodity products, whereas for new or in-season items there might be some fluctuation before a search pattern emerges. Buyer Demand Data captures these trends and helps sellers target their products to buyers.
Some of the ways eBay sources the data include:
- Extracting item specifics from buyer keywords. For example, a buyer might use the search bar and type in “Apple Macbook Pro 2019.”
- Counting clicks on search filters; such as a buyer filtering search results by the item specific “Release Year.”
- Analyzing browsing/navigation patterns. For example, buyers browsing Electronics → Computers, Tablets & More → Laptops & Netbooks → Apple Laptops → Release Year.
Important Item Specifics
Important item specifics are the ones that buyers use more often in their searches and filtering. For instance, a typical buyer may search for a Macbook Pro using “Apple Macbook Pro 15” as keywords, and then applying filters such as Release Year, Processor and SSD Capacity. If we encouraged sellers to include this information in their listings, their items would have a higher chance of showing in the search results — significantly improving listing visibility and the overall conversion rate.
Architecture and Data Flow Diagram
The following diagram shows how data flows through the system. As buyers search for listings, their queries are analyzed and broken down into structured signals allowing downstream systems to capture buyer demand. This includes extracting item specifics from the keywords based on past searches and buyer behavior, using machine learning, as well as factoring in additional filters selected by the buyer.
For instance, analyzing the sample query “Apple Macbook Pro 2019,” we can tell that buyers are searching for Brand = “Apple,” Model = “Macbook Pro” and Release Year = “2019” in the “Apple Laptops” category.
These signals are tracked and stored in a data warehouse and subsequently processed as aggregated demand over time (e.g. 2,342,535 searches for the item specific “Model” in the last 30 days as shown in Figure 1). This aggregated demand is then pushed as a feed to transactional systems, for real-time guidance to sellers around important item specifics they should provide to maximize their listings’ visibility.
The real-time guidance includes showing Buyer Demand Data to sellers at the time of creating or revising a listing, as well as promoting the visibility of high-demand item specifics in the seller experience to drive upfront engagement and the item specifics completion rate. This is achieved via the microservice “Buyer Demand Data Service” in Figure 2, which acts as a single source of truth to deliver the demand data and important item specifics throughout the cross-platform selling experiences. This integration covers both the listing creation process and the post-listing management process, where sellers are provided guidance on existing listings that are missing important item specifics.
Sellers positively respond to this valuable information, resulting in them filling out more item specifics and engaging more deeply with the platform, ultimately creating higher-quality listings that drive conversion and sales.
Guiding our sellers by transparently sharing the marketplace dynamics has allowed them to improve their listings’ visibility in search, driving conversion which is 5% higher globally in the test group compared to the control group. Besides higher sales, we also observed a 15% increase in the item specifics completion rate globally during the same period.
Data-driven insights go a long way in connecting buyers and sellers. This experience is live on eBay’s web and native (iOS and Android) selling apps in all major markets, as well as available in our public Taxonomy API.
As a pure online marketplace, eBay is committed to the success of its buyers and sellers. By transparently sharing these dynamic demand insights, we have empowered our sellers to make the best use of the platform’s capabilities to grow their businesses.
Interested in pursuing a career at eBay? We are hiring! Please take a look at our current job openings.
Feature image via Pixabay.