• Search
• Machine Learning
• Analytics and Optimization
• Trust and Reputation
• Grid Computing
• Social and Incentive Networking
• Emerging Platforms

While most people don't think of eBay as a search engine company, "finding" is fundamental to eBay. Millions of new items enter and exit the site every day, and it is crucial that our users be able to find what they are looking for, as quickly and easily as possible.

eBay's search engine is one of the largest and most complex information retrieval machines in the world. In order for the marketplace to work properly, all of the information flowing through the site needs to be updated and searchable in real-time. This includes the millions of items that enter and exit the site every day, along with every bid placed, every listing update made and every item purchased. To give you some perspective, each of the items listed on eBay is updated an average of five times during its lifetime.

No off-the-shelf product available today is capable of this job, so we had to develop our own search engine. eBay Research Labs developed a real-time search engine with the ability to scale indefinitely. It uses a flexible attribute structure and has built-in histogram functionality, allowing it to sort through the 105 million listings on the site at any given time - all in real-time.

eBay Research Labs is now working on the next generation of search technologies. Search quality and relevance are important areas of research. An initial version of relevant search is currently in use on eBay with the sort by "Best Match" feature, and we will soon be rolling out other versions on eBay.com. Through great strides we've made in applying machine learning to our search engine, search results will continue to become more relevant with each use.

The labs is also working on related areas like web-wide content aggregation, spam detection, content classification, content segmentation, productization, and high performance search.

Machine learning involves algorithms and techniques that enable computers to learn and gain knowledge over time. By identifying patterns and rules from massive amounts of data, we can build algorithms that can use this knowledge to make decisions. By building a feedback loop based upon user interactions and behaviors these algorithms are made to be adaptive. Adaptive Machine Learning is a focal area of research at the labs and influences future product offerings and features. These include areas of product classification, concept extraction, clustering, navigation, merchandizing, contextual advertising, and fraud detection. We are also working on areas of structured information extraction from web content, sentiment analysis and content summarization. Examples of machine learning in action include the relevant search currently available on eBay's new Best Match sort feature and the technology behind eBay's recently announced AdContext beta.

eBay marketplace session data contain a wealth of information. Efficiently collecting and storing this information and effectively processing it are critical. Analyzing this information in a quantitative and qualitative manner is essential for improving our user search experience. This includes the design of analytical methods that depend on user profiles, queries, click-through events and domain categories. This also includes the development of automated techniques for the identification of finer grain finding criteria such as user intention, pain points and search success or failure.

The other side of the analytics coin is using this information to help the sellers sell their items better. We are currently working on econometric models and optimization techniques to create a decision support system to build better tools for selling on eBay.

The eBay Feedback system is a unique currency for online trust and reputation. When someone finds out you're on eBay, one of the first questions they ask is "what's your feedback?" No other feedback system enjoys the broad success seen on eBay. When every item searched and every transaction become units of trust and reputation, finding new ways to establish and enhance it become prime. eBay Research Labs works on several areas of trust and reputation as applied to on-eBay and to off-eBay applications. We are exploring the deep questions of online identity, feedback, trust and reputation.

Building, leveraging, and sustaining the implicit and explicit trust and reputation system within the community is a key aspect of eBay marketplace. eBay Research Labs' work evolving trust and reputation systems, along with buyer and seller protection systems, will continue to create an open and friendly place for people to come together and transact.

eBay Research Labs is exploring new ways to demonstrate eBay's Power of 3 (the eBay Marketplaces platforms, PayPal, and Skype). We are looking at new application domains, new devices, new paradigms for finding, commerce, payment, communication, and trust and reputation.

Mobile applications are one such area. We are working on building novel user experiences and applications related to the mobile platform. Prototypical systems we've developed in the Labs include Mobile Relevance Search, a Mobile Shopping Network and a Deal Finder. In the Mobile Shopping Network we demonstrate a ticker based shopping experience based upon a user's product and category interests. The Deal Finder is an application that alerts users of good price deals on products they have expressed interest in via the mobile phone.

eBay, in a way, was the first online social network. It began as a place for people to buy and sell items of interest, and quickly developed into a place where people could share ideas, trade and communicate with people all around the world. The core of eBay's success is this community involvement and interaction. The Social Networking group in eBay Research Labs is exploring ways we can continue to foster community involvement and interaction, through research in social network analysis, affinity mining, collaborative filtering, social identity, trust and reputation, incentive networks, architecture for tagging and social search.

eBay operates tens of thousands of servers in locations all over the globe and coordinates their activities from a single location. We do this by employing a grid-style architecture. We have made incredible strides toward reducing the management costs of this infrastructure over the years, and we are now embarking on the next generation of tools that satisfy our needs.

In 2004, eBay started to formalize its efforts to construct a holistic enterprise wide architecture that achieves economies of scale through:

• Building a self healing network
• Automating problem detection and repair
• Automating management of the network
• Moving to service centric management
• Leveraging virtualization for efficiency/flexibility
• Building a horizontal layer of integration software (a meta-operating system) to turn a network of resources into a system on which you run services (applications).

To date, we've improved software development processes at eBay by creating an automatic code rollout tool for pushing new code onto the site automatically and have moved almost all of our applications onto the Grid. Currently, eBay Research Labs is working with various research organizations and standards bodies in an effort to further advances in grid computing at eBay and across the industry.