Interview with Become's CTO Yeogirl Yun: Communities, Mobile, the AIR Algorithm Advantage
Yeogirl Yun has long been on my interview wish list. His work at WiseNut, MySimon and now Become.com have shown him to be a superior architect in the shopping search space. Spend some time with Become if you haven't before.

Its usability and smart design attest to me of the quality of Mr. Yun's thought and his internal tech leadership.

I had the pleasure of interviewing him and learned a great deal about where Become is now and where it's headed. Mr. Yun shares some possible vertical directions, as well as offering the intriguing glimpse of future innovations that involve shopping communities.

I hope search marketers find this article useful for its look into where shopping search is headed, and that search engine developers find it useful for the expertise Mr. Yun shares.

General:
1) Question: What would you estimate is Become's shopping search share?

Answer: We launched just over a year ago and Become.com has already become one of the top comparison shopping sites and, according to ComScore, the fastest growing. Right now we have over 1.5 million users visiting us every month, and are looking forward to especially rapid growth during the upcoming holiday shopping season.

2) Question: What metrics are most useful to you in determining Become's success? Can you discuss where those metrics are now in relation to a year ago? Where would you like them in a year from now?

Answer: We have two key external groups to serve. First are our users, and site traffic and merchant clickout rates are good measures of the job we are doing.

The better we serve our users, the more often they will come back and the more people they will tell about us. In addition, when we do a great job of helping users through the research and shopping process, they are more likely to clickout to the merchants participating in our comparison shopping service.
The second group we serve is our merchant partners. They want highly qualified leads, so we look at both conversion rates (as reported by merchants) as well as bid prices. As we continue to create greater value for merchants, they are increasingly willing to bid up CPCs for the high ROI traffic we deliver.

One year ago, we had just launched our combined Research and Comparison Shopping, and we’ve seen huge traffic growth coupled with corresponding growth to merchant leads and CPCs. We’ve grown significantly every quarter and the upcoming holiday season should see an even more rapid expansion.

3) Question: Where is Become in its development? Where do you see it in five years?

Answer: We have built a solid platform with our AIR™ ranking technology, 3.2 billion page index and comparison shopping. We also have some very innovative features such as dynamic suggestions and our new Search Zoom feature and we are constantly innovating.

For shopping, our goal is to be the first destination people choose when they are looking to research and buy products. In five years I’d love to hear people say “When I want general information I go to Google, but when I’m looking for products to buy I go to Become.com.” AIR™ is a highly scalable vertical search technology, so, in the future, you may see us expanding into other vertical search areas like travel and health.

Also, we are looking beyond the U.S. We recently launched our site in Japan (www.become.co.jp) and have created a globally expansible technology platform. (see SEL's interview with Become Japan's CEO Masahiro Ueno)

technology:
4) Question: building a crawler using Java - how and why is this a big deal to the developer community? what languages are most crawlers built in? do you use Java in your new developments?

Answer: In hindsight, it was not a big deal to write a crawler in Java.

But back then (2 1/2 years ago), most crawlers were written in C++. (I have still not heard anyone claim that they have written a crawler in Java that crawls more than billions of web pages.)

You can write a simple crawler that crawls a million pages in a school project. But in order to scale it up to a billion pages, there are serious architectural and implementation challenges which demand a very high efficiency in IO and CPU usage.

We overcame those challenges with an innovative distributive architecture and ingenious algorithms which take advantage of Java’s clean-architecture style of programming benefits (which help avoid debugging-intensive tasks such as detailed multi-threaded implementation and memory management).

Java has advanced quite a lot ever since it was introduced and remains one of the best programming languages when it comes to a serious, scalable architectural problem. We also use Java in a variety of other projects, including our next-generation crawler platform called CyberShuttle. C++ and Java are very complementary and bring unique merits. We use about equal amounts of C++ and Java.

5) Question: You developed the crawler in - nine months was it? (it was actually three months) - this seems quite fast. Can you describe that process? Had you developed at that pace in the past? What would have been a reasonable/average pace for that project?

Answer: At Become, every project moves at lightning speed. With only about a dozen engineers, we completed development of a fully functional search engine in nine months. The crawler was operational in three months . We hired an extremely strong team that worked exceptionally hard to meet these deadlines. As a comparable, I recall reading a BusinessWeek report describing how Microsoft spent 20 months and $150M to develop their search engine (http://www.businessweek.com/technology/content/feb2005/tc2005022_6663_tc024.htm)

directions:
6) Question: In what ways are you leveraging user data to increase the value of Become to your users and advertisers?
(I'm following up on this one -G)

7) Question: in what ways is mobile/local search in Become's future? maps?

Answer: Local and mobile are highly interconnected. Local becomes much more powerful once it’s on a device you always have with you. We have already launched a Nearby Stores feature that lets users find local stores that have the products they are searching for.

Example: http://www.become.com/shop?q=ipod&pid=206234272&mid=2048&zip=94043

We have also worked to make Become.com accessible on some of the leading-edge mobile devices like Treo and BlackBerry. Our CEO Michael Yang uses Become on his Treo all the time to check online prices and information before making offline purchases.

8) Question: What are your thoughts on tracking buying behavior and providing product recommendations, or sending out pricing alerts? In other words, why can't users "join" Become for a different level of services?

Answer: First, we believe that personal privacy is very important, and we do not collect personal data without consent. We have, however, recently been investigating the power of online communities and their impact on shopping. You can look forward to announcements in this area in the future.

9) Question: Your SERPs have no entry point for my personal experience, whether that's experience with a product, a vendor, an information source or even the collection of links and text you provide. Why is this?

Answer: Many companies have tried to get their users to enter product reviews, rate vendors, etc. We initially focused on helping people read not only the reviews of Become users, but also all of the valuable reviews located across the web. To this end, we have used our search engine to crawl over three billion pages of shopping-related content and provide an extremely comprehensive set information that our competitors can’t match.

10) Question: What are your thoughts on price prediction technologies such as http://www.farecast.com/? For what products in Become would this make sense? How does this or doesn't it fit in with Become's mission?

Answer: The issues on price prediction are mostly around scalability and market forces. Each type of product has different rules that govern how prices change over time. A company like farecast is trying to back work the systems used by the airlines, but for many items there isn’t a system per se, just market forces. In a particular vertical like travel though, price prediction would fit very well with Become’s mission which is to help people make ideal buying decisions.

11) Question: Are you familiar with mpire, which shows average sale prices for products on eBay, giving shoppers the "true value" of a product? How would this or wouldn't it fit in with Become's mission?

Answer: If you have enough buyers and sellers for an item to create something close to “perfect competition,” you arrive at the true value of the product. So Become is able to offer “true value” information to our users.

Become has multiple merchants for most top products and these stores sell thousands of each item, so the prices naturally gravitate toward the “true value.” The value we show is also a more complete measure since it reflects the fact that people will pay more to some sellers than others.

Shoppers take things like brand value and quality of service into account.

12) Question: discuss the split screen for research and shopping: how did this idea evolve and develop? how do you determine if it's successful or not?

Answer: We wanted to help shoppers get all the information that they need in one place. Showing them great web search on what/how to buy alongside comparison shopping to determine where to buy was a natural integration. Since the launch, we have done user testing and focus groups to continue to improve this experience.

Our first version just had one column of products on the right side of the screen which some users mistook for static ads since that’s where Google and many others place ads. So we added price and brand filters along with a second column of products. With that change, users clearly understood both the integration and the value it provides.

AIR:
13) Question: What kind of competitive advantage does AIR give Become?

Answer: AIR is a hidden jewel in the search industry. We have not done extensive publicity yet due to our patent and IP strategy. AIR gives additional dimension to the existing link analysis algorithms. It is not just an incremental improvement over PageRank.

AIR is a systematic framework, founded on a sound mathematical model inspired by physical phenomena, to allow inclusion of vertical context and thus differential authority to the analysis of link structure on the web. Google’s PageRank projects authority purely based on the popularity of a web page derived from its static link structure.

Become’s AIR projects authority based on both the context of a web page and differential authority it receives from its dynamic link structure. I say “dynamic link structure” because AIR treats link property dynamically rather than staticly.

14) Question: Does AIR work better on the more structured data that you analyze for shopping than the "wild web" that web search engines provide?

Answer: AIR is only applicable to the “wild web.” For more structured data such as merchant data feeds, a conventional IR technology is well suited.

15) Question: In other words - what's stopping Google, MSN, Yahoo or Ask from copying your AIR concepts if they truely are superior in terms of quality?

Answer: It is the patent. We have already filed two patents around AIR technology and are doing our best to protect our core intellectual property. I am sure they will immediately appreciate the superiority of AIR as it becomes widely known. However, it is very difficult for them to just copy the algorithm and use it in their search because AIR is a vertical search algorithm. It is not very applicable to general search engines such as Yahoo, Google, and MSN.

making money:
16) Question: what percentage of Become's income is AdSense and what is direct from merchants? what is your most profitable/strongest vertical for direct sales? how about for AdSense?

Answer: The majority of Become’s income comes from merchants that participate in our comparison shopping service. AdSense links appear on the bottom of our pages and serve as a backup source of revenue.

Top categories of products include Apparel, Home&Garden, and Electronics.

17) Question: you guys have a history of flipping your projects. how does this affect your approach to building your business? what internal processes and infrastructure do you put in place to ensure that your project thrives after you sell? are there conscious processes you follow?

Answer: Great technology and businesses are often acquired. We are committed to grow this business in the long term, but at the same time we have to do what’s best for the shareholder’s interest.

In case of MySimon and WiseNut, I believe acquisition was the right choice. At Become, we have a solid management and engineering team. We also have very refined and rigorous engineering processes for development, deployment, documentation, and code releases. I am very confident that even in case of a potential acquisition, the business and its technology at Become will thrive under a right guidance and leadership of the acquiring company.

personal:
18) Question: what drew you to study law in Korea after selling WiseNut? was it hard to leave your studies and do you wish to pursue law further?

Answer: I had some interest in studying law when I was getting my undergraduate degree. After the sale of Wisenut, I had to leave the US due to my visa status. 2002 was not a very good year to start a new venture given the economic downturn and Internet bubble burst. So I thought it was very opportune to study law.

I enjoy studying for the sake of studying. Studying helps me keep my mind and body healthy. In Korea, law study is a big deal. People spend several years in a Spartan study program to pass the exam. I only spent 14 months in studying. I have not passed the test, but I learned a great deal during these 14 months. I was quite happy with my test score, which was only slightly below the cut-line.

I occasionally have the urge to go back and finish the test, but I think it’s primarily based on the pride I take in doing things well. I intend to continue my endeavor in technology innovation and entrepreneurship.

I used the following interviews and writings to develop my questions for Mr. Yun:
Q and A With Become CTO Yeogirl Yun
Become.com's Web Crawler: A Massively Scaled Java Technology Application
Competing with Google
Starting Up, Post-Bubble 2005
Become.com Founders Aiming for Google-Sized Success
Interview With Michael Yang of Become.com
Become Launches Filtering Feature Called Search Zoom
Garrett French - MarketSmart Interactive | permalink | +marktd |