A Forgotten Search Patent: Revenue Loop
Paul Graham discovers he was granted a patent in 2003 for a search algorithm that ranks results by bid price multiplied by transaction count, effectively combining relevance and revenue. He recounts a 1998 meeting with Yahoo where the idea was dismissed because advertisers were already overpaying. The patent eventually issued, but lawyers mangled his clear prose into incomprehensibility. A historical look at early search-ad optimization.
March 2006, rev August 2009A couple days ago I found to my surprise that I'd been granted a patent. It issued in 2003, but no one told me. I wouldn't know about it now except that a few months ago, while visiting Yahoo, I happened to run into a Big Cheese I knew from working there in the late nineties. He brought up something called Revenue Loop, which Viaweb had been working on when they bought us.
2006年3月，2009年8月修订。几天前我惊讶地发现，我竟然获得了一项专利。专利于2003年颁发，但没人通知我。要不是几个月前在雅虎偶遇一位九十年代末一起工作的老同事，我至今都不会知道。他提到了一个叫“收入循环”的东西，那是Viaweb被收购时正在做的项目。
The idea is basically that you sort search results not in order of textual "relevance" (as search engines did then) nor in order of how much advertisers bid (as Overture did) but in order of the bid times the number of transactions. Ordinarily you'd do this for shopping searches, though in fact one of the features of our scheme is that it automatically detects which searches are shopping searches. If you just order the results in order of bids, you can make the search results useless, because the first results could be dominated by lame sites that had bid the most. But if you order results by bid multiplied by transactions, far from selling out, you're getting a better measure of relevance. What could be a better sign that someone was satisfied with a search result than going to the site and buying something? And, of course, this algorithm automatically maximizes the revenue of the search engine.
这个想法基本是：排序搜索结果不依赖文本“相关性”（当时搜索引擎的做法），也不按广告主出价排序（Overture的做法），而是按出价乘以交易次数排序。通常这适用于购物搜索，但我们的方案一个特色是可以自动探测哪些搜索是购物搜索。如果仅按出价排序，搜索结果可能变得无用，因为靠前的条目可能被出价高但质量低的网站占据。但若按出价乘以交易数排序，非但没有出卖用户，反而得到了更好的相关性衡量。有什么比用户访问网站并购买东西更能证明他们对搜索结果的满意呢？而且，这个算法自然也会让搜索引擎收入最大化。
When Yahoo was thinking of buying us, we had a meeting with Jerry Yang in New York. For him, I now realize, this was supposed to be one of those meetings when you check out a company you've pretty much decided to buy, just to make sure they're ok guys. We weren't expected to do more than chat and seem smart and reasonable. He must have been dismayed when I jumped up to the whiteboard and launched into a presentation of our exciting new technology. I was just as dismayed when he didn't seem to care at all about it. At the time I thought, "boy, is this guy poker-faced. We present to him what has to be the optimal way of sorting product search results, and he's not even curious." I didn't realize till much later why he didn't care.
雅虎考虑收购我们时，我们在纽约与杨致远开了一次会。我现在才意识到，对他而言，那只是走个过场——确认一下这家基本上已决定收购的公司是否靠谱。我们原本只需聊聊天，表现得聪明理性即可。当我突然冲到白板前，开始介绍我们激动人心的新技术时，他一定很沮丧。而他对我的演示毫不在意，也同样让我沮丧。当时我想：“天哪，这家伙真是面无表情。我们向他展示的可是近乎最优的产品搜索结果排序方式，他却连好奇都没有。”很久以后我才明白他为什么不在意。
In 1998, if advertisers paid the maximum that traffic was worth to them, Yahoo's revenues would have decreased.
1998年，如果广告主按流量实际价值支付最高费用，雅虎的收入反而会下降。
Things are different now, of course. Now this sort of thing is all the rage. So when I ran into the Yahoo exec I knew from the old days in the Yahoo cafeteria a few months ago, the first thing he remembered was not (fortunately) all the fights I had with him, but Revenue Loop. "Well," I said, "I think we actually applied for a patent on it. I'm not sure what happened to the application after I left." "Really? That would be an important patent." So someone investigated, and sure enough, that patent application had continued in the pipeline for several years after, and finally issued in 2003.
当然现在情况不同了，这类做法已蔚然成风。所以几个月前我在雅虎餐厅偶遇那位老同事时，他首先想起的（幸好）不是我们曾有的争执，而是收入循环。我说：“嗯，我们好像为它申请过专利。但我不确定我离开后申请怎么样了。”“真的吗？那会是一项重要专利。”于是有人去查了一下，果然，那份专利申请之后又持续了好几年，最终在2003年获得了授权。
The main thing that struck me on reading it, actually, is that lawyers at some point messed up my nice clear writing. Some clever person with a spell checker reduced one section to Zen-like incomprehensibility:
Also, common spelling errors will tend to get fixed. For example, if users searching for "compact disc player" end up spending considerable money at sites offering compact disc players, then those pages will have a higher relevance for that search phrase, even though the phrase "compact disc player" is not present on those pages.
(That "compat disc player" wasn't a typo, guys.) For the fine prose of the original, see the provisional application of February 1998, back when we were still Viaweb and couldn't afford to pay lawyers to turn every "a lot of" into "considerable."
实际上，读到专利文本时最让我震惊的是，律师们把我原本清晰的行文弄得一塌糊涂。某个聪明人用拼写检查器把一段话改得跟禅宗公案一样晦涩：
“另外，常见的拼写错误也会被自动修正。例如，如果搜索‘compact disc player’的用户在提供‘compact disc player’的网站上花费了大量金钱，那么这些页面对于该搜索短语就会获得更高的相关性，即使‘compact disc player’这个短语并未出现在那些页面上。”
（那个“compat disc player”并不是拼写错误，各位。）想欣赏原始的优美文字，请看1998年2月的临时申请——那时我们还是Viaweb，还付不起律师费，用不着把每个“a lot of”都改成“considerable”。