Do you know how Google PageRank works? Lots of people out there believe they do – many of them have made turned that understanding into a career, working in SEO to get their clients on the all-important first results page. You can have talmudical-level discussion with some of these people; no question that they know their stuff. Some have even gone on the lecture circuit or written books on how PageRank works.

But can they explain how it works in three minutes or less? Probably not – no, make that definitely not, if the websites of many of these experts are any indication. “What kind of question is that,” I hear an insulted SEO expert saying; “Nobody could explain such a complicated technology in three minutes or less!”

Oh yes they can – and they (or rather he) did just that last week, during the finals of the Israel Famelab contest, a part of the British Famelab, established five years ago in Britain, and expanded to several other countries, including Israel, two years ago. Famelab is a contest designed for graduate students and researchers to “help discover the new faces of science.” According to the rules, the candidate must take an interesting (and complicated) scientific topic, and explain it to a panel of judges in just three minutes (!) in their native language. The winners get a laptop and a free trip to England, where they are honored at the Chelthenham Science Festival.

The PageRank presentation was made by Ohad Barzilai, a candidate for a doctoral degree in computer science at Tel Aviv University. Here’s a rough translation of the highlights of how he explained PageRank (original Hebrew here)

“With PageRank, Google lets the internet make its own popularity standings. The standings are determined by links and connections between sites. The sites with the most links get the highest rankings.”

Okay, even the judges knew that part! But Barzilai goes into the gory details; Google’s bots, crawling around the net, mine the data for links and connections, and builds a database over 20 billion lines long (!), with all the link data between sites correlated. But once that’s done, Google then has to determine the quality of those links:

“Here’s an example: You are an employer interviewing two potential employees. One brings with him ten letters of recommendation, and the other brings only one. But that one was written by Bill Gates. Who would you hire, based on that information? Clearly the letter with Gates’ nod is more valuable. When someone giving a recommendation has an important reputation of his own, we give those recommendations more value. But what if you discovered that Gates’ had written 10,000 such letters? Would you still value his recommendation as you did before? Most likely not; as a person gives out more and more recommendations, those recommendations are worth less.

“That is exactly how PageRank works. Google has discovered an amazing thing: If they apply a mathematical equation called a diagonal lemma to the big list of link results, they are able to get a picture of the importance of the links and the sites themselves. The astounding thing is that Google has discovered, using linear algebra, to mathematically quantify the intuitive standings of popularity and relevance. It’s all done automatically without human intervention. This is what Google does, and they do it better than anyone else in the world does today.

Not bad for three minutes! PageRank a very complicated subject, as you can see from here, a college level course that spends a semester discussing it! Making sense out of PageRank in just three minutes is truly an accomplishment. By the way, Barzilai only came in second place for this – the winner discussed memory and face recognition! Some people fear that Israel’s educational system isn’t doing the job when it comes to training students for science and hi-tech. Looks like they just might be wrong!

Tel Aviv University researchers, led by Prof. Yuval Shavit, have developed a computer algorithm that has shown a 50% chance of identifying pop music stars based on file sharing networks like iMesh. The system, first developed by a grad student as his project for a master’s degree in computer science, analyzed 500 million requests by users for file downloads on the Gnutella file sharing network, and developed a list of “most likely to succeed” unsigned musicians and groups whose songs were being downloaded – without even listening to the songs. When the list was compared later to the U.S. Billboard popularity charts, as many as half the groups picked by the algorithm were found to have succeeded, either with a monster hit or a record contract.

For example, the system “discovered” Soulja Boy Tell’em and Sean Kingston weeks before their first hits even entered the Billboard Hot 100, and Shop Boyz before they were even signed onto a label. All three of these had grammy nominated/winning songs in 2007, and Shop Boyz Party Like a Rockstar was the number one downloaded ringtone in 2007. According to Prof. Shavit, there are many factors involved in success, including geography (where the artist is located, how often they play live, etc.) that the algorithm analyzes, and he believes that even greater accuracy will be possible with more work.

The grad student, Noam Kingstein, won a prize from TAU for the most commercially viable project, and at least one record company has already spoken to him. This is maybe the first time a system has been developed to allow record companies to profit from online music sharing, which has so far devastated sales.