I'm finally getting around to putting up some notes about the CIC CIO TechForum. The best technical talk I attended was one on Search Analytics by Richard Wiggins from Michigan State. I was on the planning committee for the TechForum and I really pushed for this one.
According to Rich, he's been doing this since 1999, when their Altavista search engine began "cracking at the seams" (not scaling, returning confusing answers). What he and his team decided to do was to develop "Best Bets" where they examine what term the Web site visitor looked for (e.g. "campus map" or "admissions") and looked at the results which were successful. They then drew a frequency curve of the most used terms in searches. As you might expect (and if you've read my stuff on The Long Tail), the frequency distribution is a Zipf/Pareto curve. While this is indeed a "Long Tail" distribution, the benefits come from dealing with the "Short Head" of the frequency curve. Rich's analysis shows that only 500 terms constitute over 50% of the searches. By concentrating on the "Best Bets" in those 500 terms and their results, his team optimized the search on their "Best Bets" page. Some of his other findings include that search frequency is seasonal (think admissions, arrival, graduation, etc.) and that having a date range on the search is valuable.
Rather than steal Rich's thunder, I'll point you to his TechForum presentation which contains much more information than he could deliver in 20 minutes. He's also working on a book with Lou Rosenfeld called Search Analytics:Conversations with your Customers. They have a Web site for the project and you can read the first chapter there.
Leave a comment