Google and Search Relevance

Daniel Tunkelang gave a nice talk at Google yesterday on search relevance.  He posted the slides online.  I wasn’t at the talk, but know Daniel from some previous run-ins.  

I particularly enjoyed his comparison between Google and McDonalds.  I drew a connection between those two companies over at Jeff Jarvis’s blog a while ago for a slighly different reason.  Jeff is a sharp guy, although he may have gone off the deep end with his new book, What Would Google Do?  I’ll have to buy it and find out.  Those things don’t sell themselves, so one expects to choke down some hyperbole, but alluding to a similarity between Google and Jesus crosses over into burlesque.  Its just a company.  A clever one, for sure, but history is replete with clever companies and the small revolutions they caused.

The comparison between Google and McDonalds on the other hand has some merit.  Both are innovative companies that ought to be admired for some of what they’ve done, but neither can be accused of increasing the public’s appetite for fine food or HCIR (haute cuisine information retrieval).  Twenty or thirty years ago McDonalds was using census data and ariel surveys to determine the best location for franchises — demographics, traffic patterns, and suburban expansion.  Then there’s writing the book on fast food and franchises. 

Daniel has a few criticisms of Google’s search engine.  Consider his example on slide 23, where he searches for “IR” and demonstrates the poor results.  Just about any Googler will know that ” ‘Information Retrieval’ ” would be a better search phrase, and if they don’t they’ll learn it soon enough.  So I think this first example is not quite fair.  The example on slide 31, a search for “Steve Pollitt”, is a better criticism.  He might have added a term for research or publications, but the results would not be much improved.  Or he might do exactly what he did: go to Rexa.info where he can get an experience tailored to academic researchers.  It would be wise for Rexa to allow Google to spider their content, as LinkedIn and ACM do.  I wonder whether the best solution is when Rexa or a site like them appears high in Googles results list, and the entire experience can be tailored to the domain, for example the sidebar that lists co-authors.

So for a one-size-fits-all search engine, as Google’s web search aims to be, I’m not convinced faceted navigation is a good option.  In the case of Steve Pollitt, I don’t trust algorithms to accurately distinguish individuals.  Witness spock.com.  Faceting along other lines might wall me off from good information.  Its a rare name, I may find a personal blog where he rants about the impact of the internet on democratization in developing countries or some such thing, something I could easily skip by if I were to let myself be guided by algorithmically derived facets or even human tagging.  If the results are too noisy, Google is fast enough that adding some constraining terms is a quick process.  Products, such as the ceiling fans and shoes that Daniel demonstrates later in the talk, are simple objects with obvious facets to aid in navigation, and generally I want to navigate among options, so there its a no brainer.  Not so with most of my Google searches.  So I’m happy with its simplicity.  I enter words, it gives me back things that have the words, weighted by a variety of factors including PageRank.

I think however that methods for searching news can benefit from what Endeca has done with some of their clients.   Its a bit of a different beast, since the value of a news article or image declines rapidly with age, and the volume is quite small.  But when searching old news, admittedly a rare usage pattern, the benefit would be substantial.  I was recently searching Google News for old items on Barnard Madoff.  All I got was an ugly list.  Its an area where PageRank is apt not to help much, as with Google’s web search product, so I would have loved the sort of help that faceted navigation could offer.

One of his quotes did however ring a bell with me.  On slide 9 he quotes a Fortune 500 CEO who said that “Search on the Internet is solved”.  It reminded me of a quote from Lord Kelvin in 1900, “There is nothing new to be discovered in physics now.”  That was just when things started to get interesting.  Back then, physics at least appeared to be stable and reaching maturity.  With Maxwell’s equations, one gets a feeling of completeness, that there is nothing more except a reduction to practice.  Perhaps Kelvin can be excused.  So it is interesting that only a few years of looking at the same search page might make one think that search is a solved problem.  More revolutions will come.