OUseful Info

The End of Stateless Search? Google's "Previous Query" Refinement

We all know that web search is stateless, right? That is, you enter a query in Google and you get a result back just for that query? That is to say, the result is determined solely by the search term(s)...

Except it isn't - personlised search uses information the search engine knows about you to help tune the results... in Google's case, this could be from your previous search history, your web history, and I guess, in principle, the contents of your Reader subscription lists/attention profile, your Google docs, your email archive and your contacts list and so on...

(Okay, maybe you should regard personalised search as stateless too... and the personalisation stuff is just all part of the ranking magic...)

Anyway this moves things on: "Previous Query" Refinement Coming To Hit Google Results:

Last year, [Google started] changing the ads it displayed based on the previous query someone performed. For example, search for [spain] then do a new search for [travel], and you may notice how the ads will be targeted around Spanish travel (more...).

Google's never given this feature a formal name, but ... internally the company calls it "Previous Query" ... . Learn the name well, because Previous Query refinement is now coming to unpaid or "organic" search results...

For example, if someone were to search for [spain] and then [travel] after that, BOTH the ads and the organic results will be altered to take the previous query into account. To some degree, it will be as if the second query was for [spain travel].

Note that is also different in kind to a 'search within results' query, whereby the second search term may be used to tunnel down into the results for the first search term.

I've commented before how personalisation weakens the notion of "Google ground truth" results (Personalised Search), notwithstanding the fact that you might get different results from different Google servers from queries made around about the same time anyway, and this chips away at it even more?

For example, if you are using a public acces terminal, where maybe one user replaces another in quick succession, could your results be 'contaminated' by the results of the previous person's searches?

And what are the consequences, if any, for those of us with at least a passing interest in teaching info skills?

In a thematically relate post from SEO by the SEA, "How Search Engines Can Learn From Looking at Sequences of Search Queries", Bill Slawski introduces a new patent from Yahoo as follows:

Whenever someone searches at a search engine, they not only get information in response to their search, but they also provide information to the search engine about the things they are searching for - information which the search engine might find useful in helping other searchers.

If that searcher performs another search related to their first search, then the search engine might create an association between the two search phrases that the searcher used, if the two phrases appear to be related. If they perform a series, or sequence, of searches on a concept, then the search engine might take advantage of that information.

If a lot of people perform that first search, and then the same second search, or that same search within a search session, then the search engine might decide that the phrases are semantically related to each other. Knowing that relationship exists between search queries might help the search engine help people find things on the web, and it might help provide better advertisements from the search engine.

A patent application from Yahoo explores how the search engine might find semantically related terms by looking at queries searched for by people in search sessions, and describes some of the processes behind how the search engine might determine that phrases may be related to each other.

This approach is another step on from "previous query", essentially trying to find semantic relationships between consecutive search terms, and help the search engine 'intuit' what it is the searcher is trying to find (Jon Udell has written - and spoken - several times about the tension between being able to search for something effectively, and not knowing the terms you need to use in order to be able to search for it effectively... e.g. Hunting the elusive search strategy and Search strategies, part 2.)

Increasingly, I've been noticing search engines offering "did you mean..." and "searches related to" prompts based on my search queries, which presumably is comparing my single word query with the most popular two word queries containing that word (maybe ranked to take into account personalisation features...?)

Hmmm, i-spy "search within results" too...;-)

Interesting times...

Tags: google, search, previous query

Posted by ajh59 at May 29, 2008 05:37 PM