Tuesday, 10 February 2009

generality of query terms

An excellent blogger, named Daniel Tunkelang, recently brought up a discussion, within a discussion, on determining the exploratory nature of a query. He questions whether this is worth it. This is a very interesting focus. Ryen White has published a number of papers on determining exploratory style queries, and the effect of expertese on search style, which are very interesting and certainly related to this challenge.

The excerpt that daniel refers to, from another blog, is on how search engines should react to the terms 'vietnam travel' in comparison to 'vietnam population'. For the former, Yahoo, Google, and Live all bring up different top results, but all based around travel guides. For the latter, Google and Live try to answer the question directly. All three link to the wikipedia page on vietnamese demographics.

Term generality is an interesting case in this example. The term 'travel' has a broader network in wordnet, than population. Naturally, the more generic a term, like travel, the more broadly it will be used on the web, and so naturally generic terms bring out less specific web results, or greater variation in the highest ranked pages. Live search provides 'related queries' for both, where the travel query has many query expansions, the population query provides a series of sibling queries, like korean population.

Instead, however, determining the generality of a term, or its breadth of use on the web, calls for a good opportunity to directly and intentionally provide diversity in search results. That is, instead of letter the breadth of use on the web naturally lead to varience in results, to specifically expose the varience in results, and aim to cover them.

From an exploratory serach perspective, its interesting because it would be different to interactive query expansion, in that the search engine would be providing key results from each of the recommended query expansions, and so the interaction would be different. It, instead, may convert to exploratory behaviour, rather than directed re-querying.


Daniel Tunkelang said...

I checked with Panos, and "excellent" is worth a healthy premium when applied to a blogger. Less glibly, thanks for the shout-out!

As I said in the discussion you cited, I think that automatically determining whether the searcher is in an exploratory mood suffers from the same problem as best-first search in general: it's just another kind of mind reading. Why can't we learn from Feynman as just ask?

Max L. Wilson said...

Looks like a fun book!

I really think we should ask. But depth of interaction and visibility would indicate that you go someway to showing the user what the result is. We don't typically ask a user do they want to interactively refine a query, and then show them possible options.

do you have the current time? yes.

in fact going the opposite way would be to say, this is an example of what comes up if you were to go with this refinement. we broke it down like this because its a very broad topic that you are going for.

it would be interesting to experiment with anyway.