louisrosenfeld.com logotype

Home > Bloug Archive

Jul 11, 2006: Using locational context to beef up search queries

Here's another thought about search.

Search logs often tell us not just what a user searched, but which page they were visiting when they initiated their search. Knowing the locational context of searches can help us diagnose a variety of findability problems, such as sussing out navigational failure points.

But what about using locational context to enrich the query itself? In this case, a search engine would harvest key terms from the page where the search originates. These terms would then be added to the user's query to enhance the query.

For example, let's say a user finds herself reading an article deep in nytimes.com on Turkey, the country. She's an active traveler—and wants to know if the Times has some useful travel articles on Turkey—but she's also a lazy searcher (like the rest of us) and just submits "Turkey" as her search.

Normally, she'll retrieve documents about the bird, the latest Broadway flop, and other stuff besides the country. But by finding her way to very specific content, she's already established a narrower, more useful context, and told us much about her information need. Why shouldn't a search engine take advantage of such qualified information about the searcher, especially if it leads to more relevant results?

If we add terms from the page where her query originates—like Istanbul, sultan, Topkapi, Byzantine, and Ottoman—we might go a long way toward disambiguating her search results. (Entity extraction tools could be applied to figure out which terms to extract.)

Essentially, this approach combines a straight keyword query with "more like this," a capability common to many search engines which simply converts a page of text into a long query. Of course, as you design such an algorithm, I imagine it might be best to weight the keyword higher than the similarity factor.

Anyway, this idea is probably decades-old news for search engine algorithm experts, but I'm not one, so I'll throw it out here. Does the idea make any sense? Do you know of any examples that illustrate it?

email this entry

Comment: Peter Boersma (Jul 11, 2006)

As it happens, I discussed this issue with a client yesterday! I didn't go so far as suggest keyword extration tools, but instead we discussed the option of adding some of the metadata to the search query. In this case the meta-data would be the "target group" for which the current content was written, entered into the content-manegement system by an editor.

The problem is of course that you assume the reader really wants "more like this" and not "something else" (because she's stuck).

How do you suggest the user could influence this decision (before seeing the results)?

Comment: Lou (Jul 11, 2006)

Peter, neat approach, though definitely a bit more manual. But you rightly point out the problem behind the assumption. I think the user might be unwilling to influence this decision before seeing any results. Maybe this functionality would be available only when *revising* a search?

Comment: Eric Scheid (Jul 11, 2006)

I believe something like this was described as happening at the BBC

http://www.currybet.net/articles/day_in_the_life/index.php

Comment: Edward Vielmetti (Jul 12, 2006)

Lou, do you want these sorts of added search results to be presented in the same context as the usual search? Just thinking about presentation issues, I'd be really annoyed if I were looking at a story on one thing, typed in some search terms, and got weird results from some features unimportant to me extracted from the source text.

That said this sounds sort of similar to the Google Mail approach to throwing up ads/searches related to the content of the page you're looking at. What if...you had a sidebar that always had a feature-extracted search relevant to the page you're looking at? I'd probably get tired of it after a while but it would be a good way to test this theory.

Comment: Walter Underwood (Jul 12, 2006)

I would be really worried about this sort of silent bias confusing the user. What are they supposed to think when the same search gives different results on every page?

In the user testing at Infoseek, we found that people were not very good at getting out of "holes", either narrow categories or ineffective search terms. This seems like it digs the hole deeper just when the user is trying to get out of it to some other part of the site.

If you want related pages, list related pages in a sidebar. I wouldn't try to make search do that.

Also, beware of putting too much work into the polysemy problem (is "Java" about programming, an island, or coffee). I don't think polsemy is all that common in real queries, especially if your site covers a single topic area. Certainly, measure it before deciding it is an issue. I would guess that it shows up in far fewer queries than spelling mistakes. We don't want to do special things to search as a partial solution to a 1% problem.

Comment: Louise Hewitt (Jul 14, 2006)

I like Edwards view of creating an 'enhanced search' that leverages this kind of metadata but displaying it in a subordinate position alongside standard search responses. Anticipating users wishes is at best guesswork, albeit necessary, so generating confusion in search behaviour sounds a bit iffy.

As users would tend to use the shortest personal route to information, I'd be interested to see what happens if searching on the same terms turned out different results depending on the source page. Would users visit the source page and type the search terms again to re-visit the destination (as it is the route they know). This kind of 'misuse' would just mean adding extra steps to a user's path and not create an easier route than simply browsing the results.

A client of mine is in the practise of using this approach for contact us forms - they set the page to display a different form depending on the path taken to the page. Sounds neat, but if the user is unaware of what happened, and then revisits to find the form looks different without knowing how to recreate the original view, it could simply lead to more frustration.

How about an AJAX solution to filter search results based on associated keyword categories in a dropdown list once you reach the results? OOo, hang on, I might have to patent that :-0

Comment: Catherine van Zuylen (Jul 18, 2006)

BTW, Intellext has a sidebar product that "watches" what you're doing and recommends other content for you. (http://www.intellext.com/index.html)

[No, I'm not associated with Intellext, although I know some of the nice people over there. I'm associated with Inxight, which makes entity extraction and search extension solutions]

Comment: Austin Govella (Jul 20, 2006)

If you could be more sure about *why* the user was at that page, then I'm all for it, but I'd be worried that the user had arrived at the page by mistake and was searching to correct their path.

Say they were looking for the bird, but clicked on an article about the country.

Perhaps if you could weight meta data from that page along with data from all of the other pages theyd visited in the site (and maybe weight content vs. nav pages differently), then it seems that would be better.

Add a Comment:

Name

Email

URL (optional, but must include http://)

Required: Name, email, and comment.
Want to mention a linked URL? Include http:// before the address.
Want to include bold or italics? Sorry; just use *asterisks* instead.

DAYENU ); } else { // so comments are closed on this entry... print(<<< I_SAID_DAYENU
Comments are now closed for this entry.

Comment spam has forced me to close comment functionality for older entries. However, if you have something vital to add concerning this entry (or its associated comments), please email your sage insights to me (lou [at] louisrosenfeld dot com). I'll make sure your comments are added to the conversation. Sorry for the inconvenience.

I_SAID_DAYENU ); } ?>