October 26, 2007
@ 04:00 AM

Via Greg Linden I found the the summary of a recent survey 1,001 US adults conducted by Kelton Research. The summary was posted on Search Engine Land in an article entitled Report: 7 Out Of 10 Americans Experience 'Search Engine Fatigue' which states

The report discusses user frustration with clutter and the content of search results:

When asked to name their #1 complaint about the process, 25 percent cited a deluge of results, 24 percent cited a predominance of commercial (paid) listings, 18.8 percent blamed the search engine’s inability to understand their keywords (forcing them to try again), and 18.6 percent were most frustrated by disorganized/random results.

There was also a desire among many users that search engines be able to "read their minds":

Kelton asked survey respondents whether they wished that search engines like Google could, in effect, read their minds, delivering the results they were actually looking for. . . That capability is something that 78 percent of all survey-takers “wished” for, including 86.2 percent of 18-34 year-olds and 85 percent of those under 18.

That sounds like an argument for search personalization.

Search personalization is just one of the many ways to deal with end user frustrations with search results. For example, no amount of search personalization will be as effective as statistical analysis to discover and automatically fix mispellings in search queries (e.g. all the misspellings of britney spears Google has to deal with). Without spelling correction, a user will leave frustrated because they get few or no results when in truth they typed a misspelling into the search box.

The folks on the Live Search team have a first in a series of blog posts about how they tackle the problem of determining a user’s intent from their search queries entitled “Do what I mean, not what I say!” [Part 1 of 2]. It contains the following excerpt

We've been working on returning the very best search results for your intent, not just for the particular search terms that you happen to have chosen as a vehicle to transmit that intent.  There's an important difference there and it's been our focus for quite some time.

AutoSpell Correction

The first example of this is our new AutoSpell feature.

If we are absolutely, completely, totally, "no doubt about it" confident you misspelled one of your search terms, we automatically deliver a page that includes spell-corrected results, rather than a page of misspelled results accompanied by a "Did you mean _______?" link at the top.

For example, there's this pizza place near Microsoft called Pagliacci Pizza that is fantastic.  The problem is that I can never remember the correct spelling of the place.  My misspelled attempts are usually something along the lines of Pagliaci Pizza, Pagliaccis Pizza, or Paggliacci Pizza...

Stemming 

Another improvement in the "Do what I mean, not what I say" category is stemming.  Stemming means matching on the "stem" (or root) of the word rather than the exact word.

For example, users told us that the search half price book Redmond returned horrible results.  Searching for half price books Redmond produced much better results.   In our new release of Live Search, searches for half price book Redmond automatically include results with books in them as well.

Our team knew that tackling stemming would give us the improvements we needed for searches like these.  But we had to be careful, because you can't just stem all the time-you have to be smart about it.  An example of this is the word cable.  When you search for "cable," you could be looking for information on cable TV providers.  When you search for "cables", you could be looking for power, telephone, or network cables. 

These are just two techniques beyond just personalization which gives users better satisfaction with their search results and the impression that the search engine is “reading their mind”.

The main problem I have with personalization is that you need to give the search engine a private, personal information amount of information before it has tangible effects on search results. In the recent past, Marissa Mayer has pointed out that user studies have shown that location is the only significant factor which impacts perception of the relevance of search results when it comes to personalization. I suspect that true personalization will come from doing things like analysing my profile or my social graph friends list instead of the approach popularized by Google Personalized Search where previous search queries are analyzed.

Of course, we live in an era when 10% of the Internet population doesn’t see anything wrong with brain implants to connect them to the Web so maybe I’m being paranoid when I worry that the next major leap in search engine relevance will only occur after we allow search engines to spy on us.

Now playing: Busta Rhymes - I Know What You Want (feat. Mariah Carey)


 

Friday, October 26, 2007 6:50:12 PM (GMT Daylight Time, UTC+01:00)
Dare, I think you may be overgeneralizing Marissa's statement when you say, "Location is the only significant factor which impacts perception of the relevance of search results when it comes to personalization."

I was not at the scaling conference, so please correct me if I am wrong, but I thought Marissa had said that, of many demographic factors including age, sex, location, and education, location was most useful.

I doubt she was saying that targeting location is the only useful type of personalization given that, when talking about the more sophisticated personalization techniques Google already is using, Marissa once said, "[Personalization is] one of the biggest relevance advances in the past few years. Personalization doesn't affect all results, but when it does it makes results dramatically better."
Saturday, October 27, 2007 5:08:05 AM (GMT Daylight Time, UTC+01:00)
Perhaps other search engines have already provided as good of a result with stemming as we are going to get, and are now experimenting with personalized results. It seems like stemming should happen /before/ personalization, as it's a fairly tried and true method.

I'm a big fan of personalization as well giving privacy controls to the people. If you could see all the attention data being shared with a search results provider - and manage it - would that make you feel better about the privacy issues?
Saturday, October 27, 2007 6:05:48 PM (GMT Daylight Time, UTC+01:00)
I'd like to see search engines continue to push the envelope in terms of interpreting intent. You point out a few basics.

In terms of what I do today, I also look up what words mean or sometimes how to pronounce them using search engines. Sometimes I do lookups like I'm using an encyclopedia and sometimes I do map searches that go beyond how I'd use a printed phone book.

I think this trend can go farther. I'd like to see math pushed to the limits too. I wrote a prototype at www.TabletPCPost.com/math that takes simple handwritten math equations, recognizes them and formats them properly as text so they can be analyzed by a "search" engine.

These "search" variants can go even farther.

Comments are closed.