Catalyst Search Matters Search Marketing Blog

This is the third and final installment of our assessment of the new Ask.com. As the resident linguist here, I will discuss some of the natural language technologies that have been foregrounded in the new interface. And, as a member of the paid search team, I will discuss the impact of the new design on paid search.

As a natural language processing geek, I am always eager to see new technologies that help guide people from the keywords they initially typed-in towards what they actually wanted. People’s initial queries may take some serious massaging before yielding the intended results. Translating one’s ordinary English desires into effective search-engine-ese, can be quite a challenge for your typical, non-elite search engine user (i.e., not nerds like us).

AskJeeves.com was originally built around natural language processing technology from Alta Vista, back before the tragic murder of dear old Jeeves by the Algorithm. The idea behind it, and the reason why it had “ask” in the name, was that it supposedly interpreted ordinary, natural-language questions and delivered relevant search results. The problem is that it is damn hard to provide results based on the answer to your question, rather than based on your question itself. When I type in “Which President is on Each Bill?”, I am not really looking for pages that contain some combination of the words “which”, “president”, “is”, “on”, “each” and “bill”. Rather, I am looking for pages with the text “$1 George Washington”, “$2 Thomas Jefferson”, “$5 Abraham Lincoln”, “$10 Alexander Hamilton”, “$20 Andrew Jackson”, “$50 Ulysses S. Grant”, etc. As we well know, the easiest way to find information on a search engine is to type in exactly what you’re looking for– that is, not your question, but your answer. Short of such miracles, the folks at Ask hoped to provide a more comfortable user interface by allowing people to “ask Jeeves” questions in the same way they might talk to a real butler, intern, or personal assistant.

In my opinion, the question answering module at Ask.com still has plenty of room for improvement, but you have to applaud their efforts in this challenging undertaking. Their question-answering superiority over the other search engines is highlighted in the new Ask.com platform, through the Narrow Your Search and Expand Your Search suggestions, which tend to funnel people into natural language, question-style search queries. They achieve some success in this area. Compare results for “When was Dodger Stadium Built?” on Google and Ask. While both Ask and Google provide relevant pages, Ask gives the desired answer right up front - 1962. Still, they’ve got a long way to go.

The Expand Your Search and Related Names options are great ideas, and provide something new and fresh that the other engines don’t. But expanding and projecting names are much harder to do than narrowing, and they don’t always succeed. Narrowing the search involves adding more words to the query to get more specific, and they can do this based off of query frequency. Expanding the search involves guessing which related things you may interested in, and it sometimes suffers from the same problems as Amazon.com’s “suggestions” based on your purchase history — when the statistical sample size is small, you end up with some very off-the-wall results. See my “drugs” query, which returned the drug Related Names “Crystal Meth” and “Angel Dust” (oh, Crystal, you are a such a cruel lover!). However, some of the time, the Expand suggestions are relevant and interesting, introducing topics that the user may not have known were related.

I got a particularly mixed bag for my query “What virus causes AIDS”, which returned, among other things, an Expand suggestion for “What virus causes Influenza”. The most important part of my query was “causes AIDS”, not “what virus causes”. This result is a case of Latent Semantic Indexing gone awry, projecting relatedness based on word co-occurrence rather than similar meaning. From this SERP, I instead followed a Narrow link for “Where did AIDS come from”, and I got the following less-than-relevant results:

Of course, over time, we can expect the Expand, Narrow, and Names options to improve and work better.

An important issue for us is what this means for SEO, and the answer is pretty simple. Fewer and fewer people will be limited to big, broad queries. This means that the long tail will become more important, as more people find more specifically relevant results for their more specific queries. Also, SEOers should consider trying to get their brands to appear in the Expand or Narrow sections. This can probably be achieved by being well optimized for the keyword that is being expanded or narrowed. Consider my search for “sleeping pills”. I got a lot of very viable refinement options, based on what I could have been interested in, including one Expand suggestion for “Ambien”. Sanofi-Aventis must be happy about that.

Now, on the the issue of paid search…

As I said in my discussion of Google Universal Search,

By complicating the layout of the search results page with Related links and multimedia listings, Google is further blurring the line between paid and organic search results. With these new, more complicated results pages, the branding value of a top paid search listing is now greater than ever.

This is even more true on the new Ask.com.

The most prominent features of their paid listings are:

Finally, I would like to reiterate something very important that Sarah said in Ask.com part 2. They are boldly departing from the Google et al. template, in an effort to establish their own niche. It is nice to see someone trying something new in search. Yahoo and MSN never going to close the market share gap by becoming better and better Google imitators. They will have to offer visitors–and advertisers–something unique. People are all different. Different people like different brands of things, and they have different proclivities. If Ask.com can differentiate themselves by providing a radically different search experience, then at least some people are eventually going to like them better… Maybe even me.