One Basic Problem with Algorithmic Search

A short while back, we had the opportunity to interview Google’s Shashi Seth. This interview started with a fascinating look at basic flaws in algorithmic search. In fact, it is these limitations that has led Google to implement the Google Co-Op program.Basically, this boils down to two major issues:

1. Most user queries to not fully explain their context
2. Even if the user queries do explain their context, most web pages do not present data indicating what context they intend to address

Since many of you are probably going “Huh?”, let me explain with an example. If a user searches on “diabetes”, the search engine has a few possibilities to deal with:

1. You are looking for treatment information
2. You are a doctor looking for research information
3. You are a drug designer at a pharmaceutical company looking for drug trial data
4. You are a medical authority looking for related regulations

What makes the problem worse is that even if you type in a more specific query, such as “diabetes information for patients”, it’s hard for the search engines have a hard time using this context data to find the best authoritative resource.

One of the major ways that search engines deal with this is by deliberately offering up a diverse set of results (if we don’t know if you are a doctor or a patient, let’s make sure both types of results are available high on the first page …). It’s a workable solution for now, but Google is looking to improve on this.

There first effort was launched in May of 2006. Google invested heavily in their Topics and Subscribed Links programs. A simple search shows how Google Topics actively tries to address the concerns with search expressed above:

The links between the sponsored links and the first search results are Google Topics in action. You can see how the Topics provided includes specific contexts that will, in theory, make searching easier for the user. However, the program was not a success, because it relied on human editors to guide the output of the context filters, and the motivation for the human editors was unclear.

Thus was born the Custom Search Engine program. This program is still designed to solve the same problem. Google is looking for people that will design vertically oriented search engines for specific contexts.

The big difference is the AdSense revenue sharing. You can get paid for your work. While it remains unclear how much you will own, we now have a much more promising value proposition for tweaking search results for different contexts.

You can, of course, ask the question as to whether or not the concept works. So let’s look at an example, by comparing the results of a search engine designed to provide medical information to patients, and a search engine designed to provide medical information for doctors.:

You can see how different the results are. The one on the left presents data targeted at patients, and the one on the right presents data targeted at doctors. This is the power of human editing, addressing the basic contextual problem of search. If you like, try the patient and doctor Custom Search Engines yourself to see how they work. We don’t claim that they offer perfect results (yet), but they do illustrate the concept.