Query classification; understanding user intent

http://vervedevelopments.com/Blog/query-classification-understanding-user-intent.html

What exactly is the function of a  search engine? In simplest terms it acquires, stores and returns  information (from the web). Ok, simple enough. But we're talking  about people here, people seeking information. How they  interact with the search engine is often a huge problem. What is the  user intent? A good way of starting to pick apart that puzzle is by  classification of query types. And that's what we're going to be  looking at today.

To understand what a user truly wants  when searching for something you'd need to ask each user what it is  they are after. While that works for the offline mom and pop store,  it isn't at all feasible for a search engine. Thus an automated  approach needs to be taken. What's more limiting, is that they have  to infer intent from very few words and little in the way of  (explicit) interaction.

Why would that be important to SEOs?  Well, that's even easier. If we understand how people search for  things, we are in a far better position to actually target our  programs to ensure the highest, most relevant, levels of traffic for  our sites and our clients. Understanding how search engines attack  the problem, can be VERY useful in our own targeting and programming.

The basics of query classification and beyond

One of the reasons a search engine  looks at classifying queries is to better understand user intent. To  do that they will look at the search task process as such;

• Enter query
• Retrieve results
• Scan results
• View results (actual pages  returned)
• Refine query (if needed)

Interestingly searchers have certainly  evolved over the years beyond mere information needs, into commercial  and navigational (seeking known entity) as well. We also can consider  that, unlike say.. a library, web searchers are looking for a wide  variety of mediums (text, images, multimedia) as well as from various  locales (work, home, mobile). Search engines perform social  networking functions. They Act as dictionaries, spell checkers and  thesauruses.

This is why classification has become more and more important to search engineers over the years.  Understanding, as close as possible, the intent, is paramount. The  interesting part is the ever changing landscape of exactly what the  intent is.

Some studies make the case that users  are prone to a higher level keyword approach to simply get near the  vicinity, preferring to click through at that point and search the  local site for the exact information need. These have been referred  to as 'teleporting queries'.

Classification of query types, among  SEOs at least, have generally come in three flavours;

• Informational
• Transactional
• Navigational

But, these are simply broad  categorizations that we should play hard and fast with. Many queries  fall into more than one category and that's actually quite important  for SEOs to understand. The reason we care about these is that they  play a strong roll in keyword research and ultimately targeting and  content programs.

The following table gives you a sense  of the various classification types in this area (click for full size);

A different mind set

So let's go beyond the traditional  understanding of query types. We have looked at the core types so  far, but another paper I came across broke things down a little  differently. The reason I decided to bring this up is because we need  to understand things aren't always the same.

Here's a chart from the paper which helps understand this approach;

1.  Navigational

My  goal is to go to specific known website that I already have  in mind. The only reason I'm searching is that it's more  convenient than typing the URL, or perhaps I don't know  the URL.

aloha  airlines duke  university hospital kelly  blue book

2.  Informational	My  goal is to learn something by reading or viewing web pages
**2.1.1  Directed	I  want to learn something in particular about my topic	what  is a supercharger 2004  election dates
2.1.2  Open	Closed  I want to get an answer to a question that has a single, unambiguous  answer.	baseball  death and injury why  are metals shiny
2.2	Undirected  I want to learn anything/everything about my topic. A query for  topic X might be interpreted as "tell me about X."	color  blindness jfk  jr
2.3  Advice	I  want to get advice, ideas, suggestions, or instructions.
2.4  Locate	My  goal is to find out whether/where some real world service  or product can be obtained	pella  windows phone  card
2.5  List	My  goal is to get a list of plausible suggested web sites (I.e. the  search result list itself), each of which might be candidates  for helping me achieve some underlying, unspecified  goal	travel amsterdam  universities florida  newspapers

3. Resource	My  goal is to obtain a resource (not information) available on  web pages
3.1  Download	My  goal is to download a resource that must be on my computer  or other device to be useful	kazaa  lite mame  roms
3.2  Entertainment	My  goal is to be entertained simply by viewing items available  on the result page	xxx  porno movie free live  camera in l.a.
3.3  Interact	My  goal is to interact with a resource using another program/service  available on the web site I find	weather measure  converter
3.4  Obtain	My  goal is to obtain a resource that does not require a computer  to use. I may print it out, but I can also just look at  it on the screen. I'm not obtaining it to learn someinformation,  but because I want to use the resource itself.	free  jack o lantern patterns ellis  island lesson plans house  document no. 587

This approach certainly helps to give  you the idea of what search engineers are looking at. Does it really  matter if it is the classical  informational/transactional/navigational or resource? Of course not.  What we're doing here today is seeking to get a feel for how  classification may be done.

To a search engineer, on the larger  level, there are two simple aspects to a query; intent and  satisfaction. Each person using a search engine has a goal and  classification helps break down these goals into bite sized pieces.

Not to be taken too seriously is some  of the data this particular group found in their research;

I say not to take it to heart because  as we all know a single data set never tells us the entire picture.  This was actually taken from Alta Vista data, soooooo... take it for  what it is.

Associating Goals with Queries

If we consider goals as understanding  intent, then we can break associations into two areas familiar to  most of the search geeks reading my ramblings over the years;  implicit and explicit.

Explicit –  in most cases these would be navigational or more simplified queries.

Implicit –  would be less obvious queries or system elements such as Google's  [I'm feeling lucky] feature.

Another common element we see in query  classification is building machine learning approaches based on  training sets. As you'd imagine, the larger the data set of query  data you have, the better associations you can make between queries  and intent/satisfaction. Just because behavioural data may not be  overly-valuable for ranking elements, doesn't mean it's off the table  altogether.

In the paper I cited earlier, they talk  about using behavioural data to seek out telling signals the user  might give;

• the query itself
• results returned by the search  engine
• results clicked by the user
• further searches or other actions  by the user

There are actually other actions that  can be tracked such as;

• dwell time (on page clicked on)
• scrolling
• page depth
• saving the page to favourites
• Explicit SERP actions (such as  Google's +1)

You get the idea. Behavioural data, on  a large scale, can bring a great deal of data to further understand  the goals and potential intentions of users through implicit and  explicit data. We can also see this in recommendation engine elements  (Google Suggest, refinements etc.).

What Can SEOs Learn From Query Classification

To begin with, let us look at the core  goal of classification; assessing user intent. That of course should  be obvious as far as why we, as SEOs, would want to also understand  this. There are no tools out there that really give us this. Which  means, to some extent, we have to look at potential query spaces and  establish what user goals we're trying to service. If you use the  classic informational/transactional/navigational approach or the  above 'resource' model, is inconsequential. What we need to do is  align targeting and content programs to best serve these needs.

When do we look at it? For the most  part understanding query classification plays into one of the first  elements of an SEO program; keyword research. Out SEO programs live  and die from the efficacy of the keyword research. Keyword research  is focused on matching user intent with our (ranking) targets.

It should be noted that most queries  are informational in nature. Even quasi-classifications such as  seeking information on a product prior to purchase. In fact, much of  the research seems to show that navigational queries are often  seeking information about a product and the query is often refined to  reflect this. As such we must consider having a content program that  reflects this.

Below are some tables from a recent  research paper that shows the percentages of each (in the traditional  model) for various topic areas.

The main goal here today was to give  you a sense of how search engines are dealing with this so that you  can start to adapt your own keyword research and content programs  accordingly. If you're interested in more detailed planning, be sure  to sign up for the SEO Training Dojo as we will be putting this (and  much more) into a keyword research section being posted in the next  week or so.

I hope you enjoyed the ride... I know I  did.

Papers used/cited in this post;

Understanding User Goals in Web Search

Classifying Web Queries by Topic and  user intent (direct link to PDF)

Clustering Query Refinements by User  Intent