Saturday, March 17, 2007

Personalization, intent, and interest

Gord Hotchkiss at Search Engine Land writes about "The Pros and Cons of Personalized Search".

Some excerpts:
How do engines [disambiguate] intent without us giving it any more information to work with at the point of query?

This is where personalization comes in. In the current online reality, there are really only a few places that the search engine can look to help define intent without depending on further information from the user:
  • They can look at your past history and learn more about you by what you have already done
  • They can look at the context of the task you're currently engaged in, hoping that it will give some clues to what you're looking for
  • And finally, if they know something about you and your social, geographic and demographic cohort, the engine can hope that there is a similarity of thinking within that cohort, at least when it comes to common interests and intent
Each of these factors is being explored as a potential avenue to help with disambiguating intent.

Right now, Google is put their eggs in the past online history basket, feeling that where you have been will provide the best signal to predict where you might want to go.
When people only enter a few keywords, any additional information might help. Looking back into what they have done might allow us to better determine intent and interest and lead to more useful search results.

Different people have different interpretations of what is relevant. At some point, the only way to further improve the quality of search results will be to show different people different search results. Changing search results using long-term search history, as Google Personalized Search does, is one way of satisfying these differing views of relevance.

Personalization also seems likely to be useful when a searcher is iterating and not finding what they want. That searcher clearly is having difficulty translating intent into results. Current search engines ignore these iterations -- each search is treated as independent -- but there is valuable information in those struggles.


Eran Sandler said...


How would you define search iteration?

There are a couple of points I can think of that matches this:

1) Searches that not a single item was clicked in.

2) Searches that are similar to other searches (containing X% of the same terms - perhaps even in the same order). This might help to show that the user is trying to refine the search by adding or removing words.

3) Searches that were done in the same session (we still need to define what a session is in that context) or in a relatively small time (we still need to define that time)

I wonder if Google and other search engines considers searches that were made by the same user in a relatively small amount of time as a search iteration.

It would be good to also define a profile for users because savvy users will probably open up multiple tabs/windows and search for multiple items at the same time (I know there are various ways of distinguishing between them).

Un-savvy users that search one thing at a time are different from the search savvy people. That can also help and it's really easy to find them. Just look at the similarity of different search terms over a relatively small time span.

Romram said...

Generally when a user doesn't find the relevant documents in one go, he tries to change the search query terms. Learning from these reformulated query terms one might re-rank the next series of search results in a different manner. May be something on these lines


Greg Linden said...

Thanks, Udit, for the reference to that paper. I'll take a look.