Saturday, October 2, 2010

How predictable are you?

If I were to predict what you will buy in the supermarket today, I guess I will do pretty well by selecting the products that you buy most often. I do not know which music you will listen to tonight, but chances are odd that it will be a cd that you listened to last week as well (and the cd might still be in the cd player). And if you put on your shoes, I can safely assume that you will go out.

As in the real world, our Web behavior is quite predictable as well. We have our favorite reference sites and search engines. There are several forums and news site that we monitor and contribute to. And every now and then we return to our trusted travel planning sites to book a trip.

But how to predict what you will do next on the Web? For this, three regularities can be exploited:
  • There is a small group of pages that is visited very frequently.
  • Revisits are typically focused on pages visited very recently.
  • The more revisits, the more repetitive behavior in terms of transitions between pages.
And by combining these regularities the predictions will be even better (the PivotBar - see the preview in an earlier entry - uses such a clever combination).

But what may be even more interesting: some people are more predictable than others. You might expect that these are the people that predominantly use the Web for routine tasks, but that is not the case: people who mainly visit their favorite pages, may do this in a very chaotic and unpredictable manner; and people who like to search and explore might be very predictable in when and how they return to their favorite news sites.

Next week we will present these and other findings at the ABIS 2010 Workshop on Personalization and Recommendation on the Web and Beyond (the paper can be found here).

I bet there are many other things to discover in how we use the Web - findings that we can use for creating better tools. Help us to do so by contributing to the Web History Repository.

No comments:

Post a Comment