Where are you headed after you read this post? The previous post, maybe a related post, my Twitter profile, out outbound link, an ad? I’d like to think you would stick around for a while read a few posts and maybe subscribe to the Sciencetext newsfeed or hook up with me on Twitter or Facebook, perhaps. But, in the end, I’m easy, I don’t mind where you go next.

Millions of other websites, particularly commercial websites with something to sell see things differently, of course. They crack open the cookies, they scour their metrics, they ponder their analytics, looking for clues as to how to predict where a sites visitor will go next once they have finished on one page. It’s the secret key to getting that next sale, making their site sticky, and (bottom line) engaging with your bank account.

So, is there a magic formula for predicting which page you will visit next on any given sit? Of course there isn’t! But, there are computer models and algorithms out there that can help webmasters make an educated guess. Now, a team of mathematicians and computer scientists in Australia has developed a new model for predicting which will be the next page you visit.

Now, before readers get all fractious about privacy concerns, this concept of predicting is not quite as sinister as it may at first sound and it’s nothing more black hat than the kinds of predictions stores make about which style of dress, type of ethnic food, or pop star will be the next big thing. It’s certainly not the very sinister kind of privacy breaches allegedly propagated by the likes of Phorm, which track your every movement so that the companies they work for can spit out personally tailored advertising.

As Faten Khalil and Hua Wang of the University of Southern Queensland and Jiuyong Li of the University of South Australia point out, the likes of Amazon, Netflix, and other e-commerce sites can improve the service and experience they offer users if they have some way of predicted the path their users will take as they navigate those websites.

“It is extremely important to form some kind of interaction with web users and always be one step ahead of them when it comes to predicting next accessed pages,” Khalil explains.

“The major objective of next access page prediction is to improve customer experience,” he told Sciencetext. “It enables an e-store to provide the right recommendations at the right time. For example, this can be achieved by dynamically changing linking icons on a web page so that the most likely clicked icons are located at the top. From the commercial viewpoint, the advertising links can be reorganised accordingly.”

Khalil and colleagues have now mashed up three mathematical models that can be used to best guess the next step in a sequence – Markov Models, association rules and clustering – to devise a novel approach to prediction. Their hybrid model clusters web pages into consistent groups, then passes them through a Markov and association model to predict the next most likely page a user will click through.

The team’s research paper has some wonderful equations and they do say that their combined approach can predict next page choice much better than any of the three individual techniques. I don’t want to disrespect their efforts, but I cannot help but think that individuals will always do surprising and spontaneous things. There are so many variables to take into account, that perhaps cannot be included in any equation when making web predictions: the interjection of one’s spouse while browsing for a gift, a temporary slowing of the site, an internet outage, or perhaps even just a sudden call of nature.

“No method can predict the behavior of impulse buyer,” Khalil admits, “However, if a user easily finds his o her way through a e-store, then the user is more likely to come back.” This could be especially beneficial to uncertain users as the service provider could have a clearer picture of what the user might want even if the user doesn’t. The prediction idea could also speed up the web by allowing the webmaster to optimise the site’s structure.

But, what of those privacy concerns? Khalil dismisses them with the idea of an opt-out option. “Customer click streams have always been analyzed one way or another,” he told me. “If a customer does not want to be ‘followed’ when she or he is shopping online, the customer should be able to turn off the feature if this has been implemented to a e-store.”

F. Khalil, J. Li, & H.Wang (2009). Research collaboration crucial to meet food demands Int. J. Knowledge and Web Intelligence, 1 (1), 48-80