Predicting Customer Clicks

So we've got artificial intelligence up and running.

You didn't know that? Yeah, we do. If intelligence can be defined as a capability to learn and react to new situations on the basis of experience already gained, a wide variety of enterprises are employing intelligent programs in fields ranging from valve control in variable load boilers to predicting credit card defaulters. Voice and pattern recognition software are probably more familiar examples of artificial intelligence. But hold on, it's not yet time to get out those blasters and EMP detonators. Artificial intelligence powerful enough to accomplish the level of parallel thought to even equal a human, let alone take over the world is still very far on the design charts. But when it comes to juggling numbers and other data around, the human brain is laughable against the microchip.

But rather than take this difference in computing power personally, what we'll do is think of ways to use these same programs to make our work easier. Especially in web development.

Web development? Who'd ever want an intelligent web page, right? Wrong.

Wouldn't it be nice if I went to the MSN site and instead of all that junk, half of which I don't even read, the site displays only results which it knows I'll find interesting because it noted that I clicked more on links related to music and not culture.

Or in Google, when I type in "chess", instead of showing me every single page that has "chess" written on it, it takes me to a site that allows me to download a free chess program because it know that's what people generally want.

Or when I'm filling up forms for another of those free sites that don't allow me to see what they have without my telling them my name, age, sex et al, the form instead of asking me minor details again and again, fills in obvious things itself. For example, a person with a name like John Smith would most probably be a guy and more than 5 years old etc.

In short, wouldn't it be nice on the whole if the net behaved more like a person and less like a machine. Interactive, understanding and a lot less complicated. That's exactly what we, at Stylus Systems Pvt Ltd., propose to do for your site.

• Predict user preferences based on links and selections made by present and past users and change content on page accordingly.

• Fill in common values and perform other mundane tasks without repetitively prompting the user.

This was of course, what will be visible to the user. What about you, the owner of the site?

Every time someone visits your site, he leaves behind megabytes of information about himself and browsing patterns in general that never get recorded, never get used. Data that could tell you what people look for in your site, what turns them on and what makes them leave?

What if the administrator of your site had access to all this information in an easily understandable, maybe even graphical, manner with all the patterns and main points of interest highlighted? The administrator would be able shuffle content around to make the more in-demand articles, links and forms more accessible. Maybe even remove the redundant content and try out new material and judge the response it gets.

What we propose by data mining your site is to automate this process. An algorithm like the Microsoft decision tree algorithm (incorporated by default in the Microsoft SQL Server 2000) would be able to use this data to a higher degree of accuracy than any administrator.

Horizontal integration of data has been recognized, as the most in need feature of any interactive software. Vertical integration (through a single column) has for decades provided information as to aggregates. Horizontal integration (across columns) promises to find the patterns across those aggregates.

For example, a person who tends to always clear his credit card bills erratically could be a potential defaulter. But there are other factors influencing such behavior. Factors, which cannot be put down as a set of rules (an expert system). In such cases where the factors involved become too numerous to automate a system, the data is run through a learning algorithm like the decision tree and results are no less than spectacular.

But it should be noted here that algorithm training requires time, or data rather. The more data the program has to go through, the more accurate will be the results it predicts.

Artificial intelligence has long been touted as the fifth generation of computing and we're just beginning the journey. If you want to join in or not is entirely a business decision. A few of the possible applications and advantages have been listed; the investment factor is, of course, there. All we tell you is what is possible. Where and how you chose to use it is what will make the difference in the long run.

All in all, artificial intelligence very much exists and no longer the stuff of Doctor Who books and movies about government agents and red and blue pills. Be there.