We are at the dawn of the age of collaborative filtering. But then, what happens to our individuality?
The old adage "no two people are alike" comes to mind. While collaborative filtering is fun, we also need to go back to our individualism, at least sometimes. So, we would like to take you on a ride to individualism, but don't worry we will get back to communities in more ways than one, just bear with us. This ride will be about content discovery, but since it needs to cater to individualism, our goal is to discover algorithmically.
The first thing to notice (right after "no two people are alike") is that people have multiple personalities. Any algorithmic approach needs to capture that. Well, in the world of tagging this is easy. Let’s say we associate a tag with each personality. These tags will be a little more involved than the current notion of tags. We will need to select them a little carefully and we will need to nurture them. Eventually, this collection of tags will represent me, the individual.
So let’s say one my personalities likes cooking, and I create a tag named cooking. Now I need to define what cooking means for me. It is important to realize that keywords may be inadequate to describe my "cooking" personality. Creating a long description may be too laborious and besides my personality may evolve over time. For example, my taste in cooking may change. I need to be able to define my "cooking personality" dynamically.
So, we will define this "by association", by selecting some examples of what cooking means to me. The easiest way of doing this is to tag web pages that I like. Well, actually this is not quite enough. A good "definition by association" also needs some negative examples. So I would also need to "negatively tag" some web pages that I don't like (as far as my kind of cooking goes). Now, I feed these "positively and negatively" tagged items to an algorithm along with a whole bunch of "to be discovered" web sites, and I get back a set of web sites that "fit" my personality. I look at these results, positively tag the ones I like, and negatively tag the ones I don't like, feed them into the algorithm again, and get back a whole bunch of new web sites, hopefully with a better fit. If I do this for a whole set of dynamic tags, after a while I end up with a machine that knows exactly what I like and what I don't like, and it can discover content continuously – even while I sleep.
- cuneyt
