on searching for jaguars, behavioural targeting and reality analytics.
Q: What are you up to? A: I'm working on several things. One thing that's going to occupy me in the next few months is a new edition of my book on search engines. I'm trying to catch up on the various technologies. When you're doing research, it's very hard to keep a broad knowledge of what's happening, so it's a good opportunity to do that.
It's the issue of trying to categorise results into some top-level ontology. You type in 'jaguar' - what do you think will be the first hit? |
Yeah, I guess that's true. There is some amount of stability - obviously Google dominates the field; I mean, which search engine do you use? But Yahoo, MSN and the rest are not that far behind in the quality of search. But once you're using a search engine for a while and you're happy, you sort of get used to the way in which things are returned to you.
So there's some stability, but there are a lot of ways in which search engines can be improved. One interesting issue is personalisation, which has been haunting us for a few years now. I think it's a very difficult problem, but I think it could make things move in a certain direction.
With one of my students I'm working on an interesting idea which I think could be useful in bringing search forward. It's the issue of trying to categorise results into some top-level ontology. The usual example one gives is, you type in 'jaguar' - what do you think will be the first hit?
I imagine it would be the car.
Yeah, absolutely. There's a bias toward commercial sites. It's mainly because of the way in which the algorithm works, but the thing is, the search engine doesn't really know whether you're interested in a car or an animal. So what we're trying to do is, when you type in a query like 'jaguar,' it could classify that query in two different ways. And when the results come in, we could split up the results, and say, 'Look, you've got some results about cars and some about animals - which one are you interested in?' Obviously if there's personalisation you can go even further. I think that could be interesting for a wide range of queries.
I don't think you've got to have 100 people behind you to do good research |
I think there's quite a bit of research going on in the background. If I've thought of it, probably someone at Google has too! One thing we look at is that is very important to search engines is what's called 'behavioural targeting.' It's to do with advertising, really about giving more personalised ads to users.
For instance if you're interested in cars rather than animals, which is what most people are interested in, then obviously you'll get ads for cars. But the point is, you could be displayed different types of ads according to the categories you're interested in. So it's first of all about figuring out what categories you're interested in. Then it's figuring out what the ads should be and what the content you're looking at is about. I'm pretty sure they're working on this in the research labs of the search engines. But we're going for it - I don't think you've got to have 100 people behind you to do good research.
Does this relate to your work on trails?
No, not really. But I am still interested in that. I'm doing some work with George Roussos and a PhD student. This is sort of taking things in a completely different direction. It's what we call 'reality analytics.'
I've been doing work on web mining and trails that people follow on the web for a long time, and I'm still continuing with that. But one thing that's quite interesting is that the web is now really beyond just the computer you're sitting in front of. We're carrying around mobile phones, we're walking around with computers, we've got GPS. There are electronic traces of our movements beyond the virtual ones.
The idea here is to try to look at these, and the ways humans move, by analysing the traces or trails that people follow, using some of the same methods that we used for online trails. They're basically log files of what people are doing. So one can try and analyse that behaviour.
There's the notion of location-based services, which has been around quite a while, and is about understanding what people want, when and where they want it. Specifically we've been trying to predict what people are going to do. If you know enough about people's movements, you can try and predict what they would do in a similar situation.
There are obviously applications in location-based services, but also social interactions. A lot of social science in the past has been done of fairly small groups of people. There's an opportunity here to look at masses of people and analyse social behaviour.
That raises all sorts of interesting questions. From a computer science perspective, it means coming up with algorithms or methods to try and find the patterns in this data. I think that'll keep me busy for a while.
We're going to be testing our reality mining at the London Zoo |
I'm always looking toward the real world, so to speak. One thing that George and I have done is to patent some of the work we've done in reality analytics. I believe it's the first patent at Birkbeck. Specifically it covers a method of setting up an infrastructure to record people's movement, and present analysis of that movement. And to look at ways in which people and companies can benefit from this analysis.
It's not just about pushing ads to people. For instance in emergency situations, tracking of people might be of assistance. We haven't mapped out any single scenario. We've been talking with industry about getting some data to work with.
Yes, I've been teaching databases for a long time, and web technologies. And I've got PhD students. And I have some longstanding collaborations in web mining.
One possible project with a colleague from overseas, one thing we thought would be interesting is to look at Wikipedia and how people are using it. I think understanding social behaviour is going to be key.
We're going to be testing our reality mining at the London Zoo, where we'll have 50 people or so walking around with GPS-enabled phones. George is doing the mobile side of things, and my role is more on the analysis side. It could be interesting.
I've been associated with the Knowledge Lab for a long time, and there seem to be a lot of good projects going on.