Michael Voong HCI Researcher @ Birmingham University

PlaceRank

Picture this scenario: you have some leisure time. You want to have lunch, go to a shopping centre, or visit the theme park. With so many options on where to eat, which shopping centre to go to, or which theme park to visit, how do you decide? What would help you decide?

Friends help, and we ask those people for advice. Which place is close and serve the best tacos? This helps sometimes, but there’s a limit to the guidance you can get if your friends are like me and can often be indecisive. If you want a new place to visit, how would you go about making a wise choice?

Techno-savvy people may fire up good ol’ G in their web browser and type in “lunch “. They’ll be presented with a list of food place recommendation websites, which have sections for practically any given region of the world. If you’re lucky, you’ll get a review of the place, and you can place your own judgment on whether you want to try it out.

Geo-techno-savvy people would fire up Yahoo! Local or Google Maps to take advantage of their local business search facilities. They’d define a local area and order business searches for a given keyword by distance. These searches are very commercial, so severely limit the scope. What about that small café down the road that does the best sandwich in the area, who use ingredients they grow themselves, and use only the freshest meat unavailable to the large chains?

I was just musing in the shower this morning about the possibility of filtering recommendations by your location, social network, and the dynamics of visiting those places from other places. As far as I am aware, only the first has been done before. The remaining two are the most interesting:

• Social networks. Claim: your friends (or recursively with decreasing importance, your friends of friends) go to a particular place more often, so that place should be given a higher relative importance to other places
• Movement dynamics. Claim: guidance to a particular place B from place A should be given a higher relative importance if more people have traveled directly from place A to place B. That is, the trails of movement themselves are a factor of the recommendation relevance metric.

What inspired all of this? The PageRank algorithm (and a trademark of Google) places a higher relative importance on pages that have a lot of pages linking to it. The increase in importance given to page A from page B is a fraction of B’s PageRank depending on the number of outgoing links. Run recursively then iteratively (around 40 times), the PageRank of each page that Google crawls settles to a value.

We roughly apply this concept of the reputation of a page to the reputation of a place, built up of by the number of places that people travel from, to it. In basic terms, it’s the number of people that use that place. Then, we can incorporate ego-centric social network information (your friends, friends of friends, etc) to further determine what makes a good match, based on the activities of people in your network.

Confused? Probably, because I’m still warming up to this idea. I’ll support the aforementioned claims and expand on the PlaceRank idea in a future blog post. Probably after the next shower.

Good Stuff Mikey.
I’ll have a think about that one,

Phi

It’s probably be quite challenging. Google can implement an efficient PageRank algorithm is because it is a rather centralized network with the global information it needs. Social network is a decentralize one (i think), like a P2P network, where no one has the global information.
You’re moving closer to my field, Mike

Tien Tuan Anh Dinh, it depends how you were thinking of capturing the location data. I think the most likely source is geolocated mobile phones. Chances are, the mobile phone companies will have an accurate record of the location history of all of their customers with suitable phones. Okay, so it’s only the customers of that network (sharing data might get tricky), but it’s still a lot of people.

To produce recommendations, they can just look at where you’ve been previously. Think “people who went to the places you go also go to..”. Even better, they have a record of all your communications, so they could perhaps assume that you might share some tastes with the people you talk to most often, like Mike was suggesting.

What I find more exciting, though - the facility to find the best ‘compromise’ for a group of people without having a huge argument!