In so many ways it’s unfortunate that Mark Zuckerberg has popularized the term “social graph.” Cold and clinical, it’s an awkward phrase to describe the network of human relationships. But we’re stuck with it. It has also spawned the term “commercial graph” and now . . . “place graph.”
WHERE has coined the latter term and describes it the following way:
WHERE Place Graph is a global mapping of places and how they are related. We use it to match local lifestyles of users and predict places they will like. The PlaceGraph is the basis for our Places Recommendation Engine (which you should check out by downloading WHERE for your iPhone or Android)
By trying to understand the relationships between places WHERE hopes to offer better recommendations to its users. Many sites are using Facebook friends and Like data and some, like Bizzy, create profiles to match people and show them “collaborative filtering” style recommendations accordingly. Understanding the relationship between places adds another dimension to the recommendations algorithm.
WHERE explains its methodology in detail in a white paper on Place Graph (which the company is seeking to trademark):
The Place Graph is generated from multiple inputs ranging from general information about places (e.g. location, category) to massive amounts of user inputs and interactions (both implicit and explicit) the applying machine learning, similarity metrics, and predictive analytics. More specifically, a patent pending and proprietary indexing and retrieval algorithm, called the PlaceGraph Algorithm, is applied to compute and generate the personalized Place Graph for each user.
Inputs are categorized into three general buckets: explicit, implicit, and place information. Explicit interactions include ratings, reviews, check-ins, saving places to address books, etc. These are inputs from users who directly reveal their preferences and choices. Implicit interactions are passive user by products of browsing and searching. For example, clicking on an online detail page of a local place is considered implicit, since we don’t really know if user liked the place or not.
There may be some interest in a place based on user’s interaction, but we also have to take into account that ac lick might have been accidental or explorative. The place information (e.g. place category, location, etc.) also plays a role in computing the Place Graph.
February 21st, 2011 at 5:55 pm
[...] This post was mentioned on Twitter by Greg Sterling and Rob Bailey, Rasmus Himmelstrup . Rasmus Himmelstrup said: @gsterling pdf link in http://bit.ly/eONi3a doesn't work [...]
March 1st, 2011 at 1:28 pm
[...] This is of course qualified by the fact that there are concentrations of Foursquare users in specific places (e.g., California) and the data are not representative of all groups of people all over the globe. Having said that the data that is being “thrown off” by user behavior on apps like Foursquare and others is impressive. I’ve often said that Likes and check-ins can turn into most popular lists or be factored back into algorithms, as WHERE is now trying to do in part with its Place Graph. [...]