Because of some pretty bad-ass data science and Google’s ever-increasing awesomeness, it looks like one day in the not too distant future, we will all be able to get a free (or heavily-discounted) ride. A taxi ride, that is.
What’s the catch?.?. Well, the deal is … based on your location you’d receive an offer on your smartphone. The offer advertisement would likely be for a discount on goods or services from a local brick-and-mortar business. If you are interested in going to that part of town, then you can get a free (or discounted) ride. In order to keep people from abusing the free-ride offer, your ride-to-purchase ratio would be accounted for in your Google profile – and if you’re a ride-bum, you probably won’t get too many future offers for the free-ride.
This system isn’t just some scifi junky’s greatest fantasy… it’s on its way to becoming reality. Google was awarded a patent for this transportation-aware physical advertising conversions system back in January of 2014.
There’s tons of advanced data science that goes into a system design like this one. While I can’t cover all of the algorithms that a system like this utilizes… I’d like to discuss a powerful location-based algorithm that can be used to design systems similar to that recently patented by Google.
Quietly, behind the scenes, location-based social networking (LBSN) has been stopping the show when it comes to location-based intelligence and advanced mobile marketing. These networks have been recording and analyzing user preferences, user social influence, and user location in order to power personalized, geo-social recommendation engines that can be used to deliver mobile advertisements. Although this practice isn’t brand new, improvements are continually being made.
Recently, Dr. Jia-Dong Zhang has been working out a way to drastically improve location recommendation performance by using a “kernel density estimation approach to personalize the geographical influence on users’ check-in behaviors as individual distributions rather than a universal distribution for all users.”
If you’re not already familiar with it, kernel density estimation (KDE) is a non-parametric estimation method that can be used to calculate the probability density function of a random variable or set of variables. In spatial terms, KDE uses a kernel function to estimate a smooth tapered surface that represents clustering and density patterns of points or lines in space.
Figure 1 Kernel density estimate with diagonal bandwidth for synthetic normal mixture data
KDE is a popular method for quantifying the intensity and density of a point pattern – in other words, “hot spot” analyses. KDE is quite useful for modeling and predicting spatio-temporal trends related to interest areas like market area analysis, environmental pollution, crime, disease outbreak, and seismic risk. Since the method employs a kernel function to estimate density, there are less boundary effects than those exhibited by counting methods. KDE can be performed using R (‘ks’), Python(‘scipy’), ArcGIS (Spatial Analyst), and QGIS (heatmap plugin).
If you’d like to sharpen your skillset with respect to location-based data science and algorithms, you can check out Smoothing of Multivariate Data: Density Estimation and Visualization or Density Estimation for Statistics and Data Analysis. Both of these books provide introductory and advanced perspectives on using density estimation in advanced data analysis. I favor the first of the two books I mentioned, simply because it places a greater emphasis on data visualization. (If you decide to purchase one of these books, I receive a small commission based on that sale – just putting that out there for the sake of full disclosure.) If you decide to take on a deeper study of spatial statistics and location recommendation engine algorithms, I’d love to hear what you think of these books yourself. My email address is Lillian@LillianPierson.com, or if you leave comments in the section below then I will be sure to respond back as soon as possible.
