- What can Apple do in location services that Google doesn't already do?
Behold a new patent application by Apple that was just published, titled RELEVANCY RANKING FOR MAP-RELATED SEARCH. This patent looks at how any sort of location-based search system, such as local business search, nearby friend search, etc., can tell the difference between locations that people go to a lot and locations that people don't go to or don't stay in. In a nutshell, their method looks at logs of locations of lots of users, and from those logs, the method figures out which places get a lot of people going there and which don't. In other words, by looking at where millions of iOS users go, Apple can figure out what places are hot.
Their approach to doing this analysis is very sophisticated. They're not only looking at what where people go, they're looking at how long they stay there, how long they travel to get there, and what time of day they go. They also account for whether people go there every day (e.g., they work there) or occasionally (e.g., they shop there). So if an office building gets more people than a department store, the department store may be "hotter" because it gets lots more people that don't work there. And even if a 7-11 gets more people than a Ikea, the 7-11 gets people who drive by while the Ikea gets people that go specifically there.
Their method also takes repeat visits into account, based on the place being visited. If a restaurant is visited by lots of people, it's popular. But if a restaurant gets a lot of visitors multiple times, it's worth recommending.
Even more interestingly, their method compares the sets of places that people go, and recommends places to new users based on the similar interests that are reflected in common places visited. This is Collaborative Filtering at its finest, very similar to the technology introduced over 15 years ago for music recommendations.
The key is that all this processing is behind the scenes, analyzing logs of locations that iOS users go, and determining which places are hot and which are not. This "hotness" is then used by a variety of location-based services to rank answers to user queries.