Coconut Headphones

Back in late August of last year, some Googlers presented a paper at a conference entitled “HyperLocal, Directions Based Ranking of Places”, where they investigated the possibility of using driving directions logs from Google Maps as a ranking factor in Local Search.  In my experience, Google does not publish papers on major ideas unless they have one or more patent applications already filed.  The US Patent & Trademark office does not make applications to the public available until one year has passed.   Since this paper was presented at the 37th International Conference on Very large Data Bases in Seattle at the end of August, I would expect by mid-year in 2012 a patent application should emerge leading to a great Bill Slawski article.  If you don’t read Bill’s stuff, you’re missing out on some great analysis – he’s a “must read” – check him out at www.seobythesea.com.

Of course, as I was drafting this article, I discovered Bill had already mentioned the paper in a recent posting of his, but since he didn’t go into great detail on it, we’ll do a deep dive here and develop some SEO best practices, assuming Google is now (or will soon be) using the methods detailed in the paper.

Background

The researchers studied whether logs from web-based directions service (i.e. Google Maps) could be useful as a ranking factor.  As the paper notes, roughly twenty percent of web queries have local intent; such a high percentage makes it important for Google to work on optimizing use cases for local search.

The researchers utilized data from Google Places and Google Maps Direction searches for a particular (unnamed) geographic area of the United States, and also used human raters to benchmark the existing Google Places ranking algorithm versus the approach of using the number of driving directions to a location requested by users over a given time period.

Amazingly, the researchers only used requests over one month and ended up with 49,533,223 driving directions requests, which was *two orders of magnitude* more than the number of Google Places reviews for those businesses (550,000).  Over a year that should be *three orders of magnitude* –  very exciting potential, if the data is actually useful.

They limited their research to queries for driving directions to museums, hotels, restaurants, night clubs, and landmarks (restaurant examples figure prominently in the paper).

The direction query log data they used had the user’s location, the time of the query, the maximum distance the user was willing to travel, and a query string (for instance, an address or business name).

Results

The researchers presented the results in Figure 5 of their paper, and said that there was a “clear correlation” between driving directions and number of reviews, but surprisingly didn’t go into any detail on this correlation.  I have eyeballed the numbers and reproduced the curve here in Figure 1, and if you fit a power law curve to it, the R squared value is *astonishingly* high, at .9865:

Clearly driving directions are extremely useful for ranking purposes (since it’s already been well proven that reviews are, the transitive property should hold).  Also, 22.5% of the places for which the researchers had directions queries had *no reviews*, so driving direction have great potential to provide better coverage.

The researchers drew these conclusions:

“This study shows that driving directions logs can serve as a strong signal, on par with reviews, for place ranking….the fact that the directions-based signal is comparable to the review-based signals is surprising.   It is up to a scorer that takes into account multiple signals to decide how the signals should be combined based on their characteristics….further, the logs provide near real-time evidence of changing sentiment, an aspect that is usually hard to capture with other signals (e.g., the reviews that a place has received; even a newly added web page of a restaurant will need time to increase its PageRank), and they are available for broader types of locations.”

An Interesting Implication They Didn’t Write About: Identifying Review Spam

Although the researchers bundled multiple locations together into this aggregate graph (and for individual locations it’s probably much more variant) on average, if Google knows the number of driving directions to a business that have been executed, it can very easily calculate how many reviews the business *ought* to have.  This is very interesting and could be used for much more than just local search ranking.   For instance, imagine that a particular large business should have 2,000 reviews but only has 50, or a particular small business should only have 10 but has 200.  This could easily power a methodology for Google to identify suspicious review spam by SEO professionals!

Location of a User Matters

In the study, they even checked out sensitivity by locations of the users.  At close distances the users’ locations didn’t matter, but as you get further and further away, ranking that took into account the users locations caused the ranking to differ.  I eyeballed the box-and-whisker plot data from Figure 13, used it to come up with my own estimates of what the 10th percentile and 90th percentile would be for variation based on distance, and show the results in Table 1 here:

The way to read the table is, for distances of 1/4 mile from a location, on average, search ranking results should differ only between 3.0% and 18.8% for most businesses depending on the searchers locations; results will differ more widely with larger distances.  If you’re a business in the middle of a city that gets most of your business from the nearby few blocks, locations of the searchers matters less than it will matter for a business in a suburb whose customers drive 5-10 miles to reach it.

SEO Best Practices

I believe Google is already using this, although to be clear, I have not done any experiments to verify this.   At a minimum, Google has certainly telegraphed that if they’re not already using it, they certainly will be.

Either way, I expect the SEO industry to rapidly adopt some best practices around this.  Here’s some proposed best practices, based on what’s in the paper and how we know Google has guarded against people “gaming” their system with regard to reviews and links as signals in the past.

#1   Driving directions should be from unique machines and unique users
#2   Driving directions should be from a mix of mobile *and* desktop searches.
#3   Searches from the location of the business are probably not helpful.
#4   Directions should be requested from different locations and distances.
#5   Driving directions should occur from where your customers can be found.
#6  Driving directions should have a natural distribution of timing that match customer’s search patterns.
#7  If you obtain a lot of reviews without a lot of direction searches, that could be flagged as review spam.
#8   Driving directions should be from a mix of search entry paths on Google:
– From a search for the address
– From a search for the product or service, then a selection of the Google places entry
for the business, and then clicking “directions”.

If you are closed on Sundays, driving directions requested on a Sunday may look unnatural for instance (as the paper implies), as would having directions from the same location, or all from mobile phone, and so on.  Just try to think like Google – how would *you* identify people trying to game the system?

#9: Don’t Make Directions Too Easy For Your Users

Before anyone proposes it, I wouldn’t recommend bothering embedding a form or a link on your website (with a large orange button) that generates a driving directions URL for Google Maps.    If you use the API for this, Google is probably checking where the query came from.  If instead you’re thinking of actually just using a Google Maps URL, it would still be very easy for Google to check the referrer.   So any approach like this will probably be filtered out.

In fact, if you provide such an experience for your users, you’re actually hurting your rankings – better for you that they copy the address and paste it into a Google search themselves.  If you already have a link providing Google Maps Directions to your users, you should remove it and simply list your address instead .  This is one case where a Google ranking variable argues for a worse user experience, unfortunately.

Conclusion

Clearly driving directions make a great ranking variable, and it’s hard to imagine Google not using it.  It’s  inevitable that our industry is going to have to develop best practices around it.  Please post any thoughts, comments, or anecdotal experiences with this approach below!