When you’re using an online map as a consumer, one of the things you will undoubtedly do is search for places — or as the mapping industry likes to call them ‘places of interest’, ‘points of interest’ or ‘POIs’ ¹. 

In performing these searches there are generally two factors: one factor is almost always proximity and the other is commonly quality.

Unless it’s something mundane like a fuel station this quality factor is incredibly important, especially for categories like restaurants, hotels and service providers.

But have you ever gone to a restaurant with tons of great reviews and then found yourself totally underwhelmed… questioning why you even went there in the first place?

Maybe it’s just my curmudgeonly self, but I’m guessing I’m not alone in this regard.

Apple Maps and Google Maps have been adorned with ratings and reviews for places pretty much since their inception. After an initial affair with Zagat’s, Google quickly went their own way and built their own home grown rating system — almost destroying Zagat’s in the process. For many years Apple Maps took a partnering approach, first with Yelp, and then later with organizations like TripAdvisor, OpenTable, Booking.com and LaFourchette² . More recently you’re starting to see Apple’s own rating system creep gingerly into the picture, so it’s becoming a real melting pot. 

The fundamental problem with all these rating systems is that none of them take your own preferences into account, so the ratings you see are this amorphous, unwieldy glob of data that provides little information that is tuned to the individual. Ipso facto you get good recommendations for crappy places and you may also get poor recommendations for places you actually think are quite good. 

This, in my mind at least, makes all the rating systems pretty much useless. 

Now of course there are work arounds: 

• Work around number one: invest copious amounts of time delving into the reviews, and in doing so trying to pull out little nuggets of information that might indicate someone has said something that resonates with your own tastes or concerns. At the same try to guess which of these reviews were actually written by a human and whether or not someone was nefariously incentivized to submit the review in the first place. Ugh. If you’re like me you probably don’t have the energy to do this, especially as there’s no guaranteed success after expending all the effort.

• Work around number two: ferret through curated reviews from publishers. Now here you’re onto something perhaps a little more reliable. If you happen to know a place on one of these curated lists then you can use that nugget to deduce whether you trust their recommendations as a whole. So, for example, if a particular hotel is highly rated and you agree with their rating then the level of trust you can put in the rest of the ratings from that publisher might increase. Conversely, if the highly rated hotel was, in your opinion, ‘meh’ or an armpit then you can probably ignore this entire set of recommendations and treat them as places to avoid. The fact that Apple Maps incorporates dozens of curated guides from well known publishers provides useful fodder for this work around. 

But it all kind of sucks.

So what have organizations done for rating systems in other industries?

If you look at the big, nefarious world of online retail — and Amazon in particular — then it used to be that you’d see “People who liked X also liked Y”. However I see that this approach is now being replaced by advertising (which also sucks): 

If we look at music, Apple Music has something called “Similar Artists”, but it’s not clear what algorithm they use to determine the recommendation. I’m not a Spotify user, but it does appear they might do a better job at providing recommendations.

In the accommodations world, Airbnb doesn’t appear to offer any recommendation features other than generic ratings and reviews. In the restaurant world OpenTable now has a “news for you” feed, but like Amazon’s product recommendation feature it seems to be driven by promotional advertising.

Moving over to social media networks it gets a little more interesting. By having the ‘follow’ concept built right into their foundations these networks provide a framework to get recommendations from people who you know or at least trust in some way. For example, perhaps your friend Yevgeny might create his favorite list of Paris restaurants and share it as a guide on Instagram³. Like Julia Moon who uses TikTok, you might stumble across a particularly enticing video about some new doughnut shop, but frankly this a approach still a bit of a crap shoot and it all takes time and effort. 

It seems that only Snapchat comes close to a genuinely useful concept — it combines the set of people I know, their activity (and therefore their likes) with a map, specifically a Snap Map. But again it’s not perfect as it all takes energy to sleuth out where my friends are going and what they’re doing. And the whole thing breaks down completely if I elect not to participate in social media networks. 

What I’m really asking for is the perfect automatic recommendation machine that takes zero energy and is not sullied or adulterated by advertising. 

So is there a solution or am I in fairy tale land?

Well perhaps I might posit one approach based on a classic Venn diagram:

There is the set of places that I like. If you combine that set of places with another person’s set of places that they like then they might overlap. So, for example, we might both like some of the same restaurants in London. But I might also like some restaurants in Paris that the other person does not know. Equally the other person might have liked some restaurants in Berlin that I don’t know. So the system recommends the Paris restaurants I like to the other person and recommends the Berlin restaurants that I don’t know to me. 

I think the parlance for this kind of analysis is called ‘collaborative filtering‘.

The approach has several advantages:

• You don’t have to be a member of a social network
• You don’t have to know anything about the people from whom the recommendations are drawn
• You don’t have to spend endless time and energy rummaging through reviews trying to determine their pedigree
• It maintains privacy — other users don’t get to see my lists of favorites
• The recommendations are inherently personalized based on the set of people who have similar tastes to you.

Now obviously an actual system would need to be more complex and would need to process a ton of data. But in today’s world that’s not out of the question. 

Why has nobody taken this approach for a rating or recommendation system for places? 

Perhaps the reason is that there simply isn’t enough raw data from which to provide any useful recommendations.

To solve the data volume problem you need to make it super easy to rate places. Apple is doing this a little bit with their new rating system in Maps. Opentable encourages ratings after you’ve been to a restaurant booked via their platform. Airbnb does something similar. Google Maps has it even easier—  they have so much mindshare that businesses themselves encourage you to rate them on Google. So surely there must be enough data?

Perhaps, like online retailers, companies are now much more interested in chasing advertising dollars than providing useful recommendations.

The lack of any decent recommendation system for places feels like I’m missing a wheel on my bicycle. I can’t get anywhere useful. 

So in closing I have a number of questions to ask the Map Happenings audience:

• do you feel the same way?
• why is collaborative filtering not used more broadly?
• is there a better solution than collaborative filtering?

---

¹  I recently rented a Nissan Murano and was surprised to see their navigation UI used the term ‘POI’. Are you kidding me, Nissan? Do you really expect consumers to know the meaning of that TLA?

² Now part of TripAdvisor

³ Although the Yevgeny I know would never, ever do that because he wouldn’t want to let the world know about his secret culinary haunts.