The Internet is great at helping us find stuff that what we are looking for. But what if you don’t know what you are looking for?
Nobody has properly solved the issue of how to find entertainment on the internet.
What Options Do We Have?
So what are the options we have at hand to find entertainment recommendations?
The first option that comes to mind is the search engine (and these days, the word search engine is synonym to Google).
We use Google for most things in our lives, but the one thing that we don’t use if for is “what film should I watch tonight?” or “What would be a good book to take with me on holiday”, etc.
Why? Because it does not work.
In-store Recommendation Engines
The second option is to go to use an in-store recommendation engine.
One of the most known ones is the Amazon one. Say you buy Ewan McGregor’s “A Long Way Down”. After checkout Amazon will show you a section called “Customers who bought this also bought” which, in this example, will contain some other motorcycle adventure documentaries, and in particular other ones with Ewan McGregor.
Is it likely that I will like them? Possibly.
Is that the next film I want to watch? Probably not.
The reason why this method is fairly poor is because it clusters the content and assumes there is a relationship between the items, based on other people’s purchases.
The Problem with this Approach
Whilst under certain circumstances this might yield good results, there are two issues with this approach:
1- Using purchases as a driver for recommendations instead of ratings is not ideal. In my example of the Ewan McGregor film, let’s assume that most customers buy both “the Long Way Down” and “the Long Way Round” at the same time, but only like the first one.
In this case Amazon would establish a strong connection between the two and recommend it, but is unaware that most people actually were disappointed with the sequel.
The answer here is: switch from actions to actual ratings by users.
2- The second issue is related to datapoints, or put differently, the more information the engine has about things you like and dislike, the better it will be a giving you recommendations on other things.
That’s why Amazon isn’t great. Not only does it not know whether you liked the item you purchased; it also has a limited set of entertainment products that you have bought through Amazon that it can use as a base.
3- The 3rd option is slightly more tailored, although it still has the same disadvantages to the recommendation engine. And here I’m talking about services such as iTunes or LastFM for music, and LoveFilm or NetFlix for films and series.
The recommendations are more accurate, because it is a specific entertainment vertical, but it still suffers because of the methodology of clustering. And because of the fact that these services are for a specific entertainment verticals, they miss out on cross category recommendations (for example: because you liked this book, you’ll probably like this film...)
There is a Solution
The solution is to use a service that matches individual tastes of people (both stuff they like and dislike). Once the connections have been established (i.e. who is similar to whom), it is easy to find entertainment: just check the recommendations from people similar to you.
This is the reason why we decided to launch itcher: to connect people with similar tastes and generate recommendations based on people with similar tastes and interests.