Should A Website Design Show Its Precious Algorithm?
One of the great things about web2.0 is the use of social power to create product recommendations. For example, Amazon.com recommends books from similar authors and Last.fm recommends similar artists.
It’s more than magic
Behind those recommendations there are [usually] complicated algorithms looking for recommendations patterns and similarities among users.
In my opinion there are 3 important issues regarding recommendation algorithms:
1. The more users a website has, the better the recommendations.
2. The more users a website has, the higher the calculation price,
3. Should a website show users how the algorithm works?
There are different approaches regarding the last point. Should the algorithm be kept as something precious that other websites should not access? An algorithm could improve its performance if the user gets at least some information about the way it works (see Amazon example). On the other side, letting users know about the algorithm could make the recommendation system hackable and someone could take advantage of that to promote his products.
This issues should be kept in mind at the moment to decide to show information or questions to improve the algorithm:
1. Making questions to users could make the system hackable (negative).
2. Hiding those questions would make a website look smarter and additional usability problems could be avoided (positive).
3. Not making those questions could decrease recommendation quality (negative).
Below there are some examples of recommendation algorithms embedded in websites design.
|
The Amazon.com approach Amazon show recommendations everywhere, they do that also as a marketing tool because they know that users love recommendations. Amazon is so good recommending because they have a huge base of customers buying all those products to recommend. They also have a good algorithm but they don’t hide it as a precious secret, note "These recommendations are based on items you own".
This is very important: Amazon is recommending a book because the user bought a "similar" one before. Amazon knows that recommending this way is dangerous and it offers a way to fix it.
Here there is a pop-up window to solve the potential problem. Amazon might be recommending a book because the user bought a similar one, but that one could have been a gift for another person.
Amazon is not only showing a part of how its algorithm works but it also asks users for input to improve results. Here there is another important thing to check: is it OK to ask users questions to improve website recommendation algorithms or should these algorithms be intelligent enough to avoid bothering users?
The Amazon recommendation algorithm in action:
The Last.fm approach Last.fm, the music recommendation website, adopted a different design solution. At it, users don’t know where the recommendations come from. One might guess but nobody really knows if those artists are there because the user likes other similar artists or because someone is paying to push artists into that list. There is no message like "we recommend Shakira because you like Beyonce". A valid approach? Last.fm makes a good job recommending music without disturbing users. On the other side this kind of state of user ignorance could be translated into a drop in trust.
Last.fm also has a radio set for each user with special recommendations. Here the recommendation algorithm behavior could be easier guessed, the player has controls to "love" or to ban a song. There is a serious problem here, once you ban a son there is no way back.
StumbleUpon StumbleUpon is a web browser plugin that let users discover new websites based on recommendations and ratings done by other users and the user himself. StumbleUpon works fine and improves the recommendation quality as the user rates more websites. Like Last.fm, StumbleUpon also decided not to show the user why a particular website was recommended to him. In this case the algorithm might be too complicated and [I guess] they decided to improve usability taking out additional information. But keep in mind that a lack of information also negatively impacts on usability and trust.
|
External Resources About Recommendation Systems
- Amazon.com Recommendations: Item-to-Item Collaborative Filtering (PDF file)
- How Netflix Recommendations Works
- Recommender Systems on Wikipedia
Popularity: 62% [?]
