Recommender systems have become an important force in online commerce. Sites such as Amazon and Netflix have huge databases that record the purchases and preferences of every user. This allows these companies to match users with similar interests and preferences.

When a particular user visits the site, the recommendation system uses this information to recommend products that others with similar preferences also like. That’s a process called collaborative filtering and most commercial recommendation systems rely on it. 

However, there’s another way to make recommendations which has had far less less attention, say Shang Shang at Princeton University in New Jersey and a few buddies. 

These guys point out that there is plenty of evidence that preferences are contagious. That means they can flow through social networks in the same way as epidemics spread.  

So an alternative way to make recommendations is to look at the structure of an individual’s social network and predict how certain preferences are likely to spread through it. 

In the past, the factor that has limited the success of this type of prediction is a detailed knowledge of the structure of the network. But all that has changed in recent years with the huge popularity of online social networks. It’s now straightforward to see how individuals are linked.

Shang and co’s basic assumption is that if Adam likes a film, that preference will spread to his nearest neighbours on his social network–his friends–with a certain probability.  If enough people share this preference, it can cascade through the network like a does of flu.

So one way to predict that Eve will like this film is to see how close she is to Adam and how likely this preference will reach her. If Adam and Eve are close friends, this may be a relatively high probability. 

That’s an interesting idea but the acid test will be whether it works in practice. One important question is how commercially useful it will be. The social contagion model might be predictive but will it influence buying decisions in practice? Will Eve be as influenced by the recommender system as she is by the word of mouth opinion of her friend Adam?   

Shang and co don’t know but they plan to find out using data from Yelp.com, which provides user ratings of restaurants, spas etc. If it works, I’m sure we’ll be hearing from them again.

Ref: arxiv.org/abs/1208.0782: Wisdom of the Crowd: Incorporating Social Influence in Recommendation Models