a visualization of a social network

The Perils of Collaborative Filtering Algorithms

Because of the many wonderful years I spent living in Florida, many of my close friends are there. Which means many of my Facebook and LinkedIn contacts are there as well.

As many of you probably know, when social networks are choosing which ads to show me, they depend on this network of contacts to gain some insight into what I might be interested in. There are as many strategies for these collaborative filtering algorithms as there are ad networks. They benefit tremendously from good human factors expertise in developing the user models that define what ads are likely to resonate with different target markets. Some people and some domains are more influenced by social network affinities than other kinds, such as product functionality or brand relationship.

Collaborative filtering also leads to some amusing errors, which brings me back to my Florida network. Even though Facebook knows I live in Boston now, for some reason they keep showing me ads to political candidates in Florida. Apparently, my contacts there are clicking on political ads or reading political articles and Facebook’s ad network is generalizing that interest to me.

I am not sure which is worse, these completely unrelated political ads or the ones for the florists that I have been getting ever since I was “caught” browsing for a Mother’s Day gift.

Image credit: “Social Network Analysis Visualization” by Calvinius used under CC BY-SA 3.0

2 thoughts on “The Perils of Collaborative Filtering Algorithms”

Leave a Reply

Your email address will not be published. Required fields are marked *