17 April 2010

Ideas for better friend suggestions

‘Social’ is with little doubt the latest online trend. Some sites try to integrate social features because everyone else is doing it, even if doesn’t really adding anything to their business model. At least that’s what I thought when I first discovered a leading online job search site in Romania added contacts and suggestions style. It feels a wrong move on many levels, starting with the fact that this is far from an original idea and people are not about to leave established networks behind to sign up for one with far less features. Not to mention a job search site is probably the last place where you want to connect with work colleagues, especially if you are planning on leaving your current company.

There is another thing that struck me when I saw their contact suggestions: they were very bad! From a page of probably 20 people, I knew only 2, and most of the others had nothing in common with my professional background. That got me thinking about how these sites come up with these suggestions: having a detailed history of my education and professional experience, is this really the best they can do?

Other bigger and better known networks also struggle with this aspect. I didn’t use Xing very much, but LinkedIn also has a poor suggestion system. Even when you search for a specific company, there are few useful results in my experience. They all seem to rely on the model of common connections and on the contacts you supply them by uploading e-mail addresses from your various address books.

It’s not that farfetched to imagine all the information you supply on your profile can used to find people you might know in a more meaningful way. All you need is an algorithm that matches your location, the schools you attended and the companies you worked for with other people and makes suggestions based on that. It should also take into account the year(s) when you worked for that company and remove people (or at least assign them a lower probability) that left before you were hired or joined the firm after you already moved on. For bigger, multinational companies, location can also be factored in, by highlighting people that worked in the same town or country as you, since it’s more likely you met or had frequent contact with them.

Social networks connectionsI drew a small, very simple example, of how this could look like based on the life of five fictional people. The social network is in the last column, four of the people joined, and you can see the connections between them, color-coded by how they met in real life. If Michelle joins, the algorithm should suggest adding Christopher as a connection, since they grew up together and also were coworkers at their first job. Current systems would instead wait for her to supply emails manually. If she adds Christopher, the next suggestions would probably be Lisa and James, because they are strongly connected to Christopher, although Michelle probably never met them.

One potential problem with this model could be that people don’t share some of the information or input it in different forms (different spelling, abbreviations, etc.). But most of the sites have at least some standard required fields and the search could compensate by matching partial strings (it’s very likely that “Google” and “Google Inc.” are the same company, don’t you think?).

This model could be applied successfully on professional networking or job search sites, where people tend to share as much data about their background as possible. It’s less relevant for social networks like Facebook, where such details are secondary for the site’s experience.

As a side note, I think Facebook made some changes to it’s recommendation algorithm lately. It keeps suggesting people I don’t know and are unlikely to have my e-mail address, and also first time users that have no friends! Maybe they are trying to help them get started on the network, but annoying long time users with poor suggestions is not a very user-friendly move. My reason for being on Facebook is to (re)connect with people I know in real-life and receiving noise from complete strangers will slowly but surely destroy that experience.

Post a Comment