Rich Morin :: tchotchkes

Graph-related notions about LinkedIn

LinkedIn bills itself as "an online network of more than 4.8 million experienced professionals from around the world, representing 130 industries". My spouse Vicki Brown has a weblog entry that gives a general introduction, but it says little about the graph-related aspects of the network. So, here are some initial observations, based on a day or two of my own explorations...

LinkedIn's "online network of professionals" is supported by a dynamic (and presumably, database-backed) web site. As a member of the network, I get several pages, listing my "contacts", showing my "profile", etc. The links between these pages are the usual ones supported by HTML, but the underlying structure is rather different. Specifically, it's a binary, sparsely-connected, typeless, undirected graph:

• binary

  Connections (read, edges) are made between pairs of members. There is no provision for defining triads, etc. However, this isn't a problem, partially because members can link to groups, etc.

• sparsely-connected

  The LinkedIn graph is very large (five million and growing), but each member only connects to a tiny fraction of it. Some members have hundreds or even thousands of links, but most have only a few dozen.

• typeless

  Edges have no "type", aside from that conferred by the nodes they connect. So, for example, there is no way to differentiate an "acquaintance" from a "friend". There is, however, a way to give "endorsements".

• undirected

  Edges can be traversed in either direction. Thus, if I have a connection to Fred, he will have a connection to me.

The connections go to other members, but it is also possible to "link" to institutions (e.g., schools, companies), groups, and interests (e.g., astronomy), by mentioning them in the profile. This allows me to search for people in an abstract manner (e.g., people who worked at XYZ Company at the same time I did, people that work in the ABC industry).

By adding connections, I increase the size and/or connectivity of my portion of the graph. This increases the network effects, but the sparsity of the graph limits this to less than that predicted by Metcalfe's Law. That is, unlike the Internet, a member can't (trivially) connect to any other arbitrary member. So, the fact that someone I don't know has joined an distantly connected subgraph will have little bearing on my own connectivity.

So much for half-baked theory. In practice, LinkedIn seems to work well for re-connecting with folks. Connecting with folks you've never met is also possible, but your mileage may vary. The recipient of a contact request may evaluate it by looking over your profile, the intervening contacts, etc. If they aren't convinced that you're "worthy", you won't get the connection.

Even though this is billed as a professional network, contacts tend to be a mix of family, friends, peers, etc. Links to peers are probably going to be more useful most of the time, but links to family and friends connect you "outside of your own circle", which can be useful when you want to make a contact.

Despite its strongly graph-based architecture, LinkedIn has few resources for graph exploration, visualization, etc. However, I expect to see this deficiency modified soon: if LinkedIn doesn't provide these facilities, others will (e.g., by means of Firefox plug-ins).

Technorati Tags: graph, LinkedIn, network effect

Graph-related notions about LinkedIn in Computers , Science , Technology - posted at Thu, 02 Feb, 10:18 Pacific | «e» | TrackBack