another six (or seven) degrees

Jure Leskovec (CMU) and Eric Horvitz (Microsoft) have just published this paper analysing 30 billion IM conversations* on Microsoft Messager over one month.

• Staying power & age: Young folk have more conversations but older folk have longer conversations with more messages.
• US, Canada, Spain, Scandanavia & Australasia are heavy per capita users of Messenger.
• Paths within between people in the communications network have a median length of 7 nodes.

Interesting links to the work of Duncan Watts.

Thanks: Graham

social network analysis fun & games (2)

Krackhardt's Graph Theoretical Dimensions of Hierarchy; Density; E-I index; Cliques; N-cliques; N-clans; K-plexes; K-cores; F-groups; Lambda sets and bridges; Automorphic equivalence; Optimization by Tabu search; Two-mode SVD analysis...

The list goes on. SNA has its own arcane set of techniques and an equally arcane language to decribe them. Other forms of data mining have their own obscure vocabularies (CHAID, CART, k-Nearest-Neighbour) and they have been heavily used in the CRM world. As we have tended to think of customers as isolated actors, this kinda made sense - but if that assumption was once correct (and I doubt it was) then it is decreasingly true.

So can we use N-clans & E-I indexes to make more money? I mean sure we want to engage with customers, improve service, yadda yadda yadda but can they earn us cold, hard cash?

social network analysis fun & games (1)

Currently playing with various SNA tools:

• NetDraw - download it for free and have a play. I've pretty much got my head round the VNA data format and it generates cute pictures such as this:

• Maje offers nifty 3D visualisations of social networks like this*:

• And now I am wrestling with UCInet. Whereas NetDraw gives you lots of pretty pictures, UCInet is all about social network ANALYSIS. The screen shots from UCInet would involve lots of numbers formatted in ASCII text. Not so much fun, oh no. Get your harcore SNA jollies here.

*My only beef with Maje is that whilst 3D visualisations look cool, they are pretty much useless from an analytics perspective.

how long is a piece of string? (and who does it join?)

So the recent posts by Gav & Katie have mingled in my brain with the SNA work I have been doing.

Trad market research is based demographics. Although it lumps people together in groups (by age, location, wealth, gender, race, etc), it tells us little about how they interact with each other.

As Gav well knows, audience 2.0 isn't really an audience. An audience sits in the dark, only joined to each other by what they observe. This audience is noisy, they jump out of their seats. They interrupt the play. The critical thing about social media is that it's, well, social. What you have is more like a football terrace or a dance hall than a theatre/cinema/TV audience.

In SNA, you have two sets of measures:

1. Attributes of nodes - i.e. characteristics of entities or demographics if we're talking about people (which we may not be).
2. Attributes of links between nodes. Now a link between nodes is simply an interaction between them. We might have edited the same wiki page, we might communication in some fashion once a week about tennis, we might be having sex every night.

As a network analyst, you are often concerned with the interplay between node & link data - e.g. do 20-something males talk to more people more frequently on the topic of, say, organisational change. Or aftershave.

Link data is normally collected in 2 ways:

• Surveys of network participants - with all the problems of surveys (e.g. deliberate lying, wishful thinking, inaccurate recall).
• Automated collection of data (e.g. email usage patterns). In absolute terms, these are more reliable but give you no access to subjective evaluations of interactions - e.g. "that email was useful for me".

If we are going to understand the social media environment we need a solid understanding of 2. as well 1. There are several impediments to this:

• Collecting link data has traditionally been a lot harder than demographic data.
• Link data that is easy to collect is often difficult to interpret ("was that email exchange positive or negative?").
• Network-based metrics are poorly understood by marketers (in fact, by everyone).
• SNAs have tended to break down when you get more than 200+ participants - which in the consumer space is piddling. There are ways round this however.

Understanding social media and herd behaviour requires us to revolutionize our measurement techniques. Are we ready for that?

my own private archipelago

I'm doing fair bit of ONA stuff at work at the moment - using Laurie & Cai's getting-there onasurveys and the wonderful NetDraw. And in doing so, I'm reminded of this post from August concerning Facebook's Friend Wheel that highlights for me the strengths and weaknesses of network mapping.

The strengths: The "Friend Wheel" does a decent job of identifying the main groups in my life from the last 10 years. The clusters from IBM/PwC, Oracle, Calcutta Rescue, Ankali, London/Barbelith, the Australian music heads & poets that leave slight traces, the huge tangle of KM/blogging folk that I know. All beautiful people whom I am lucky to have met.

The weaknesses: You would know nothing of the people I went to schools and universities with (OK - with one exception, Mike). You couldn't see my oldest friends. The people I would die (& maybe kill) for. The vast underground of love and hate and need you will never know. And even with the folk you can see, you don't know about the joy and the pain and the people I've helped, been helped by, screwed over & been screwed over by. You see from 30,000 (air-brushed) feet.

The humility this instills in the network analyst is critical. We see but through a glass darkly - and that is better than not seeing at all. Let's not confuse the map for the territory.

More on social software visualisation

Laurel goes nuts for socialistics. Facebook sees more & more innovation around "consumer SNA" - already noted here. These aren't analytics per se - no betweenness or centrality measures here. But they are giving people the new best friend of analytics - visualisations.

Ross talks about some specialist social networking sites here and their enterprise brethren here.

The collision of social networking tools with visualisation tools will accelerate the uptake of both. Those who have been banging on about SNA for the last 20 years may find themselves the flavour of the month. Interest in SNA/ONA got a resurgence in the corporate world a couple of years ago but now seems to be moving into the consumer environment*. I think this new round of social network mapping will be extremely messy & lacking in rigour. But it will also be a fascinating group experiment as individuals try to make sense of this stuff. And manipulate for their own ends (human beings are like that). And get caught out by incorrect inferences (as anyone with SNA experience can tell you - the map is not the territory).

*For once moving in the opposite direction to other social software trends.

Facebook network mapping

Ton Zijlstra points me towards the SNS application I have been waiting for 3 years for. Someone has twigged that not only can you build social network maps from social networking software - but that this can be useful for individuals too.

Jack Vinson has also picked up on this and comes up with some good upgrade suggestions:

• navigate my network's connections (see their wheels), mostly for fun
• merge the wheels of a limited number of people, again for entertainment

This isn't a full-blown social network mapping application - in effect it only models an ego-centric network (modelled around me in the example below). Where is might become interesting is to map the members of a particular group or network and those immediately related to them - i.e. those who might be potential members of that group...