The data below encodes an acquaintance network. The challenge is to divide it into three groups of similar sizes (not necessarily identical) aiming to maximise familiarity among groups. The idea is that it is very hard to do it from just being presented with a list of edges such as below; it becomes easier when visualising it, especially with a force directed layout. This can be tied to a much larger network, for which visualisation becomes less helpful again. The list of edges is also intended to impress on the audience that that is what computer algorithms deal with; the input is raw unstructured data. Especially with clustering, it can be deceptive to explain the difficulty, as any visualisation of cluster structure suggests the problem is easy. So my approach is to use visualisation as an aid, but stress that such perception is not available to algorithms in the first place and that it does not generally scale in the second place. This data is basically a toy example I have used in many places to explain mcl clustering (e.g. Figure 2 page 9 of http://micans.org/mcl/lit/INS-R0010.ps.Z). The best grouping is (using the first letter of each name) {a, b, f, g, j}, {b, c, e}, {d, h, i, k, l}, leading to only four 'cut' acquaintance pairs. Data:
ann bob
ann fred
ann gillian
ann john
bob ann
bob charlie
bob esme
charlie bob
charlie daniel
charlie esme
daniel charlie
daniel harriet
daniel indy
daniel kim
esme bob
esme charlie
esme gillian
esme harriet
fred ann
fred john
gillian ann
gillian esme
gillian john
harriet daniel
harriet esme
harriet indy
harriet kim
indy daniel
indy harriet
indy kim
indy lucy
john ann
john fred
john gillian
kim daniel
kim harriet
kim indy
kim lucy
lucy indy
lucy kim