> it shows us once again how clustering algorithms fail What do you mean by 'fail'? It looks like the algorithm ultimately did what they wanted it to, so do yo

> it shows us once again how clustering algorithms fail What do you mean by 'fail'? It looks like the algorithm ultimately did what they wanted it to, so do you mean some sort of relevance or practical failure? All clustering algorithms do is propose groupings in which intra-group links are stronger than inter-group links. The researchers measured the modularity of their clusters, where 0 is 'no better than random' and 1 is 'networks with strong community structure', and scored over 0.9 (though they had to make some changes to eliminate long-distance linkages). I agree that Google would have much better commuting data, if they chose not to anonymise their maps data collection, because they could determine where your phone usually spends the night and the middle hours of the day... as well as the route taken to get there and back.