<br><div class="gmail_quote">On Mon, Jul 14, 2008 at 3:49 PM, Michael Stone <<a href="mailto:michael@laptop.org">michael@laptop.org</a>> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div class="Ih2E3d">> Otherwise how can we reasonable sort/group the activities in any way<br>
> that makes sense?<br>
<br>
</div>I suggested one (stupidly slow, but very general) approach based on the<br>
Travelling Salesman problem. To recap:<br>
<br>
Regard all activities as nodes in a fully connected graph. Let<br>
activities state that they are close to some collection of other<br>
activities according to any system you please. (e.g. specify a metric on<br>
activities, plug in some heuristics on names and numbers, give a list of<br>
'similar' activities, do cosine similarity on keyword vectors, etc.)<br>
<br>
When you learn that A thinks it should be close to B, shorten the edge<br>
between A and B.<br>
<br>
"Solve" the TSP for the graph. Approximate solutions are fine.<br></blockquote><div><br></div><div>An interesting idea, for sure, but not ultimately what concerns me. We actually know the "grouping"...it will be determined by the signature of the bundle, with broken signatures splitting the grouping. I was more concerned with sorting the available versions in a "meaningful" way. Perhaps this is too lofty a goal.</div>
<div><br></div><div>- Eben</div><div><br></div></div>