1
$\begingroup$

Let us say I know the number of pages users generally looked at in the same session over a period of time. I can obtain data like this:

P1, P2, 1000
P1, P3, 10
P1, P4, 0
P2, P3, 100
P2, P4, 0

Here

P1, P2, 1000

Means that Page1 and Page2 were looked at 1000 time by all users in the chosen time period.

Looking at the tiny sample above one could, for example, deduce that Page4 seems to be disjointed/very distant to the other pages, whereas P1 and P2 are close/similar.

I am just wondering whether the above could somehow be exploited to cluster similar pages. Ultimately, a page pair with a large number corresponds to some closeness.

What clustering algorithm could I use for this? I am only familiar with SOMs/k-means. Any feedback would be very much appreciated. Thanks.

$\endgroup$
4
  • $\begingroup$ What is your overall goal, and what is the full dataset available, eg do you have the pages viewed in each session? $\endgroup$
    – seanv507
    Dec 8, 2016 at 17:50
  • $\begingroup$ The overall is to put pages viewed together into the same cluster. Sorry if this is not clear. $\endgroup$
    – cs0815
    Dec 8, 2016 at 18:06
  • $\begingroup$ My question was why do you want to cluster the web pages together - if you don't specify that then its hard to identify the correct clustering. $\endgroup$
    – seanv507
    Dec 8, 2016 at 18:26
  • $\begingroup$ You tagged your Q as hierarchical clustering. What prevents you to consider the option in your Q? $\endgroup$
    – ttnphns
    Dec 8, 2016 at 22:08

1 Answer 1

1
$\begingroup$

I would first think of the data as a graph i.e. the edge (P1, P2) would have weight 1000 in your example. You might already be doing this given the way your formatted the question.

After this, there are several graph clustering algorithms and libraries in the world. The Markov Clustering Algorithm seems to be the way to go about this even though I have never done it myself

http://www.cs.ucsb.edu/~xyan/classes/CS595D-2009winter/MCL_Presentation2.pdf

$\endgroup$
1
  • $\begingroup$ Thanks. I thought about this as a graph as well. I will have a look. $\endgroup$
    – cs0815
    Dec 8, 2016 at 18:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.