entropy again (case of multilinks)

[holtzermann17]

In #29 we talked some about entropy of words (interesting reading)!

I noticed one potential simply application. This article contains 5 links for the term "equivalent":

• http://planetmath.org/filterbasis
• http://planetmath.org/equivalenceofforcingnotions
• http://planetmath.org/equivalentmachines
• http://planetmath.org/equivalencerelation
• http://planetmath.org/metricequivalence

(AFAICT, none of them are correct.)

I think that when we have multi-links like this, it's a pretty clear case of a "diluted" meaning. This is very different from the idea of entropy I was promoting in #29, where I was thinking about words like "group". The solution proposed in #44 might work better for "group".

But I feel like getting 5 links back is clear evidence that the system is overlinking (overthinking?). Can we run some tests (perhaps using the build system)?

For now I think the goal would be only be one of exploration and descriptive statistics.

[dginev]

I would say the equivalencerelation link is appropriate.

As to the overlinking, I have no clue about the MSC classes - if they were in the same MSC class as some of the other links on the page (which seem ok), then there is no way for my current algorithm to figure out they don't apply.

Each of the non-generic links for equivalence should only be used in their very specific domain (e.g. forcing notions), which they currently aren't.

I think I have an idea how to prevent very specific concepts appearing in other classes, by introducing a handicap for specificity (in a way adding some of the old MSC steering logic back in). If a concept is defined within a very concrete MSC class, it should add a penalty for being added to the group of concepts in the top-level MSC classes being compared. Hm...

[dginev]

Moving to the 3.0 release milestone.

[holtzermann17]

Here's another example that in my view pushes the limit a bit too far:
http://planetmath.org/proofthatgrouphomomorphismspreserveidentity#comment-19582
Eight outbound links for homomorphism.

This is a comment, so in order to get it to link correctly, we'd have to find a way (on the Planetary side) to send in MSC information about the parent article.

[dginev]

Haha, that's remarkable :) Maybe I should take your report as a sign to get back to improving NNexus (especially with December rapidly approaching).

As to sending in the MSC information, my implementation currently can't use any MSC info on input (I think), but maybe that is something we can enhance.