The science of science | Hacker News Reader

The science of science

economist.com

34 pointspama12 comments

They are talking about Latent Dirichlet Allocation which was made by Blei in 2002 ( http://en.wikipedia.org/wiki/Latent_Dirichlet_allocation )

Here is video by him where he explains the basic model as well timeline-related hacks the article talks about - http://video.google.com/videoplay?docid=3077213787166426672#

You can very easily play with LDA yourself with this toolkit - http://mallet.cs.umass.edu/

Was I just looking at formulae or some pages taken from the Odyssey?

I guess we could get the computer to categorise it ...

I think they're talking about this one:

http://projecteuclid.org/euclid.aoas/1183143727

This work uses the correlated topic model, which extends LDA to model correlations among the extracted topics.

Edit: and this one [pdf]:

http://www.icml2010.org/papers/384.pdf

"If It Has "Science" in the Name, It's Not"

Unfortunately true. This seems to be a pop-sci cover of some data mining and semantic clustering algorithm for scientific publications. The author claims the groupings may be better than citations (less political, for instance) in showing how ideas impact a given research community.

People with last names in the "lower" part of the alphabet are also cited more than their higher-alphabetized counterparts.

[citation needed]

Trivial counterexample: science has "science" in the name, and is science.

You have problems with climate science?

"It has proven difficult to provide a definitive account of scientific method that can decisively serve to distinguish science from non-science. Thus there are legitimate arguments about exactly where the borders are, which is known as the problem of demarcation." http://en.wikipedia.org/wiki/Science#Philosophy_of_science

Science is not currently a iron-cast term, irrespective of pithy aphorism.

materials science isn't science?

There are others.

Classifying science sounds like a boring application of this algorithm. I'd be more interested to see it used for spontaneously identifying groups of related publications any kind, such as blogs.

Blogs are a simple case because the default is to refer to each other through inline hyperlinks.

I've also seen linguistic data mining studies done that suggest that blogs which commonly agree with each other, commonly agree with each other. For purposes of politeness, I acted astonished at this revelation.

Reply on news.ycombinator.com