Comments for Aperte.org

Comment on So you want an elected Senate? Think twice. by Jeremy Handcock

Jeremy Handcock — Tue, 30 Nov 2010 07:18:42 +0000

@Sabrina S.: Yes, I compiled the graph using data obtained from the LEGISinfo database. Feel free to cite these data, however, note that the numbers haven't been reviewed by anyone but myself.

Comment on So you want an elected Senate? Think twice. by Sabrina S.

Sabrina S. — Mon, 29 Nov 2010 17:17:58 +0000

Did you compile the graph? i need to do something similar for a paper and if I use your graph as part of my research then I need to cite it properly. Thanks!

Comment on Learning from strangers by Learning from strangers | Catenary

Learning from strangers | Catenary — Thu, 25 Nov 2010 02:04:20 +0000

[...] got a copy of "Learning from Strangers," by Robert S. Weiss, after reading about it in Jeremy Handcock’s blog. It’s such a good book that I wanted to bang my head against a wall for not reading it [...]

Comment on Learning from strangers by Whitney Hess

Whitney Hess — Mon, 09 Aug 2010 02:39:59 +0000

Thanks so much for the link love!

Comment on Design that disappears by Jeremy Handcock

Jeremy Handcock — Sun, 08 Aug 2010 16:24:31 +0000

Bang on! I didn't know anything about Heidegger, but Weiser actually references "ready-to-hand" in his article:

Whenever people learn something sufficiently well, they cease to be aware of it. [...] Computer scientist, economist, and Nobelist Herb Simon calls this phenomenon "compiling"; philosopher Michael Polanyi calls it the "tacit dimension"; psychologist TK Gibson calls it "visual invariants"; philosophers Georg Gadamer and Martin Heidegger call it "the horizon" and the "ready-to-hand", John Seely Brown at PARC calls it the "periphery".

Indeed, the challenge is in the execution, which I guess is the reason why Weiser's article sounds like it could have been written yesterday. These are great thoughts to tuck away in a back pocket when getting down to the details of a design, though. Another reason that his words sound so present may be that such design thinking has only recently entered the mainstream in software development.

Comment on Design that disappears by Jorge Aranda

Jorge Aranda — Sun, 08 Aug 2010 14:26:55 +0000

These "designs that disappear" concept sounds like Heidegger's idea of readiness of objects; see for instance the entry on "Ready-to-hand" here: http://en.wikipedia.org/wiki/Heideggerian_terminology

Of course, the trick is to bring the concept to fruition...

Comment on Almost famous by Jeremy Handcock

Jeremy Handcock — Fri, 11 Jun 2010 01:55:47 +0000

Hi Neil, thanks for the tip on ArXiv.org. Yeah, an ICSE workshop might be a possibility. I'll watch to see what comes up for 2011.

Comment on Almost famous by Neil

Neil — Wed, 09 Jun 2010 13:56:58 +0000

Jeremy, you should consider posting this on ArXiv.org. That way it will be more accessible and longer-lived (should you not renew your domain for example).

I wonder if another venue would be suitable. Maybe an ICSE workshop?

Comment on Cheap data mining tricks by Jeremy Handcock

Jeremy Handcock — Wed, 12 May 2010 00:38:41 +0000

Hi George, thanks for your comment. You're correct in that a distance derived from the Normalized Google Distance (NGD) isn't really comparable to a distance derived from LSI, so it's inappropriate to classify any technique using NGD as an "alternative" to LSI. Instead, NGD is a single technique in the complete set of semantic analysis techniques (as is LSI). Obviously the appropriateness of each technique depends on the application. My apologies for the confusion. The point about scalability is not necessarily with respect to the application discussed in the paper. Instead, it's about employing a semantic distance measure at huge scale (e.g., the scale of the world wide web) in which case LSI would not be appropriate. Consider the example of computing a semantic distance between 'horse' and 'saddle' using all the documents in the Google search index as the document set.

Comment on Cheap data mining tricks by George

George — Tue, 11 May 2010 21:21:41 +0000

So you have 4323 disceases and 2585 proteins? And in order to compute the co-occurence counts you need to do a scan over 16,000,000 documents? But the hypothetical term document matrix would only end up being 4323 by 2585? Am I understanding this correctly? If I am, why is there any problem at all in running the most naive matrix factorization method on the small matrix you make? Furthermore, your post sets up what seems to be a false dichotomy between using LSI and using the NMD formula. You could easily run SVD on a matrix with entries computed using the NMD formula. What is your actual point about scalability?

As far as I can tell, the paper just picks a formula for computing similarity based on simple count statistics. This isn't really an LSI alternative except in the sense that "not doing LSI" is an LSI alternative. What I mean by that, is that the statistics you compute would have been the input to LSI. So a situation where someone might want to use LSI would almost be by definition a situation in which the pairwise NMD matrix isn't sufficient for the task.