OK, this is a tricky one. By selecting only one work by each author, you’re not really looking to see clustering based on the author’s styles so much as . . . .something else. But what? I’d be curious both about your assumptions of what you expected the clusters to look like, as well as what “signals” you think are being expressed in the clust…[Read more]

Yes, I think you got some very legible results off the bat. Clearly, most of these authors are clustering at the level of “author signal.” I would wonder about the larger break you have between the group on the top and the one on the bottom—is that explained by time period? genre? sometihg else?

I’m also unsure what you mean by “Africa…[Read more]

I think you got some very legible visualizations. I’m not at all surprised to see Austen cluster so differently from Twain and Dickens, although the fact that those two cluster together using 100 MFW does raise the question of what part of the “signal” distinguishes her: is it her gender? genre? time period? Dickens was a little earlier than Auste…[Read more]

I like the idea of a corpus of fairy tales because it raises the question of how to define that as a genre. Like . . . what do “christmas stories” have in common with Greek myths, and can see those similarities happening at the level of most frequent words? Because you’re dealing with a variety of languages and translators and countries and time…[Read more]

Great experimentation with trying different numbers of words. Now you’re left with the question: which one is right?! Obviously that’s a trick question, so instead we should consider what we’re seeing when we see 100 vs. 1000 words: at these two different scales what details emerge that result in different clustering?

I like that doing your…[Read more]

I also find the cluster analysis easier to look at, but I think that raises an interesting question of “why”: are just more used to seeing visualizations like that? Or does the dendrogram convey certain kinds of information in a more useful way than the consensus tree?

Did you try the cluster analysis with only 100 words? One important thi…[Read more]

I wonder what kind of “popularity” Gutenberg captures, and if there’s a way to measure that. I think you’re saying something about how popularity “lasts” over time, with some authors remaining perennially popular and others (who we don’t remember) fading from view.

The idea of “isms” is pretty interesting, and I think speaks to a theme of thi…[Read more]

I wonder if the Gutenberg people have published anything about their downloads; like if they know what kind of people tend to use their site and what that might say about why some downloads are more popular.

One challenge for you down the road will be to consider what this data set represents; that is, what exactly is it that you have…[Read more]

Your search took quite a turn! I think it will be interesting to think about the associations you draw between short stories and online writing and political speeches. One thing that will be interesting to add later in the semester are notes about the speech writers—you may find some patterns in language use that presidents use, but you may also f…[Read more]

I think you made a smart choice in selecting your data set; when you’re more comfortable later in the semester you might consider pursuing some of the other approaches you mentioned to give context. We might also see if we can’t find anthologies or collections that would give your data set a little more shape, or make it feel less random. You…[Read more]

Great topic! I think one interesting aspect is that you’ll be getting a mixture of genres; this might make things confusing, or it might help you sort out the genre signal more easily. By adding in fictional accounts of the witch trials, I’m guessing you would see a very clear separation when we do clustering.

As you start to explore the dat…[Read more]

Hi there.