Tag Archives: Mallet

Merry White Christmas~~

Dear Digitalists,

I have to say, this course is absolutely one the most fascinating courses I have ever taken (and I’m finishing my PhD—so I’ve probably taken the greatest number of courses here!). And I feel lucky to have met you all—you were such an inspiring group! Also a big round of applause to our two amazing professors—thank you for masterminding this seminar (a year ago I believe?); your pedagogical conceptions and curriculum designs are truly visionary.

Christmas is only a few days away, and I thought of posting something fun and Christmasy that is also related to my final project “Production of Desire, Consumption of Pleasure, and Creation of National Identity: Broadway Musicals as Cartography of US Sociocultural Values, 1920s-2010s.” In that spirit, why not run a data analysis of White Christmas, the Broadway musical adapted from a movie musical by Irving Berlin? It is by no means my favorite musical; in fact it is a pretty cheesy saccharine piece (with its own adorable moments). But so what? Christmas is all about eating candies and having some damn feel-good fun! So here we go:

Christmas_dvd_white_christmas_irving_berlin

What I’d like to see is which words stand out as topics/key words in this musical. Having been told that Mallet is best at handling topic modelling, I spent one afternoon teaching myself how to use Mallet.

I start by installing both Mallet and Java developer’s kit. Then I pull the data (all the lyrics of the 18 songs in White Christmas) into one folder under Mallet, so it’s ready to be imported. I run Mallet using the Command Line and type in commands such as “bin\mallet import-dir –help” to test it. Then I import the data and command the Mallet to create a file called “tutorial.mallet.”

1

Then Mallet does its job and picks out the key words:

3

4

I make another command to open this file, and by typing in this command “bin\mallet train-topics –input tutorial.mallet –num-topics 20 –output-state topic-state.gz –output-topic-keys tutorial_keys.txt –output-doc-topics tutorial_compostion.txt” I ask the Mallet to find 20 topics, and it generates 3 documents:
1. Topic-state.gz
2. Tutorial composition
3. Tutorial keys

5

The first one is a compressed file that outputs every word in the corpus of my input and the topic it belongs to. And here is what it looks like after extraction:

7

The second one is the breakdown, by percentage, of each topic within each original text file.

0

The third file shows what the top key words are for each topic.

00

I clean up the data, and the result looks like this:

6

10

Now since Mallet is known for generating a slightly different result each time, I have to try it at least twice. In my second try, I use “optimize-interval” to lead to better results.

9

What this does is it indicates the weight of that topic! (Under item 8, “0.04778” is the “weight” of the topic “white,” followed by key words such as “bells” “card” “snow” and “sleigh.”)

11

This topic-modelling process sounds really simple, but it in fact takes quite some time to familiarize with. This is a try-out example of one musical; for a larger corpus of musicals, Mallet’s power should be more evident.

As for the musical data analysis of my project, I’m thinking of combining Schenkerian analysis with automating chord progression using idiomatic analysis. It is a musicological approach rather than audio signal processing. However, I’m not shutting down the latter option, since it might turn out to be more comprehensible to the general public—our eventual target audience. Also a shout-out, musicians in the group (I know there are several), come talk with me!

Merry Christmas everyone! (Looking at these key words makes my mouth covet sweetness; now where is my Twix?! …. nom nom…)

~Sissi

Desperately Seeking Sustainability

Yale’s Digital Collections Center image

Apologies, Desperately Seeking Susan, for the poor pun.

 

 

 

Hi DHP14.

I wish I were having the dataset success of Liam, Sarah, James, and, I imagine, others (in my vivid imagination you are all succeeding marvelously). Liam, I tried to follow your line of investigation and wound up in a Mallet vortex that left me feeling more out of my depth than before. In order to find a tool that felt a little bit more manageable, I poked around on diRT again. Since the scope of possibilities felt overwhelmingly vast, I looked to the dhcommons directory of projects to see if some might bring me some amazing idea. Beside seemingly active projects (Entity Mapper, Boston Marathon archive, Modernist Versions Project, Pulp), there were many forgotten ends (Forget Me Not’s sadly forgotten guest book, Bulgarian dialectology) or unrealized projects (Kanon Foundation archive) or proposals unlinked to their outcomes (Fordham DH pedagogy). While I grant that this database, an initiative of CenterNet, might not be their primary focus, the seemingly short shelf-life of some of these projects seems relevant to the approaching Tom Scheinfeldt visit. In his webinar on October 14th, he discussed generating funding and the human needs of maintaining these projects both in terms of community and of maintenance.

I googled digital sustainability (I know I’m not the only anxious person). About 50,400,000 results. Jisc, historically Joint Information Systems Committee– now just Jisc, has a guide to sustainability, but it hasn’t relieved my mind much.

Now back to the task I (data)set out to accomplish. If anyone has good suggestions for text analysis tools for the tech-challenged (beyond the Manovich-maligned tag cloud), please point me in the right direction.

-Jojo

 

Also, I really enjoyed this image (even though I’m not talking about sustainability in terms of digital decomposition….)

Kyle Bean, The Future of Books

Data Set: Topic Modeling DfR

Hello Praxisers, I’m writing today about a dataset I’ve found. I’ll be really interested to hear any thoughts on how best to proceed, or more general comments.

I queried JSTOR’s dfr.jstor.org Data for Research for citations, keywords, bigrams, trigrams and quadgrams for the full run of PMLA. JSTOR gives this data upon request for all archived content. To do this I had to request an extension of the standard 1000 docs you can request from DfR. I then submitted the query and received an email notification several hours later that the dataset was ready for download at the DfR site. Both the query and the download are managed through the “Dataset Requests” tab at the top right of the website. It was a little over a gig, and I unzipped it and began looking at the files one by one in R.

Here’s where I ran into my first problem. I basically have thousands of small documents, with citation info for one issue per file, or a list of 40 trigrams from a single issue. My next step is to figure out how to prepare these files so that I’m working with a single large dataset instead of thousands of small ones.

I googled “DfR R analysis” and found a scholar, Andrew Goldstone, who has been working on analyzing the history of literary studies with DfR sets. His GitHub  contains a lot of the code and methodology for this analysis, including a description of his use of Mallet topic modeling through an R package. Not only is the methodology available, but so is the resulting artifact, a forthcoming article in New Literary History. My strategy now is simply to try to replicate some of his processes with my own dataset.