So at 6AM this morning, I began following the instructions
that
Jon Goodwin kindly sent me for topic modeling with data from
JSTOR data for researchers using his stuff plus
some stuff written by
Andrew Goldstone.
Subtract the 3 hours hours I spent dealing w/ life, and by noon I had, with huge huge huge support from both Jon and Andrew (like way more than you would expect from anyone who hadn’t actually been your BFF since grad school) topic modeled 1000 randomly sampled articles from
Off OurBacks (I really want the full 14000+ but must see if the fine folks @jstor will give them to me!)
Picture me people staring at my mac terminal yet not
realizing I’m running linux rather than the MS-DOS of the ‘90. Yes it took me 30 minutes to change a
directory. Then it turned out a line of code had to be changed. Oh thank you
@proghist2 for the preliminary lessons I did. I opened file in komido, made change, and saved. I was SO proud of myself. Glitch again when it turned out I downloaded xml not csv files from JSTOR. DOH, back again. Another glitch when unzipped JSTOR file put things in subfolders, but I figured out on my own that I needed to move them. Too bad I didn't remember the correct spacing for putting back together a line of code that was split by email. At this point, Jon and Andrew had each responded to approximately 100000 emails from me, that I'm sure had them snorting coffee out their noses (as at one point I put in the $ that is used to denote a line of code to be entered at the command prompt.) I got right down to the end and the final line of code would not run. I erased the last file output, regenerated, and voila it did.
Trust me when I say, if I can do this, anyone can. I am such a novice at doing anything without
the aid of a GUI that newbie doesn’t even accurately capture my lack of
knowledge.
So JSTOR limits downloads without special permission to 1000 randomly sampled articles, and since I was eager to start I didn't fuss with parameters at all. Still the results astounded me.
I absolutely did not expect clusters to appear so clearly. I cannot wait to go back to JSTOR, to play with search terms and date limits to see if it remains as clear.
update here
0 5 women men sexual sex male woman rape
girls female man pornography violence abuse prostitution gender sexuality
rights harassment human violence against women
1 5 people time feel life children work
things lot find make years change woman talk important part good thing jewish maybe women's life cycle?
2 5 lesbian lesbians feminist gay oob
issue book read writing letter books dear paper radical reading article letters
collective written maybe women's writing, maybe lesbian feminism?
3 5 women group movement political
feminist conference issues feminists black groups feminism work con working
discussion meeting white oppression present internal
movement dynamics
4 5 call send national center dc community
write box university studies contact po program college black information june
education info maybe women and education?
5 5 court state law abortion rights case
bill legal gay federal public child judge action supreme decision news order
tion women and the law
6 5 health women woman medical found
control study birth list drug cancer aids research breast pregnancy treatment
doctors high clinic women and health
7 5 women work workers union year money
support pay care leave working jobs job labor national home members million
welfare women
and work
8 5 back day time love house street days
play night called made young good place home left head hear music
9 5 women police international united
rights war violence world people peace government human groups country backs
march held page report International women’s issues
I was curious about whether a sample, again of 1000, from a shorter time period would shift the results. 1970-1980 yielded
0 5 lesbian feminist lesbians issue gay
oob movement feminists book conference issues political radical writing
feminism read books dear letter possibly lesbian
feminism
1 5 court state abortion law rights bill
case legal federal gay judge action supreme prison discrimination public
decision york states women and the law
2 5 women groups national support
government rights international united war union workers march police human
members peace held group con International women’s issues
3 5 children child years family woman time
mother home life day young mothers year age love back parents end told possibly family life
4 5 people time work feel good things make
group lot find part talk thought wanted thing place change movement felt
5 5 call dc write send box community show
po contact june national information make st gay held art center area
6 5 women men sexual sex male violence
rape woman pornography man human victims prostitution female sexuality power
harassment death society violence against women
7 5 health women medical care hospital
birth drug found woman control study cancer aids treatment percent high breast
pregnancy doctors women and health
8 5 black white woman made jewish world
life oppression days speak support struggle color racism page times country
face stand internal movement
dynamics
9 5 women work working class issues social
university education studies group center list discussion girls program ing
public project world women and work
combined with women and education
five of the themes clearly carried over, indicated in large red above. Two collapsed, women and work and women and education into one indicated in orange, and two new possibly themes emerged, indicated in blue.
ok it is late and I'm getting tired, but I just realized here is where the human interpretation of this process is faltering for me.
In run one at approx noon today I labeled this as
2 5 lesbian lesbians feminist gay oob issue book read writing letter books dear paper radical reading article letters collective written maybe women's writing
at 4PM i labeled as possibly lesbian feminism
0 5 lesbian feminist lesbians issue gay oob movement feminists book conference issues political radical writing feminism read books dear letter possibly lesbian feminism
sigh
I'm bleary eyed, but I can't resist one run at Feminist Studies a more academically oritented publication than OOB with similar run 1972 v 1970 start but through to present
0 5 women family work labor social men
workers children economic class working families domestic home gender male
control household mothers work
1 5 women medical abortion health care law
pregnancy child legal state rights welfare reproductive children policy pregnant
social prison woman health
2 5 sexual female woman male men sex life
social human mother world love part sexuality nature fact sense desire
experience women’s sexuality
3 5 women black white american african
history sexual south race racial native colonial political indian middle
western university north cultural race/ethnicity
4 5 feminist gender women social feminism
lesbian studies work political cultural identity university feminists theory
politics history analysis critical power academic feminism?
5 5 pp university paper feminist women
studies ed american eds published mary history susan gender book judith social
press essays
6 5 art body work film disability images
chicana beauty image female cultural artists people music woman space visual
story bodies images of women
7 5 women movement political feminist studies
national issues research rights work groups international american program
conference issue organizations public organization organzing
8 5 women de history female la woman mary
early men marriage french century public life nineteenth political young
american wrote
9 5 time back mother told people years
life day make felt home knew feel long man good eyes left woman
I need to look at this more, but I'm intrigued by not only the different themes such as sexuality race/ethnicity, but also by how the categories that overlap like work and health are constituted from different key words (academic activist divide?). However its been twelve long hours of this work, so I think that is enough for now.
Would love input into how I labeled and about emerging methodology
I'm glad it worked out! One tip I have is that if you want extra stop words, you can create a file in the mallet directory called "stop.txt" (or whatever you want). Open it in a text editor and add the extra stop words (like "con," maybe, or "lot"). Then, when you run the first mallet command, add "--extra-stopwords stop.txt".
ReplyDeleteI've been experimenting with using Python Natural Language Toolkit's WordNet lemmatizer to reduce words to a common form. It seems to work well, but it can be somewhat slow for larger files. It's also possible to tag only nouns, for instance, which is what some of the computer science papers using DfR have done.