CUNY dissertation topics
summary
topic modeling 16,738 dissertation abstracts across 45 years.
overview
I trained a topic model on 16,738 dissertation abstracts from the CUNY graduate center, from the last 45 years. out of 25 topics I followed the six that changed the most.
six of the 25 topics found in 45 years of CUNY dissertations. the bigger the circle, the more that year's dissertations wrote about it. top row: topics that grew. bottom row: topics that shrank. scroll to go from 2000 to 2023.
now each circle becomes a line, retracing its topic's weight year by year from 2000 to 2023.
three topics climbed and three fell. social, race & community gained the most (+4.77 pp); chemical & material processes lost the most (−3.49 pp).
reading the circles
each circle is one topic, and its size is how much of that year's abstracts went to it. the number under it is how much it went up or down between 2000–2009 and 2017–2024, which is how I picked them.
what stands out.
- social, race & community roughly tripled, from 3.68% in 2000 to 11.15% in 2023, its highest year.
- literary, music & cultural works was the heaviest of the six every year until 2021, when social, race & community overtook it. it did so again in 2023.
- chemical & material processes roughly halved, from 7.60% to 3.91%, with a low of 2.90% in 2022.
- education & schools and politics, state & policy both peaked in 2017.
the colors match the dashboard below, which follows all six topics across the whole dataset.
cleaning
what surprised me about the data was the artifacts I had to remove, such as lcub or rcub for left curly brackets, right curly brackets. they were prominent in number enough that they could skew the results, and needed to be cleaned.
doubts
the names are my interpretation, and anything before 1980 is too thin to compare. something I would tell to someone who would doubt topic models is to experience using one to get a look at where the reasoning is coming from.