What can decades of dissertation research reveal about the intellectual landscape of a university? I explored nearly 19,000 dissertation records and built an interactive topic network from a focused set of 196 abstracts. The network makes the model's overlapping themes, characteristic terms, and representative research explorable.
Finding a focus
The collection spans disciplines and contains titles, abstracts, programs, years, and language information. English dominated the language distribution, while Spanish was the largest non-English category with 215 dissertations. Of the 209 Spanish-classified dissertations with abstracts, 196 abstracts were in English and 13 were in Spanish. I used the 196 English-language abstracts so the modeled text would be comparable.
From text to topics
I cleaned and preprocessed the abstracts in Python, removed stopwords, stemmed terms, and represented documents as bags of words. I used Latent Dirichlet Allocation to identify recurring combinations of terms, then examined prominent terms alongside the dissertations most strongly associated with each topic. I interpreted the five areas as translation and textual tradition; subjectivity, politics, and narrative; literature, history, and political change; culture, politics, and nation; and Spanish language and sociolinguistics.
Reading the model
The network connects each theme to its prominent terms. Shared words stay connected to every relevant theme; thicker links and larger word nodes reflect their relative weights in the model. Select a topic to see its primary dissertation count and examples, or select a word to see where themes overlap. These labels are interpretations developed from model output and abstracts, not definitive classifications. The network describes patterns in this 196-abstract subset; it does not measure trends across all nearly 19,000 records.