Conversational English Topic Modeling

This was a term project I worked on for the course 36-468 - Special Topics: Text Analysis at Carnegie Mellon University in the Fall 2023 semester. In this project, I constructed structural topic models from transcriptions of 1-on-1 video calls from the CANDOR American English corpus. With these models, I observed similarities and differences in prevalent discussion topics between people of the same or different demographic groups (based on sex/gender, race, and age) and inferred explanations of such through the lens of American culture.

Some key findings I found were that most conversations discussed things such as entertainment, education, and places, and that more prevalent keywords in different topic models may be affected by social norms of gender or life stages in age (i.e. video games being discussed more among men and younger people, television being discussed more among women, side hustles being discussed more among those in the workforce).

I would like to give my greatest thanks to Dr. David Brown for being very helpful and supportive of my efforts in and out of class, and to my classmate Marion Haney for helping me figure out how to use the relevant libraries in R to create and visualize the topic models.