Methods

One strength of the HathiTrust Analytics Research site is its use of the general HathiTrust search site to offer collections, which serve as a corpora-building tool. The collections tool allows a researcher to query multiple terms across historical dates, restricted by media types including journals, newspapers, serials, and books, and then save those items to a digital collection. A researcher can then download any public collection as a workset through the HathiTrust Analytics research center website. It was this accessibility measure which drew me to HathiTrust as a possible site to model shellshock across newspapers.

There are multiple tools available through the HathiTrust Analytics site, including a Token Count and Tag Cloud Creator, an Extracted Features Download Helper, and a Named Entity Recognizer, which generates a list of all of the names of people and places in a workset. These tools may be helpful for further analysis, but for my purposes, I chose to focus on the InPho Topic Explorer, which trains multiple Latent Dirichlet Allocations, or LDAs, to extract topics which occur across a given corpus. This method “allows you to export files containing the word-topic and topic-document distributions, along with an interactive visualization,” according to HathiTrust’s documentation.

This tool draws upon a topic modeling technique known as Latent Dirichlet Allocation, which allows a researcher to extract and cluster topics from a workset or corpus. The term “latent” conveys something that exists, but is unmarked or potentially invisible. My aim was to render visible trends in my corpora, in comparison with one another.