Possible Applications and Limitations

There are a number of potential limitations in relying so completely on HathiTrust for a digital humanities project. One limitation lies in the reliability of servers (there is a limit to storing your files). At one point, while I was still learning how to navigate the research center, the server went down for maintenance, leaving me unable to continue my research for the day.

Another limitation lies in HathiTrust’s search system and its tendency to include duplications. According to Eleanor Dickson, Harriett Green, Leanne Nay, Angela Courtney, and Robert McDonald’s white paper, there are some existing issues with metadata access within HathiTrust’s workset building tool. A keyword search can index multiple full-text versions of the same document. However, when doing keyword searches, multiple plain-text duplicates from the same title are not separated when downloading corpora from a preselected collection, meaning that sometimes data can represent accidentally duplicated volumes, instead of meaningful trends across different texts (5).

I appreciated HathiTrust’s visualizer, though I found it frustrating that there is no way to export the.json files produced and embed the visualizations on a website. Thus, I settled for taking screenshots and embracing the somewhat more clunky aspect of this project. I see this clunkiness as a form of transparency, as I am still learning digital humanities modeling techniques. This brings me to one big factor in favor of HathiTrust’s Analytics Research Center: it allows humanities scholars who may not be remotely versed in topic modeling to get started immediately. Through the familiar format of a keyword search, scholars can create worksets and working within a research capsule, can apply algorithms using HathiTrust’s documentation.

As I worked with this form of topic modeling, I found it helpful in noting overall trends and unseen patterns across the four corpora I addressed. One strength of LDA topic modeling was its ability to help me embrace the ambiguous, the confusing, and the just-plain what is that? moments which enliven my research. In effect, a disembodied gaze can assist in breaking documents up into topics, to show how certain terms may be used with much more frequency than others, and in allowing me to notice trends as they emerged throughout the topic models.