Engaging in the text mining praxis assignment was very interesting as I haven’t ever actually initiated a text mining project. Working with large volumes of text can be tedious, but the importance of the contents of passages deserve adequate analysis, which text mining programs are useful for.

The introductory materials to the assignment, Getting Started with Text Mining, was a helpful overview which explained not only what text mining is, but why it matters and the basics of how it can be done. Transforming text to data can be a fun process, and the programs make it accessible to first time text miners, such as myself. As a student who work largely in statistical analysis of large data sets, it was cool to see the statement “Getting your data to look nice takes FAR LONGER than you want it to, than you think it should, than you think it deserves to. It is arguably the most difficult and crucial part of the process.” This is something that I have always had to remind myself when going through the data cleaning process: it is tedious and laborious, but absolutely essential.

After playing around with a few of the programs, I jumped into Voyant, as this seemed to be the most approachable. The text that I chose was a publication by Corina Boar and Simon Mongey, “Dynamic trade-offs and labor supply under the CARES Act”, published in August of 2020. This publication explores a statistical analysis of the impact of the CARES Act on workers tendency to return to work after being furloughed or laid off at the outset of the pandemic. After a few analyses, the researchers found that the CARES act, in a dynamic model rather than a static one, did not have a causational relationship with workers not returning to the labor force. There were a number of reasons for this, if you have interest I highly recommend looking it over here.

The outcomes were interesting, and it took a bit to digest the output of the Voyant program as a first time user. The dashboard looked like this:

The dashboard is highly colorful and includes a lot of information that is useful for analyzing the text. The text was given a readability index of 10.978 and a vocabulary density of 0.187. The two aspects of the dashboard that I found most interesting to toy with were the Cirrus, which is a word cloud highlighting the density of terms in the text, and the frequency chart, displaying a similar bit of information in a more interactive and more quantitatively-presenting way (images below).

This was a rather dense article I have been working through to analyze the economic impacts of the CARES Act, and this text mining tool was helpful in presenting data which showed the importance of topics within the paper. Playing around with the tools showed a new way of analyzing texts which I was previously unfamiliar with and was glad to be exposed to.