Sentiment analysis is an important component to anyone interested in studying large datasets of text. Like most topics in this course, James has set forth to highlight some of the underlying functionality that generates sentiment analysis output.
First, it is interesting to note that part of speech tagging, lemmatization, and rate words, elements essential to sentiment analysis are direct descendants of the work linguists have been doing for many years. The field of computational linguistics interfaces quite nicely with the work being done in digital humanities particularly for this reason. This again, is a moment of innate interdisciplinary work.
Sentiment analysis is also hyper-reliant on the human component. Interpretation of data is brought center stage in this type of research as we can not determine the accuracy of the output without interpreting the results. For the Mets vs. Royals example that James gives in his slides, there is contextual information regarding the World Series, the stakes of the game, the date, etc that come into play in understanding the data. Having accurate results is an important step, but the researchers interpretation is fundamentally important. For computer scientists/computational linguists to create more accurate sentiment analysis, digital humanities scholars and the like must utilize and improve upon the output.
From a professional standpoint, I have come across sentiment analysis in brand management platforms. One such example is Trackur, which allows a brand to see positive, negative, and neutral mentions of it’s products/services. Essentially, the platform will aggregate mentions of specific keywords allowing the brand to see the sentiment associated with it. The announcement for this feature includes the important caveat: “It’s my philosophy that only a human can accurately navigate the nuances of the human language and understand with 100% accuracy whether a tweet, update, or posts is positive or negative.” The kicker is that this is an expensive platform. This speaks to James’ frustrations with access to data. The value of sentiment analysis in the workplace, like all random KPIs is being able to tell a valuable story with it, strategize next steps, and create active plans to make improvements to said KPIs.
In my research in particular, sentiment analysis will have value as I begin to work with the textual elements of a large set of picturebooks. Looking at a corpus of say Dr. Suess books, it would be interesting to see the sentiment analysis of the entirety of his texts. The lemmatization has potential problems considering how much invented language is used, but that’s a problem that we can tackle with a Suess specific dictionary.