Nikhil and Raghava are both members of UCLA DataRes, which is “UCLA’s first Data Science club focused on providing data-driven insights in a variety of domains.” (Go Bruins!).
The project's goal is ”to bring to light how complete an artist Curry is, [by examining] his lyrics with text mining methods to uncover specific lyrical insights”, which I think it totally suceeds in.
I want to blame it on the loss of context that comes with streaming releases (no album back cover art visible), but I had no idea TA13OO had 3 acts when I heard it last year. This write up gave me a new appreciation for the acts, especially some of the specific song topics that it pointed out.
Sentiment analysis isn't perfect and the project points out some examples of its short-comings like taking into account context; one example of this is how the word "cash" applies to many categories such as anger, anticipation, fear, joy, positivity and trust.
This example reminds me of a basic sentiment analysis tool I created in college which gave positive/negative sentiment counts on any input text. I used it on rap lyrics and saw that curse word heavy songs showed up as negative even if the curse words themselves weren't being used in a negative context. I don’t think that specific short-coming applies to this TA13OO project since most the results make sense intuitively.
I haven't even scratched the surface of the findings and the authors' analysis so please check it out and also check out the continued discussion (on topics like the validity of sentiment analysis) on the Reddit thread.