The moment they mentioned sentiment analysis as a core tool and attaching sentiment to individual words, it became evident that this approach is fundamentally f

The moment they mentioned sentiment analysis as a core tool and attaching sentiment to individual words, it became evident that this approach is fundamentally flawed. Their list of most negative words included words like "rape" and "terrorist". Consider a story about a soldier deployed in war who is fighting for his life and then triumphantly says, "I killed the terrorist." Leaving aside politics and considering this as a point in a plot arc, that would be extremely positive - but rated as extremely negative. Another go-to example for me is the sentence "I hope you die" being often rated as "positive" or "hopeful" by sentiment analysis systems. I suspect that this study has deep flaws, and that the graphs of stories meant to demonstrate correlation between story arcs instead illustrate systemic shortcomings of the model.