So there seems to be a lot of skepticism about whether our algorithm can actually measure emotional impact accurately. For the long answer, I'll refer you to the about page[1] for EffectCheck. For the short answer:
My co-founders are an AI PhD and a Clinical Psychology PhD. They spent three years curating a huge dictionary of words using a methodology similar to the Harvard Psychosocial Dictionary [2], but with the twist that they were focused on lexical impact of words. The dictionary is pretty accurate at measuring both impact and sentiment [3]. For example, we can predict Amazon reviews as being positive or negative, using the stars to validate if we are correct-- blog post on that coming soon.
Regarding context of the word usage: Both behavioral studies and fMRI scans have confirmed that context is not as important as one might believe. Our brains process multiple meanings of words in parallel, and the emotions associated with those words linger in our subconscious even after we know the correct context. Similarly, in cases of reviews and comments, people who use hostility-evoking words are often hostile themselves (angry people tend to make others angry) and the same is true of the other five fundamental emotions we measure.
Happy to answer more questions. Also happy to analyze any data that you would like to see in order to verify accuracy of the algorithm-- just give me a link or the text. :)
[1] http://effectcheck.com/about
[2] http://www.wjh.harvard.edu/~inquirer/homecat.htm
[3] Note that sentiment is correlated to impact only if the writer or speaker is writing without detailed attention to word choice. For example, political speech writers comb over every word to make sure they have the desired impact-- thus, the final text likely has little correlation to the sentiment of the orator.