Approach: We started with the transcript at http://www.vox.com/2016/9/26/130... and then passed the candidates' text (not the audio or the video) through a lexical / sentiment analysis layer, built on top of a part-of-speech tagger from third-party nlp libraries and the sentiment / emotion classifier provided by IBM Watson's AlchemyLanguage API.
@andreacoravos AlchemyLanguage takes a supervised learning approach to sentiment calculation. A neural net is trained on in-context English word data, labeled as positive / negative / neutral, and so feeding in a test set (e.g. the candidates' speech transcript) results in an output that represents positive / negative / neutral along with a confidence interval, which we communicate in this analysis as the 'certainty'.
MOB: Mother Of all Bots
Ummo
MOB: Mother Of all Bots