Subscribe
Sign in
Start new thread
Ummo
p/ummo
Your personalized speech coach
Visit Product
Anshul Bhagi
Clinton vs Trump Debate Analysis — Lexical and sentiment analysis of Presidential Debate 1
3
•
Replies
Anshul Bhagi
Maker
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.
Andrea Coravos
@anshulbhagi Curious, how is "sentiment" calculated?
Anshul Bhagi
Maker
@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'.