This project initially aimed to perform a sentiment analysis on Board of Directors' minutes regarding country matters, but I soon realized my expectations, that canned dictionaries don't work for the IMF which has it's own "Fund-glish".
Hence, with this project, I trained a corpus using IMF language, and performed some basic NLP such as word similarities and analogies. I demonstrate the results of the algorithm using a Fund-invented term, "evenhandedness", and am fairly pleased with the results.
Unfortunately, sentiment analysis could not be performed because the data has to be labeled...
The code is viewable here.