Work done in adapting general purpose sentiment lexicons for domain specific use. We specifically used word embeddings from the political domain to automatically learn new polarity values using SentProp.
The methods used are available in this repository, but they rely heavily on the socialsent package which needs:
- python (2.7)
- theano (0.8.0)
- keras (0.3.3)
- sklearn (0.19)
- numpy
- scipy
The lexicon, embeddings, tweet data are all available and can be downloaded and tested by running example.sh.