Implemented target-dependent BERT which achieves results equivalent to state of the art reported results.

Reverted back to old method of annotation. Annotated around 50 reviews with 605 opinion targets. Implemented function to write these into standardised xml file form for SA

Target-dependent features implemented, improvement of 0.0071 in accuracy compared to just target-indep features

Noun phrase parsing now returns tree instead of string. TODO: Use the tree in opinion target replacement

Lots of changes, started working on ConceptNet, CoreNLP, in order to prepare SemEval data for training classifier.