Training a model to learn word representations using gradient decent and negative sampling and then try use those representations in intrinsic tasks that measure word similarity and an extrinsic task for sentiment analysis.
