This specification describes methods and systems for sentiment analysis. One of the methods includes: receiving a plurality of documents, each document having text data and for each of the documents: (1) representing at least part of the document's text data in a multi-dimensional vector space to produce vectorized text data; (2) applying a neural network to the vectorized text data to calculate a sentiment score, wherein the neural network has been trained using a two step process including (a) training the neural network with a non-domain specific training set; and (b) training the neural network with a domain specific training set; and (3) determining a sentiment score for an entity based at least in part on the sentiment scores for the plurality of documents.