Sentiment analysis on Twitter using word2vec and keras

In this post I am exploring a new way of doing sentiment analysis. This model takes as input a large corpus of documents like tweets or news articles and generates a vector space of typically several hundred dimensions. Each word in the corpus is being assigned a unique vector in the vector space. The powerful concept behind word2vec is that word vectors that are close to each other in the vector space represent words that are not only of the same meaning but of the same context as well. What I find interesting about the vector representation of words is that it automatically embeds several features that we would normally have to handcraft ourselves. Since word2vec relies on Deep Neural Nets to detect patterns, we can rely on it to detect multiple features on different levels of abstractions.