Updated Nov 2, 2017: New version with LSTM model, as presented at Turku.ai.
Samuel Rönnqvist
In deep learning, a key principle is to learn abstract representations of data, rather than relying on manual feature engineering, as suitable input for modeling tasks. For natural language processing, this approach may provide means of obtaining representations from data that are more complete, generalizing and flexible to construct. This tutorial focuses on word vectors (semantic space embeddings) as a representation of word-level semantics, for use in classification tasks. We will use some helpful Python libraries to train word vectors from unlabeled text, and use them as features for classification, using neural networks in both steps.
Read more: http://users.abo.fi/sronnqvi/deepnlp/
Download data for sentiment analysis at: https://www.kaggle.com/c/word2vec-nlp-tutorial