Recently, Deep learning has enjoyed remarkable success in Natural Processing Language , establishing new state-of-the-art performance in language modeling, machine translation, document classification, and sentiment information retrieval. This breakthrough in artificial intelligence is associated with the availability of massive amount datasets and the high computing power. In this study, we address the problem of sentiment prediction with zero prior knowledge using a state-of-the-art deep learning techniques : autoencoders and denoising autoen-coders. We first survey the latest deep learning models for learning semantic representation. Then, we develop a theoretical and experimental study of these techniques that cover sentiment analysis and semantic representation.
Keywords : manifold learning, computational semantic, learning representation, Dimensionality reduction, autoencoders, cross-entropy, semi-supervised learning.
Link to the class : https://perso.limsi.fr/xtannier/teaching/aic_rei/index.php