Target/Aspect Based Sentiment Analysis using Deep Learning

Recently, deep learning has been very effectively used to represent words/sentences/paragraphs as vectors and these have been used as features for sentiment classification. These approaches have been proven to be very powerful and have achieved state-of-the-art performances. I was wondering if any work has been done on aspect-based or target-dependent sentiment classification using Deep Learning. I would rephrase the task as follows: Assuming that we are given a sentence and a target, the goal is to find the sentiment label (pos, neg, neutral) towards that target, rather than the sentiment label for the entire sentence.