I have recently started to learn about deep Q-Learning.
If deep reinforcement learning is new for you, I think these blog posts are really useful in giving you an idea of what it really is with proper amount of details 🙂
Deep Q-Learning has been pretty effective for playing games where as a deep neural network is given an image representing the state of the game in each step and has to predict the Q-value. However, they are not yet used very much to solve language tasks. In order to get familiar with these models, I started by training an LSTM network for sentiment analysis with deep Q-learning. The result was not very good and it needs further work. Here is the GitHub link to the codes for this project.