A Study on Dialogue Reward Prediction for Open-Ended Conversational Agents