Text this: Dynamic Fusion of LSTM Predictions Using Reinforcement Learning-Based GOWLA for Human Activity Recognition