Balancing stability and plasticity when learning topic models from short and noisy text streams