This is the type of question other people can hardly help you with. You need to find out why your model generates different results. For that, you usually need to dig deep. Along which dimensions does your model differ from the one you cite? Can any of those explain the difference in results? That often involves a lot of work in the form of adding and removing features from your model.
It would probably be worth thinking about the differences between the model you want to work with and the one in the paper you cited. Can you replicate the paper’s results with their model? If so, you could take that as your starting point and then do as @jfeifer suggests: once you have something you are reasonably certain is correct you can add things to get to the preferred specification.