Designing a movie recommender system with two cnn_lstm networks

My project is recommender system using combination of two deep networks cnn-lstm. Dataset used movielens 100k. The written code runs, but the output (MAE and RMSE) is very low and better compared to all the articles. And this makes me suspect that the dataset may have been manipulated and not all the dataset has been used. What I need is for the written code to be analyzed and checked if everything is correct, everything means data and output and codes. And if there is a problem, correct it properly. The programming language used is MATLAB.