Recommendation models play a significant part on the internet. They combine user and information data to create personalized recommendations for various users. But this may be hidden by bias. Exposure bias is a kind of bias that blurs the distinction between a disliked item and an unshown one that may be likable. Countless research has been done to minimize this kind of bias. Our research will focus on using a new algorithm called ExpoMF++", an enhanced version of Exposure Matrix Factorization. This enhancement is done in two ways: replacing the simple dot product with neural collaborative filtering and adding an optimization with the Gaussian mixture model. We tested this new model on different datasets of various sizes and attributes. Our model proves functional and works better than the original ExpoMF, significantly reducing problems caused by exposure bias. We then compared our model with other standard methods for exposure bias and found sound results.