Comparison of NCF and PMF using Movie Lens dataset

Two well-liked methods for creating movie recommendation systems are probabilistic matrix factorization (PMF) and neural collaborative filtering (NCF). PMF is a probabilistic method of matrix factorization, whereas NCF uses neural networks to learn user-item interactions. The performance of NCF and PMF on the MovieLens 100 K and 1M datasets, which comprise user-rated movies, is compared in this abstract. The main conclusions are as follows: PMF can be a good option for non-commercial implementations where novelty is not a requirement because it combines simplicity with accuracy. In terms of diversity, uniqueness, prediction accuracy, and recommendation accuracy (for both unordered and ordered lists), NCF performs better than PMF. When applied to the MovieLens datasets, NCF yields better overall quality results than PMF, but it is more sophisticated and requires more training data. In conclusion, NCF outperforms PMF in terms of accuracy and beyond-accuracy metrics like novelty and diversity, but at the expense of greater complexity. Both models are useful for suggesting movies. The recommendation system’s particular criteria will determine which option is best.