Neural Network-Based Collaborative Filtering for Recommender Systems
Ananya Singh, Debajyoty Banik · 2023
From studying through online courses to online shopping to watching movies and TV series with the comfort of being at home, recommender systems play a major role. This chapter analyses different algorithms developed and used in the collaborative filtering (CF) based recommender systems, such as graph matching-based collaborative filtering (GMCF), neural collaborative filtering (NCF), neural factorization machines (NFM), Wide & Deep (W&D), attentional factorization machines (AFM), deep factorization machines (DeepFM), L0-statistical interaction graph neural network (L0-SIGN), automatic feature interaction learning (AutoInt) and feature interaction graph neural network (Li-GNN), which use neural networks and generally achieve higher accuracy than other approaches and can be trained on multiple objectives. The comparison is done on the MovieLens 1M dataset, and the basis of the NDCG@10 accuracy metric. This will help in the further development of this field and develop models that are more efficient than ever. The results of this chapter show that the GMCF performs significantly better than the other algorithms, mainly because of its different approaches to using inner and cross interactions because the inability of the present efforts to use attribute interactions for making good joint judgments is unavoidably hampered by a lack of understanding of attribute interaction kinds.