Neural Collaborative Filtering with Pairwise Learning for Implicit Feedback Data
Aakash Swami, Tirumala V · 2024
Deep neural networks have found wide applications in fields such as natural language processing, language translation, computer vision, and speech recognition, including recommendation systems. Among different types of deep neural networks used in recommendation systems, neural collaborative filtering (2017), which is a multilayer perceptron based recommender system, is widely explored as a general model for user-item interactions. Neural collaborative filtering is generic and can express and generalize matrix factorization under its framework. However, pairwise learning of neural collaborative filtering is less explored. In this paper, we present a framework NCF-PL, short for neural collaborative filtering with pairwise learning to model the pairwise preference of the items for the users using implicit feedback data. To model user-item interaction, both dot product and MLP are explored. Using dot product to model user-item interaction generalizes matrix factorization under the NCF-PL framework. We perform experiments on two real-world datasets to demonstrate the improvements of our NCF-PL framework over the state-of-the-art Bayesian personalized ranking framework.