HNF: Hybrid Neural Filtering Based on Centrality-Aware Random Walk for Personalized Recommendation

Abstract

Social computing which analyzes users’ behaviors can help personalized recommender system to extract preferences of users. Most of personalized recommender systems exploit a user-item rating matrix to learn representations of users and items for predicting users’ ratings on items. In this paper, we design a new framework, called HNF, to learn two kinds of representations and fuse them for recommendation. Our HNF consists of a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">topological neural filtering</i> (TNF) module, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">collaborative neural filtering</i> (CNF) module and prediction module. The TNF module is to learn topological representations of user-item interactions from a user-item bipartite graph constructed based on a user-item rating matrix. The CNF module is to learn collaborative representations of user-item interactions. The prediction module aims to fuse the topological representations and collaborative representations to generate hybrid representations for rating prediction. We conduct experiments on three real-world public datasets. Results validate that our proposed HNF algorithm outperforms the state-of-the-art algorithms in terms of higher evaluation metrics.