Abstract:<p>Personalized recommendation has been widely applied to many real-world services. Many of recent studies focus on collaborative filtering (CF) by deep neural networks, which pursue to predict users' preference on items based on the past user-item interactions (e.g., a user rates an item). A general CF approach consists of two key modules, embedding representation learning and interaction modeling. In most existing methods, the embedding module is followed by the interaction modeling module, and the user-item interaction information is only emploited in interaction modeling directly. Existing methods, however, defectively overlook the correlation between users and items, as well as the inherent connection between embedding learning and the interaction information. To fill this gap, we propose neural graph personalized ranking (NGPR) which directly makes use of the user-item interaction information in embedding learning by incorporating the user-item interaction graph in embedding learning. Specifically, we construct the user-item interaction graph using <em>de facto</em> interaction between a user and an item. Correlation between users and items can also be reserved by concatenating representations of users and items in the entire procedure of embedding learning. Moreover, more complicated structures like multilayer perceptron (MLP) can be used in interaction modeling to make the most use of the representations, rather than simple linear transformation. We conduct extensive experiments on three public benchmarks and demonstrate the superior performance of the proposed NGPR model on personalized ranking task. In addition, our ablation studies verify that our novel design to incorporate the user-item interaction graph in embedding learning is effective.</p>

What problem does this paper attempt to address?

The paper attempts to address the problem of how to more effectively utilize the interaction information between users and items in personalized recommendation systems to improve recommendation accuracy. Specifically, existing deep learning-based collaborative filtering methods have two main issues in the embedding learning and interaction modeling processes: 1. **The relevance between users and items is not directly encoded into the embedding representation learning**: Existing methods typically learn the representations of users and items in isolation, leading to the loss of relevance between users and items. This relevance is crucial for accurately modeling user preferences. 2. **The connection between embedding learning and interaction modeling is ignored**: Most existing methods use only simple inner product or bilinear decoders for interaction modeling, without fully leveraging the discriminative embedding features of neural network models. Therefore, interaction modeling can adopt more complex structures rather than simple linear transformations. To address these issues, the authors propose a new recommendation model—Neural Graph Personalized Ranking (NGPR). This model improves existing methods in the following ways: - **Embedding Learning**: By constructing a user-item interaction graph and incorporating user-item interaction information during the embedding learning process, the model directly utilizes user-item interaction information to learn representations. Additionally, by connecting the representations of users and items throughout the embedding learning process, the relevance between users and items is preserved. - **Interaction Modeling**: Using a multi-layer perceptron (MLP) model endows the interaction modeling with nonlinear transformation capabilities, thereby more effectively extracting the nonlinear collaborative signals between user embeddings and item embeddings. Through extensive experiments on three public benchmark datasets, the authors demonstrate the superior performance of the proposed NGPR model in personalized ranking tasks. Furthermore, ablation studies validate the effectiveness of the new design that incorporates the user-item interaction graph into embedding learning.

Neural graph personalized ranking for Top-N Recommendation

Multi-Behavior Enhanced Heterogeneous Graph Convolutional Networks Recommendation Algorithm based on Feature-Interaction

Enhanced Sub-graph Reconstruction Graph Neural Network for Recommendation

Personalized Graph Neural Networks with Attention Mechanism for Session-Aware Recommendation

Node-personalized multi-graph convolutional networks for recommendation

Research on Personalized Recommendation Algorithms for the Web with Knowledge Representation

HNF: Hybrid Neural Filtering based on Centrality-aware Random Walk for Personalized Recommendation

Bayesian personalized ranking based on multiple-layer neighborhoods

Personalized Session-Based Recommendation Using Graph Attention Networks.

Personalized Recommendation Via Inductive Spatiotemporal Graph Neural Network

Knowledge-aware fine-grained attention networks with refined knowledge graph embedding for personalized recommendation

Recommendation Ranking Method Combining Graph Convolutional Network and Factorization Machine

IHGNN: Interactive Hypergraph Neural Network for Personalized Product Search

Research of Personalized Recommendation System Based on Multi-view Deep Neural Networks

ITEM: Improving Training and Evaluation of Message-Passing based GNNs for top-k recommendation

Cluster-Enhanced Federated Graph Neural Network for Recommendation

Interaction Graph Neural Network For News Recommendation

A Graph-Neural-Network-Based Social Network Recommendation Algorithm Using High-Order Neighbor Information

NGAT4Rec: Neighbor-Aware Graph Attention Network For Recommendation

Graph neural networks for preference social recommendation

Fusion of Personalized Implicit Relations for Social Recommendation