COMPARATIVE ANALYSIS OF THE APPLICATION OF TRADITIONAL METHODS AND GRAPH NEURAL NETWORKS IN THE AREA OF RECOMMENDER SYSTEMS
DOI:
https://doi.org/10.24867/22BE19DosticKeywords:
recommendation systems, graph neural networks, similarity matrix, graph convolutionAbstract
Recommendation systems have increasingly become part of our everyday life and influence our choices and activities. Data used in recommendation systems can naturally be represented as a graph in which we present users and objects as nodes of that graph and their interactions as links between them. This paper compares the effectiveness of traditional recommendation systems and recommendation systems based on graph neural networks. For comparison, two systems were implemented - a traditional hybrid system and a system based on graph neural networks. Performance and results of experiments performed on both systems were evaluated and compared based on precision, recall, and F1 measure, but also mean square error in the case of continuous values.
References
[2] Neural Graph Collaborative Filtering https://arxiv.org/abs/1905.08108 [2019]
[3] GraphSage: Inductive Representation Learning on Large Graphs https://arxiv.org/abs/1706.02216 [2017]
[4] PinSage: Graph Convolutional Neural Networks for Web-Scale Recommender Systems https://arxiv.org/abs/1806.01973 [2018]
[5] Multi-Component Graph Convolutional Collaborative Filtering https://arxiv.org/abs/1911.10699 [2019]
[6] DiffNet++: A Neural Influence and Interest Diffusion Network for Social Recommendation https://arxiv.org/abs/2002.00844 [2020]
[7] GraphRec: Graph Neural Networks for Social Recommendation https://arxiv.org/abs/1902.07243 [2019]
[8] KGAT: Knowledge Graph Attention Network for Recommendation https://arxiv.org/abs/1905.07854 [2019]
[9] Deep Graph Library https://www.dgl.ai/
[10] Graph Convolutional Matrix Completion https://arxiv.org/pdf/1706.02263v2.pdf
