Handling information loss of graph convolutional networks in collaborative filtering

Xiong, Xin; Li, Xunkai; Hu, Youpeng; Wu, Yixuan; Yin, Jianwei

doi:10.1016/j.is.2022.102051

Cited by 9 publications

(14 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, the bifurcated graph B in Figure 1, cold-start user u1 interacts only with item i1. In this paper, we use the term "graph" to refer to the graph/network structure of the data and "network" to refer to the structure of the machine learning model [5]. Since u1 shares i2 with user u2 and user u3, all three items (i2, i3, and i4) connected to u2 or u3 can be recommended to u1 by the cf-based model.…”

Section: Introductionmentioning

confidence: 99%

Research on the Spectral Domain Graph Convolution Collaborative Filtering Algorithm Based on Reinforcement Learning and Chebyshev

Wang

Zeng

Song

et al. 2022

Wireless Communications and Mobile Computing

View full text Add to dashboard Cite

To solve the abovementioned problem, we propose a collaborative filtering recommendation algorithm that incorporates singular value decomposition (SVD) and Chebyshev truncation in spectral domain convolution. Firstly, the SVD algorithm is used to optimize the adjacency matrix, mine the potential association information between users and items, and expand the user-item adjacency matrix. Finally, based on the MovieLens-1M public dataset, the proposed algorithm (CBSVD-SCF) is compared with other commonly used algorithms. The results show that the article optimizes the recommendation effect of the algorithm based on the traditional collaborative filtering algorithm by combining the temporal order and sequence of user interaction information, as well as the popularity of items and the activity of users; the experimental results on MovieLens show that the optimized collaborative filtering recommendation algorithm can effectively improve the recommendation effect.

show abstract

Section: Introductionmentioning

confidence: 99%

Research on the Spectral Domain Graph Convolution Collaborative Filtering Algorithm Based on Reinforcement Learning and Chebyshev

Wang

Zeng

Song

et al. 2022

Wireless Communications and Mobile Computing

View full text Add to dashboard Cite

show abstract

“…We empirically test the performance of FPS-T on graph reconstruction and node classification tasks. We compare the performance to existing baselines such as message passing-based Euclidean (GCN [29], GAT [50], SAGE [26], SGC [54]), hyperbolic (HGCN [5], HGNN [37], HAT [61]), and mixed-curvature (κ-GCN [2], Q-GCN [56]) GCNs. We also add TokenGT as our baseline, which is equivalent to FPS-T with fixed zero curvatures.…”

Section: Methodsmentioning

confidence: 99%

“…To allow smooth connections between the three constant-curvature spaces, [2] proposed a model of constant-curvature space called the stereographic model, on which geometric operations such as distances and inner products are differentiable at all curvature values including zero. Incorporating pseudo-Riemannian manifolds with the GCN architecture also showed promising results [56], but its performance is sensitive to the time dimension of the manifold, which requires extensive hyperparameter tuning.…”

Section: Related Workmentioning

confidence: 99%

“…Let h u denote the encoded representation of node u ∈ V given a graph G = (V, E). Given the continous representations h, we minimize the loss function that aims for preserving the local connections [56]:…”

Section: Graph Reconstructionmentioning

confidence: 99%

“…While most GCNs learn features that lie on the typical Euclidean space with zero curvature, real-world graphs often comprise of complex structures such as hierarchical trees and cycles that Euclidean space requires excessive dimensions to accurately embed [44]. In response, the graph learning community has developed generalizations of GCNs to spaces with non-zero curvature such as hyperbolic, spherical, or mixed-curvature spaces with both negative and positive curvatures [5,37,61,2,56].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Endothelial PTP4A1 mitigates vascular inflammation via USF1/A20 axis-mediated NF-κB inactivation

Cho

Lee

et al. 2022

Cardiovascular Research

View full text Add to dashboard Cite

Aims The nuclear factor-κB (NF-κB) signalling pathway plays a critical role in the pathogenesis of multiple vascular diseases. However, in endothelial cells (ECs), the molecular mechanisms responsible for the negative regulation of the NF-κB pathway are poorly understood. In this study, we investigated a novel role for protein tyrosine phosphatase type IVA1 (PTP4A1) in NF-κB signalling in ECs. Methods and results In human tissues, human umbilical artery ECs, and mouse models for loss of function and gain of function of PTP4A1, we conducted histological analysis, immunostaining, laser-captured microdissection assay, lentiviral infection, small interfering RNA transfection, quantitative real-time PCR and reverse transcription-PCR, as well as luciferase reporter gene and chromatin immunoprecipitation assays. Short hairpin RNA-mediated knockdown of PTP4A1 and overexpression of PTP4A1 in ECs indicated that PTP4A1 is critical for inhibiting the expression of cell adhesion molecules (CAMs). PTP4A1 increased the transcriptional activity of upstream stimulatory factor 1 (USF1) by dephosphorylating its S309 residue and subsequently inducing the transcription of tumour necrosis factor-alpha-induced protein 3 (TNFAIP3/A20) and the inhibition of NF-κB activity. Studies on Ptp4a1 knockout or transgenic mice demonstrated that PTP4A1 potently regulates the interleukin 1β-induced expression of CAMs in vivo. In addition, we verified that PTP4A1 deficiency in apolipoprotein E knockout mice exacerbated high-fat high-cholesterol diet-induced atherogenesis with upregulated expression of CAMs. Conclusion Our data indicate that PTP4A1 is a novel negative regulator of vascular inflammation by inducing USF1/A20 axis-mediated NF-κB inactivation. Therefore, the expression and/or activation of PTP4A1 in ECs might be useful for the treatment of vascular inflammatory diseases. Translational Perspective The inhibitory effect of PTP4A1 in the dysfunction of vascular endothelial cells could apply to develop a drug for vascular inflammatory diseases including atherosclerosis. This study highlights an alternative as well as an effective therapeutic approach to increase A20 expression, which is an endogenous negative regulator of NF-κB signalling in vascular inflammatory disorders.

show abstract

Improvement of Graph Convolution Network of Missing Data Based on P Systems

Chi,

Liu

2023

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Handling information loss of graph convolutional networks in collaborative filtering

Cited by 9 publications

References 13 publications

Research on the Spectral Domain Graph Convolution Collaborative Filtering Algorithm Based on Reinforcement Learning and Chebyshev

Research on the Spectral Domain Graph Convolution Collaborative Filtering Algorithm Based on Reinforcement Learning and Chebyshev

Endothelial PTP4A1 mitigates vascular inflammation via USF1/A20 axis-mediated NF-κB inactivation

Improvement of Graph Convolution Network of Missing Data Based on P Systems

Contact Info

Product

Resources

About