Acceso abierto

SEGNN4SLP: Structure Enhanced Graph Neural Networks for Service Link Prediction

Yuxi Lin

Yuxi Lin

School of Computer Science and Technology, Hainan UniversityHainan, China
Buscar este autor en
Sciendo|Google Scholar
Lin, Yuxi
, 
Mengfei1 Li

Mengfei1 Li

School of Computer Science and Technology, Hainan UniversityHainan, China
Buscar este autor en
Sciendo|Google Scholar
Li, Mengfei1
 y 
Nuo Chen

Nuo Chen

School of Computer Science and Technology, Hainan UniversityHainan, China
Buscar este autor en
Sciendo|Google Scholar
Chen, Nuo
  
31 dic 2024
SEGNN4SLP: Structure Enhanced Graph Neural Networks for Service Link Prediction's Cover Image
Acerca de este artículo
Artículo anterior
Artículo siguiente
Cite
Descargar portada
Publicado en línea: 31 dic 2024
Páginas: 9 - 18
DOI: https://doi.org/10.2478/ijanmc-2024-0032
Palabras clave
© 2024 Yuxi Lin et al., published by Sciendo
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1

Maup example schematic
Maup example schematic

Figure 2

The SEGNN4SLP structure
The SEGNN4SLP structure

Figure 3

The SEGNN4SLP framework demonstrates the use of route labeling (PL) on a 1-hop subgraph that includes nodes a and b. In (a), we notice a route with a length of 2, in (b) a route with a length of 3, and in (c) a route with a length of 4. Every unique pathway is depicted using a distinct hue. The nodes are labeled and the labels are presented above the nodes. Nodes with different labels are shown in distinct colors.
The SEGNN4SLP framework demonstrates the use of route labeling (PL) on a 1-hop subgraph that includes nodes a and b. In (a), we notice a route with a length of 2, in (b) a route with a length of 3, and in (c) a route with a length of 4. Every unique pathway is depicted using a distinct hue. The nodes are labeled and the labels are presented above the nodes. Nodes with different labels are shown in distinct colors.

Figure 4

Assign different coefficients between nodes by GAT.
Assign different coefficients between nodes by GAT.

Figure 5

Architecture of SEGNN4SLP
Architecture of SEGNN4SLP

Figure 6

Impact of different embedding size d
Impact of different embedding size d

Figure 7

Impact of different path length λ
Impact of different path length λ

j_jhp-2024-0019_utab_002

input: target edge (i,j); input graph G; node characterizes X
output: forecast score s,
1 /* extracts enclosing subgraph */ 2 G ← G
3 zu ← Structural Encoding (Gs, i, j), ∀u ∈ Gs; 4 Sstructure ← MLP (Zi ° zj)
5/* feature fusion */cu←{ pi,j,Gs },∀u∈Gs;xu∼(0)←xu(0),∀u∈Gs;x˜u←MLP(∼(0)xu),∀u∈Gs;hu(0)←x˜u,∀u∈Gs;
6 /* GNN message passing */for k=1,2…K do: for u ∈ G do: hu(k) ← Equation 4; endend
7 hG ← SortPool(hu(k)| u ∈ Gs,k = 1,…,K); 8 Ssemantic ← MLP(hG)
9 S = Sstructure + Ssemantic

j_jhp-2024-0019_utab_001

input: target nodes vi, vj; enclosing subgraph Gs
output: node embedding z
1 /*extracts the routes*/ 2 pi,j (G;i,j)
3 /* generate node structural features */ 4 cu ← {pi.j,GS}, ∀u ∈ GS;
5 Zu(0)← one −hot(min(cu,λ)),∀u∈Gs; 6 /* encode with a GCN layer and a MLP */
7 for u ∈ GS do 8 Zu(I)←AGGREGATE(Zv(0),∀u∈ N(u));
9 end for 10 Zu → MLP(zu), ∀u ∈ Gs;

Comparison of different methods in nDCG@k_

K=5 K=10 K=15 K=20 K=25
Node2vec 0.2314 0.2786 0.3278 0.3529 0.3604
GCN 0.2811 0.3378 0.3588 0.3687 0.3786
GraphSAGE 0.2823 0.3468 0.3770 0.3819 0.3793
GAT 0.2811 0.3398 0.3764 0.3987 0.3859
SEAL 0.2994 0.3410 0.3896 0.4055 0.3986
SEGNN4SLP 0.3516 0.3814 0.4156 0.4258 0.4288

Results for SEGNN4SLP, SEGNN4SLP-1, SEGNN4SLP-2_

Methods Recall nDCG
Recall@5 Recall@25 nDCG@5 nDCG@25
SEGNN4SLP-1 0.3389 0.4844 0.3486 0.3855
SEGNN4SLP-2 0.3284 0.4964 0.3357 0.3746
SEGNN4SLP 0.3598 0.5287 0.3617 0.4137

Comparison of different methods in Recall@k_

K=5 K=10 K=15 K=20 K=25
Node2vec 0.2185 0.2915 0.3473 0.3761 0.4012
GCN 0.2729 0.3461 0.3684 0.4561 0.4716
GraphSAGE 0.2816 0.3553 0.3941 0.4611 0.4933
GAT 0.2810 0.3513 0.3902 0.4687 0.4910
SEAL 0.2984 0.3588 0.4013 0.4701 0.4987
SEGNN4SLP 0.3514 0.3981 0.4586 0.4981 0.5231
Idioma:
Inglés
Calendario de la edición:
4 veces al año
Temas de la revista:
Informática, Informática, otros
RSS Feed de revista