Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MLP-Enhanced Nonnegative Tensor RESCAL Decomposition for Dynamic Community Detection

Chaojun Li, Hao Fang

TL;DR

MLP-NTD addresses the rigidity of RESCAL based dynamic community detection by decoupling the decomposition rank from the number of communities through an MLP mapping after RESCAL. It combines a RESCAL based tensor factorization with an MLP that outputs community indicators $B_t$, trained with a reconstruction loss that couples the latent factors and with temporal smoothing, refined via modularity optimization. Experiments on Chess and Cellphone demonstrate superior modularity and more coherent community structures compared with state-of-the-art baselines. This approach enhances flexibility and robustness for detecting evolving communities in dynamic networks.

Abstract

Dynamic community detection plays a crucial role in understanding the temporal evolution of community structures in complex networks. Existing methods based on nonnegative tensor RESCAL decomposition typically require the decomposition rank to equal the number of communities, which limits model flexibility. This paper proposes an improved MLP-enhanced nonnegative tensor decomposition model (MLP-NTD) that incorporates a multilayer perceptron (MLP) after RESCAL decomposition for community mapping, thereby decoupling the decomposition rank from the number of communities. The framework optimizes model parameters through a reconstruction loss function, which preserves the ability to capture dynamic community evolution while significantly improving the accuracy and robustness of community partitioning. Experimental results on multiple real-world dynamic network datasets demonstrate that MLP-NTD outperforms state-of-the-art methods in terms of modularity, validating the effectiveness of the proposed approach.

MLP-Enhanced Nonnegative Tensor RESCAL Decomposition for Dynamic Community Detection

TL;DR

MLP-NTD addresses the rigidity of RESCAL based dynamic community detection by decoupling the decomposition rank from the number of communities through an MLP mapping after RESCAL. It combines a RESCAL based tensor factorization with an MLP that outputs community indicators , trained with a reconstruction loss that couples the latent factors and with temporal smoothing, refined via modularity optimization. Experiments on Chess and Cellphone demonstrate superior modularity and more coherent community structures compared with state-of-the-art baselines. This approach enhances flexibility and robustness for detecting evolving communities in dynamic networks.

Abstract

Dynamic community detection plays a crucial role in understanding the temporal evolution of community structures in complex networks. Existing methods based on nonnegative tensor RESCAL decomposition typically require the decomposition rank to equal the number of communities, which limits model flexibility. This paper proposes an improved MLP-enhanced nonnegative tensor decomposition model (MLP-NTD) that incorporates a multilayer perceptron (MLP) after RESCAL decomposition for community mapping, thereby decoupling the decomposition rank from the number of communities. The framework optimizes model parameters through a reconstruction loss function, which preserves the ability to capture dynamic community evolution while significantly improving the accuracy and robustness of community partitioning. Experimental results on multiple real-world dynamic network datasets demonstrate that MLP-NTD outperforms state-of-the-art methods in terms of modularity, validating the effectiveness of the proposed approach.
Paper Structure (25 sections, 7 equations, 4 figures, 1 table, 1 algorithm)

This paper contains 25 sections, 7 equations, 4 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Notations
  4. Problem Description
  5. The MLP-NTD Model
  6. Model Architecture
  7. Nonnegative Tensor RESCAL Decomposition
  8. MLP Community Mapping
  9. Modularity Maximization Refinement
  10. Loss Function
  11. Optimization Algorithm
  12. Algorithm Analysis
  13. Computational Complexity
  14. Memory Complexity
  15. Experiments
  16. ...and 10 more sections

Figures (4)

  • Figure 1: Framework of the MLP-NTD model
  • Figure 2: Modularity comparison across time slices: (a) Dataset 1 (Chess), (b) Dataset 2 (Cellphone). MLP-NTD consistently achieves higher modularity values compared to baseline methods.
  • Figure 3: Average Modularity Comparison across Different Dataset.
  • Figure 4: t-SNE visualization of community structure at time slice 4. MLP-NTD exhibits clearer community separation and more compact clustering compared to the original MNTD.