Ensemble graph-transfer learning under Gaussian graphical models on heterogeneous network-linked data

This repository is the official implementation of "Ensemble graph-transfer learning under Gaussian graphical models on heterogeneous network-linked data".

File Structure:

• data_preprocessing.R
  Processes academic publication data through three steps and builds final author-term matrices and coauthor network matrices for term graph construction.

• method_functions.R
  Defines core functions for:

  • Data generation;
  • Precision matrix estimation using GTrans-NCMA.

• run_example.R
  Demonstrates complete workflow:

  1. Data generation;
  2. Precision matrix estimation with GTrans-NCMA;
  3. Performance evaluation using:
     • KL divergence;
     • Frobenius norm error.

Requirements

First of all, make sure you have installed the R language environment (it is recommended to use R version 4.1 or higher).

In R, run the following command to install the required packages:

install.packages(c("glasso", "Matrix", "igraph"))

For data preprocessing, we suggest downloading the MADStat dataset and the file 'DataForGNC-Plot-Combined.Rda' contains the top 300 terms selected by tf-idf scores, as described in the work by Li et al. (2020) on high-dimensional Gaussian graphical models for network-linked data. The MADStat dataset is publicly available at https://github.com/ZhengTracyKe/MADStat. The file 'DataForGNC-Plot-Combined.Rda' is publicly available in the GNC repository at https://github.com/tianxili/GNC/blob/master/GNC-lasso.R.

Training

To train the model in the paper, run the entire run_example.R script.

Evaluation

To evaluate my model, run:

KL_divergence <- -log(det(Omega_hat)) + sum(diag(Omega_hat %*% Sigma_true)) - (-log(det(Omega_true)) + p)
Frob_norm_error <- sum((Omega_hat - Omega_true)^2) / p  

Results

The final datasets obtained after preprocessing by running data_preprocessing.R include:

target journals	source journals
1090 authors and 300 terms	8616 authors and 300 terms