Markov Process-Based Graph Convolutional Networks for Entity Classification in Knowledge Graphs

TL;DR

The paper addresses incomplete entity-type annotations in Knowledge Graphs by introducing MPERL, a Markov process–augmented GCN with an evidential loss. The model learns adaptive computation steps via a geometric halting distribution $p_n$ and produces calibrated predictions from a Dirichlet posterior with parameters $\alpha^{(i)}=\phi(h^{(i)})+1$, enabling uncertainty quantification. Empirically, MPERL+R-GCN improves over strong baselines on small KG benchmarks and remains competitive on larger graphs, with ablations showing the complementary benefits of the Markov process and evidential loss. The approach offers uncertainty-aware predictions while allowing cost-efficient training through tunable Markov steps, making it effective for KG completion tasks.

Abstract

Despite the vast amount of information encoded in Knowledge Graphs (KGs), information about the class affiliation of entities remains often incomplete. Graph Convolutional Networks (GCNs) have been shown to be effective predictors of complete information about the class affiliation of entities in KGs. However, these models do not learn the class affiliation of entities in KGs incorporating the complexity of the task, which negatively affects the models prediction capabilities. To address this problem, we introduce a Markov process-based architecture into well-known GCN architectures. This end-to-end network learns the prediction of class affiliation of entities in KGs within a Markov process. The number of computational steps is learned during training using a geometric distribution. At the same time, the loss function combines insights from the field of evidential learning. The experiments show a performance improvement over existing models in several studied architectures and datasets. Based on the chosen hyperparameters for the geometric distribution, the expected number of computation steps can be adjusted to improve efficiency and accuracy during training.

Figures (4)

• Figure 1: Approach for entity classification using MPERL. MPERL gets as input a one-hot encoding of the entity ID (denoted $x^{(i)}$) and the learned hidden features from the previous Markov step. The GCN-based model uses the structure of the KG to compute the hidden features $h^{(i)}_{n}$ and the halting probability $\lambda_{n}^{(i)}$. The prediction $\hat{y}^{(i)}$ is based on the Dirichlet parameters $\alpha^{(i)}$.
• Figure 2: Degree distribution of entities in datasets
• Figure 3: Learning curves for the AIFB dataset for different $\lambda_p$ values
• Figure 4: Learning curves for the MUTAG dataset for different $\lambda_p$ values

Theorems & Definitions (1)

• Definition 1