Extending TWIG: Zero-Shot Predictive Hyperparameter Selection for KGEs based on Graph Structure

TL;DR

This paper extends TWIG to a multi-KG setting to enable zero-shot predictive hyperparameter selection for KGEs based on graph structure. By modeling how KG structure and KGE components interact, TWIG can accurately predict ComplEx performance across unseen hyperparameters and entirely unseen graphs. Finetuning TWIG on small amounts of data further enhances zero-shot and few-shot performance, suggesting a path toward pre-hoc hyperparameter optimization without exhaustive searches. The work demonstrates domain-agnostic structural cues as a driver of KGE learnability and highlights future work to generalize across more models and larger graphs. The approach has practical implications for efficient KG-based learning in diverse domains.

Abstract

Knowledge Graphs (KGs) have seen increasing use across various domains -- from biomedicine and linguistics to general knowledge modelling. In order to facilitate the analysis of knowledge graphs, Knowledge Graph Embeddings (KGEs) have been developed to automatically analyse KGs and predict new facts based on the information in a KG, a task called "link prediction". Many existing studies have documented that the structure of a KG, KGE model components, and KGE hyperparameters can significantly change how well KGEs perform and what relationships they are able to learn. Recently, the Topologically-Weighted Intelligence Generation (TWIG) model has been proposed as a solution to modelling how each of these elements relate. In this work, we extend the previous research on TWIG and evaluate its ability to simulate the output of the KGE model ComplEx in the cross-KG setting. Our results are twofold. First, TWIG is able to summarise KGE performance on a wide range of hyperparameter settings and KGs being learned, suggesting that it represents a general knowledge of how to predict KGE performance from KG structure. Second, we show that TWIG can successfully predict hyperparameter performance on unseen KGs in the zero-shot setting. This second observation leads us to propose that, with additional research, optimal hyperparameter selection for KGE models could be determined in a pre-hoc manner using TWIG-like methods, rather than by using a full hyperparameter search.

Figures (5)

• Figure 1: A sample Knowledge graph (left) and an example of link prediction (right).
• Figure 2: An schematic overview of how rank-based evaluation of KGE models is performed.
• Figure 3: An overview of link prediction (using KGEs) and of KGE simulation (using TWIG).
• Figure 4: A visualisation of the TWIG neural network architecture. KG structure (in magenta) and hyperparameter influence (in yellow) are first learned in separate blocks, then combined in an integration component (in red) before predicted ranks are output.
• Figure 5: A visualisation of the region around a triple that TWIG uses to calculate structural features. Details statistics are gathered for the main triple (shown in red), and aggregate statistics are gathered for all triples connecting to that triple (shown in blue). Other triples (shown in black) are ignored.