Graph-Linguistic Fusion: Using Language Models for Wikidata Vandalism Detection

TL;DR

This paper presents a novel vandalism detection system for Wikidata that unifies structured and textual edit evaluation using a multilingual language model, improving detection coverage and simplifying maintenance.

Contribution

The paper introduces Graph2Text, a method converting all Wikidata edits into a unified textual format for vandalism detection with language models, enhancing performance and scalability.

Findings

Outperforms existing production vandalism detection system

Achieves higher coverage in detecting vandalism

Provides open-source code and dataset for research

Abstract

We introduce a next-generation vandalism detection system for Wikidata, one of the largest open-source structured knowledge bases on the Web. Wikidata is highly complex: its items incorporate an ever-expanding universe of factual triples and multilingual texts. While edits can alter both structured and textual content, our approach converts all edits into a single space using a method we call Graph2Text. This allows for evaluating all content changes for potential vandalism using a single multilingual language model. This unified approach improves coverage and simplifies maintenance. Experiments demonstrate that our solution outperforms the current production system. Additionally, we are releasing the code under an open license along with a large dataset of various human-generated knowledge alterations, enabling further research.