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

2025-05-23

AI Summary

To address vandalism detection in Wikidata—the world’s largest open-source structured knowledge base—where concurrent structural triple edits and multilingual textual edits introduce complex cross-modal threats, this paper proposes the first end-to-end cross-modal vandalism detection system. Methodologically, it introduces (1) Graph2Text, a novel graph-to-sequence transformation that uniformly encodes structural edits as natural language sequences; (2) a multilingual large language model (MLLM) backbone for joint representation learning and discriminative classification of triples and multilingual text; and (3) the first large-scale, human-annotated knowledge editing dataset with a standardized evaluation framework. Experiments demonstrate significant improvements over Wikidata’s production system across F1-score, recall, and cross-lingual robustness. All code and data are publicly released.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Mining of Visual, Multimedia & Multimodal Data
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

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.

Problem

• Detecting vandalism in Wikidata's complex structured and textual content
• Unifying content changes into a single space using Graph2Text
• Improving vandalism detection coverage and maintenance with multilingual models

Innovation

• Graph2Text converts edits into unified space
• Uses multilingual language model for vandalism detection
• Open license code and dataset released