The Wikidata Quality Toolkit (WQT, https://king-s-knowledge-graph-lab.github.io/WikidataQualityToolkit/) is a set of tools that various researchers, developers, and open knowledge activists and enthusiasts are currently working on to improve the quality of content in Wikidata and improve the workflows of editors in their everyday tasks. These tools are built on top of successful, recently published research about Wikidata (including a paper that received the Wikimedia Foundation Research of the Year Award in 2022). Our aim is to help transition these tools from the lab into the community and the world.

The WQT contains three tools covering the spectrum of Wikidata content quality and editor workflow improvement, and solve three tasks: verifying the quality of references in supporting claims, recommending items based on expertise automatically, and generating schemas for item completion.

• Reference Quality Verification (RQV). RQV provides an automated pipeline that verifies whether Knowledge Graph triples are supported by their documented sources. It involves text extraction, triple verbalization, sentence selection, and claim verification using rule-based methods and machine learning models. The users can verify the reference quality of specific document or wikidata item by using this tool. Futhermore, this tool supports to verify a batch of documents and wikidata tiems automatically.

• Wikidata Game+. Wikidata Game+ builds upon the Wikidata Game by incorporating a novel recommendation system that provides personalised recommendation items for the editors, relying on both item features and item-editor previous interactions. It utilises users', items' content, and items' relations representations using matrix factorization, ELMo, and TransR embedding techniques.

• EntitySchema Generator. There are numerous issues with Wikidata modeling and data quality, with inconsistent modeling of EntitySchemas being one of the most significant challenges nowadays. The EntitySchema generator addresses this by generating reference patterns of entity schemas for specific topics of entities based on Large Language Models (LLMs). By training on both good and bad examples, it can generate reference patterns and evaluate the quality of entity schemas. Additionally, it can modify inconsistent entity schemas based on the generated best patterns, and provide explanations and additional comments leveraging the capabilities of LLMs.

Session recording: https://youtu.be/BbGrkYK8FEk?list=PLhV3K_DS5YfJ1xyY0LNDNX3RKyRQEXOdB&t=22855