WikiResearch

Research news on @Wikipedia @Wikidata @Wikimedia, hot off the press. Volunteer-curated by @tilmanbayer & @mad_astronaut. Read our newsletter for full coverage

[WikiResearch @WikiResearch]

In the September issue of our newsletter:
* Google's AI overviews may have reduced search engine referrals to English Wikipedia by 5% (according to a preprint that has seen substantial changes)
And other recent research publications https://t.co/6Xra3x7M9T

[WikiResearch @WikiResearch]

"WikiFlow: Predicting Wikipedia Editor Dropout at Scale Using Distributed Machine Learning on Google Cloud" https://t.co/jlp42po88z
#paywall💰
4 "editor archetypes":
Frustrated burst editors
Casual contributors ("most valuable")
Power/talk-page editors
Focused/bot-like accounts

[WikiResearch @WikiResearch]

Dataset: https://t.co/Q4wuuoLiTF ("an exploratory, fine-grained dataset designed to capture implicit, context-dependent semantic biases in French political biographies on Wikipedia")

[WikiResearch @WikiResearch]

"A Corpus for Evaluating Neutrality in Political Biographies on Wikipedia: NeutralWikiPol-Fr" https://t.co/QZ57QRGahx #paywall💰
"LLMs identify biased words effectively [but] performance drops when bias depends on the broader context"
Human experts' "low agreement" on neutrality

[WikiResearch @WikiResearch]

"Linking Speakers of the German Parliament to Wikidata: Scope and Coverage of Metadata" https://t.co/ccJhALzU9x
For Bundestag members, "Wikidata is almost complete" for basic properties like gender or DOB, but coverage of "CV" fields like "educated at" is "significantly weaker"

[WikiResearch @WikiResearch]

"Can Language Models Play the Wikipedia Game?" (get from one article to another via article links) https://t.co/Yl4SqVXhIR
"No model significantly outperformed humans, although the best models (GPT-o3 mini, GPT-4o, QwQ:32B, and DeepSeek-R1:70B) achieved comparable success."

[Alexander Doria @Dorialexander]

An new Pleias paper: in collaboration with AI Bridges and @Wikimedia Deutschland we developed a new harness for complex semantic search over the largest knowledge database, @wikidata

[Manoel @manoelribeiro]

What happens when a Wikipedia language edition starts using LLMs?

Unlike in other languages, Korean Wikipedia embraced LLMs through WikiVault: only 0.1% of editors use it weekly, but it accounts for ~30% of added bytes!

New preprint led by @song_inhwa_ (who's at @ACM_CSCW)

🧵

[WikiResearch @WikiResearch]

"WikiCoach: Scaffolding Learning for Novices in Online Apprenticeship Communities" https://t.co/ovhik7cYwE
"a Q&A–style LLM system designed to scaffold learning-through-doing in Wikipedia"

[WikiResearch @WikiResearch]

Web interface for the resulting knowledge graph: https://t.co/Wu3fna0XrL

[WikiResearch @WikiResearch]

"WiCleanData: Guaranteeing the Type Consistency of Wikidata by Taxonomy Refinement and Constraint Enforcement" https://t.co/Nmu0QZA8bI
Pipeline "cleans the taxonomy with [LLM] assistance, then simplifies type constraints by hierarchical aggregation, and finally filters facts"

[WikiResearch @WikiResearch]

"Sanctioned by Others: Remediating Knowledge Exclusion on Wikipedia Through Design" https://t.co/Krdz89Sq5H
"a more inclusive Wikipedia is one in which editors can acquire the skills to invoke, bend, and even break the platform policies that dictate encyclopedic content."

[WikiResearch @WikiResearch]

"The Evolution of Coordination in a Collective Intelligence System: 25 Years of English Wikipedia and the Emergence of Generative AI" https://t.co/MN0mFuUY0G
"participation in coordination spaces has declined relative to content production, particularly in governance areas"

[Shlomit Lir @shlomitlir]

My new research on the experiences of Jewish editors on English Wikipedia has just been published.
The study exposes the harassment Jewish contributors face when trying to keep Israel‑related topics neutral. Reforming Wikipedia matters.
https://t.co/aPduylWH4i

[WikiResearch @WikiResearch]

"Wikimedia research methodologies: a bibliometric history of peer production and AI (2005–2025)" https://t.co/53tKT7xR2v
Argues that "industry-funded research popularized the idea that Wikipedia existed as a semantic resource to be mined"

[WikiResearch @WikiResearch]

"AI Use among [Swiss] Wikipedians in Science-Related Information & Communication" https://t.co/7AyFzWaXcx
* Wikipedians much more cautious about using AI than scientists or journalists
* Only 22.6% find AI useful for their Wikipedia contributions
* Polarized on AI disclosure

[WikiResearch @WikiResearch]

"WikiStar: A System for Shedding Light on the Hidden History of Scientific Wikipedia Articles", detecting edit types such as "technical term added" using an LLM classifier https://t.co/BeFTTmW41I
(with human-annotated benchmark dataset)

[WikiResearch @WikiResearch]

"SimpleWikiSearch: A Clean Offline Wikipedia Environment for Agentic Search" https://t.co/P00z3ORvyJ
(Code: https://t.co/jmxjiLsyd7 )
#RAG

[WikiResearch @WikiResearch]

"Scalable Methods for Storing and Retrieving Wikipedia Revision Histories for Large-Scale Analysis" https://t.co/UDVnigBjLg

New "WikECD" Python library can compress them by as much as 94% https://t.co/05xxYtevR1

[WikiResearch @WikiResearch]

"Festivities at the edge of data: an interdisciplinary study of Andean festivals in wikidata" https://t.co/aWQ6Dvevzc
"Wikidata’s semantic model, classification properties, linking practices, and editorial accumulation make some cultural relations more visible than others"