@WikiResearch

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

 San Francisco
meta.wikimedia.org/wiki/Research:Newsletter
Joined December 2011
  • Tweets8K
  • Following1.4K
  • Followers17.4K
2.2KMedia
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 meta.wikimedia.org/wiki/Rese…
"Google AI Overview results.png" (via Wikimedia Commons, uploader: Belbury, PD)

ALT "Google AI Overview results.png" (via Wikimedia Commons, uploader: Belbury, PD)

2
3
660
"WikiFlow: Predicting Wikipedia Editor Dropout at Scale Using Distributed Machine Learning on Google Cloud" doi.org/10.1007/978-3-032-42… #paywall💰 4 "editor archetypes": Frustrated burst editors Casual contributors ("most valuable") Power/talk-page editors Focused/bot-like accounts
7.3 K-Means Editor Archetypes (from Meena et al. 2026, WikiFlow: Predicting Wikipedia Editor Dropout at Scale... )

ALT 7.3 K-Means Editor Archetypes (from Meena et al. 2026, WikiFlow: Predicting Wikipedia Editor Dropout at Scale... )

Fig. 1. WikiFlow system architecture. (from Meena et al. 2026, WikiFlow: Predicting Wikipedia Editor Dropout at Scale... )

ALT Fig. 1. WikiFlow system architecture. (from Meena et al. 2026, WikiFlow: Predicting Wikipedia Editor Dropout at Scale... )

Table 3. WikiFlow feature set: 13 behavioral features derived from each editor’s first 10 edits (from Meena et al. 2026, WikiFlow: Predicting Wikipedia Editor Dropout at Scale... )

ALT Table 3. WikiFlow feature set: 13 behavioral features derived from each editor’s first 10 edits (from Meena et al. 2026, WikiFlow: Predicting Wikipedia Editor Dropout at Scale... )

4
121
"A Corpus for Evaluating Neutrality in Political Biographies on Wikipedia: NeutralWikiPol-Fr" dx.doi.org/10.1007/978-3-032… #paywall💰 "LLMs identify biased words effectively [but] performance drops when bias depends on the broader context" Human experts' "low agreement" on neutrality
"Table 1. Comparison of human and LLM alignment rates with Wikipedia editors and the impact of context-dependency on bias detection" (from Plance et al. 2026, A Corpus for Evaluating Neutrality in Political Biographies on Wikipedia: NeutralWikiPol-Fr)

ALT "Table 1. Comparison of human and LLM alignment rates with Wikipedia editors and the impact of context-dependency on bias detection" (from Plance et al. 2026, A Corpus for Evaluating Neutrality in Political Biographies on Wikipedia: NeutralWikiPol-Fr)

1
3
283
Dataset: github.com/Vic-SP/NWP-fr ("an exploratory, fine-grained dataset designed to capture implicit, context-dependent semantic biases in French political biographies on Wikipedia")

GitHub - Vic-SP/NWP-fr: NeutralWikiPol-fr is an exploratory, fine-grained dataset designed to...

NeutralWikiPol-fr is an exploratory, fine-grained dataset designed to capture implicit, context-dependent semantic biases in French political biographies on Wikipedia. - Vic-SP/NWP-fr

github.com
71
"Linking Speakers of the German Parliament to Wikidata: Scope and Coverage of Metadata" arxiv.org/pdf/2609.18289 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"
Excerpt from 3 Results - 3.1 Properties (from Haider et al. 2026, Linking Speakers of the German Parliament to Wikidata..., CC BY-SA 4.0).png

ALT Excerpt from 3 Results - 3.1 Properties (from Haider et al. 2026, Linking Speakers of the German Parliament to Wikidata..., CC BY-SA 4.0).png

Figure 1: Proportion of speeches given by women across legislative periods (LP 1–19). (from Haider et al. 2026, Linking Speakers of the German Parliament to Wikidata..., CC BY-SA 4.0)

ALT Figure 1: Proportion of speeches given by women across legislative periods (LP 1–19). (from Haider et al. 2026, Linking Speakers of the German Parliament to Wikidata..., CC BY-SA 4.0)

1
2
398
"Can Language Models Play the Wikipedia Game?" (get from one article to another via article links) arxiv.org/html/2610.13948v1 "No model significantly outperformed humans, although the best models (GPT-o3 mini, GPT-4o, QwQ:32B, and DeepSeek-R1:70B) achieved comparable success."
1
1
8
593
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
11
21
154
5,130
WikiResearch retweeted
Manoel@manoelribeiroOct 12
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) 🧵
1
10
1
21
1,219
"WikiCoach: Scaffolding Learning for Novices in Online Apprenticeship Communities" doi.org/10.1145/3774935.3806… "a Q&A–style LLM system designed to scaffold learning-through-doing in Wikipedia"
Figure 1: WikiCoach interaction pipeline and interface. (from Zhou et al 2026, CC BY 4.0)

ALT Figure 1: WikiCoach interaction pipeline and interface. (from Zhou et al 2026, CC BY 4.0)

4
9
970
"WiCleanData: Guaranteeing the Type Consistency of Wikidata by Taxonomy Refinement and Constraint Enforcement" arxiv.org/abs/2609.20057 Pipeline "cleans the taxonomy with [LLM] assistance, then simplifies type constraints by hierarchical aggregation, and finally filters facts"
"Figure 1: Repairing type constraint violations of Wikidata by taxonomy cleaning" (from Peng et al. 2026)

ALT "Figure 1: Repairing type constraint violations of Wikidata by taxonomy cleaning" (from Peng et al. 2026)

1
4
6
589
Web interface for the resulting knowledge graph: wicleandata.r2.enst.fr/
1
110
"Sanctioned by Others: Remediating Knowledge Exclusion on Wikipedia Through Design" doi.org/10.1145/3816994 "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."
Four design personas (purple) mapped onto the engagement-adherence matrix (Figure 4 from AjmaniLeah et al. 2026, Sanctioned by Others-Remediating Knowledge Exclusion on Wikipedia Through Design, CC BY 4.0)

ALT Four design personas (purple) mapped onto the engagement-adherence matrix (Figure 4 from AjmaniLeah et al. 2026, Sanctioned by Others-Remediating Knowledge Exclusion on Wikipedia Through Design, CC BY 4.0)

Low fidelity prototype of Wikipedia forking and merging (Figure 16 from AjmaniLeah et al. 2026, Sanctioned by Others-Remediating Knowledge Exclusion on Wikipedia Through Design, CC BY 4.0)

ALT Low fidelity prototype of Wikipedia forking and merging (Figure 16 from AjmaniLeah et al. 2026, Sanctioned by Others-Remediating Knowledge Exclusion on Wikipedia Through Design, CC BY 4.0)

1
5
573
"The Evolution of Coordination in a Collective Intelligence System: 25 Years of English Wikipedia and the Emergence of Generative AI" arxiv.org/html/2609.16856v1 "participation in coordination spaces has declined relative to content production, particularly in governance areas"
Longitudinal Analysis of Edits (Left) and Users (Right) per NamespaceLongitudinal analysis of Edits and Users per Namespace.Lo… (Figure 1b from Reeves et al. 2026, The Evolution of Coordination…, arXiv.org perpetual non-exclusive license)

ALT Longitudinal Analysis of Edits (Left) and Users (Right) per NamespaceLongitudinal analysis of Edits and Users per Namespace.Lo… (Figure 1b from Reeves et al. 2026, The Evolution of Coordination…, arXiv.org perpetual non-exclusive license)

1
5
524
WikiResearch retweeted
Shlomit Lir
@shlomitlirJun 25
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. journals.sagepub.com/doi/10.…

Between conflict and contribution: Jewish editors and the politics of knowledge on Wikipedia -...

Wikipedia is one of the most influential platforms for digital knowledge production, organized around community-based consensus and open participation. These id...

journals.sagepub.com
2
20
1
39
1,193
"Wikimedia research methodologies: a bibliometric history of peer production and AI (2005–2025)" doi.org/10.1007/s00146-026-0… Argues that "industry-funded research popularized the idea that Wikipedia existed as a semantic resource to be mined"
The four waves of Wikimedia research and their concerns (2005–2025) [from Jankowski 2026]

ALT The four waves of Wikimedia research and their concerns (2005–2025) [from Jankowski 2026]

4
7
1,000
"AI Use among [Swiss] Wikipedians in Science-Related Information & Communication" upload.wikimedia.org/wikiped… * 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
Compared with other groups, Wikipedians are more cautious about future AI use (from Schäfer et al., "AI Use among Wikipedians in Science-Related Information & Communication", CC BY-SA 4.0)

ALT Compared with other groups, Wikipedians are more cautious about future AI use (from Schäfer et al., "AI Use among Wikipedians in Science-Related Information & Communication", CC BY-SA 4.0)

AI is mainly used as a language and support tool (from Schäfer et al., "AI Use among Wikipedians in Science-Related Information & Communication", CC BY-SA 4.0)

ALT AI is mainly used as a language and support tool (from Schäfer et al., "AI Use among Wikipedians in Science-Related Information & Communication", CC BY-SA 4.0)

Disclosure of AI use is polarized (from Schäfer et al., "AI Use among Wikipedians in Science-Related Information & Communication", CC BY-SA 4.0)

ALT Disclosure of AI use is polarized (from Schäfer et al., "AI Use among Wikipedians in Science-Related Information & Communication", CC BY-SA 4.0)

2
623
"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 arxiv.org/html/2607.12441v3 (with human-annotated benchmark dataset)
Overview of the WikiStar system (Figure 2 from Ehrlich et al. 2026, WikiStar-A System for Shedding Light on the Hidden History of Scientific Wikipedia Articles, CC BY 4.0)

ALT Overview of the WikiStar system (Figure 2 from Ehrlich et al. 2026, WikiStar-A System for Shedding Light on the Hidden History of Scientific Wikipedia Articles, CC BY 4.0)

2
11
814
"SimpleWikiSearch: A Clean Offline Wikipedia Environment for Agentic Search" arxiv.org/pdf/2607.26070 (Code: github.com/JimXiongGM/simple… ) #RAG
1
1
1
4
583
"Scalable Methods for Storing and Retrieving Wikipedia Revision Histories for Large-Scale Analysis" dl.acm.org/doi/10.1145/37874… New "WikECD" Python library can compress them by as much as 94% github.com/descentis/WikECD
Editors distribution by quality class (from Scalable Methods for Storing and Retrieving Wikipedia Revision Histories for Large-Scale Analysis, CC BY 4.0)

ALT Editors distribution by quality class (from Scalable Methods for Storing and Retrieving Wikipedia Revision Histories for Large-Scale Analysis, CC BY 4.0)

3
598
"Festivities at the edge of data: an interdisciplinary study of Andean festivals in wikidata" nature.com/articles/s41599-0… "Wikidata’s semantic model, classification properties, linking practices, and editorial accumulation make some cultural relations more visible than others"
Table 1 Overview of entity types and their corresponding Wikidata relations used to capture ritual, geographic, and heritage dimensions (from Zúñiga and Meza 2026, CC BY-NC-ND 4.0)

ALT Table 1 Overview of entity types and their corresponding Wikidata relations used to capture ritual, geographic, and heritage dimensions (from Zúñiga and Meza 2026, CC BY-NC-ND 4.0)

Community detection (excerpt from Zúñiga and Meza 2026, CC BY-NC-ND 4.0)

ALT Community detection (excerpt from Zúñiga and Meza 2026, CC BY-NC-ND 4.0)

2
588
Load more