Jump to content

Wikimedia CH/Grant apply/Support for open-access publication of the Code2Code Search system

Add links
From Meta, a Wikimedia project coordination wiki

Please provide the following information for your micro-grant request:

Your Username: Francesco Tosoni (SSSA)

Your Country of residence: Italy

Amount requested: 1,000 CHF

Planned activity:

To cover part of the publication costs for the research article "MediaWiki Code2Code Search: Neural Retrieval for the Semantic Discovery of Open-Source Software Entities", submitted on 5 June 2026 to the journal SoftwareX (Elsevier). The total Article Processing Charge (APC) is 1,400 EUR, of which this micro-grant would cover approximately 1,000 CHF. The article presents a fully open-source, semantic code-search tool for the Wikimedia technical infrastructure, and this grant would enable its open-access publication.

People involved (e.g. How many people can benefit from the micro grant you request):

The primary beneficiary is the Wikimedia technical community, including developers, maintainers, and contributors across more than 2,500 MediaWiki repositories. By improving code discovery and lowering technical barriers, the tool benefits the entire ecosystem of contributors who maintain software for Wikipedia, Wikidata, and other sister projects. The interface is localised into 17 languages (including 14 Indian languages) to support international contributors, with a specific focus on Indian developer communities.

Please explain what influence your activity will have on Wikimedia Projects (e.g. quantitative: number of new pictures uploaded to commons, no of new articles created / qualitative: strengthen the community of german-speaking wikipedia authors etc,):

The project strengthens the Wikimedia technical ecosystem by providing a scalable, semantic code-search tool that addresses the lexical gap hindering current development workflows. Traditional keyword-based search fails when a function's name does not literally match the query: for example, a contributor looking for least-recently-used cache eviction logic will not find a function named evictOldest, nor will a search for rate-limiting retrieve isRateLimited or pingLimiter. Code2Code Search retrieves these by computational intent rather than surface tokens, succeeding precisely where keyword tools return nothing.

Qualitatively, the search engine improves volunteer onboarding and developer efficiency, and it operationalises the UNESCO Recommendation on Open Science through citable, ISO/IEC 18670:2025-compliant SWHIDs. Quantitatively, the tool has already indexed 1.29 million structural entities (functions, types, and templates) across over 2,500 repositories, all of which are now searchable and citable for research and development. On these name-obfuscated queries (category B of the benchmark), it achieves a Precision@10 of 0.91, compared with 0.54 for a traditional keyword (BM25) baseline, and 0.87 versus 0.64 across the full 27-query benchmark.

Please provide links to your contributions after your project or activity ended: Please provide the links to the categories on Wikimedia Commons, to the articles on Wikipedia or any other contributions on Wikimedia projects:

Important: AFTER having submitted your request, please write an eMail to info@wikimedia.ch with your contact details (at least: Name, Surname, Email address)

Decision

[edit]

The grant is accepted and confirmed.--Ilario (talk) 21:12, 5 June 2026 (UTC)reply

Retrieved from "https://meta.wikimedia.org/w/index.php?title=Wikimedia_CH/Grant_apply/Support_for_open-access_publication_of_the_Code2Code_Search_system&oldid=30637860"