Wikipedia is a free, collaboratively edited online encyclopaedia available in dozens of language editions. Wikipedia page-view data provide a measure of how frequently users access Wikipedia content and can therefore be used as an indicator of interest in particular topics over time.
The Wikimedia Foundation publishes page-view data for Wikimedia projects, including Wikipedia, through downloadable files containing aggregated information on page requests. The visualisations on OECD.AI use these data to identify Wikipedia pages related to artificial intelligence (AI) and measure their page views over time.
AI-related Wikipedia pages are identified using information from Wikidata. The methodology combines text-based identification of pages referring to artificial intelligence with the semantic relationships contained in Wikidata’s knowledge graph. This approach aims to capture both pages explicitly referring to AI and pages covering concepts that are classified as subclasses of artificial intelligence.
Coverage
The dataset covers Wikipedia page views for AI-related Wikipedia pages. The analysis is based on publicly available Wikimedia page-view data and therefore measures access to Wikipedia content. The data subsequently inherit biases stemming from cross-country differences in Wikipedia usage.
The page-view data cover all Wikipedia language editions and are aggregated by date and country. The resulting indicators capture the number of page views associated with pages identified as AI-related according to the methodology described below.
The dataset does not measure unique users. Multiple views of the same page by the same user may therefore contribute multiple page views. Similarly, page views should not be interpreted as a direct measure of the number of people interested in or knowledgeable about AI.
The identification of AI-related pages is based on Wikipedia and Wikidata classifications and is therefore dependent on the coverage and structure of these sources. Pages may be added to or removed from relevant classifications over time, and the underlying Wikidata structures are continuously edited by their contributors.
Update frequency
Wikipedia data on OECD.AI will be updated quarterly.
Dimensions
The dataset contains several dimensions used to disaggregate the indicators.
- Country/territory: The country or territory associated with the page-view request, based on the geographic information provided in the Wikimedia page-view data. This dimension represents the location from which the request originated and should not be interpreted as the nationality or residence of the user.
- Date: The date on which the Wikipedia page view occurred.
- Page language: Wikipedia is organised into separate language editions, each represented by a Wikimedia project. A page may therefore exist in multiple language editions, with each edition having its own page title and revision history.
- Page: The individual Wikipedia article receiving the page view. Pages are identified using the English page title.
Page views
Page views are measured using the number of requests to Wikipedia pages recorded in the Wikimedia page-view data.
The source files are processed to obtain the total number of Wikipedia page views for each combination of country and date. The data are then filtered to retain page views associated with pages identified as AI-related.
The resulting indicator is calculated as the sum of page views for AI-related pages at the relevant level of aggregation. The total views across all Wikipedia pages for each country and month is recorded, to normalise the data as a share of total visits. This indicator allows for a better comparison of AI-related page views across countries by controlling for the popularity of Wikipedia in each country.
Page views represent requests to Wikipedia pages and are not equivalent to unique visitors. They may therefore reflect repeated access to the same page by the same or different users.
Identification of AI-related pages
AI-related pages are identified using two complementary approaches: title-based identification and semantic classification using Wikidata.
The purpose of combining these approaches is to identify both pages whose titles explicitly refer to artificial intelligence and pages covering concepts that are semantically related to artificial intelligence through Wikipedia and Wikidata’s classification structure.
Title-based identification
The first approach identifies pages whose titles contain a reference to artificial intelligence.
Because Wikipedia operates across multiple language editions, the expression for artificial intelligence varies by language. The methodology therefore uses the relevant translation of “Artificial intelligence” for the language of each Wikipedia project rather than searching for the English expression alone. Differences in terminology between languages may result in some relevant pages being identified or excluded.
Title-based identification is complemented by the semantic approach described below because many AI-related pages do not contain the words “artificial intelligence” in their title. For example, pages covering specific AI techniques or fields may have titles such as machine learning, AI agent or natural language processing without explicitly referring to artificial intelligence.
Wikidata-based identification
Wikidata contains a collaboratively maintained knowledge graph that provides structured information about entities and concepts. Concepts in Wikidata can be connected through various properties. For AI pages in this data collection, two base concept pages were referenced (“Artificial Intelligence” and “Large language model”), and pages with the following relations were classified as AI-related:
- Instance of (P31)
- Subclass of (P279)
- Part of (P361)
- Facet of (P1296)
When there are duplicates in the relationships, for example, when a page is a “instance of” one base concept but a “part of” the other, the priority is given to the relation higher in the list above. Two other relations, “base concept” and “keyword search” are also included and are at the top and bottom of the relationship list, respectively.
Subclasses of the two base concepts at all levels of the hierarchy are included. For example, the conceptual structure can include relationships of the following form:
→ Artificial intelligence
→ Machine learning
→ Ensemble learning
A page associated with a Wikidata item further down this hierarchy can therefore be identified as AI-related even when the Wikipedia page title does not contain the expression “artificial intelligence”. This approach provided 792 unique pages, 304 of which have regular visits included in the final data.
The Wikidata identifier (QID) is used as the stable identifier connecting the conceptual classification to Wikipedia pages. Where a Wikidata item has Wikipedia sitelinks, these links identify the corresponding Wikipedia page in each language edition. The English page title is used for the OECD.AI visualisations, and if no English version exists for a QID in the Wikidata, the title is translated via the Google Translate API.