About

Wikipedia does not have to be truthful, but it is important that the information must be confirmed by reliable sources. Additionally, one of the most important factors affecting the quality of Wikipedia articles is the availability of reliable sources. By following the links in the references (footnotes), readers can check facts or find more information on a topic described.

The BestRef shows importance of information sources in references of Wikipedia Articles in different languages. Data extraction based on complex method using Wikimedia dumps in July 2026. To find the most important sources we used information about more than 400 million references of Wikipedia articles.

More details you can find here and in the scientific publications:

Watch the video featuring the top websites cited in Wikipedia references:

Models for assessing website importance

Importance of each of websites was assessed in the BestRef using three models (based on the research published in 2020):

F-model: based on frequency (F) of source usage.
PR-model: based on popularity (P) of the article in which source appears divided by number of the references (R) in this article.
AR-model: based on number of authors (A) of the article in which source appears divided by number of the references (R) in this article.

So, each model approaches website importance from a different perspective: frequency of occurrence in references, readership of the articles in which the references appear, and the number of editors involved in developing those articles.

The following notation is used consistently in all three models:

w – the website being analyzed;
a – a Wikipedia article;
R_a – the total number of references in article a;
R_a,w – the number of references in article a that link to website w;
P_a – the corrected cumulative number of pageviews received by article a during the selected 12-month period;
E_a – the number of unique registered non-bot editors who edited article a.

The set 𝒜_ref contains only articles with at least one reference:

𝒜_{ref} = {a \in 𝒜 : R_{a} > 0}

This restriction is particularly important in the PR and AR models because it prevents division by zero.

The examples for all three models use the same hypothetical website w, which appears in the references of two articles:

Article	R_a	R_a,w	P_a	E_a
a₁	10	2	10,000	25
a₂	5	1	400	4

F model: frequency of occurrence

The F model is the simplest method for assessing website importance. It counts the total number of references linking to a given website across all analyzed articles. If the same website appears in several references within a single article, each occurrence is counted separately.

F (w) = \sum_{a \in 𝒜_{ref}} R_{a, w}

In this example, website w appears in two references in the first article and one reference in the second article:

F (w) = 2 + 1 = 3

The resulting F value is therefore 3. This model does not account for article popularity, the number of editors, or the proportion of an article’s references that link to the website.

PR model: reader popularity

The PR model accounts for the popularity of articles among readers. Within each article, the website’s contribution is determined by the proportion of references linking to that website and is then weighted by the square root of the article’s corrected pageview count.

PR (w) = \sum_{a \in 𝒜_{ref}} (\frac{\sqrt{P_{a}}}{R_{a}} \times R_{a, w})

Taking pageviews into account gives greater weight to references appearing in more popular articles. At the same time, using the square root of the pageview count reduces the disproportionate influence of articles with exceptionally high traffic.

For the example data, the PR value is calculated as follows:

PR (w) = \frac{\sqrt{10000}}{10} \times 2 + \frac{\sqrt{400}}{5} \times 1 = 20 + 4 = 24

The first article contributes 20 points to the result, while the second contributes 4 points. The difference is primarily due to the greater popularity of the first article.

AR model: editor participation

The AR model accounts for the number of people involved in developing an article. The proportion of references linking to the analyzed website is weighted by the number of unique registered non-bot editors who edited the article.

AR (w) = \sum_{a \in 𝒜_{ref}} (\frac{E_{a}}{R_{a}} \times R_{a, w})

This model assigns greater weight to websites appearing in articles created and developed by larger numbers of editors. However, this does not automatically indicate higher source quality, because the number of editors measures community participation rather than the reliability of the source itself.

For the example data, the AR value is calculated as follows:

AR (w) = \frac{25}{10} \times 2 + \frac{4}{5} \times 1 = 5 + 0.8 = 5.8

The first article contributes 5 points to the result, while the second contributes 0.8 points. The larger contribution of the first article results from the greater number of editors involved in its development.

Interpretation of the models

The F model indicates how frequently a website appears in Wikipedia references. The PR model highlights websites that appear primarily in articles popular among readers, whereas the AR model gives greater prominence to websites used in articles developed by larger numbers of editors.

Values produced by different models should not be compared directly because they are calculated on different scales. Each model produces a separate ranking, and comparing their results provides a more comprehensive view of the role played by individual websites within Wikipedia’s referencing system.

Please note that none of these models directly measures the quality or reliability of a website. Their results describe its importance within the analyzed collection of references according to the criteria applied by each model.

Chrome extension

There is also BestRef extension for browser on Chrome Web Store. See short video on how it works:

Why reliability is important?

Information about the reliability of sources can help to improve models for quality assessment of Wikipedia articles. This can be especially useful when comparing inconsistent facts between language versions of Wikipedia articles. In addition, one of the promising areas of upcoming research is the creation of publicly available tools that would make it possible to recommend the best sources for individual statements and on selected topics in different language versions of Wikipedia.

More information on research in this field can be found on the WikiQ project.