Multilingual Ranking of Wikipedia Articles with Quality and Popularity Assessment in Different Topics

Multilingual Ranking of Wikipedia Articles with Quality and Popularity Assessment in Different Topics Włodzimierz Lewoniewski1,†,‡ *, Krzysztof Węcel1,‡ and Witold Abramowicz1,‡ 1 Poznań University of Economics and Business; {wlodzimierz.lewoniewski,krzysztof.wecel,witold.abramowicz}@ue.poznan.pl * Correspondence: wlodzimierz.lewoniewski@ue.poznan.pl; Tel.: +48 (61) 639-27-93 † Current address: al. Niepodległości 10, 61-875 Poznań, Poland ‡ These authors contributed equally to this work.


Introduction
Nowadays, in order to make the right economic decisions, one needs to analyze and interpret vast amount of information.The quantity and quality of information to a large extent determine the quality of decisions in various branches of the economy.On the one hand, one must take care of access to proper sources of information.On the other hand, the quality of information determined by various characteristics is also important.High-quality information is essential for effective operation and decision-making in organizations [1].Inaccurate and incomplete information may have a negative impact on a company's competitive edge [2].
The Internet enables cooperation and exchange of information on a global scale.Useful information can be found both in specialized sources as well as in general online resources.Nowadays, everyone can also contribute to the development of common human knowledge on the Internet.One of the best examples of such online repositories is Wikipedia, in which content can be created from the level of a web browser.This online encyclopedia has been available for approximately 20 years as a freely available resource, and anyone willing can co-create content.Wikipedia relatively quickly became an important source of information around the world.It contains over 50 million articles in over 300 different languages [3].The English language version is the largest and contains over 5.8 million articles.Currently, Wikipedia is placed on the fifth place in the ranking of the most visited websites on the Internet [4], giving way only to Google, YouTube, Facebook, and Baidu.
The popularity of Wikipedia is even reflected language that scientists use in their works [5].
Despite its popularity, Wikipedia is often criticized for the low quality of content [6].Articles on a specific subject (a thing, a human, an event etc.) can be created and edited independently in each language version.Therefore, quality of information about the same subject often varies depending on the language [7][8][9][10].It should also be noted that the topic described in one language version can be translated into other languages.However, a relatively small number of users with knowledge of two or more languages take up such an initiative by transferring content between different language versions [11].
Even the largest English Wikipedia does not contain information about all subjects.As we can see in Figure 1, there are over 15 million unique subjects described in at least one of 55 considered language versions.This can be explained by the fact that some issues may be more common in smaller geographical areas, hence the probability of finding more information on a given topic in the relevant language versions (other than English).Overall, we can find almost 10 million subjects that are not covered in English and appear in less-developed versions of Wikipedia [7,12].When a subject is not described in the analyzed language version or information about the subject is of low quality, we can try to find information about it in other Wikipedia languages.However, identifying a language version best describing the subject may require significant effort from userpopular subjects are available in several dozen language versions.Automatic quality assessment of Wikipedia articles is a known challenge in the scientific community.Existing works have some limitations, e.g. they focus mostly on the biggest edition (English) or other popular language versions of Wikipedia.Usually the measurement of quality is reduced to analysis of volume of content -number of important elements that the article must contain (such as references, images, sections).However, for quality assessment content must be checked by other users in terms of the neutral point of view, timeliness, quality of sources and other important elements that can be challenging even with current approaches.Therefore, the popularity of the article may be another factor to be considered for quality assessment -the more users read the content, the greater probability of introducing amendments to the article, especially when incorrect or outdated information is detected.
In this paper, we present the assessment of quality and popularity of Wikipedia articles in different languages related to selected topics.This assessment was performed for articles on two levels: within each considered language version (local) and for all languages combined (global).
For the purpose of this study we selected 55 language versions of Wikipedia that in 2018 and 2019 had at least 100 thousand articles and the depth indicator was at least 5.The depth (or editing depth) shows how frequently articles are updated in a specific language version [13].Table 1 presents basic statistics about 55 language versions of Wikipedia that were considered in the study.

Category Classification
Wikipedia has extensive category network and each article can be annotated with multiple categories, organized into an "ontology of topics" [14].Each language version can define own structure and hierarchy of categories.Moreover, in some language versions that structure is often too fine-grained to be directly analyzed [15].All this may make it difficult to determine the number of possible topics to deal with.
Category structure and alignment of articles to each category can be analyzed based on files from Wikipedia dumps.There are three files that has to be used (example for English Wikipedia): • enwiki-latest-category.sql.gz-category information; here we use category identifiers and their names; • en-latest-categorylinks.sql.gz-wiki category membership link records; here we use information about source page ID and destination category name; • en-latest-page.sql.gz-base per-page data; here we use pages ID, title and information about namespaces to identify articles (ns 0) and category (ns 14) pages.
For further research we extracted information about over 10 million articles in 55 language versions and analyzed about 400 million links from articles to categories and over 26 million links between categories.General statistics about categories are presented in  We can also notice that in some language versions of Wikipedia there is a large number of categories that do not have own page that describes these categories and point to the parent category.
The highest values has Vietnamese, Chinese and Indonesian Wikipedia -about 100 thousand categories without pages.For first two languages with about 1 million articles this is one fourth and one third of all categories respectively.In Indonesian with about 460 thousand articles it is about half of all categories.
For comparison, the largest English version with over 5 million articles has only 97 categories without a page.
The so called main categories are present in majority of considered languages.This applies mainly to those categories that are at highest levels in the polyhierarchy.One of the main categories are presented at special page "Category:Main topic classifications" [16].Based on this page, we can identify 38 categories on specific topics in the English Wikipedia.As mentioned before, the category structure is a complex and ever-changing, as it can be edited by 107 any person -users can add or change a category assignment to other category.The resulting category 108 structure is noisy [14], sparse and it contains duplications and oversights [15].So, we can also face the 109 situation that categories are repeated at different levels of the tree, in which the root can be another 110 main category (one of the 27 considered).In order to avoid such situations, we cut off those branches 111 that were found at higher levels.Figure 2 shows an example of such procedure, when subcategory 112 "Food and Drink" is found at different levels of the tree and only one remains, which is at the highest 113 level.By counting articles in English Wikipedia in each of considered main categories we discovered that almost 15% of them are about people.Pie chart in Figure 3 shows shares of articles in English Wikipedia in 27 considered categories.As we mentioned before, in some language versions there is a relatively high average number of categories assigned to each article.This may increase the possibility of an article falling into more than one main category.We studied this issue for the leading language versions (Arabic, English, French) with regard to the number of categories per article.Results are presented in Figure 6.

Semantic Classification
The second approach to category assignment to Wikipedia articles is based on Wikidata and DBpedia.Wikidata is a collaboratively edited knowledge base [17].DBpedia is the semantic database resulting from extraction of structured, multilingual knowledge from Wikipedia [18,19].The data from this open databases are widely used in a number of domains: web search, life sciences, maritime domain, art market, digital libraries, business networks and others [20][21][22][23].
DBpedia uses its own ontology with defined properties and classes organized into a hierarchy.
DBpedia provides English names to each class, such as "Place", "Species", "Person" etc. Wikidata gives unique identifier to each class, for example class "city" is marked as Q515, "human" as Q5, "Organization" as Q43229.Another difference between these databases lies in the number of classes and placing these classes in an ontology.Wikidata has over 300 thousand classes [24], while DBpedia ontology consist of about 800 classes [25].
A significantly larger number of classes in Wikidata can lead to difficulties in finding a list of objects on a particular topic.For example, if we want to find all cities, it is not enough to take into account only one class Q515 (city), because city can also be described by Q1637706 (city with millions of inhabitants), Q5119 (capital), Q2264924 (port city), Q58339717 (city of India), Q174844 (megacity) and other identifiers.This variety of classes leads to significantly fewer instances in each class in Wikidata than in DBpedia [24].
We should consider also way of assigning a class to objects in these semantic databases.DBpedia extracts information from Wikipedia infoboxes and identifies classes based on name of the infobox and values of some special parameters.Thus, articles with the same infobox name often go to the same class.In Wikidata, items can be edited by everyone, therefore different classes can be assigned to similar objects.
There are some papers that study differences between DBpedia and Wikidata [24,26,27].Each has own advantages, so we decided to use combined data to divide articles into separate classes: actor, automobile, business, city, film, football player, human, programming, university, videogame, and website.One of the advantages of such a classification approach by topic is that we are dealing here with more explicit assignment of articles to specific classes and each language version has at least several representatives of each class.

Quality Measures
In order to discern the quality of content, the Wikipedia community created a grading system for articles.However, each language version can use its own standards and grading scale [28,29] Even though the grading system is available, still the big challenge is a large number of unassessed articles.For example, German and Polish Wikipedia has less than 1% of articles with quality grades.
Moreover, articles about the same topic in different languages can also be graded using different criteria.The above facts not only pose problems for comparing the quality of articles in the same language but also for evaluating and comparing different language versions of articles on the same topic.
Using machine learning techniques it is possible to solve the problem of quality assessment of Wikipedia articles as a classification task.In order to build such models, various features can be taken into the account, for example length of an article, number of references, number of images or sections [30][31][32][33][34][35].
One of the universal approaches for quality assessment of multilingual articles is Objective Revision Evaluation Service (ORES) [36].This service automates tasks like detection of vandalism and removal of edits made in bad faith [37].Additionally the service can evaluate articles on a scale between 0 and 1 in some language versions.However, automatic quality assessment of an article by the ORES is currently limited to nine language version of the Wikipedia and it does not include such developed language chapters as German, Spanish, Italian, Polish, Japanese, or Chinese.
In our previous studies [28,38] we defined the synthetic measure to combine several features of articles to allow ranking of Wikipedia articles on a scale between 0 and 100.It is based on the most universal features inferred from machine learning models built for several languages.In the paper we present conclusions from an assessment of over 39 million articles.Additional focus of this work is analysis of demand for information about various topics in different languages from the point of view of readers, as well as from the authors of Wikipedia content.The intersection of those two dimensions is also considered.
Our previous study [39] showed that popularity of the Wikipedia articles can be measured by different SEO metrics from other websites.Such indicators as social signals from Facebook, Twitter, Pinterest, Youtube and others can help to determine also the quality the content in multilingual encyclopedia from the external sources.In this work we decided to use internal popularity measures from the point of view of readers and writers of the Wikipedia articles.Additionally we decided to provide cumulative (global) values of these measures over the language versions about various subjects.
Diverse approaches to defining information by researchers lead also to inconsistencies in defining the notion of its quality.According to the most popular definition, quality of information can be defined as fitness for use [40,41].
In order to define the quality dimensions in Wikipedia, one should take into account the similarity of this website with traditional encyclopedias and Web 2.0 services.On the one hand, content in Wikipedia is created to be a reference point, in an encyclopedic style.According to various studies it has comparable accuracy to other traditional encyclopedias [42,43].The quality of an article in a traditional encyclopedia can be defined by 7 dimensions: authority, completeness, format, objectivity, style, timeliness, uniqueness [44,45].On the other hand, Wikipedia is built in a way to allow collaboration between users.It it therefore based on Web 2.0 technologies, which have the following quality dimensions: accessibility, completeness, credibility, involvement, objectivity, readability, relevance, reputation, style, timeliness, uniqueness, usefulness [45,46].
Considering the quality criteria adopted by the Wikipedia community and previously described characteristics of traditional encyclopedia and Web 2.0 documents, we can choose the following quality  7 shows coverage between quality dimensions of the Web 2.0, traditional encyclopedia and Wikipedia.Length of text can be measured in various ways -most often it is represented by the length in bytes, the number of letters or words [28,38,[47][48][49][50][51][52][53][54][55][56][57][58].Length of an article is related to completeness and may indicate the presence of relevant facts and details in its articles.
High-quality articles are expected to use reliable sources [59].Readers of encyclopedias must be able to check where the information comes from [60].Therefore, one of the most commonly used reliability measures is the number of references in a Wikipedia article [28,34,38,[48][49][50]56,58,[61][62][63][64].References are related to credibility of the article.Our previous research has shown that it is advantageous to analyze not only the quantity but also the quality of the references [39].
Length of text can be positively correlated with the number of references but it is important that all relevant facts in Wikipedia should be supported by reliable sources.For this purpose, the reference density can be calculated as the number of references divided by the length of text.
Wikipedia articles must provide information in a fair and impartial manner.In this case, we can take into account information presented graphically -images [28,34,38,47,50,[55][56][57]61,62,65,66].On the one hand, pictures can help to assess the objectivity of the presented material.On the other hand we can also measure completeness (because articles on a specific topic should contain images) and style (because the authors decided to add more photos instead of writing long text).
High-quality content must be prepared in accordance with the guidelines of Wikipedia regarding the style that applies to, among others, organization and structure of the article.Therefore, one of the simplest and most popular measures of this dimension is the number of sections in the article [28,32,34,50,52,56,58,[61][62][63].
Quality measures mentioned before can be combined to build a synthetic measure for evaluation of Wikipedia articles.Unlike most methods in this domain, the synthetic measure can assess the quality of Wikipedia articles on a scale from 0 to 100 [38].Thus, we can compare quality of articles between different language versions, which can have own quality grading scheme.For each language version of Wikipedia, each feature could play an important role in assessing the quality; therefore we first counted the normalized metrics average (NMA) by the following formula: where mi is a normalized measure m i and c is the number of measures.
Next we took into account the number of quality flaw templates (QFT) in the considered article (if they existed) and our final formula for the quality measure reads as follows: Previous research [29] revealed that the synthetic measure was one of the most significant among 100 variables used in quality model of Wikipedia.

Popularity Measures
Popularity of an article can be determined with measures reflecting the demand for information contained in it by the readers and Wikipedia authors.Popularity can play an important role in quality estimation in specific language versions of Wikipedia [29,34].Larger number of users reading an article can contribute to faster identification and correction of errors, therefore amendments can be made more often (including update of the information).
Popularity of an article can be measured based on the number of visits [34,38].For example, one of the studies compared reptiles species' page view numbers across languages and in their spatial distribution along with various biological attributes [67].
For assessment of popularity we decided to use features available in Wikipedia database -page views and number of unique authors of an article.We also provided local and global measurements characterizing articles, which took into account semantic links between language versions.
For each page of Wikipedia, daily page views statistics are available in a dedicated online service [68] and Wikimedia dumps [69].We used dumps to analyze popularity of over 39 million articles in considered language versions of Wikipedia.
Popularity measure in this study were calculated as a median of number of page visits per day, as it was proposed in the previous study [38].If the measurement concerns only selected language version, then we call it local popularity.We can also calculate the global popularity, which takes into account popularity of articles about the same topic in different languages (the so called interwiki links are considered).The global popularity of an article is calculated according to the following formula: and n is number of the language versions of the selected article.
For quality improvement even more important than the number of page views is the number of real edits.Authors' interest (AI) can be measured as the number of unique authors of the Wikipedia The number of authors can be extracted from article history. Figure 8 shows part of the article history about Game of Thrones (season 8) in English and German Wikipedia with highlighted authors.Similarly to measuring popularity, AI can also be calculated for a specific language version (local AI) and as a cumulative value for all languages (global AI).Authors are identified by names or IP addresses.So, if the same user edited the article in different language versions, in the global AI it will be counted as one author.Calculation of this measure can be carried out using the flowing formula: Authors lang (article) , where Authors means a set of authors' names, lang is the index of specific language version and n is the number of language versions of the article.

Quality and Popularity Assessment
Following the procedures described in previous sections, we extracted over 100 million values of features characterizing articles in all analyzed languages.These values were then used to calculate the synthetic measure that assesses quality of the content.We next grouped articles by 27 main categories This is due to the fact that in this language version only a few articles fall into this category and they are generally well written according to studied features.Articles about crime also have relatively higher quality scores in English (en) and Chinese (zh) Wikipedia.
Second place in the ranking are taken by articles about events in Uzbek Wikipedia (uz) -43.96 points.Again, this main category does not contain much content -there are only 31 articles.If we take into account the development of the Uzbek Wikipedia (about 130 thousand articles), we can conclude that this category is rather important for local community of editors.Articles about events also have relatively higher quality scores in Hungarian (hu), Slovak (sk), Hebrew (he), and Chinese (zh) Wikipedia.
Third place regarding the quality is taken by articles about mathematics in Volapük Wikipedia -39.63 points.However, in this language chapter the category contains only 2 articles.Latin Wikipedia (la) has the fourth place with average quality of articles about religion -37.77.
If we take into account the most developed English Wikipedia, the highest average quality of articles can be found in categories: Philosophy, Crime, Military, and History.Generally, we can conclude that English Wikipedia articles usually have high value of average quality measure in different topics.Generally, page views values are higher for the most popular languages.This led to the fact that the first 11 positions in the rank are occupied by English (en) Wikipedia.The most popular topic in this language is Philosophy.One of the highest average popularity in this language characterizes also articles about crime, technology, entertainment, mathematics, culture, and health.All these categories had at least 20 thousand page views in year 2018.
Second most popular language version is Spanish (es).Similarly to English, the most visited category is Philosophy.It is also worth to mention two other popular categories in this language: Mathematics and Health.Articles in three mentioned main categories of Spanish Wikipedia have at least 14 thousand page views per year.Second language version that has most active authors is Hebrew (he) Wikipedia with articles about entertainment.During a year at least 6 authors have edited each article in this topic.Entertainment is also popular among authors in Italian (it), Spanish (es) and Chinese (zh) Wikipedia.At the same time Italian Wikipedia we can met as the third language in the authors' interest ranking.Depending on Wikipedia language version, we observed different categories with the highest average quality, popularity and AI.For example in English Wikipedia articles in category "Crime" have the highest average quality, but articles from category "Philosophy" has the highest average popularity and AI.Another example: Arabic Wikipedia has the articles from Religion category as the best for these three measures.Similar applies to Latin Wikipedia.In Persian Wikipedia there is also a similar situation, with exception to popularity -here category "Philosophy" has the highest values.
Articles in Russian Wikipedia from category "Entertainment" are the most popular and has the highest average quality, at the same time from authors point of view is most popular "Events" category.Similar applies to German Wikipedia.Category "Government" Azerbaijan, Finnish, Slovenian Wikipedia occupies a leading position.
Finally, we do the similar calculations for articles in semantic classes: actor, automobile, business, city, film, football player, human, programming, university, videogame, website.Figure 12 shows average quality, authors interest and page views in 2018 per article in each semantic class and language version of Wikipedia.Darker colors in heatmaps represent higher values of the selected measures.As for page views, we have similar situation as it was in the case of main category classifications -   Profiles of Wikipedia articles can also be used to compare the demand for a specific product between various language communities.For example video game Dota 2 is the most popular in English, Russian, Chinese, German, and Spanish [83].Based on obtained measures for the action-adventure video game Grand Theft Auto V (GTA 5) we can see relatively large demand from English, Russian, Arabic, Spanish, and Chinese language community [84].

Results and Discussion
During the research we encountered several restrictions, mainly related to the differences between language versions of Wikipedia.For example, as we showed in Table 3, some main categories do not have links to all considered language versions.This is also true for developed languages.For example, category "Art" in English Wikipedia does not have direct equivalent in German Wikipedia, which uses category "Kunst und Kultur" [85] ("Arts and Culture") to describe part of this topic.
Regarding categories, our experiments showed that each language version has specific ratio between number of articles and number of categories.Additionally, some language versions can have a lot of undefined pages for the categories.There is also a difference in the number of categories that are assigned to each article.Some languages can use an average of 30 categories to describe one article, while the others are limited to 2-3 categories per article.
Depending on Wikipedia language version, we observed different categories with the highest average quality, popularity, and authors' interest.For example in English Wikipedia articles in category "Crime" have the highest average quality, but articles from category "Philosophy" have the highest average popularity and AI.Another example, Arabic Wikipedia has the articles from Religion category as the best for these three measures.Articles in Russian Wikipedia from category "Entertainment" are the most popular and have the highest average quality, while from authors point of view the most popular is "Events" category.
Results for authors popularity can be sometimes biased due to temporal or permanent restrictions.
According to one of the main principles of Wikipedia anyone can edit content.However, in some particular situations this right can be revoked to protect content from unwanted changes (vandalism) [86].Each language version can define own levels of page protection.For example, in English Wikipedia there is a full protection, where only administrators can edit an article, and semi-protection, which prevents editing by unregistered users or users that are not confirmed.Each article can be protected for a specified period.Figure 17 shows an example of the protected Wikipedia article about Bitcoin with a marked level of protection.As a result, some articles can have less authors' interest than it would in the situation without protection.
In our work, we provided classification of articles by main categories according to structure of categories in English Wikipedia.However, each language can have own definition of main categories.
In future, we plan to develop more sophisticated methods to take into account refined category structures.Supplementing research results are available online at WikiRank service [78].In research we used some tools that are available on GitHub [88].

Conclusions and Future Work
In this paper we presented results of quality and popularity assessment of articles in multilingual Wikipedia.For this purpose we calculated over 200 million values characterizing quality and popularity of articles in 55 language versions of Wikipedia.Additionally, we analyzed over 10 million categories, over 26 million links between them, and about 400 million links from articles to categories in order to determine assignment of articles to one of the topics in main classification.In order to assign articles from different languages to various topics we also used semantic databases -Wikidata and DBpedia.
We combined data from these sources to obtain more comprehensive classifications of articles.
Results of the research showed not only how quality and popularity differ for articles from various topics and languages but also how the same topic is developed in different languages of Wikipedia in terms of quality and popularity of content.We observed that articles from topics that are popular in a given language are characterized by a relatively higher quality.For instance articles related to main category 'Religion' have relatively higher quality and popularity in Arabic and Latin Wikipedia.
Likewise, articles from main category 'Goverment' have relatively higher quality and popularity in Azerbaijani, Finnish, Armenian, Romanian, and Slovenian language version of Wikipedia.Articles related to main category 'Entertainment' are more popular in Chinese, Russian, German Wikipedia.At the same time, articles in those three language versions has relatively the highest quality compared to other main categories.
Additionally to categories, we also studied semantic classes as defined by DBpedia ontology and their relation to quality and popularity.The highest average number of page views among different classes in almost all considered language versions had articles that described websites, e.g.Facebook, YouTube, Google.However, popular articles from this class rarely were assessed as articles of high quality.Articles about cities were relatively better described in English, German, Czech, Hindi, Polish, and Spanish Wikipedia.Actors were described better than other classes in Bulgarian, Catalan, Danish, Greek, French, Hebrew, Croatian, Indonesian, Italian, Malay, Portuguese, Serbian, Tamil, Thai, Turkish, Ukrainian, and Chinese language versions.
With regard to popularity, we proposed to pay attention not only to how often users visits certain articles but also what is authors' interest in them.The authors' interest measure can be calculated for a language version or can be combined across studied languages.Sometimes both popularity measures show similar leader in main categories and semantic classes.For example, Slovenian Wikpedia has the most popular articles related to main category 'Government', while for readers and authors of English Wikipedia articles have higher preference related to 'Philosophy'.If we consider semantic classes, we can conclude that among analyzed languages the most popular articles for Wikipedians are related to cities and automobiles.We also aggregated numbers for all considered languages so that global demand for specific products, such as films, video games, cars, can be studied.
Additional analyses of popularity measures allowed to find priorities and preferences of Wikipedians and readers in relation to temporal dimension.Often the most popular subjects of the readers differed from leading subjects from authors point of view in the same periods of time.This can be explained by the fact that popular articles are protected and cannot be edited by anonymous users.Additionally, some Wikipedia authors may choose articles based on various initiatives related to improvement of specific topics at certain period of time.
Presented results can be used to build more complex models for quality assessment of information in Wikipedia in different languages and topics.In the future, they can help not only to automatically enrich less-developed language versions of Wikipedia but also can be used to build massive semantic databases with powerful inference system, creating new knowledge for humanity in a relatively short time.
The work towards more precise assessment of Wikipedia quality will be continued, especially different measures and approaches for quality assessment in Wikipedia and other collaborative knowledge bases will be studied.As of April 2019, based on our calculations, there were over 70 thousand wiki services in the Internet, which potentially can be used to enrich various knowledge bases used in enterprises.Additionally, there are over 1300 linked databases [89] that use data from open sources.We can also take into account dedicated web portals that allow companies and individuals to share their databases for research, such as Kaggle [90].Local and global AI measurements can be improved by including different additional features.For example, it is possible to divide all users into three categories: anonymous users, registered users, and bots.We can also take into account reputation and experience of each author of the article.For this purpose we can use information provided by services like GUC [91] or WikiTop [92].

Figure 1 .
Figure 1.Subjects overlaps of articles in various language versions of Wikipedia.Source: own calculation based on Wikipedia dumps in April, 2019.Over 175 thousand of interactive combinations of these Venn diagrams can be found on the Web page: http://data.lewoniewski.info/computers/vn1/

Figure 2 .
Figure 2. Occurrence of similar sub-categories in the English Wikipedia category polyhierarchy.Source: own work based on Wikipedia dumps from April 2019.

Figure 3 .
Figure 3. Shares of articles in each category in English Wikipedia.Source: own calculation based on Wikipedia dumps in April, 2019.

Figure 4
Figure 4 shows distribution of articles by category within each considered language version of Wikipedia.Darker colors in the heatmap represent higher share of articles in particular main category within the selected Wikipedia languages.

Figure 4 .
Figure 4. Share of articles in main categories within each of 55 language versions of Wikipedia.Source: own calculation based on Wikipedia dumps in April, 2019.More detailed and interactive chart can be found on the Web page: http://data.lewoniewski.info/computers/heatmap-cat-artAftercombining articles from all considered language versions to particular category we concluded that the largest number of articles are in one of two categories: Geography (12.68%) and People (11.48%).Pie chart in Figure5presents how articles in all considered Wikipedia languages are distributed among 27 main categories.

Figure 5 .
Figure 5. Shares of articles in each category in 55 language versions of Wikipedia.Source: own calculation based on Wikipedia dumps in April, 2019.

PreprintsFigure 6 .
Figure 6.Overlap of articles between selected main categories in Arabic, English and French Wikipedia.Source: own calculation based on Wikipedia dumps in April, 2019.Over million of interactive combinations of these Venn diagrams (each main categories and language versions) can be found on the Web page: http://data.lewoniewski.info/computers/vn2/.

Figure 7 .
Figure 7. Quality dimensions of Web 2.0 portals, encyclopedias and Wikipedia.Source: own work based on [45] Each quality dimension contains a specific set of features (measures).Some features can be related to multiple quality dimensions.There are different ways to define and extract features of the Wikipedia articles.Based on the literature and own experiments, we focused on one of the important features, which can show quality of Wikipedia article from different dimensions.
articles.Each user editing articles on Wikipedia has own experience, level of knowledge and can adhere to a certain world view.In this regard, it can be assumed that larger number of authors can positively influence the objectivity of the article, since it may contain different points of view on a particular question.At the same time, the number of authors of an article can also indicate the relevance of the article to the Wikipedia community.To sum up, articles created by a larger number of people may be more objective, hence one of the measures leveraged in our research is the number of unique authors[28,34,47,[55][56][57][58][63][64][65][70][71][72][73][74][75].

Figure 8 .
Figure 8. Part of the article history about Game of Thrones (season 8) in English (en) and German (de)Wikipedia with highlighted authors.Source:[76,77]

Figure 9 .
Figure 9. Average quality of articles in each category and language version of Wikipedia.Source: own calculation based on Wikipedia dumps in April, 2019.More detailed and interactive chart can be found on the Web page: http://data.lewoniewski.info/computers/heatmap-cat-quality.The highest average quality have articles in category Crime in Slovak Wikipedia (sk) -63.92 points.

Figure 10
Figure 10 shows average number of page views per article in year 2018 for each category and language version of Wikipedia.Darker colors in the heatmap represent higher average number of page views of articles in specific category and language version.

Figure 10 .
Figure 10.Average page views per article in year 2018 for each main category and language version of Wikipedia.Source: own calculation based on Wikipedia dumps.More detailed and interactive chart can be found on the Web page: http://data.lewoniewski.info/computers/heatmap-cat-views Third place is taken by Russian (ru) Wikipedia and category Entertainment, with about 16 thousand page views per year.Entertainment is also the most popular topic in Chinese (zh) Wikipedia.Finally, Figure 11 shows average number of authors (authors' interest) per article in 2018 in each category and language version of Wikipedia.Darker colors in the heatmap represent higher values of average number of authors of articles in specific category and language version.

Figure 11 .
Figure 11.Average number of authors per article during 2018 in each main category and language version of Wikipedia.Source: own calculation based on Wikipedia dumps.More detailed and interactive chart can be found on the Web page: http://data.lewoniewski.info/computers/heatmapcat-authorsAs in the case of the popularity of page views, in the ranking of authors' interests categories in English Wikipedia topped the ranking.Here we have such popular categories as Crime, Philosophy, Entertainment.Articles about topics were edited at least by 8 authors during the 2018 year.

Figure 12 .
Figure 12.Average quality, authors interest and page views during 2018 per article in each class and language version of Wikipedia.More detailed and interactive chart can be found on the Web page: http://data.lewoniewski.info/computers/heatmap-classesThe leader in terms of the value of average quality is Tamil (ta) Wikipedia with articles that describe cars (automobiles) -43.22 points.The second place in this ranking occupy articles about football players in Hindi (hi) Wikipedia -40.35 points for quality per article.The third place in quality took English (en) Wikipedia with articles about cars -37.39 points.Articles about cars have also relative high quality un Hebrew (he), Hindi (hi) and Chinese (zh) Wikipedia -over 31 points.In this qualityranking most often we can met articles about cities in English (en), Latin (la), German (de), Slovenian (sl), Serbo-Croatian (sh), Greek (el) Wikipedia -over 30 points per article.

Figure 14 .Figure 15 .Figure 16 .
Figure 14.Local ranking with quality distribution of all articles in English Wikipedia in WikiRank service.Source:[80]

Table 2 .
Category ratio shows the number of unique categories per number of articles in a particular language version.The highest value of this indicator has Urdu Wikipedia -1.23.The largest English Wikipedia is in the middle in the ranking regarding the value of this indicator.

Table 2 .
Number of categories, number of links from articles to categories and between categories in 55 lanquage versions of Wikipedia (sorted by category density).Source: own calculations in April, 2019

Table 3 .
List of the categories in "Category:Main topic classifications" in English Wikipedia with number of the considered language versions (April 2019) Articles (FA), Good Article (GA), A-class, B-class, C-class, Start, Stub.Russian Wikipedia has also 7 quality grades but with other names and criteria: Izbrannaja Stat'ja (similar to FA), Horoshaja Stat'ja (similar to GA), Dobrotnaja Stat'ja, I, II, III, IV (similar to Stub).German Wikipedia uses only two quality grades (Exzellente Artikel and Lesenswerte Artikel) which has similar criteria to FA and GA grades respectively.Polish Wikipedia defined 5 quality grades: Artykuł na Medal (similar to FA), . For example, in English Wikipedia, articles can get one of 7 grades (from highest to lowest): Featured Preprints (www.preprints.org)| NOT PEER-REVIEWED | Posted: 5 August 2019 doi:10.20944/preprints201905.0144.v2Peer-reviewed version available at Computers 2019, 8, 60; doi:10.3390/computers8030060

preprints.org) | NOT PEER-REVIEWED | Posted: 5 August 2019 doi:10.20944/preprints201905.0144.v2
[38]-reviewed version available at Computers 2019, 8, 60; doi:10.3390/computers8030060 of Wikipedia has a special distinction for articles of the highest quality -equivalents to FA and GA grades in English version.Normalization of the 5 selected features depends on language chapter of Wikipedia, since it uses thresholds, which depend on the best articles in the considered language version[38].Normalization of each feature was conducted according to the following rule: if value of a given feature in a given language exceeded the threshold of median value of the best articles in the same language version, it was set to 100 points; otherwise its value was linearly scaled to reflect the relation of the value to the median value.For example, if the median for the number of references in Polish Wikipedia was 97, any article with a larger number of references would score 100 for this feature; an article with 59 references would score proportionally 60.82 (59/97) points after normalizing.Changing the value of any metric in a particular Wikipedia language version would have a different effect on the normalized value.
Synthetic measure encompasses normalized values of the following five features: length, number of references, reference density, number of images, and number of sections.Every considered language Preprints (www.

Table 4
presents main categories that have the highest value of average quality, average popularity and authors' interest in each language version of Wikipedia.

Table 4 .
Main category of articles with the highest value of average quality, average popularity and authors' interest in each language version of Wikipedia.Source: own calculations.
English Wikipedia has here the highest values.The most popular class in this language versions is

Table 5 .
Classes of articles with the highest value of average quality, average popularity and authors' interest in each language version of Wikipedia.Source: own calculations.

Table 6 .
Top 3 articles about cars with highest number of page views and authors' interest in multilingual ranking, monthly.Source: own calculations.

Table 8 .
Top 3 articles about video games with the highest number of page views and authors' interest in multilingual ranking, monthly.Source: own calculations.