Research talk:Automated classification of article importance/Work log/2017-04-28

Friday, April 28, 2017

Today I'll focus on the WikiProject candidates I've selected, to understand more about them.

WikiProject candidates

I created a list of candidates. Before we start processing their articles and building models, I want to look at how many articles they have in the different importance classes. Here's a table with info on that:

proj_name	n_top	n_high	n_mid	n_low	n_na	n_unknown	n_unassessed	n_articles
WikiProject Africa	2,289	1,261	4,097	25,952	2,220	34,945	10,173	80,937
WikiProject Albums	138	963	4,133	59,107	11,507	71,281	24,896	172,025
WikiProject Beauty Pageants	12	36	131	742	349	3,192	1,522	5,984
WikiProject Buddhism	115	167	396	1,347	149	2,514	1	4,689
WikiProject Chicago	63	142	1,447	12,113	647	21,918	7,274	43,604
WikiProject China	418	1,728	9,824	13,768	2,260	19,114	3,734	50,846
WikiProject Cycling	13	67	463	9,202	312	10,602	551	21,210
WikiProject Dungeons & Dragons	14	139	357	1,346	1,168	1,032	6	4,062
WikiProject Europe	22	79	332	1,393	80	2,448	737	5,091
WikiProject Historic sites	24	125	1,001	4,051	78	3,263	11	8,553
WikiProject Horror	30	169	514	5,167	346	5,169	707	12,102
WikiProject Iran	108	617	769	2,976	8,019	68,961	8,170	89,620
WikiProject Judaism	235	513	1,636	4,083	364	3,337	868	11,036
WikiProject Korea	66	545	1,916	9,295	501	8,469	843	21,635
WikiProject Malaysia	24	463	1,619	2,681	134	2,500	1,625	9,046
WikiProject Motorsport	51	130	751	3,197	148	3,018	2,320	9,615
WikiProject National Football League	365	516	3,008	4,425	338	19,112	2	27,766
WikiProject Olympics	25	647	7,051	47,797	373	47,662	5,038	108,593
WikiProject Pharmacology	116	583	1,901	3,152	285	4,456	464	10,957
WikiProject Politics	115	1,123	4,133	20,015	1,037	13,818	7,315	47,556
WikiProject Politics of the United Kingdom	175	319	1,469	9,322	418	19,385	6,077	37,165
WikiProject Rock music	118	581	1,642	4,665	358	5,171	2,811	15,346
WikiProject Rugby league	23	175	951	6,296	753	4,684	1,162	14,044
WikiProject Television	76	605	3,013	32,237	14,832	37,081	19,721	107,565
WikiProject Television Stations	5	12	186	624	1,853	3,331	3,049	9,060
WikiProject United Nations	0	3	92	21	19	3,290	1,697	5,122
WikiProject Yugoslavia	32	139	305	989	48	892	317	2,722

Sorting the table on the number of Top-importance articles (in ascending order), we can make some notes about the WikiProjects with low number of such articles:

• WikiProject United Nations is somewhat of a meta-project, appears as a collaboration between United Nations and Wikimedia. On talk pages, their tag shows up as part of WikiProject International relations. The project does not have a definition of its importance ratings, but do use them.
• WikiProject Television Stations refer to the WP 1.0 importance scale in their talk page templates. The project does not define its own importance scale.
• WikiProject Beauty Pageants also refers to the WP 1.0 importance scale instead of its own.
• WikiProject Cycling has its own importance scale. They define importance through what geographic level a topic is notable at (e.g. internationally means "top importance", locally means "low importance"). Note that they also specifically point out that importance is orthogonal to article quality.
• WikiProject Dungeons & Dragons also defines their own importance scale, which adds "Bottom importance" to the existing set of categories.
• WikiProject Europe both defines and describes their importance scale. Their description specifically mentions the "average reader of Wikipedia needing to look up the topic", suggesting that our measure of viewership should fit well in their case. Unsure whether the project only having 22 articles (and 1 portal) rated Top-importance will be a problem or not, though.
• WikiProject Rugby league appears to refer to the WP 1.0 importance scale.
• WikiProject Historic sites defines their own importance scale, which appears to separate classes mainly based on geography.
• WikiProject Malaysia defines their own importance scale, it mainly separates between international/national notability when it comes to Top/High importance. They also feature a list of certain categories of topics and their general importance ratings right below their importance scale, e.g. companies are generally classed as low to mid importance.
• WikiProject Olympics have a detailed definition of how various categories of Olympics-related articles should be rated.
• WikiProject Horror both defines their importance scale and describes in more detail how geographic regions can play into notability, suggesting that something that might not be particularly popular in the U.S. (thereby perhaps being popular on English Wikipedia) can be highly important because it is highly notable in a different area.
• WikiProject Yugoslavia also define their own importance scale, which focuses on a core set of topics and their related pages. They also set a cap on the number of Top-importance articles to approximately 100, and limit the number of biographies. High-importance articles are also related to how well-known the topic is outside of Yugoslavia.
• WikiProject Motorsport define their own importance scale, and it seems like the scale is mainly divided by coverage area. Internationally notable topics are Top-importance, those that are notable on a specific continent are High-importance, and so on.
• WikiProject Chicago define their own importance scale and describe each importance level in great detail. Similar as for WikiProject Europe, the description mentions the need of the average reader of Wikipedia to look up the topic, suggesting that our importance criteria can be a good fit for this project.
• WikiProject Korea also define their importance scale. The descriptions are fairly short, and appears to be mainly based on fitness for an encyclopedia (Top-importance) and whether the topic is notable on an international or various national levels.
• WikiProject Television define their own importance scale, but mention that assessing importance is oftentimes not necessary (e.g. for "less influential television articles"). That can explain why they have 37,081 articles of "unknown" importance.
• WikiProject Iran define their importance scale, where they define Top-importance mainly based on whether the topic is mentioned on Iran. High-importance articles are "vital to understanding Iran", while Mid-importance are articles about topics that are "well known".
• WikiProject Buddhism define their importance scale in much the same way as WikiProject Iran, but of course centering it around Buddhism instead.
• WikiProject Politics has an importance scale that is described overall through the notion of the "average reader of Wikipedia", but for individual classes depending on whether the topic has a global, international, or local notability.
• WikiProject Pharmacology's importance scale defines importance mainly through how specific the subject is (e.g. a major class of drugs is Top-importance, while individual drugs are lower) and whether it is well-known, commonly prescribed, etc…
• WikiProject Rock music has a fairly short definition of their importance scale, mainly centered around the concept of "key" articles.
• WikiProject Albums defines their importance scale partly relative to whether the article can attain Featured Article or Good Article status, and partly whether the album is historically and culturally notable. How the albums performed with regards to sales and/or charts is secondary to historical/cultural impact.
• WikiProject Politics of the United Kingdom are working on rewriting their importance criteria, and have split them up into individual sections for constituencies, parties, and politicians. Constituencies are Low-importance by definition, and there should be no exceptions (this is specifically spelt out on the page. Importance for political parties spans the whole spectrum and is mainly dependent upon representation in UK government, European parliaments, etc… When it comes to politicians, importance relates to their (elected) position, as well as the importance of the party they represented.
• WikiProject Judaism also define their importance scale using the "average reader of Wikipedia", allowing them to define more popular topics as more important.
• WikiProject National Football League also uses the average reader to define importance. Their list of examples also specifically differentiate between general and biographic articles, meaning we might have to train two models in their case?
• WikiProject China has a short definition of importance. It does not describe the "average Wikipedia reader" like some others do, but are making references to the same concept in the descriptions. Requests Top-importance ratings to be discussed on the project's talk page. Descriptions of importance appears to also be somewhat tied to article quality (e.g. "contributes a depth of knowledge to the encyclopedia").
• WikiProject Africa also has a fairly short definition of importance. Importance is mainly defined through the importance to a specific field, or if the subject is internationally notable.