There isn't too much consistency in the statements of media outlets (see a selection below)
Most have one or more P31 (instance of) statements that are one of these: (or a subclass thereof)

• Q11033 # mass media
• Q1193236 # news media
• Q2943864 # news satire
• Q27881073 # fake news website
• Q1331793 # media company
• Q24354647 # editorial team
• Q2001305 # television channel
• Q56611639 # media industry
• Q38926 # news
• Q11578774 # broadcasting program

There are 1506831 of these.

2 other properties are also commonly found to describe media outlets: P452 (industry) and P136 (genre), where the value is often one of the Q-ids mentioned earlier, or one of:

• Q11033 # mass media
• Q56611639 # media industry
• Q11030 = journalism
• Q25245117 = telecommunications
• Q3972943 = publishing

Combining all these properties & items, we get this query that returns 1686161 results:

SELECT DISTINCT ?item WHERE {
  { wd:Q11033 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = mass media
  UNION
  { wd:Q1193236 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = news media
  UNION
  { wd:Q2943864 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = news satire
  UNION
  { wd:Q27881073 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = fake news website
  UNION
  { wd:Q1331793 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = media company
  UNION
  { wd:Q24354647 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = editorial team
  UNION
  { wd:Q2001305 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = television channel
  UNION
  { wd:Q56611639 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = media industry
  UNION
  { wd:Q38926 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = news
  UNION
  { wd:Q11578774 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = broadcasting program
  UNION
  { wd:Q56611639 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = media industry
  UNION
  { wd:Q11030 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = journalism
  UNION
  { wd:Q25245117 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = telecommunications
  UNION
  { wd:Q3972943 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = publishing
  UNION
  { wd:Q1002697 ^wdt:P279*/^(wdt:P31|wdt:P136|wdt:P452) ?item . } # instance of/genre/industry = periodical
}

This casts a very wide net and it's quite impossible to validate how relevant all results are. I went over a couple hundred results, and none of the ones I did recognize seemed false positives, though.

Does above list of properties/items to identify media outlets (and exclude such topics) make sense? Ok to move forward with this list?

Alternatively, we could check for items that have a property that we only expect in media outlets (e.g. P449 (original broadcaster)). I haven't found that to be particularly helpful, though - they're so specific that they're missing for many items, in which case we're back to also including P31 etc. I don't think we should pursue this.

A selection of media outlet items and their relevant-ish statements:

Reuters:

• instance of = news agency
• industry = telecommunications

Associated press:

• instance of = news agency
• instance of = photo agency
• industry = news media
• industry = media industry

CNN:

• instance of = United States Cable news
• industry = journalism

ABC News:

• instance of = television station
• industry = journalism

Fox News:

• instance of = United States cable news
• industry = journalism

Late Night with Conan O'Brien:

• instance of = television series
• genre = talk show

The Onion:

• instance of = weekly newspaper
• instance of = news satire
• industry = publishing

Playboy:

• instance of = magazine
• instance of = men's magazine
• instance of = nude magazine

BBC News:

• instance of = news desk
• instance of = news broadcasting
• industry = journalism

Bild:

• instance of = daily newspaper
• genre = tabloid journalism

Al Jazeera:

• instance of = broadcaster
• instance of = television station

France 24:

• instance of = television channel
• instance of = television station
• industry = journalism

Charlie Hebdo:

• instance of = newspaper
• instance of = satirical newspaper
• genre = political satire
• genre = satirical newspaper

Het Nieuwsblad:

• instance of = daily newspaper

Het Journaal:

• instance of = television program
• genre = news program

VTM Nieuws:

• instance of = television program

Terzake:

• instance of = television program
• genre = current affairs

Vive le Vélo:

• instance of = television program
• genre = talk show

Humo:

• instance of = magazine

Dag Allemaal:

• instance of = periodical

Does above list of properties/items to identify media outlets (and exclude such topics) make sense? Ok to move forward with this list?

LGTM -- let's get @mfossati's and @AUgolnikova-WMF's eyes on it.

Thanks @matthiasmullie for the deep dive in the Wikidata ontology!
Totally agree we can't come up with a complete solution.

I made some explorations as well:

• the most generic class looks like media and has 18,387 subclasses - query. Probably the widest net we could cast
• querying all media instances times out, way too many
• the more specific mass media class has 792 subclasses - query
• there are 271,628 mass media instances - query. Perhaps a too narrow net

Conclusion: OK to move forward with @matthiasmullie's net, looks like the best trade-off.

mlitn opened https://gitlab.wikimedia.org/repos/structured-data/section-topics/-/merge_requests/18

Exclude media topics from section topics dataset

Thanks! LGTM for the current version

mlitn merged https://gitlab.wikimedia.org/repos/structured-data/section-topics/-/merge_requests/18

Exclude media topics from section topics dataset

In T323036#8621701, @mfossati wrote:

It's merged, and the bot agrees! 😄 Closing.

Yes, but what happens the day the bot disagrees?!