In T312900#8107795, @mfossati wrote:

We're using mwparserfromhell. Given a wikitext, current Research code is: mwparserfromhell.parse(wikitext).get_sections(levels=[2], include_headings=True). This should extract h2-level sections and won't include the lead section. We can modify the behavior, but the lead section potentially contains infobox templates and the like, so not sure we want to parse that as well, what do you think?

It contains a lot of templates, but also a concise summary of the whole article, and skipping the templates should be straightforward with mwparserfromhell. Not sure how useful it would be to have topics for that - on one hand, the topic of the lead section should be the same as the topic of the whole article; conceptually, it doesn't have its own topics. But then, we don't have article-level topics so topics for section 0 could be a nice approximation for that. (Of course, in practice the lead section could be missing or poorly written and not really reflective of the article.) Not sure what use cases we have for section 0 - I assume for image suggestions we wouldn't use it as the current image suggestion pipeline already does top-level image suggestions?

Unrelated to section 0, maybe it's worth noting that MediaWiki considers raw <h2> tags in the wikitext as section titles but mwparserfromhell doesn't, so in theory they don't always agree on what sections an article has. It's very unlikely you'd encunter raw <h2> tags in wikitext though, most wikis consider it a code style violation.

In T313901#8109067, @Tgr wrote:

It contains a lot of templates, but also a concise summary of the whole article, and skipping the templates should be straightforward with mwparserfromhell.

True, tracking below one way to remove templates:

zero = wikitext.get_sections()[0]
templates = zero.filter_templates()
for t in templates:
    zero.remove(t)

Not sure how useful it would be to have topics for that - on one hand, the topic of the lead section should be the same as the topic of the whole article; conceptually, it doesn't have its own topics. But then, we don't have article-level topics so topics for section 0 could be a nice approximation for that.
Not sure what use cases we have for section 0

Section 0 topics may be a great starting point for the schema.org use case, see T302735: [SPIKE] Investigate effects of schema.org section metadata on Google search results, CC @CBogen @AUgolnikova-WMF

I assume for image suggestions we wouldn't use it as the current image suggestion pipeline already does top-level image suggestions?

In principle yes.

Unrelated to section 0, maybe it's worth noting that MediaWiki considers raw <h2> tags in the wikitext as section titles but mwparserfromhell doesn't, so in theory they don't always agree on what sections an article has. It's very unlikely you'd encunter raw <h2> tags in wikitext though, most wikis consider it a code style violation.

Got it, thanks for the clarification.