Comment by dredmorbius | Hacker News Reader

Comment by dredmorbius | Hacker News Reader

There are a few capabiliti es lacking from Algolia which I'd really like to see in a replacement:

- Negative search / exclusion: the ability to exclude terms from a search, as in "procfs -linux", which would look for any references to "procfs" which did not also reference "linux".

Edit: This exists, see dang's reply below.

- Replies to a specific user, e.g., "by:dredmorbius inreplyto:skeptrune <search terms>". I'm often looking for a specific context of my own previous comments.

- An improved date-bounding interface. If there's one thing that frustrates me about Algolia's interface, it's the GUI (and syntax) for defining dates. It's cumbersome, and at least on my browser, the dates are generally hard to read or invisible. Going back years is especially cumbersome.

I'll add: Algolia has been massively useful, and the fact that I can search HN, especially for my own content, has been a huge part of the value of the site, and is worlds ahead of other online platforms. (Mastodon / the Fediverse is catching up here, Diaspora*'s lack of search was among my main frustrations with the site and explains my absence there after more than a decade of participation.)

Algolia does do search term exclusion. Compare https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que... and https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu....

On the second point - HN has an undocumented endpoint https://news.ycombinator.com/replies?id=skeptrune&by=dredmor... but that doesn't give you search of course.

whoah

Just guessed another two undocumented endpoints "deadcomments" and "deadstories", these take you to a special admin login page if you're logged out, or say "Unknown." if logged in:

https://news.ycombinator.com/deadcomments

https://news.ycombinator.com/deadstories

And the endpoint "flagged", which is empty for me:

https://news.ycombinator.com/flagged

But if you add a username, e.g. "?id=dang", it says "Can't display that." instead:

https://news.ycombinator.com/flagged?id=dang

Interesting to stumble upon these even if they do nothing for a non-admin user!

`/flagged` is the list of stories you've flagged.

That makes sense, thank you.

/flagged, /upvoted, and /favorites, with an "?id=<your_userID>" are how your own flagged, upvoted, and favourited posts/comments are displayed. "&kind=comments" specifically returns flagged comments, otherwise posts. For upvotes and faves, the argument is "&comments=t", for consistency one presumes...

Favourites can be displayed for other users, upvoted and flagged cannot (for mere mortals).

I suspect downvotes can also be displayed, though I don't know that syntax.

I recently found out that "/vouched?id=<your_userID>" also works.

The replies/by is particularly handy in sorting out why a dead account has been banned, or if there've been previous moderator interactions. That's my own principle use.

Search term exclusion: TIL! I swear I've tried that w/o luck before. Possibly was confounding that with OR pairing, usually given as "(termA|termB)". I'm pretty sure that doesn't work as expected (and just tested it). It's also annoyingly absent from DDG search, which is out of scope for here.

I'd been made aware of the undocumented endpoint but didn't want to spill the beans ;-) It's apparently expensive. OK to run manually, but don't script it.

Oh yes, OR clauses would be great.

Maybe Trieve can give us propositional logic over search terms!

Can do!

Is it possible to add the alternative ("OR") operator? https://github.com/algolia/hn-search/issues/169

>Replies to a specific user

Will definitely make sure to add that to our search before launch! That is a really good idea.

>Improved date-bounding interface

Note taken :)

We also have negated terms which will work the same way.

While you're taking notes ...

Comments to a posts from a specific domain might also be useful. HN exposes that in its story syntax and I'm pretty sure that's in the API as well.

Yep, already did that one! Making it performant was a pain, but it should be useful.

Other feature requests/ideas would definitely be appreciated.

The one other: a URL-based selector for comments vs. posts. That's one thing Algolia seems to be missing. It defaults to post search, I'm usually looking for comments. That's one additional step, which isn't fatal but is vaguely annoying.

I think that's my main gripes / wants.

And I'm a heavy user of Algolia and other search tools, so I think that's a reasonably complete set.

I would like to sort comments by the level of the author’s expertise in whatever they are discussing. HN is a goldmine, but finding valuable knowledge within heated or elaborate discussions requires too much commitment to read through everything.

A weighted number of a comment’s upvotes is one signal. However, I can often tell when an author has deep knowledge or comprehensive experience with a subject just by reading their comment.

Do you think it might be possible to automate that kind of judgment?

I would really love that to be possible. It is ultimately, I suspect, one of the Hard Problems of epistemology / epistemic systems.

Diverging slightly: truth is not a popularity contest. The "wisdom of crowds" concept argues that crowds are, on average, more intelligent than individuals, even expert individuals. In practice ... crowds are subject to their own biases and failures. While uninformed (or lightly-informed) opinion may be better than no opinion, expert opinion tends to be superior to both ... though of course it is also subject to biases (co-option of motives, ideological and academic conservativism, etc.). Still, there are times when the popular winner is quite evidently not the most informative or relevant winner. Reddit is especially subject to this (and more so in the past couple of years than previously based on my very rare sojourns there).

Ultimately the question of a rating / moderation / ranking system is what do you want to optimise for? I'd written on this about a decade back now:

<https://web.archive.org/web/20200629055317/https://www.reddi...>

LLM AI seems like it might offer either a way of weighting individual votes in their appropriate areas of expertise, or offering its own assessment of relevance based on specific criteria (say: truth valance, significance, novelty). I still suspect it's not the sort of thing that's easily obtained. And is probably beyond the scope of an HN search tool.

But I love the suggestion.

And so long as we're all divulging secrets here ...

I've hacked the HN CSS to my own liking, links in my profile. Most of that's styling and such.

What's not included there is something I find useful: some visual tweaks to not specific contexts (users/sites) of interest.

As examples, it might be handy to recognise admin comments and posts immediately. Or YC hiring notices. Or people or sites you find particularly clueful. Or perhaps not.

I've found it useful, and a little classification goes a long way (long tails, Zipf functions, etc., etc.).

As long as people are throwing suggestions your way, it might cool if we could sort by either Votes or Comments rather than just "popularity" which is seemingly just sorting by votes. A lot of posts will only have 30-50 votes but over a hundred comments which makes them hard to find even by searching.

Reply on news.ycombinator.com