Hacker News Reader

Comment by Cyberdog | Hacker News Reader

> Search is a good example of this, HN's search is basic keyword-based search, and I find it practically unusable because I don't always know the right keywords. A significantly smarter version, of the calibre that FAANG style companies could put out, would be great! Of course it would take more effort than it's worth for this user base to create it because investment like that only makes sense at scale.

Elasticsearch would be a good fit for this, and it's such a delight to use, once you understand how it works. Just tweak some parameters, and you get a fast search engine with immensely useful features like stemming, parameter ranking, and so on. It's one of those things that I love to use on client sites because it makes me look like more of a genius than I really am.

View on HN

> of the calibre that FAANG style companies could put out

Facebook search is abysmal

Amazon search is (now) abysmal (bloated with promoted shit)

Netflix search is... kinda ok? (but rarely use it)

Apple I'm not familiar with.

Google search is the best we've got, and these days is also pretty shit a lot of the time. (https://searchengineland.com/google-search-quality-crisis-27...)

Elasticsearch is absolute garbage for text search such as articles or comments. Text contains 'foocompany.com', and you search for 'company.com'? No results. It also tends to ignore punctuation (not sure whether this is configurable or not).

Please let me keep simple word search.

> Text contains 'foocompany.com', and you search for 'company.com'? No results.

Well… yes… I mean, you could probably make that causing a hit with some stemming if you wanted to, but that seems like reasonable behavior to me.

If you meant to give an example more along the lines of "singer" and "singing" not matching, again, that's a job for stemming. The feature is there; it just needs to be enabled.

> I mean, you could probably make that causing a hit with some stemming if you wanted to

A simple String.contains() gets you there, and that is the behaviour of grep, and most tools behaved that way too in the past (and still do, if they don't use ES).

> The feature is there; it just needs to be enabled.

It's not enabled in Discord (afaik uses ES), Slack (afaik uses ES?) nor Kibana (ES). What use is a feature if it's never used?

I'm sure there's plenty good things you can do with elasticsearch, but the things I use daily that use it are absolutely infuriating.

How it treats punctuation is indeed configurable, you can index data in multiple ways to allow the choice to be made at search time. It can also be contextually aware to distinguish for instance end-of-sentence dots from domain name or math expression dots.

Matching "foocompany.com" from "company.com" can be achieved with a wildcard or fuzzy search. If you specify an exact match then that's what you get.

Agreed here too... Unstructured data search tools leave too much for spammers to work with (especially when intense moderation and grooming of sources is not involved)...

The accuracy of results (In cases like HN search) is inherently flawed with unstructured data in mission critical settings unless there is an intentionally limited set of resources and contributors... Limiting contributors and resources also often defeats the purpose of maintaining a wide variety of accurate and diverse resources and results.

I was going to suggest that search should just redirect to DDG or Brave search with the !hn tag, but I tried it out and both of those just redirect to the same search page that HN uses (hn.algolia.com). That's both really nice because why not have your site-specific search results reference the search page that site uses and really annoying if the problem is that that page's search doesn't work very well.

It looks like redirecting the search to "site:news.ycombinator.com" does leverage their individual search results instead of going to algolia.com, so that's still an option.

Reply on news.ycombinator.com