Hacker News new | ask | show | jobs
by faceshapeapp 2439 days ago
Congrats to the team!

Although, am I the only one who finds that searching HN through Google gives better results than searching it through the Algolia powered HN search?

For example, search "ml" in both, Algolia results are years old and don't seem that relevant, whereas Google picks up more recent threads.

10 comments

I have found their search interface for HN (https://hn.algolia.com) useful. They have a drop-down for selecting how you'd want to sort the results. From how I see it Google is doing more blackbox magic whereas Algolia is being a faithful/predictable search engine.
Probably because Algolia isn't using yet any machine learning, but more old-school (yet still very good) tries-based search.

And it's probably why also Algolia shines as search engine for a website vs searching the internet like Google. The former has a small scope (looking for videogames on twich, polo on lacoste) while the latter must be personalized for the user (the snake python vs the programming language python).

Sample size of 1, but yes, I also tend to have better results using `query site:news.ycombinator.com` than using the built-in Algolia search.
I do this sometimes as well when Angolia isn't giving me the results I'm looking for.
Not really. The sorting is based on popularity by default, which you can change. To compare:

https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...

https://www.google.com/search?q=news.ycombinator.com%3A+ml&o...

It feels ironic, I searched "Algolia" from both, today's news is ranked as #7 in Algolia powered HN. From Algolia: https://imgur.com/a/atRMnwj From Google: https://imgur.com/a/d3EKdUp

Some people says it's popularity based, but if I change it to date based, it's broken? https://imgur.com/a/HGzL6fO

Simply sorting by date doesn't really seem to be that useful either.

For example, the poster below showed the following link for the "ml" query: https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que....

If you look at the results, the first 2 results only matched part of the poster's username and many of the top results here are empty threads which are not that useful since HN is mostly for discussions.

To be clear, I'm not saying it's all bad, just pointing out that there are low hanging fruits which can improve results quite a bit.

Also, Algolia seems misspelled in your query which is why it fails when reordering by date: https://imgur.com/a/pO9fBWr, it works otherwise. Although, it says the date is 1 day old when it's just 2 hours old.

️ oops, a typo in the query, but why they have query correction for popularity based, not date based?
Yes, when sorting by date it's currently configured to disable typo-tolerance. This is to avoid having an old (but approximative, like with typos) result BEFORE the correct ones.
Sort by date works fine. Your last search has a typo `agolia`.
You can change the filters and ranking in the HN search page. I find it faster and easier than Google for comment content.
It's only as good as the HN team have configured the ranking and other algolia settings. Be nice for them to expose these.
Pretty sure HN's search is implemented by Algolia themselves. It's served from algolia.com.
Correct, and the source code is here: https://github.com/algolia/hn-search
You cant self host algolia
No, that wasn't my point. You can't host the service, but you can self-host a site that uses it behind the scenes.
I came to this discussion just to say something similar. Algolia doesn’t seem to be great for HN, and off late I see it not loading search results (maybe something to do with my network, browser and the site) or being quite slow compared to say, Google Search.
Another data point: Have encountered the same problem.
I prefer much more Algolia than Google for searching things in HN.

The customizable stuff are very helpful -- search by user, by type, order stories by popularity or date, etc.

It all depends on what you’re trying to find when you search. Google knows your background, Algolia doesn’t.

I too built a search engine:

https://demo.insideropinion.com/meta_profiles?utf8=%E2%9C%93...

I find it works 100x better for me, when I weight results by my expertise. Basically, returning results based on what I already am familiar with.

The results aren't necessarily better because of the Google knowledge of you, because they are pretty good results even in Private browsing without being logged in.

It just seems to me that HN search does a simple keyword matching and reordering based on points without any ranking (please someone correct if I'm wrong).

So freshness doesn't seem to have any value, but the reality is that, at least in tech, very few content is evergreen. Also, when you try to reorder by date for example, you'll end up with a lot of empty threads which again are not that useful. I'm not sure if there's a way of eliminating empty threads from search results.

FWIW, Private browsing doesn't prevent Google to tailor results to what it can guess about you: https://spreadprivacy.com/google-filter-bubble-study/
They rank by popularity by default (up votes), but I agree generally their API doesn’t seem to do ranking.
You compare a generic solution vs a highly tuned mission critical search engine.