HN Search Updated | Hacker News Reader

HN Search Updated

hn.algolia.io

88 pointsfjk58 comments

It has lost a lot of usability for searches I used to do. (And, alas, since many of my examples of searches were stored in Google Chrome as form entries for the previous interface, I'll have to think for a while about what my typical searches used to look like.) How does one do usability tests on changes like this?

AFTER EDIT: Thanks for the comment posted previously that points out that I can, for the moment, use HNSearch

https://www.hnsearch.com/

to find threads the way I'm used to finding them.

I think it's great that YCombinator is "eating its own dogfood" and the new site has some cool features, like instant search results.

I have a few concerns, though.

I think that there is too much whitespace, which makes content discovery take longer. An infinite scroll interface is not appropriate for HN. Content is not added as quickly as Twitter, and infinite scroll makes searching for the oldest comment or story (is that still possible?) a nightmare. Separating the username from the points/# of comments by placing it on the top right also makes scansion more difficult.

The previous search engine's syntax no longer works and the functionality it offered seems to not have been replaced (If that's not true, please let me know).

How do you:

Search for comments written by a certain author

Look at an individual comment's score

Search by date

Search exclusively for stories or comments

Thank you for the feedback, we'll improve it based on all of them in the next few days.

I'm legitimately terrified that you're going to replace HNSearch, and that therefore HNSearch may be shut down.

Using your version, I get the feeling you've never used HNSearch extensively, or at least not in the way a power user would. We at HN are power users. We need a minimum of glitter and a maximum of features. But there's not even a basic feature like by:tptacek.

I'm getting the vibe that you care more about testing your ideas than delivering features, otherwise there wouldn't be nonfeatures like image thumbnails. I would love to be wrong about that.

At a minimum, your search engine must be capable of most of what HNSearch was capable of. A simple test is: are you able to search for all of PG's downvoted comments? It should be possible to specify the search parameters: "comments only. Author: pg. sort by points. Sort ascending."

Where is the "order by time" functionality. This is pretty much useless now for searching old stuff you recall. I still use Google "site:news.ycombinator.com keyword" query to search on HN as many of those homemade search engines are not handling synonyms and plurals etc well.

The order by time and filtering to only stories/comments was the reason I used HNSearch before. The new interface takes away all of that flexibility and makes finding things a lot harder.

We will make changes in the next couple of days and we will ask feedback again. We prefer to launch sooner with few features and get community feedback rather than developing tons of features that we are the only to use :)

I mostly used the sort by date to find a recent (a few days, or last week) submission that I remember but it’s too deep to find in the normal page.

I'd found things from over a year ago with sort by date functionality (because of lack of good synonym matching and the fact that hn's url structure makes searching for comments less likely from an 'SEO' perspective)

The new engine is currently unusable because:

It does not clearly allow searching for stories, comments, or users.

It does not clearly allow sorting by relevance, date or score.

The screenshots are useless and take up valuable screen space, disrupting the flow of text and making it hard to scan. It is especially bad on a mobile phone.

The excerpts of text are too long, requiring you to scroll pages and pages when all you need are titles.

There is too much wasted blank space on the pages.

The search results look more like a daily digest you'd read in email, than search results you'd use to quickly find what you're looking for.

At least hnsearch.com still works, for now.

As someone who uses hnsearch literally every day, this makes me a little concerned.

I'm active in the Rust community, and my usecase involves searching for the keyword "Rust" and sorting by date in order to figure out where the current discussions are happening. Without the capability to quickly zero-in on recent comments (especially given the lack of any automatic notification of comment replies), HN will be much diminished for me.

Are there plans to both begin indexing comments and to add a sort-by-date feature?

What about: https://www.hnsearch.com/search#request/comments&q=Rust&sort... ?

That's what I'm using currently, but I have no idea how much longer hnsearch will be around for.

The next iteration is planned for tomorrow, we are taking into account the maximum of feature request possible.

We will ask for new feedback after :)

I always found HNSearch[1] to be excellent, but it's good to see a little competition.

[1]: https://www.hnsearch.com/

Thanks! It's great to hear you found HNSearch useful. We were very happy to power search for HN but we haven't been able to dedicate much time to making it better. It's exciting to see Algolia iterating on HN's search functionality.

If it ain't broke, don't fix it.

If anyone from HNSearch is listening:

You should default to HTTPS for all links to HN.

Right, working on it.

I see you felt the need to add a footnote when there was only one link[1].

[1]: Unnecessary

Fully unnecessary comment, knowing you otherwise understand.

I use hnsearch quite a bit as a general search/opinion tool. I use it less as a gateway to HN and more as filtering tool (wherein I expect readable, non-italic versions of the comments on the result page). For instance if I was curious if Don DeLillo ever popped up in HN comments.

Compare the results:

https://hn.algolia.io/?q=Don%20Delillo https://www.hnsearch.com/search#request/all&q=Don+Delillo

Algolia's manages to find a result that hnsearch doesn't, but on the other hand all of the white noise surrounding each result and false positives further down the page feels anti-utilitarian, and frankly is difficult to visually parse.

The HN comments are useful body of knowledge if you use it the right way. Having a fast, easy to read search that lets you scan for the best comments on a subject is really handy. Focus on the text and readability because that's all that's important.

I also use Pinboard and Quora (and to some degree Reddit) for this kind of opinion/commentary based research.

false positives further down the page feels anti-utilitarian

Those aren't false positives, those are additional helpful results in case you accidentally typed "Don Delillo" instead of "Delicious don't" or "dealloc done". I know I do that all the time!

Unhelpful sarcasm aside, it's nice that the results that contain the actual search are at least above the non-results. But I really don't like trend of search engines to include results that don't include any of the search terms above results that contain some of them. If you search for just "Delillo", you'll come up with a lot of results that are probably relevant to the user's intention.

Offering "If you didn't find what you were looking, maybe you meant:" alternatives could be helpful, but if not hidden (preferable), at the least these results should be clearly delineated from the actual matches. Google's approach of showing certain terms with a strike-through is the minimum I'd accept happily.

edit: Yuck, when I wrote this I hadn't even realized that it keeps loading non-relevant results in an endless scroll.

I miss the old version, as it made it possible to search stories/users/etc., and fit with the design of the site.

This is the first iteration, we will improve it with your feedback

How easy is it to add some of the filters that were in the old version? Usually I care about titles but sometimes I want to lookup an old insightful comment (as some comments are as length and as thoughtful as the OPs). The search here seems to be a little more simplified than what many users here (well, i.e. "hacker" types who don't mind configuring their own searches) would need

Yeah, it is just ui/ux ; the core search speed is awesome.

I m a search engineer among other things. New update is weird, especially ranking. try 'machine learning', gets something from a year ago then Stanford class from 2011, really ? If you re dealing with 10M entries in the index this is low enough that ranking should be better.

I know how perfect tuning these engine as is difficult, and this one requires some more work ;) good luck and congrats for YC 14 anyways!

For now, the ranking is only based on the number of points/comments of an item. We're actually working on a way to include the freshness of a post. Thanks for the support ;)

solr/ES would do that almost out of the box. how does your stuff differ or make it simpler (besides cloud) ?

Wow that was really unexpected!

- IMHO it does not fit in the spartanic style of HN (which I like!)

- there is no "search by date" which I use very often

Is there still an old version of search available?

I love how fast it is! Two things I rely on heavily with the old hnsearch are (a) restrict to stories or comments, and (b) sort by date (most recent first). Can you please do those? Especially the second one.

The API is available at https://hn.algolia.io/api

Unfortunately, it's not any more powerful than the previous HNSearch API (search query is limited to 1000 entries: http://hn.algolia.io/api/v1/search?query=pg&hitsPerPage=2000 )

Is this a real "update" or did you just create your own unofficial alternative? I thought the original was by the Octopart guys?

Yes this is a real "update", we explain details on our blog post: http://blog.algolia.com/hacker-news-search-algolia/

From that blog, it seems more like a new "alternative" to it rather than an "update" that replaces the existing one, but if it's purely a matter of semantics, I guess that's fine - just making sure I understand it right.

Congrats on getting into YC by the way! :)

Does this mean that Octopart provided search APIs will go away? If so, when?

The set of features is different with for example typo tolerance, if the limit of 1000 entries is a problem, we can increase it.

We have written a blog post about the change if you want more details: http://blog.algolia.com/hacker-news-search-algolia/

That post explains a lot. Thanks!

Is there a way to query on the created_at parameter?

No yet, we have only indexed text. We will collect feedback to improve it, create_at will be part of the next update.

Relevancy is all wrong. I searched for spark and the top result is from two years ago. I want the most recent result first, like email. Relevancy should be a function of tf/idf, comment count and recency, biased towards the latter.

But the instant search and screenshot is good.

Sadly, I don't like this much at all. I don't see any way to filter by whether the searched for keyword appears in the title or in comments, or to sort by date, etc. This appears to be much less useful for the kinds of searches I do.

I noticed its been updated so here are a few concerns still:

-I prefer the old date search compare to the new options that mimic how Google does it. I prefer the ability to sort by relevance vs date (in descending from most recent) the way HN search does it. Simply restricting to most recent by a certain timeframe doesn't let me expand beyond the period specified and currently it doesn't sort the date in order (at least not when choosing forever)

-Negative keyword parameter no longer available

-Title search parameter is also gone

-Not sure how the relevancy algorithm is calculated but for tests I've ran, it is showing some really old posts (2 years ago) at the top which makes no sense

can we have an option to use the old version as well? I use your search all the time and this new version doesn't work well enough yet.

I like to search by date (newest at top) and then toggle between stories and comments. also would be nice to allow stemming or not, i.e. search for "julia" comes up with stuff about juliaN assange.

This is really great, love the speed.

Where does the comment karma come from? Doesn't seem like you can get it by crawling HN directly.

Are you intercepting URLs typed in the address bar and reverting back to your own site or is Chrome being weird?

Yes I've noticed the same strange issues using Chrome. Working on it.

Hopefully one day the default HN search will understand the # in C# or F# or the . in .NET

Can anyone comment on why Swiftype (a YC search-as-a-service) startup isn't providing search for HN? It seems like an obvious choice (although the Octopart-powered search seemed to work great).

Some things that are missing include (as far as I can tell): karma totals for comments, sort by karma/date, syntax to limit search for {stories, comments}

Thanks for your feedback, we are collecting missing features and will improve it.

Bring back the HN Search that actually worked!

I've just deployed a new version based on your feedback. We'll continue to improve it. Thanks for your help.

nice ui and fast, but the search results could improve

Test: Search for a "story" about "Firefox" "today"

Result:

http://i.imgur.com/xQwc7WP.png

Yes, that's a consequence of our typo-tolerance handling. We're computing a damerau-levenshtein edit-distance and are currently allowing 2 typos if the word is at least composed by 7 chars. Regarding the corpus, we'll probably increase it to 8 or 9 (instead of 7 by default) to allow 2 typos and 4 (instead of 3 by default) to allow 1 typo.

love it! The speed is amazing. One minor nitpick would be to improve the experience on mobile. The search works great, but the display is hard to read. Really nice work though!

How do I sort by date or votes? Filter comments or posts?

That's fast

Reply on news.ycombinator.com