Hacker News new | ask | show | jobs
by anigbrowl 1858 days ago
When I tell people I work on Google Search, I’m sometimes asked, "Is there any work left to be done?" The short answer is an emphatic “Yes!” There are countless challenges we're trying to solve so Google Search works better for you.

Sorry to be off-topic but it's hard to get excited about blue sky ventures when the search UI offers no capability for simple things like delivering search results in date order. You can filter results by date, but not sort them.

5 comments

I would bet that sorting isn’t so simple, at least if you want it to be any good. If you did a naive chronological sort, I imagine you would end up with a whole lot of irrelevant results at the top. There is just too much stuff out there.

To be useful, your “sort” would really just need to be another parameter to the existing relevancy model. And if you did that, then people would probably complain that “it’s not a real sort” and we’re back to square one.

Edit: You know what, this probably is simple for Google, because they’re freakin Google. To your point, I guess they probably don’t do this because money.

> I guess they probably don’t do this because money.

Exactly this. There are many controls they could have given us to trivially improve search for end users without needing this AI, but they would have made search less good for their customers, the advertisers.

It's really hard to design something like that that anyone would get value out of using because matching the query isn't a binary notion.
Another simple thing is that there's no way to not get localized results.

I'm currently in Spain. I'm not Spanish. If I want results that don't have to do with that country, and aren't in Spanish, I need to use Duckduckgo. Google is unable to not give localized results.

google.com/ncr or use a VPN
google.com/ncr doesn't work anymore. Both the interface and results are localized in Spanish when I try it.
Should you always be able to sort results by date? If I search for "California", doesn't really ever make sense to date-sort all the pages that match?
Is a good way to find things you had seen in the past, but don't recall exact date.
The range filter exists for that.

Sorting by date wouldn't work for that. You'll have a ridiculous number of pages of oldest search results (or newest, depending on sort order).

It depends what you're searching for. And searching within a range yielding lots of unsorted results is often unhelpful too. Of course, you can get around this with the API, but that is a lot of extra work for a student or non-computational researcher if they don't happen to already have those skills.
Within the range, the results are sorted by relevance.
Yes, that's what I'm complaining about. Pagerank's idea of relevance is often not what I need.
But if it is sorted by range, you can start binary searching on events.

I do this quite often with photos.

Sure it does. I might want things written about California in the past week, or in October last year, or in 2005.
You can already do that on Google -- press "Tools", then use the dropdown to change "Any Time" to "Past Week", or use the Custom Range option to select whatever time period you want.

The question is whether you should also be able to do a strict order by date, without regard to ranking. I'm not sure if you'd ever really want that.

---

[1] California in the past week: https://www.google.com/search?q=california&tbs=qdr:w

[2] California in October of last year: https://www.google.com/search?q=california&tbs=cdr%3A1%2Ccd_...

Yes, I use the filtering tool all the time. As a use case, consider where you're trying to find information about an incident or criminal case, where you can specify your search quite narrowly. But although news reports generally have the publication date extracted (and highlighted) as metadata by Google, There's no way to sort on this. It's a huge pain in the ass from a research perspective, because reports of crimes and arrests usually get far less coverage than reports of sentences or appeals, so the top results tend to more sensational but less informative.
s/less/more, sorry for the confusion.
The way search works:

1) terms are split into tokens 2) tokens are looked up to find documents 3) documents are ranked by scoring functions

I suspect sorting by chronological order might require too many document metadatas to be retrieved at step 2. (A lot of filtering occurs between steps 2 and 3.)

A simple explanation is product-wise this feature request doesn't make sense. The number of users will be using the said feature is not worth the effort the amount of effort needed to implement, maintain and operationalize in a product of literally billion users.
I have the same feeling every time Amazon announces some amazing thing and their Top Rated sort still isn't weighted by review count.