back

by softwaredoug·7y ago·view on hn ↗
I work on (non Google) search engines. This kind of thing has been happening for decades. Whether it’s synonyms, entity extraction, or embeddings to capture conceptual relationships, I’m surprised this is surprising... but then again I’m a search nerd :)
3 comments
On Google, at least a few years ago and back, when you put a query "X" you were guaranteed to see X in the HTML contents of the page. Or at least I can't remember getting a result that didn't satisfy that condition. Now it's common to not find X there.
Yes you’re right on exact matches.

For basic keyword searches. It’s not uncommon for search users to want inexact equivalencies, synonyms or conceptually close search results. (Search for cold, get back rhinovirus). Every user and use case has different definitions/tolerances of “this term appeared on the page”

It’s a complicated topic, without simple solutions. I wrote about it here.

https://opensourceconnections.com/blog/2018/12/07/synonyms-b...

If users want equivalencies then they shouldn't use quotes, which are for exact match. Let's not play semantics here. Search syntax is clear - put things in quotes, get 'exact match' in return. Equivalent words are not exact match. Get smart with me, decrease usability for me. This is what is happening.

We can argue whether there is place for more operators, or whether it's better on average for an average user to still be "outsmarted" by the engine, but it's clear what's going on - these aren't exact match results and the results often have zero relevance with what user is looking for precisely because the user knows the specific word WILL appear on the page they need and a page without that word WON'T be what they need. I don't know how much clearer we can be :) It used to work (or seemed to). Now it doesn't. It frustrates some people. That's it.

Yes you’re right. I missed the “” in your comment and failed at reading comprehension
But there's more to it anyway.

I've interpreted searches. Database searches, not web pages, but my problems should apply to web search too: For example, a user who types "märz 2019" with quotes might mean that string, in German, or might on the other hand mean that particular month and use the quotes to eliminate february 2019, march 2018, etc. And people enter the exact string they remember in the hope of avoiding a sea of mismatches, but then they either mistype or don't remember quite the right wording.

This has nothing with the quality of the search results. This belongs in the "did you mean" box at the top.
It's extremely common to become very frustrated and angry when you cannot get an exact match for the search terms you've entered after explicitly indicating to the search engine that you want an exact match and nothing else.

Don't mess with my search terms! Bring exact matches back! Only show pages that match the search terms!

Verbatim mode as default?

The amount of automatically created synonyms is getting out of hand, it decided the name of where I work is a synonym for another company in the same business that's 15 miles away

I never figured out what verbatim does. It certainly doesn't seem to force exact matches.
Verbatim mode does not resolve the issue.
From the customers point of view, those two things are pretty similar.
Verbatim mode has been broken more than once, too :-/
Is it possible that the times you're frustrated that verbatim search isn't working are the same times there are no decent results to your query anywhere on the internet?

It seems to me the main failing of Google is they have no good way to say "We understood your question, but the answer isn't on the internet".

Instead they just return a set of not very relevant results.

at least Google seems to be better at that than DDG
The texts other sites use to link to given site have affected the search results on Google for many many years.

This has been abused many times by using derogatory link texts when linking to politicians web pages.

See https://en.wikipedia.org/wiki/Google_bomb

I would so love access to Google database, but with Boolean search and some simple Regex / Grep type functionality. We can but dream.
The closest to this is to use BigQuery over HTTPArchive, it's not the Google index but a pretty solid base though and requires no setup at all.

https://github.com/HTTPArchive/legacy.httparchive.org/blob/m...

You can search for anything in the source-code of pages, using any regex or grep you can imagine.

Get a job at Google and then you can directly query the underlying data with as complex a query as you like (and don't have to use indexes etc).

Obviously running your custom filter logic across every byte of data that has ever existed on the internet is compute heavy... But Google has lots of that!

> Now it's common to not find X there.

Do you have an example? Exact match has been working fine for me.

Just guessing, but maybe there just aren't enough results featuring that exact match? Would you prefer to just see less results in such a case?
Yes, because then I know I need to broaden my query slightly. Feed me a bunch of junk I didn't ask for and my first instinct is to restrict my query more.
Absolutely. I want to see what I searched for, not what some algorithm thinks might be related. A lack of results is often useful information when searching for an error, for instance. It tells me whether I've run into a common situation or something that's likely related to my specific configuration or use case.
Can you provide an example of this? People always say this, but never provide the query.
This is an example from yesterday:

"azure" devops agent "checkout" "freeze"

I'm obviously searching for something very specific here, and Google helpfully decided to show me results omitting "freeze" and "checkout" so I had to quote them.

I also notice that it looks like Google thinks that check-up is the same thing as checkout, which in context are not synonyms, though this would require Google to infer "git checkout" but I didn't include git because that would introduce another universe of unhelpful results.

https://dzone.com/articles/getting-started-with-jenkins-the-... was included which doesn't include "azure" at all, though this is obviously Google being cheeky.

Of course I could also call this somewhat sinister, as Google is basically saying I should ditch a direct competitor to their services and instead use a self-hosted service. Or maybe I'm projecting since I would like to ditch Azure DevOps and use Jenkins, but in any case there's an example.

Feel free to play with it: https://www.google.com/search?rlz=1C1CHBF_enUS820US820&biw=1...

Of course, in this case it was a networking problem so I do have to admit that usually when Google starts ignoring what I typed it's because the answer to the question is something else entirely. Though sometimes I know exactly what I'm asking for and Google doesn't get it.

Wow THANK YOU! I have been asking for an example of this for literally years.

I think possibly what is happening here is that the index believes "azure" is on the page.

Consider the second result here: https://www.google.com/search?q=%22azure%22+%22Allows+adding...

(the one that starts java.dzone.com but leads to the same article).

It appears to have captured at index time a phrase "Microservices and Serverless on Azure" that is no longer in the artcile (possibly a link to another article).

Perhaps a Googler can explain exactly what is happening.

Can you provide an example of this? People always say this, but never provide the query.

Because anyone who's used Google in the last 10 years knows that Google tailors its results to what it thinks you want to see.

So a search query that fails for one HN reader may work for another, or at least work differently.

A repeated frustration in my life has been explaining to bosses that the search results that appear on their personal phone are not the same as what will appear on a client's computer.

It's pointless to provide examples, because Google will quickly "recall" anything that is being searched by multiple different people.

Unfortunately, it is no longer possible to use Google Search for searching in Internet. You are only allowed to look up viral memes and query neural networks, trained by other people's searches. Searching for rare and unique things will quickly get you banned. An interesting side-effect: using modern Google to check if name is vacant is meaningless. I used to google for names to see, if something else used them, but Google's search repeatedly returned me 0 results, even when there were multiple pages with that particular name in title, many of them years old. Searching for the same thing couple of weeks later have suddenly returned hundreds of hits.

I wonder, how much it costs to actually query Google's database instead of some distant neural network approximation. Apparently, Google's own employees can't afford it anymore — some of them are using DuckDuckGo instead (they even added it to Chrome, lol)

I... What? This is so disconnected from any experience I've ever had searching anything I don't even know how to approach it. What on Earth do viral memes have to do with anything? What does "banned" even mean in this context? There's no such thing as a ban from Google search.

And if I were to hazard a guess I think you dramatically overestimate the role of neural networks in search in general.

"banned" in this context probably means that idiotic popup about "unusual traffic detected from your machine" and if i would "pretty please run some anti-virus scanning software please".

This happens often when creating complex search terms with 'inurl' and 'intext' operators. This happens to me at least once a month.

Just curious, do you live in a western country? Do you have a Google account? I use funny queries like that often but I've seen that message maybe only once or twice before. There must be some additional factors involved.
Western country while logged in on my google account. I've given it some thought myself, I have that account logged in from a variety of systems: linux boxes and windows boxes some behind a vpn, some not.

I suppose it sometimes looks like my account is used for botting, but then again: why wouldn't I be allowed to connect my bot to their results, given that google uses bots to index the content in the first place?

Despite the fact that I could probably adjust to better fit their 'nothing weird going on here' preconceptions, I think I'd rather complain and hope that over time it's google that will change.

This is so disconnected from any experience I've ever had searching anything I don't even know how to approach it.

Then consider yourself lucky. The post was a little rambling, but it wasn't wrong.

> What does "banned" even mean in this context? There's no such thing as a ban from Google search.

You get warnings and captchas.

> You get warnings and captchas.

Frequently they are unsolvable captchas that just waste your time. :(

If you want exact then put it in quotes. Sometimes you may not know the exact language to use or proper language to ask and in those cases their current functionality is extremely useful.
The problem is, quotes don't work anymore. Even the verbatim mode doesn't restrict searches to exact matches. No matter what I've tried, Google still messes with my search results.
In fact, some of the recent google queries returned results that have nothing to do with the query! I had to go to - sorry - bing to find what I was looking for.
Not really true: Google bombing is almost as old as Google itself.
The article clearly explains the phone number is related to Uber, and the top post is people worried about Uber phone number scams. So why are they surprised the 3rd results is also about uber phone number scams?

I'd hazard a random guess that 90% of the time someone looks up a phone number for an unlisted number, they want to know if it's a scammer/spammer. So why the author seems to think this is the "wrong" result to show seems pretty strange to me.

Not wrong, just strange :)

And the algorithm in general appears to frustrate many people, if the comments here are anything to go by.

Any books, blogs or courses you recommend to learn about search?
Well yes a ton :) And these are the ones off the top of my head... others feel free to list ones I'm missing

Books:

Lucene in Action http://manning.com/books/lucene-in-action Solr in Action https://www.manning.com/books/solr-in-action Relevant Search http://manning.com/books/relevant-search (disclaimer, my book) Elasticsearch Definitive Guide (free/online) https://www.elastic.co/guide/en/elasticsearch/guide/current/... Introduction to Info. Retrieval https://nlp.stanford.edu/IR-book/information-retrieval-book.... Deep Learning for Search http://manning.com/books/deep-learning-for-search

Training:

Lucidworks Training https://lucidworks.com/resources/solr-training-and-consultin... Elastic's training https://training.elastic.co/ OpenSource Connection's training (disclaimer, my company's training) http://o19s.com/events/training

Conferences:

Haystack http://haystackconf.com (disclaimer I'm a co-organizer) Activate http://activate.conf SIGIR http://sigir.org ECIR http://ecir2019.org/ Search Solutions https://irsg.bcs.org/SearchSolutions/2018/ss2018tutorials.ph... Berlin Buzzwords https://berlinbuzzwords.de/ MICES E-Commerce Search - http://mices.co

Blogs:

OpenSource Connections blog - http://o19s.com/blog (disclaimer - my companys blog) Lucidworks Blog - https://lucidworks.com/blog/ Sematext Blog - https://sematext.com/blog/ Elastic's blog http://elastic.co/blog