Hacker News Reader

The question is: Did it ever happen to you when google actually had this exact combination in their index? For me, so far, getting these kinds of results is just synonymous to 'no match found', so I move on. In some edge cases it might be helpful to get these results, in most cases I just change my query. But if google were actually hiding good results from you this way, they'd be really bad at their job wouldn't they?

> Did it ever happen to you when google actually had this exact combination in their index?

Yup!

Try searching for "ruby net::ldap search", without the quotes. For me, a CPAN page is the second result. (This is even logged out, in a chrome incognito window, so no cookies are sent to google.) That page doesn't have "ruby" or anything related to "ruby" on it, at all.

But even if I was searching for something esoteric like an error message, it's much more helpful to give me no results than it is to give me irrelevant ones. Irrelevant results just waste my time clicking/skimming through them, where an empty result set will signal to me very clearly that what I'm searching for doesn't make sense.

FWIW, when I do the same in a chrome incognito window, all of the results are ruby (the language) related. But us seeing two different sets of results may make the problem worse, not better :)

I get GP's result when using DuckDuckGo with g! (which should be the same as incognito), so there is that. However I don't see what's the big deal with that CPAN result - it's not like the useful results are drowning in these links here and you can always use quoted if you don't want that. In fact I think it's sometimes useful information that there are very similarly named things than the one we've searched for.

I see cpan in 4th postion, but I'm not sure I see the problem?

Net::LDAP is a perl module too, one of the things it does is search, so 3/4 of the words are not just matched literally, but are the right kind of net (not fishing or stockings), the right kind of ldap , the right kind of search.

You claim there is nothing related to ruby on that page, yet how many times must people have linked to both those languages from a webpage, perhaps alongside other related terms like python?

Its not perfect, and it's not magic, but is that really the level we expect from Google, so much so that we're in a huff if we have to scroll past one easily discounted search result?

I expect that if I use a search term, Google won't ignore it. I used the word ruby in there because I specifically care about ruby-related search results, not perl ones. When google ignores one of my search terms in the name of giving me more results, it gives me less relevant results.

They had to put a lot of effort into a search system that can ignore words that you're searching for. I'm wondering why that's necessary when I can clearly get better search results when all of my search terms are included.

The problem with judging this is confirmation bias, though. I only notice it when google has infuriatingly dropped the key term in my query, but I search for things at least ten or twenty more times each day. Google may be doing the same thing there and I just don't notice, and it's actually helpful.

I wonder if it would help if they added some heuristic where, especially if it looks like a technical topic, it would try to find the word that was most domain-defining (in your example: "ruby") and never drop that one. I wonder how difficult that would be to solve. Even with a lot of false positives, the worst that you'd get is the results quality of not dropping any of the terms.

They do try search combinations with some words omitted, and right at the bottom of each query they tell you which part of your phrase they have excluded.

I also get this behaviour all the time, and I have no idea how it could improve the quality of my search results.

Yes, if I search for "Windows 7 NAT", dropping the NAT keyword returns more results. But none of them will give me the information I'm looking for.

Exactly. Searches with less results are nearly always better than searches with more... I don't know why Google would think it's helpful to include more, but clearly less relevant results. (By definition, too. If a result is lacking one of your search terms, it's less relevant to what you searched for, strictly speaking.)

This is especially true when what I'm searching for yields almost no results. That's how I typically know that what I'm searching for was probably bogus to begin with. But when the first page is full of a bunch of fuzzed-out results that are only tangentially related to what I searched for, I now have to skim through a page full of results to realize that Google just gave up and decided to omit several of my terms.

Right, and it told me that the CPAN pages were with "ruby" omitted. But why would I want those search results? Those are obviously not what I searched for.

And they weren't at the bottom, they were actually the first result on the list.

That's because being a generalist consumer-facing product, Google is a recall-oriented rather than precision-oriented, which means that it is in their best interest to never have their users in front of a blank page, even if it means having less relevant results in the front page.

> They do understand that people want less - but more relevant - results, instead of just having more useless ones

I would argue that since they have access to their data and they do not do anything without having data supporting the change, they would know better what people want. You and I may very well just not be in their main demographics.

Under their search tools menu they have a "verbatim" option. There's probably a bookmarklet out there somewhere to make it a one click option.

My post wasn't really about technical support, it's about the merits of Google's decision to fuzz your search terms in the name of better results. I'm aware of verbatim search and the fact that I can quote words to make them required for the search results. I'm saying that it's a recent phenomenon for Google to intentionally ignore words in my query in order to include more results. I'm not really sure why anyone at Google would ever think that to be a good idea.

I can guess that it's because most of their queries are closer to natural language ones and not to a specific term search. If you think that way it's clear why more related results can be useful. Though on a specific term search it is a disaster.

I believe (but have no evidence) that Google does this to redirect users towards searches which are more likely to serve ads. The more results a search yields --> the more likely it is someone is paying to target it. This likely applies more to general users and not HN users (turns out queries for error messages are hard-to-monetize).

> turns out queries for error messages are hard-to-monetize

Really? You would think it would be the perfect opportunity for competitors to advertise. Here's the user frustrated with their existing vendor because their computer is emitting some kind of gibberish instead of doing what they paid for it to do, and here you are offering a different product that presumably makes the error go away.

Double really! error msg == probably developer or tech person with lots of disposable income. Editors, debuggers, conferences, books, geekchandise, etc.

In corporate software, perhaps. I'm not going to ditch python-ldap just because an ad on a search page had… I don't even know who would advertise on that kind of page. Software consulting maybe? Outsourcing?

This totally makes sense unless you actually think about it. Then it's just shit.