> But it is boneheaded to serve up results that completely ignore one of or more search terms.
Then you've never worked on a search engine. There's a real bias in the comments here because the way technical people search for technical topics is dramatically different from the way the general audience uses a search engine. Most people use natural-language sentence fragments in search. Requiring exact match for every term in a query would vastly reduce usability.
No major search engine has worked that way since the 90s, I don't know why this is suddenly becoming a narrative on this site now.
>> I can abide synonymous substitutions, for instance if I search for "interesting articles" and the results include "fascinating", "intriguing", etc. But it is boneheaded to serve up results that completely ignore one of or more search terms.
> Most people use natural-language sentence fragments in search. Requiring exact match for every term in a query would vastly reduce usability.
He wasn't talking about that, and actually said he was fine with synonyms (and I would assume stemming, etc).
His problem was Google flat-out ignoring some of his search terms, thus giving irrelevant results. I don't have examples handy, but I'm under the impression that Google often ignores some of the more relevant terms in my queries to give me more popular results.
Which is ironic since most pre-Google search engines defaulted to OR'ing terms together, and one of Google's innovations was to switch to defaulting to AND. Now they seem to be revisiting the mistake of their competitors and wandering back to something like default-OR.
now I have to control F a term to visually ignore results without my term.
what is really amazing is how bad youtube search is. I can search for the exact name of a video, and it wont show up in the results. many times I have to use google search to search youtube
> dramatically different from the way the general audience uses a search engine
How many people have you actually talked to? Over the last few months I've had more than a few non-tech friends/family/etc complain to me about Google's broken search results, often asking if I knew of a way to fix it. The "missing: <query_term>" behavior is particularly enraging: I've seen multiple people yell at Google's search results variations of "F* you, Google! Stop telling me how you're ignoring half of my search!"
> Most people use natural-language sentence fragments in search.
Most people have a wide variety of behaviors that cannot (and shouldn't) be reduced into s single group. That said, a lot of people have learned to use Google in ways that work for them, even if it sometimes seems inefficient or unusual to those of use used to using query languages. Assuming you know what they "meant to ask" is often wrong and usually considered offensive. Priding tools people can choose to use is great; attempting to DWIM[1] - interpreting what the user intended instead of what they said - has been a terrible idea that ruins tools even since it was first attempted >50 years ago at Xerox PARC[2].
Yeah, Google has steadily become more and more infuriating to use to answer questions that aren't already answered on Stack Overflow. If they're going to use NLP, how about they use NLP to prioritize things that aren't exactly like the 10 or 15 results pages which give a 10 year old solution to my problem or which are written for an esoteric package and not the one I'm mentioning in my search results?
They're probably running web queries through the same or similar NLP engine as the Google Home and Android voice query to stress test it with additional input. That roughly coincides with the time frame that results have gotten worse.
> No major search engine has worked that way since the 90s, I don't know why this is suddenly becoming a narrative on this site now.
I understand the idea, but you are wrong or at least not completely right (I can't speak for all markets).
I know someone who learned to spell as a kid using Google, because if he didn't spell the words correctly he wouldn't get to see pictures of the moon or whatever he wanted to see.
Also after they learned to guess what people meant, Google used to be able to respect +, "" and also the verbatim option hidden beneath a button.
For a while they would also ask politely: did you mean x?
Since then it has become worse and worse: silently rewriting, fuzzing without asking etc.
No, really. Especially if Google is using search queries and clicks to update ML models - non technical people "pollute" the algorithms with poorly formed queries and overwhelming numbers of clicks on non technical articles.
There's probably enough of a niche now for a genuine technical search engine. Which treats keywords like Google used to, maybe with some regex thrown in and what not. None of this full question nonsense.
I personally believe that allowing people to search with full queries has had a negative effect on society - very little critical thought goes into which parts of your question are actually important, and searching is no longer a learnable skill; just ask a literal question and let Google do all the thinking for you.
It is an interesting viewpoint. One of the advantages of Twitter is that people can weight, or Twitter itself can weight who is posting and sharing that information. For instance, if Fred Wilson writes something about startups or VC, I generally find it interesting, if he writes about public markets, I largely ignore it. Originally, page rank probably solved quality of the source well enough, but as more interesting content is appearing in podcasts or niche, subscription blogs, does the approach need to be drastically different? What I mean is rather than Google's algorithm likely boosting the popularly clicked and shared post, "10 Things I learned as a VC Intern", should Google's ML algorithm weight that click based on the expertise of the search user or subject matter density, etc.? Like should Linus Torvalds clicking on a link be given far more importance on a technical/software related query than myself?
Or at least, rather than factoring in the expertise of the clicker, factor in the similarity of the clicker to the new searcher, so as to predict the new searcher's behaviour.
Having never worked on a search engine myself I will avoid guessing as to why Google's results are the way they are. As a search engine user I noticed Google's results seem to be less useful than they were before. After a particularly poor series of results I tried DuckDuckGo and found the results better. It seems similar to when I switched to Google from MetaCrawler. Whatever the reasons I just want to find what I'm looking for.
And then went on to ignore both that as well as the verbatim option and serve up stupid results despite our carefully crafted query strings.
I mentioned this a couple of weeks ago and I can repeat it: if they pay my tickets and a fair price I'll be happy to hold my course "how to continue being best by not nerfing your market leading product".
Personally I gave up last year and I'm now on DDG. Not perfect but less annoying.
I've gone over almost completely to DDG as well, for at least the last few years, but have noticed that DDG has been ignoring required terms more often lately. But they're still better than Alphagoog.
DDG has a feedback mechanism (the almost invisible "Send Feedback" thing in the lower right corner), but I'm unsure if anyone there actually sees the feedback and cares enough to do anything about it -- as I said, the problem is getting worse over time.
And while I'm on my soapbox, I'm going to continue complaining about having to use quote pairs instead of + for required terms. That still _really_ grinds my gears.
Sadly, I've been seeing similar behavior from DDG lately. In some cases it's been even worse than Google. I really wish there was a search engine optimized for technical work.
Having a list of stopwords is different from discarding half the query.
Yes, most search engines have ignored commonly-used words since at least the late 90s. But Google only started doing its idiotic "Missing: [key word from your search]" fairly recently. And it's a step back.
Although I personally don't think this is the case, it's something that's going to provide ammunition to the conspiracy theorists who'll swear they're just doing it to drive clicks to pages with Google ads.
Just today I had to ask Google Scholar for a lot of things since Google itself was completely useless. So it's not like it can't be done. Maybe that's the way going forward, one general search for grandma and specialized search engines for everyone else.
Then you've never worked on a search engine. There's a real bias in the comments here because the way technical people search for technical topics is dramatically different from the way the general audience uses a search engine. Most people use natural-language sentence fragments in search. Requiring exact match for every term in a query would vastly reduce usability.
No major search engine has worked that way since the 90s, I don't know why this is suddenly becoming a narrative on this site now.