Hacker News Reader

I thought I was the only one going insane here. It's become legitimately difficult to discover blog posts/articles on programming topics that don't include the asinine "missing term" suffix. There are plenty of great blog posts/guides out there which DO include the term and are relevant, but are ignored for whatever reason. I end doing "xx yy mdn" with the search range to 1 year for most of my JS needs now. But god forbid you search for an object property (i.e. Object.assign) - you'll end up with 75% of results having "object", "assign", but not both.

... there is no "not" operator, which is desperately needed

Not true. It can be achieved by simply adding the

Operator to a search. For example, to get Google Search results about Tesla, minus all the stuff on the internet about the cars / motor company try...

  tesla -"tesla motors" -car

Can you give some examples of searches that don't work for you? I'm also a professional and an engineer and generally find a quick Google search to point me in the right direction every time.

Search for:

rsync version freebsd 9.2

First, note that the first four results (for me, anyway) do not contain the string "9.2" at all.

Second, note that google does not give you the "missing: 9.2" notifier on those in the results page.

Later results (for me, 6 and onward) do contain all of the terms. Google decided that search results that did NOT contain all of my terms were more "relevant" than ones that did.

Bonus laughs: changing the search to:

allinsite:rsync version freebsd 9.2

does change things up a bit, but results 1,3,4 and 5 still don't contain the string "9.2".

So, 5 exact matches and 5 related pages is considered poor result quality?

I have no idea why I would ever want "related pages".

I am not waiting in line with my girlfriends and searching for Britney Spears Songs. Related pages do nothing for me.

They are bad results.

Follow-up question: Are you logged in?

Google search will generally learn your preferences and get better at vending you the results you can actually use if you're logged in so it can build that search history for you.

I see what you mean. Yeah, write that blog post.

I think most of your points are wrong. Quoted strings always work for me (symbols are still stripped -- Google is a natural language search engine, not a code search engine), and there is a "not" operator, if I understand what you mean by that. From the article you linked:

When you use a dash before a word or site, it excludes sites with that info from your results. This is useful for words with multiple meanings, like Jaguar the car brand and jaguar the animal. Examples: jaguar speed -car or pandas -site:wikipedia.org

"Quoted strings always work for me (symbols are still stripped..."

That is the definition of not working. The point of the quotes is to allow searching for literal strings.

You have elucidated the entire problem. Unless, of course, you're just searching for movie star names, in which case the rubbery nature of the search results really doesn't matter.

It appears that -negative searching does work, which is amazing - that's nice to know. But it immediately reminds me of a terrible, glaring problem: searching for command line switches. You cannot effectively search for things like:

rsync --ignore-existing

... since they are interpreted as -negative keywords, even if you quote them.

Use quotes and search for rsync "--ignore-existing" or "rsync --ignore-existing". They are two different searches but both work fine and return useful results.

You can also remove the double dash and search for rsync ignore-existing

I appreciate the point of the posted article, but for the most part I don't actually have any trouble using Google as a programmer. You just need to format your searches correctly.

sure you can - just use "rsync ignore-existing" . Might be slightly inconvenient, but people who use command line switches should generally be able to figure this stuff out?

> I think most of your points are wrong.

I believe s/he makes a good case.

Google provides no concept of time. Best case conception of time seems to be either provided by the site directly, e.g. "article date" or something to this effect, or t0 = when google first learned about the page.

Top anser for: "nodejs" + "2016" mongo api

returns top hit: 2015, 2nd hit 2014.

and that I can't give it context myself:

Qouted strings can't possibly work. There is almost certainly an algorithm in the background to provide confidence that they could work before being run. Even with the best case optimization, how could they check every page in index for:

"does google respect qoutes?", I mean even if it finds pages with "google", "respect", "quotes" then runs query on all of them, how could it be fast enough for a human.

In above case, it is possible, however smaller phrases with common words can't ever be respected.

> Google is a natural language search engine

Eeryone has a google search profile. Allowing someone to calibrate theirs would provide them best results. Outline down thread, I explain why it would be nearly impossible for google to do this better than you.

Your appraisal of the not or "-" operator falls into the same problems in terms of time. I can't define my percentage of crawl index that is authoritative, have to have the applied "-" operator applied to a larger base than google and this may not be possible in limited time.

One idea I had is that google could actually take longer, and update itself on the rolling basis. So immediate best guess would come up, and then it would continue running in bg and rerank if you waited another 3 seconds for better results.

You can restrict results to the past day, week, month, year or custom range, using the search tools that appear underneath the search box.

> The worst part is, the strikethrough "missing: search term" identifier does not always appear, and you click through to a page that is useless without knowing it.

I always assumed that such cases mean that the webpage changed since Google crawler's last visit. Did you try ctrl-f'ing cached version of the webpage in such situations?