Live data from Hacker News

How is search so bad? A case study

svilentodorov.xyz

71–80 of 416 posts

Re: How is search so bad? A case study

#71

Google has definitely stopped being able to find the things I need. Pasting stack traces and error messages. Needle in a haystack phrases from an article or book. None of it works anymore. Does this mean they are ripe for disruption or has search gotten harder?

> Pasting stack traces and error messages.

I cannot fathom the number of times I've pasted an error message enclosed by quotes and got garbage results, and then an hour of troubleshooting and searching later I come across a Github/bugtracker issue, which was nowhere in the search results, were the exact error message appears verbatim.

The garbage results are generally completely unrelated stuff (a lot of Windows forum posts) or pages were a few of the words, or similar words, appear. Despite the search query being a fixed string not only does Google fail to find a verbatim instance of it, but instead of admitting this, they return nonsense results.

> Needle in a haystack phrases from an article or book.

I can confirm this part as well, searching for a very specific phrase will generally find anything but the article in question, despite it being in the search index.

Zero Recall Search.

Re: How is search so bad? A case study

#72
It's a hard problem because what is relevant is inherently subjective and context specific and only a minority of users uses the advanced search functionality so it is also not a big priority to solve it. Both Google and Duck Duck Go optimize for the simple use case where there's a bit of user context and some short query that the user typed. That's what needs to work well. For that Google is still pretty good. I try duck duck go once in a while but it's just not good enough for me right now. And of course when Google fails me, that's probably also a hard case for Duck Duck Go.

The other problem is that websites provide very inconsistent meta-data, and worse, are actively trying to game the system by abusing that metadata. So, things like timestamps are not standardized at all (well, a little bit via things like microformats). So recency of data is important as one of many relevance signals but not necessarily super accurate. And given that it's a relevance signal, you have people doing SEO trying to game that as well.

Anyway, Hacker News could also do with some search improvements to its ranking. It always pulls up some ancient article as the most relevant thing as opposed as the article from last week that I remembered and wanted to find back. I consult people on building search engines with Elasticsearch, so I have some idea what I'm talking about. It seems the ranking is basically "sort by points". Probably not that hard to fix that with some additional ranking signals. I just searched for "search" expecting to find this article near the top 5 (because it is recent and has search in the title). Nope; not a thing.

Re: How is search so bad? A case study

#73

You're located in Berlin. It found a page on Reddit about buying phones in Berlin. Not saying you're wrong about the dates but... I dunno... seems like an odd query. "Phone to buy?" And why not just search Reddit? This site feels like we're just complaining about nothing these days. (Downvote away!)

I'm downvoting you because your response is totally useless.

He wasn't looking for where to buy a phone in Berlin, and besides, it's an old reddit thread.

Re: How is search so bad? A case study

#74
post #55

Earlier quoted context omitted.

But those pretend complaints aren't his complaints. His complaint is "why does this archived reddit page from six years ago without any updates come up on search results for 'things within the past month'?" Which is... reasonable.

It is reasonable. It is also likely that whatever meta information reddit is sending back (in headers or tags) is probably not dated correctly for the time of the origin post. Google COULD offer more time machine features and perform diffing on pages. But a reddit "page" will always have content changes, as everything is generated from a database and kept fresh on the page. The ONLY metric therefore Google could use…

Seems like an easy solution to this problem would use two functions.

One function that takes the output of the page, and renders it so only what's user visible, actually gets indexed. So no headers, no JSON data, no nothing, unless it's actually in the final outcome of the page when rendered. This would require jsdom or some other DOM implementation. Hardly hard for Google (Chrome) to achieve this, and been done multiple times.

Second function is a function that does the same call twice, passing the page to function one each time, then compare them two. If you make two calls right next to each other, and some data is different, you discard that from your search index. Instead you only index data that appears in both calls.

Now you don't have the issue of "dynamic content" anymore...

Re: How is search so bad? A case study

#75

Google has definitely stopped being able to find the things I need. Pasting stack traces and error messages. Needle in a haystack phrases from an article or book. None of it works anymore. Does this mean they are ripe for disruption or has search gotten harder?

Perhaps expectations have risen over time

Re: How is search so bad? A case study

#76

Google has definitely stopped being able to find the things I need. Pasting stack traces and error messages. Needle in a haystack phrases from an article or book. None of it works anymore. Does this mean they are ripe for disruption or has search gotten harder?

I've been using DDG as a good enough search engine for most things, but when I sometimes fall back to Google, it blows me away how many ads are on the page pretending to be results!

Same here, I actually prefer DDG to Google now, even for regional (Germany) results.

When I switched, about a year and a half ago, I felt like I was switching to a lesser quality search engine (it was an ethical choice and done because I can), that, however, gradually and constantly got better, whereas Google went the opposite path.

Nowdays I only really use Google to leech bandwidth off their maps services. Despite there being a very good alternative available, OpenStreetMaps, they unfortunately appear to have limited (or at least, way less than Google) bandwidth at their disposal... A pity though, because their maps are so awesome, the bicycle map layer with elevation lines is any boy scout's wet dream... but yeah, to find the next hairdresser, Google'll do.

Speaking of bandwidth and OSM reminds me, is there an "SETI-but-for-bandwidth-not-CPU-cycles" kind of thing one could help out with? Like a torrent for map data?

EDIT: Maybe their bandwidth problems are also more the result of a different philosophy about these things. OSM is likely "Download your own offline copy, save everybody's bandwidth and resources" (highly recommended for smartphones, especially in bandwidth-poor Germany) whereas Google is "I don't care about bandwith, your data is worth it".

Re: How is search so bad? A case study

#78
post #34

Earlier quoted context omitted.

It wouldn’t matter. 0.0001% of content authors would employ semantic markup. Everyone else would continue to serve up puréed tag soup.

If WordPress outputs semantic output that instantly gives you a lot more than 0.0001%. The rest would follow as soon as it improves discoverability of their content

Wordpress can't magically infer semantic meaning from user input any better than Google can. The whole point of the semantic web is to have humans specifically mark their intention. A better UI for semantic tagging would help for that, but it would still be reliant on the user clicking the right buttons rather than just using whichever thing results in the correct visual appearance.

Re: How is search so bad? A case study

#79

Earlier quoted context omitted.

Boutique hand crafted artisanal catalogs? Not joking I have a feeling subject specific topics will be further distributed based on expertise & trust.

That's exactly what github's Awesome lists are: Decentralized, democratized handcrafted subject-specific catalogs

If they became important sources of information outside technically competent people I suppose we would end up with a bunch of Awesome lists of Content Farms!

Re: How is search so bad? A case study

#80
post #7

Google has definitely stopped being able to find the things I need. Pasting stack traces and error messages. Needle in a haystack phrases from an article or book. None of it works anymore. Does this mean they are ripe for disruption or has search gotten harder?

My guess is that suppressing spammy pages got too hard. So they applied some kind of big hammer that has a high false positive rate. You're getting the best of what's left. Maybe also some quality decline in their gradual shift to less hand weighted attributes and more ML.

Two anecdotes: It’s really fascinating.

1. My work got some attention at CES so I tried to find articles about it. Filtering for items that were from the last X days and searching for a product name found pages and pages of plagiarized content from our help center. Loading any one of the pages showed an OS appropriate fake “your system is compromised! Install this update” box.

What’s the game here? Is someone trying to suppress our legit pages, or piggybacking on the content, or is that just what happens now?

2. I was looking for some OpenCV stuff and found a blog walking through a tutorial - except my spidey sense kept going off because the write up simply didn’t make sense with the code. Looking a bit further I found that some guys really well written blog had been completely plagiarized and posted on some “code academy tutorial” sort of site - with no attribution. What have we come to?

Post reply on HN