Hacker News Reader

Comment by overshard | Hacker News Reader

I've been using DuckDuckGo for months now and over the past few weeks it's results have gotten considerably better. I wonder if this is because a massive amount of new users are consistently using it overall improving it's results?

Overall I think this is a great change though!

View on HN

I've been reading these comments about DDG for what feels like years now. Was it so terrible to begin with or is considerable an overstatement?

It can't be improving so quickly and still pretty shitty for such a long period of time.

I'd say it's improving for different categories of searches. Because for a long time users were largely technical, it's search for tech stuff has been pretty good for a long time (in general, there are a few sub-categories in that that have lagged[1]). Where, searching for an author or book might have terrible results. So, when these people say it has improved, they mean it has improved for the kind of searches that that person does. As more people use it, and suggest improvements[2], it will approach being good for all kinds of searches.

[1] It still isn't very good at searching for specific error messages. It definitely used to show obscure projects on GitHub way too often. You'd do a search for 'getty' and some 3-commit project with a name containing getty would be the first hit. Also, I think it's actually gotten worse at finding Rails API stuff.

[2] It's been a while since I tried, but if you tweet at them with a search term with a poor result, they will try to figure out how to tweak the algorithm to make it better. It's really cool to see them act so directly on your suggestions.

I don't think it was _terrible_.

It just doesn't give as accurate results as Google in some searches, I assume due to the rafts of information collected about users - a trade off I will make happily.

It's not even that they're less accurate, I think for a lot of searches we expect certain results.

If the search engine isn't giving me results I expect isn't that the same if not worse than it being inaccurate? If I type cucumber in google, it knows since I'm always search technical information I probably mean Cucumber BDD and it gives it to me as the second result right after the wikipedia article about the gourd, that's what I expect. DDG, however, having no clue who I am, gives me 8 results about the gourd before finally, below the fold mind you, giving me Cucumber BDD.

It's definitely uncomfortable, I don't know if it's better or worse but for me I couldn't adjust and found myself going through !g until I just switched back.

That's like searching for CakePHP by typing "cake", or for information on french braids by typing "french".

I don't expect my search engine to be staffed by psychics, or people who have gone through my emails and bookshelves. That's the problem, really.

But having a search engine staffed by psychics is almost undeniably more useful.

Certainly more creepy, and not a tradeoff many people would be willing to make, but incredibly useful.

Ah but Google does return me CakePHP if I search cake (and CoffeeScript for some reason, coffee and cake go together I guess I dunno). I don't agree with your example about french braids, at some point you are getting a little too generic.

You're fooling yourself. Google and Startpage return Cucumber BDD as a second result whether you're logged in or not. And in foreign countries such as France, it's even the first result.

Compare to Bing: in France, "cucumber" returns a porn(!) site as the first result (note: I don't visit porn sites or use bing).

DuckDuckGo and Bing simply aren't good enough, but it has nothing to do with whether they respect your privacy.

I wonder if there's an element of users learning how to optimize their query for Duck Duck Go. I've been using it as a default for some months now, but I often fall back to Google for software error message strings and other things.

Honest question: if DuckDuckGo doesn't track the users, how can it improve when more users are using it?

Not tracking the user's personal information doesn't mean you're not tracking the number of clicks, or calculate geoip stats.

How is that different from using Google over https without logging in?

Google can still track you, cookies are just one example. So DDG uses only the data which can not related to a single user, essentially "users who searched X clicked most often on the second link" while Google builds a profile, that is "the user with cookie value Y usually searches for computers."

You can dump your google cookies frequently or blacklist google.com from setting cookies.

IIRC DDG was vocally against personalization and "filter bubbling".

Even if they collected these stats, they wouldn't be able to refine their results unless they've changed their position.

Optimizing their search results globally based on user clicks is not personalization or filter bubbling.

That's assuming that search preferences are homogeneous, that everyone in the world uses DDG, or that the people currently providing click data are a representative sample of the world. The first and last things might be true for a few terms, but even then, the order of the search results is a very tricky thing (e.g. NYTimes or WSJ editorial page first? the difference in traffic between 1st and even 2nd or 3rd on the search results page is huge).

The only real benefit is that you and your neighbor are in the same filter bubble.

There are upper limits to global optimization. We all don't want the same results for a given query; to get beyond some particular bound, you can't treat all users the same.

In addition to aggregate usage stats, there's also the indirect (and slower) route: more users -> more ad impressions -> higher revenue -> more employees or other resources available to improve the service.

Working for a company that takes privacy very seriously, I can say that it can be a PITA. Worth doing, but still a PITA. We had to resort to some clever trickery just to count monthly active users without any sort of unique identifier. I bet DDG faces similar challenges.

Spot on, but worth it!

Out of curiosity, what solution did you hit upon?

Google uses a cookie to tie visits to a certain user for analytics.

The other solution I've come up with is to use a hash of IP + other identifying info, so not recording the IP or anytihng to tie it to a person, but still recording a relatively unique ID. What sort of techniques are you using?

Well, we have daily update requests (it's not a web app), so we knew the number of daily users already. But we really wanted to know the monthly users, gives a way better idea of the actual user numbers. What we ended up doing is to transmit the timestamp of the last update check with each update check, that way we can see which requests are the first over an arbitrary period.

Granted, the situation is probably not that similar to what DDG faces, but it was just one issue that came up. What we use for web analytics is based on awstats, which operates on access logs with anonymised IPs. Not as good as Google Analytics, but as useful as it gets without JS/tracking, I guess.

Thanks

I'm not sure if they are still as responsive on Twitter as they used to be, but if you tweeted at them with a specific search term that gave poor results, they would act directly on that. It was cool to see your own input make a difference.

After many months of zero Japanese support, DDG seems to now have support. It's been great!

Reply on news.ycombinator.com