Detecting 'stealth' web-crawlers

Question

What options are there to detect web-crawlers that do not want to be detected?

(I know that listing detection techniques will allow the smart stealth-crawler programmer to make a better spider, but I do not think that we will ever be able to block smart stealth-crawlers anyway, only the ones that make mistakes.)

I'm not talking about the nice crawlers such as googlebot and Yahoo! Slurp. I consider a bot nice if it:

1. identifies itself as a bot in the user agent string
2. reads robots.txt (and obeys it)

I'm talking about the bad crawlers, hiding behind common user agents, using my bandwidth and never giving me anything in return.

There are some trapdoors that can be constructed updated list (thanks Chris, gs):

1. Adding a directory only listed (marked as disallow) in the robots.txt,
2. Adding invisible links (possibly marked as rel="nofollow"?),
   • style="display: none;" on link or parent container
   • placed underneath another element with higher z-index
3. detect who doesn't understand CaPiTaLiSaTioN,
4. detect who tries to post replies but always fail the Captcha.
5. detect GET requests to POST-only resources
6. detect interval between requests
7. detect order of pages requested
8. detect who (consistently) requests https resources over http
9. detect who does not request image file (this in combination with a list of user-agents of known image capable browsers works surprisingly nice)

Some traps would be triggered by both 'good' and 'bad' bots. you could combine those with a whitelist:

1. It trigger a trap
2. It request robots.txt?
3. It doest not trigger another trap because it obeyed robots.txt

One other important thing here is:
Please consider blind people using a screen readers: give people a way to contact you, or solve a (non-image) Captcha to continue browsing.

What methods are there to automatically detect the web crawlers trying to mask themselves as normal human visitors.

Update
The question is not: How do I catch every crawler. The question is: How can I maximize the chance of detecting a crawler.

Some spiders are really good, and actually parse and understand html, xhtml, css javascript, VB script etc...
I have no illusions: I won't be able to beat them.

You would however be surprised how stupid some crawlers are. With the best example of stupidity (in my opinion) being: cast all URLs to lower case before requesting them.

And then there is a whole bunch of crawlers that are just 'not good enough' to avoid the various trapdoors.

Answer 1

One thing you didn't list, that are used commonly to detect bad crawlers.

Hit speed, good web crawlers will break their hits up so they don't deluge a site with requests. Bad ones will do one of three things:

1. hit sequential links one after the other
2. hit sequential links in some paralell sequence (2 or more at a time.)
3. hit sequential links at a fixed interval

Also, some offline browsing programs will slurp up a number of pages, I'm not sure what kind of threshold you'd want to use, to start blocking by IP address.

This method will also catch mirroring programs like fmirror or wget.

If the bot randomizes the time interval, you could check to see if the links are traversed in a sequential or depth-first manner, or you can see if the bot is traversing a huge amount of text (as in words to read) in a too-short period of time. Some sites limit the number of requests per hour, also.

Actually, I heard an idea somewhere, I don't remember where, that if a user gets too much data, in terms of kilobytes, they can be presented with a captcha asking them to prove they aren't a bot. I've never seen that implemented though.

Update on Hiding Links

As far as hiding links goes, you can put a div under another, with CSS (placing it first in the draw order) and possibly setting the z-order. A bot could not ignore that, without parsing all your javascript to see if it is a menu. To some extent, links inside invisible DIV elements also can't be ignored without the bot parsing all the javascript.

Taking that idea to completion, uncalled javascript which could potentially show the hidden elements would possilby fool a subset of javascript parsing bots. And, it is not a lot of work to implement.

Answer 2

An easy solution is to create a link and make it invisible

<a href="iamabot.script" style="display:none;">Don't click me!</a>

Of course you should expect that some people who look at the source code follow that link just to see where it leads. But you could present those users with a captcha...

Valid crawlers would, of course, also follow the link. But you should not implement a rel=nofollow, but look for the sign of a valid crawler. (like the user agent)

Answer 3

It's not actually that easy to keep up with the good user agent strings. Browser versions come and go. Making a statistic about user agent strings by different behaviors can reveal interesting things.

I don't know how far this could be automated, but at least it is one differentiating thing.

Answer 4

See Project Honeypot - they're setting up bot traps on large scale (and have DNSRBL with their IPs).

Use tricky URLs and HTML:

<a href="//example.com/"> = http://example.com/ on http pages.
<a href="page&amp;&#x23;hash"> = page& + #hash

In HTML you can use plenty of tricks with comments, CDATA elements, entities, etc:

<a href="foo<!--bar-->"> (comment should not be removed)
<script>var haha = '<a href="bot">'</script>
<script>// <!-- </script> <!--><a href="bot"> <!-->

Answer 5

A while back, I worked with a smallish hosting company to help them implement a solution to this. The system I developed examined web server logs for excessive activity from any given IP address and issued firewall rules to block offenders. It included whitelists of IP addresses/ranges based on http://www.iplists.com/, which were then updated automatically as needed by checking claimed user-agent strings and, if the client claimed to be a legitimate spider but not on the whitelist, it performed DNS/reverse-DNS lookups to verify that the source IP address corresponds to the claimed owner of the bot. As a failsafe, these actions were reported to the admin by email, along with links to black/whitelist the address in case of an incorrect assessment.

I haven't talked to that client in 6 months or so, but, last I heard, the system was performing quite effectively.

Side point: If you're thinking about doing a similar detection system based on hit-rate-limiting, be sure to use at least one-minute (and preferably at least five-minute) totals. I see a lot of people talking about these kinds of schemes who want to block anyone who tops 5-10 hits in a second, which may generate false positives on image-heavy pages (unless images are excluded from the tally) and will generate false positives when someone like me finds an interesting site that he wants to read all of, so he opens up all the links in tabs to load in the background while he reads the first one.

Answer 6

One simple bot detection method I've heard of for forms is the hidden input technique. If you are trying to secure a form put a input in the form with an id that looks completely legit. Then use css in an external file to hide it. Or if you are really paranoid, setup something like jquery to hide the input box on page load. If you do this right I imagine it would be very hard for a bot to figure out. You know those bots have it in there nature to fill out everything on a page especially if you give your hidden input an id of something like id="fname", etc.

Answer 7

There is an API available at http://www.atlbl.com/en/webcrawlers.html that provides detection/identification of stealth webcrawlers and other bad bots.

Answer 8

Untested, but here is a nice list of user-agents you could make a regular expression out of. Could get you most of the way there:

ADSARobot|ah-ha|almaden|aktuelles|Anarchie|amzn_assoc|ASPSeek|ASSORT|ATHENS|Atomz|attach|attache|autoemailspider|BackWeb|Bandit|BatchFTP|bdfetch|big.brother|BlackWidow|bmclient|Boston\ Project|BravoBrian\ SpiderEngine\ MarcoPolo|Bot\ mailto:[email protected]|Buddy|Bullseye|bumblebee|capture|CherryPicker|ChinaClaw|CICC|clipping|Collector|Copier|Crescent|Crescent\ Internet\ ToolPak|Custo|cyberalert|DA$|Deweb|diagem|Digger|Digimarc|DIIbot|DISCo|DISCo\ Pump|DISCoFinder|Download\ Demon|Download\ Wonder|Downloader|Drip|DSurf15a|DTS.Agent|EasyDL|eCatch|ecollector|efp@gmx\.net|Email\ Extractor|EirGrabber|email|EmailCollector|EmailSiphon|EmailWolf|Express\ WebPictures|ExtractorPro|EyeNetIE|FavOrg|fastlwspider|Favorites\ Sweeper|Fetch|FEZhead|FileHound|FlashGet\ WebWasher|FlickBot|fluffy|FrontPage|GalaxyBot|Generic|Getleft|GetRight|GetSmart|GetWeb!|GetWebPage|gigabaz|Girafabot|Go\!Zilla|Go!Zilla|Go-Ahead-Got-It|GornKer|gotit|Grabber|GrabNet|Grafula|Green\ Research|grub-client|Harvest|hhjhj@yahoo|hloader|HMView|HomePageSearch|http\ generic|HTTrack|httpdown|httrack|ia_archiver|IBM_Planetwide|Image\ Stripper|Image\ Sucker|imagefetch|IncyWincy|Indy*Library|Indy\ Library|informant|Ingelin|InterGET|Internet\ Ninja|InternetLinkagent|Internet\ Ninja|InternetSeer\.com|Iria|Irvine|JBH*agent|JetCar|JOC|JOC\ Web\ Spider|JustView|KWebGet|Lachesis|larbin|LeechFTP|LexiBot|lftp|libwww|likse|Link|Link*Sleuth|LINKS\ ARoMATIZED|LinkWalker|LWP|lwp-trivial|Mag-Net|Magnet|Mac\ Finder|Mag-Net|Mass\ Downloader|MCspider|Memo|Microsoft.URL|MIDown\ tool|Mirror|Missigua\ Locator|Mister\ PiX|MMMtoCrawl\/UrlDispatcherLLL|^Mozilla$|Mozilla.*Indy|Mozilla.*NEWT|Mozilla*MSIECrawler|MS\ FrontPage*|MSFrontPage|MSIECrawler|MSProxy|multithreaddb|nationaldirectory|Navroad|NearSite|NetAnts|NetCarta|NetMechanic|netprospector|NetResearchServer|NetSpider|Net\ Vampire|NetZIP|NetZip\ Downloader|NetZippy|NEWT|NICErsPRO|Ninja|NPBot|Octopus|Offline\ Explorer|Offline\ Navigator|OpaL|Openfind|OpenTextSiteCrawler|OrangeBot|PageGrabber|Papa\ Foto|PackRat|pavuk|pcBrowser|PersonaPilot|Ping|PingALink|Pockey|Proxy|psbot|PSurf|puf|Pump|PushSite|QRVA|RealDownload|Reaper|Recorder|ReGet|replacer|RepoMonkey|Robozilla|Rover|RPT-HTTPClient|Rsync|Scooter|SearchExpress|searchhippo|searchterms\.it|Second\ Street\ Research|Seeker|Shai|Siphon|sitecheck|sitecheck.internetseer.com|SiteSnagger|SlySearch|SmartDownload|snagger|Snake|SpaceBison|Spegla|SpiderBot|sproose|SqWorm|Stripper|Sucker|SuperBot|SuperHTTP|Surfbot|SurfWalker|Szukacz|tAkeOut|tarspider|Teleport\ Pro|Templeton|TrueRobot|TV33_Mercator|UIowaCrawler|UtilMind|URLSpiderPro|URL_Spider_Pro|Vacuum|vagabondo|vayala|visibilitygap|VoidEYE|vspider|Web\ Downloader|w3mir|Web\ Data\ Extractor|Web\ Image\ Collector|Web\ Sucker|Wweb|WebAuto|WebBandit|web\.by\.mail|Webclipping|webcollage|webcollector|WebCopier|webcraft@bea|webdevil|webdownloader|Webdup|WebEMailExtrac|WebFetch|WebGo\ IS|WebHook|Webinator|WebLeacher|WEBMASTERS|WebMiner|WebMirror|webmole|WebReaper|WebSauger|Website|Website\ eXtractor|Website\ Quester|WebSnake|Webster|WebStripper|websucker|webvac|webwalk|webweasel|WebWhacker|WebZIP|Wget|Whacker|whizbang|WhosTalking|Widow|WISEbot|WWWOFFLE|x-Tractor|^Xaldon\ WebSpider|WUMPUS|Xenu|XGET|Zeus.*Webster|Zeus [NC]

Taken from: http://perishablepress.com/press/2007/10/15/ultimate-htaccess-blacklist-2-compressed-version/

Answer 9

I currently work for a company that scans web sites in order to classify them. We also check sites for malware.

In my experience the number one blockers of our web crawler (which of course uses a IE or Firefox UA and does not obey robots.txt. Duh.) are sites intentionally hosting malware. It's a pain because the site then falls back to a human who has to manually load the site, classify it and check it for malware.

I'm just saying, by blocking web crawlers you're putting yourself in some bad company.

Of course, if they are horribly rude and suck up tons of your bandwidth it's a different story because then you've got a good reason.