Unlike Ruby and Python, Java and Scala don't expose the URI. We should expose it.

Ticket based on #48.

Believe the bad URI fed in was: http://search.tb.ask.com/search/GGmain.jhtml

@aposashenko could you provide more information (e.g. on error message) and exact URL used here?

referer-parser could take a cue from ua-parser by adding a YAML (or JSON) file with test cases consisting of example referer urls for each of the referers in search.yml and the corresponding parse results.

This would greatly increase the confidence level in the consistency of future ports to other programming languages.

Is there any reason why duckduckgo.com is labelled as DuckDuckGoL ?

See #18

1. Simplify API to just return medium, source, term
2. Add support for webmail, social, internal, unknown referers (see Java/Scala impl for details)

Came across this tonight. I am interested in the php to parse domain and search terms out of url logs. Problem is I cannot use composer with my current server (WHM/Cpanel setup)... I came across this situation once before, however, they also offered a phar file of everything bundled which I could use without problems and eliminated the need for composer.

Any chance of this happening?

Java should be built with Gradle, no Scala deps anywhere.

Basically, the v1 library was looking for exact matches even www.google.com vs google.com

That approach is actually flawed:

1. Social networks often let users have their own subdomain, and you can't obviously list all of them
2. Yahoo! puts its search engine on load-balanced subdomains, and that's an unpredictable list

So this version instead tries multiple lookups:

• It tries the host, then the host + path, then the host + one-level path
• Then strips off a subdomain and tries again

We should write some explicit tests into the Specs2 test suite to check that this recursion logic works correctly.

/cc @donspaulding

Composer (http://getcomposer.org/) is currently the fashionable way to manage PHP dependencies. Unfortunately, the composer.json needs to live in the root directory of a repository so that https://packagist.org/ allows publishing it. I see two ways:

• Put it in the root directory
• Publish a subtree split of php/ as referer-parser-php and publish that repository

@alexanderdean what do you think?

The main things I'm thinking here are:

1. URL shorteners - e.g. if referer is t.co, we know that this is a click from a Twitter client (mobile or web)
2. Webmail clients - e.g. if a referer is "36ohk6dgmcd1n-c.c.yom.mail.yahoo.net", we know they came from an email received to their Yahoo Mail

/cc @yalisassoon @kingo55, see also snowplow/snowplow#130

Usage information is incomplete vs other examples.

No need to backfill

Unfortunately the bump to httpclient 4.3.3 has broken referer-parser on Hadoop.

Specifically it's this line of code:

81b88ff#diff-729b6a9a457c5e4a0b244bf130a1e08eR192

This is the error on Hadoop:

Caused by: java.lang.NoSuchMethodError: org.apache.http.client.utils.URLEncodedUtils.parse(Ljava/lang/String;Ljava/nio/charset/Charset;)Ljava/util/List;
    at com.snowplowanalytics.refererparser.Parser.extractSearchTerm(Parser.java:205)
    at com.snowplowanalytics.refererparser.Parser.parse(Parser.java:154)
    at com.snowplowanalytics.refererparser.Parser.parse(Parser.java:116)
    at com.snowplowanalytics.refererparser.scala.Parser$.parse(Parser.scala:153)
    at com.snowplowanalytics.snowplow.enrich.common.enrichments.registry.RefererParserEnrichment.extractRefererDetails(RefererParserEnrichment.scala:107)
    at com.snowplowanalytics.snowplow.enrich.common.enrichments.EnrichmentManager$$anonfun$enrichEvent$2$$anonfun$apply$5.apply(EnrichmentManager.scala:291)
    at com.snowplowanalytics.snowplow.enrich.common.enrichments.EnrichmentManager$$anonfun$enrichEvent$2$$anonfun$apply$5.apply(EnrichmentManager.scala:277)
    at scala.Option.foreach(Option.scala:236)
    at com.snowplowanalytics.snowplow.enrich.common.enrichments.EnrichmentManager$$anonfun$enrichEvent$2.apply(EnrichmentManager.scala:277)
    at com.snowplowanalytics.snowplow.enrich.common.enrichments.EnrichmentManager$$anonfun$enrichEvent$2.apply(EnrichmentManager.scala:277)
    at scalaz.Validation$class.foreach(Validation.scala:126)
    at scalaz.Success.foreach(Validation.scala:329)
    at com.snowplowanalytics.snowplow.enrich.common.enrichments.EnrichmentManager$.enrichEvent(EnrichmentManager.scala:277)
    at com.snowplowanalytics.snowplow.enrich.hadoop.EtlJob$$anonfun$toCanonicalOutput$1$$anonfun$apply$2.apply(EtlJob.scala:70)
    at com.snowplowanalytics.snowplow.enrich.hadoop.EtlJob$$anonfun$toCanonicalOutput$1$$anonfun$apply$2.apply(EtlJob.scala:70)
    at scalaz.std.OptionFunctions$class.cata(Option.scala:157)
    at scalaz.std.option$.cata(Option.scala:209)
    at scalaz.syntax.std.OptionOps$class.cata(OptionOps.scala:9)
    at scalaz.syntax.std.ToOptionOps$$anon$1.cata(OptionOps.scala:103)
    at com.snowplowanalytics.snowplow.enrich.hadoop.EtlJob$$anonfun$toCanonicalOutput$1.apply(EtlJob.scala:70)
    at com.snowplowanalytics.snowplow.enrich.hadoop.EtlJob$$anonfun$toCanonicalOutput$1.apply(EtlJob.scala:70)
    at scalaz.Validation$class.flatMap(Validation.scala:141)
    at scalaz.Success.flatMap(Validation.scala:329)
    at com.snowplowanalytics.snowplow.enrich.hadoop.EtlJob$.toCanonicalOutput(EtlJob.scala:69)
    at com.snowplowanalytics.snowplow.enrich.hadoop.EtlJob$$anonfun$7.apply(EtlJob.scala:170)
    at com.snowplowanalytics.snowplow.enrich.hadoop.EtlJob$$anonfun$7.apply(EtlJob.scala:169)
    at com.twitter.scalding.MapFunction.operate(Operations.scala:58)
    at cascading.flow.stream.FunctionEachStage.receive(FunctionEachStage.java:99)
    ... 11 more

Hadoop bundles an old version of httpclient which doesn't have parse(String, Charset). There is talk of Hadoop removing that dependency but in any case we and EMR use an oldish version of Hadoop.

The problem is not using 4.3.3 per se, but using parse(String, Charset). /cc @squeed

Architecture is very OO and convoluted.

Move towards similar architecture to the Scala version: a Parser module which includes one-time instantiation, and then a static parse() method which returns a Referer object for a given URL.

@Tombar has moved it in the right direction with a public parse() method (Tombar@a280d9e).

I got the error:

from referer_parser import Referer
Traceback (most recent call last):
File "", line 1, in
File "C:\Python27\lib\site-packages\referer_parser__init__.py", line 32, in <
module>
REFERERS = load_referers(JSON_FILE)
File "C:\Python27\lib\site-packages\referer_parser__init__.py", line 21, in l
oad_referers
params = list(map(text_type.lower, config['parameters']))
TypeError: descriptor 'lower' requires a 'unicode' object but received a 'str'

i fix it temporaly modifing __init__.py, load_referers function:

if 'parameters' in config:
    p = config['parameters']
    if text_type == unicode:
        p = unicode(config['parameters'])
    params = list(map(text_type.lower, p))

Python 2.7, SO:Windows 7

Most search engines are UTF8 so a nice to have.

Some search engines operate load balancing etc on subdomains, leading to referers which can't be found in search.yml. For example, a Yahoo! refererer URL might be "us.yhs4.search.yahoo.com"

The most performant way of supporting this is probably this algo:

1. Lookup the full domain in search.yml. Found? Finish
2. Not found? Strip off first sub-portion ("us."). Lookup. Found? Finish
3. Not found? Strip off next sub-portion ("yhs4"). Lookup. Found? Finish
4. Continue till found or no parts left!

This should be a lot faster than switching to a regexp-based approach.

org.junit.ComparisonFailure: Internal subdomain HTTP medium
Expected :internal
Actual :unknown

at org.junit.Assert.assertEquals(Assert.java:125)
at com.snowplowanalytics.refererparser.ParserTest.refererTests(ParserTest.java:36)
at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
at org.junit.runners.model.FrameworkMethod$1.runReflectiveCall(FrameworkMethod.java:45)
at org.junit.internal.runners.model.ReflectiveCallable.run(ReflectiveCallable.java:15)
at org.junit.runners.model.FrameworkMethod.invokeExplosively(FrameworkMethod.java:42)
at org.junit.internal.runners.statements.InvokeMethod.evaluate(InvokeMethod.java:20)
at org.junit.internal.runners.statements.RunBefores.evaluate(RunBefores.java:28)
at org.junit.runners.ParentRunner.runLeaf(ParentRunner.java:263)
at org.junit.runners.BlockJUnit4ClassRunner.runChild(BlockJUnit4ClassRunner.java:68)
at org.junit.runners.BlockJUnit4ClassRunner.runChild(BlockJUnit4ClassRunner.java:47)
at org.junit.runners.ParentRunner$3.run(ParentRunner.java:231)
at org.junit.runners.ParentRunner$1.schedule(ParentRunner.java:60)
at org.junit.runners.ParentRunner.runChildren(ParentRunner.java:229)
at org.junit.runners.ParentRunner.access$000(ParentRunner.java:50)
at org.junit.runners.ParentRunner$2.evaluate(ParentRunner.java:222)
at org.junit.runners.ParentRunner.run(ParentRunner.java:300)
at org.junit.runner.JUnitCore.run(JUnitCore.java:157)
at com.intellij.junit4.JUnit4IdeaTestRunner.startRunnerWithArgs(JUnit4IdeaTestRunner.java:74)
at com.intellij.rt.execution.junit.JUnitStarter.prepareStreamsAndStart(JUnitStarter.java:211)
at com.intellij.rt.execution.junit.JUnitStarter.main(JUnitStarter.java:67)
at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
at com.intellij.rt.execution.application.AppMain.main(AppMain.java:134)

Looks like a way of avoiding having to specify all TLDs for a given search engine.

Also seems to work at the beginning of a domain too.

https://github.com/piwik/piwik/blob/master/core/DataFiles/SearchEngines.php#L906
https://github.com/piwik/piwik/blob/master/core/DataFiles/SearchEngines.php#L907

referer_tests.json now has two new tests for custom internal domains functionality:
https://github.com/snowplow/referer-parser/blob/feature/json-tests/resources/referer-tests.json#L235-L248

The Referer-Parser is configured with a list of domains which should be counted as internal:
https://github.com/snowplow/referer-parser/blob/feature/json-tests/java-scala/src/test/scala/com/snowplowanalytics/refererparser/scala/JsonParseTest.scala#L41

The PHP version of the Referer-Parser already has support for internal hosts, so it should be possible to get it working with the new tests.

When done, please update sync_data.py to automatically copy the master copy of referer-tests.json into the PHP subfolder: https://github.com/snowplow/referer-parser/blob/23e3fd9f3bfaa8947fcb456ed8fbdb22f271dabc/sync_data.py#L58

Basically, a Java URI can be invalid to the point that .getHost() and/or .getPath() fail - so we need to catch any errors here.

Example:

http://bigcommerce%20wordpress%20plugin/

See snowplow/snowplow#314 for further details.

, crossScalaVersions := Seq("2.9.2", "2.9.3", "2.10.0", "2.10.1")

And add into .travis.yml

Because we never released it to Maven.

Hi Guys,

It'd be really useful if you could update pip with 0.2.0 - currently it only has 0.1.0 available.

Thanks a lot!

E.g. extract product/banner ID from affiliate IDs

Got this rare referrer:
http://www.dogpile.com/info.dogpl/search/web?fcoid=417&fcop=topnav&fpid=27&q=HELP+speaker+dock+not+working+since+6.1.2+update+avforums&ql=

Despite I see this domain in installed 'referer_parser-0.1.0-py2.7.egg/referer_parser/data/search.json', the library does not detect it:

In [21]: r.known
Out[21]: False

Can you check with your updated copy? Cheers

Using Argonaut to process the tests.

A test to ensure that a google.com/product referer returns "Google Product Search" not "Google".

/cc @donspaulding as I wasn't sure that this was handled in the Python version.

The relevant line in the Ruby is:

https://github.com/snowplow/referer-parser/blob/master/ruby/lib/referer-parser/referers.rb#L28

The relevant line in the Java is:

https://github.com/snowplow/referer-parser/blob/master/java-scala/src/main/java/com/snowplowanalytics/refererparser/Parser.java#L78

See impl in #43

/cc @iblue

Hi, all, please see this referer:
http://www.google.com/search?hl=en&as_q=&as_epq=carbonite+offer+code&as_oq=&as_eq=&num=100&lr=lang_en&as_filetype=&ft=i&as_sitesearch=&as_qdr=all&as_rights=&as_occt=any&cr=&as_nlo=&as_nhi=&safe=images

I've added parameters as_q, as_epq, as_eq to my local referers.yml.
and run

irb(main):002:0> require 'referer-parser'
=> true
irb(main):003:0> ref = "http://www.google.com/search?hl=en&as_q=&as_epq=carbonite+offer+code&as_oq=&as_eq=&num=100&lr=lang_en&as_filetype=&ft=i&as_sitesearch=&as_qdr=all&as_rights=&as_occt=any&cr=&as_nlo=&as_nhi=&safe=images"
=> "http://www.google.com/search?hl=en&as_q=&as_epq=carbonite+offer+code&as_oq=&as_eq=&num=100&lr=lang_en&as_filetype=&ft=i&as_sitesearch=&as_qdr=all&as_rights=&as_occt=any&cr=&as_nlo=&as_nhi=&safe=images"
irb(main):004:0> st = RefererParser::Referer.new(ref, 'referers.yml')
=> #<RefererParser::Referer:0x25feb227 @search_term="", @known=true, @referer="Google", @uri=#<URI::HTTP:0x746231ed URL:http://www.google.com/search?hl=en&as_q=&as_epq=carbonite+offer+code&as_oq=&as_eq=&num=100&lr=lang_en&as_filetype=&ft=i&as_sitesearch=&as_qdr=all&as_rights=&as_occt=any&cr=&as_nlo=&as_nhi=&safe=images>, @search_parameter="as_q">
irb(main):005:0> st.search_term
=> ""

I think that "carbonite offer code" is the better result.
https://github.com/snowplow/referer-parser/blob/master/ruby/lib/referer-parser/referer.rb#L85
If the referer contains two or more parameters, I prefer to return searchterm that is not nil or empty rather than the one when parameters first match.

Hi guyz. Would you be so kind to update npm module :3. It's a bit outdated ('0.0.2': '2013-08-16T20:29:36.351Z')

Potentially because the &url= isn't being escaped? (Why isn't it being escaped?)

sa=t&rct=j&q=g+star&source=web&cd=3&ved=0CEEQFjAA&url=http://www.gstars.co.uk/?ito=GAG5362963510&itc=GAC19854885430&itkw=g-stars&itawnw=search&ei=8eMQUt_hAvTSpgLjqQE&usg=AFQjCNFFNpW7yF9pcqCfOpYvqafYS94p_Q

1. Simplify API to just return medium, source, term
2. Add support for webmail, social, internal, unknown referers (see Java/Scala impl for details)

Suggestion from Peter O'Neill based on the blog post A new method to track keyword ranking using Google Analytics.

On occasion, Google search exposes the position of the keyword that drove the click to your website in the page_referrer as a cd= parameter. We should extract this from the referrer_url so that it can be stored in SnowPlow, and used to track:

1. The average position of keywords over periods of time. (Is search engine rank for particular terms getting better or worse?)
2. Grouping keywords together around specific e.g. products and categories and creating an performance index for those buckets. (As per the blog post.)

In order to implement this, we'll need to:

• Extend the search engines YAML so it includes not just the query string parameter to identify the keywords, but also the parameter to identify the location of the search result. (Where this is available.)

Then in SnowPlow we'll need an addition mkt_xxx field to store the result e.g. mkt_rank.

Are there any interesting specific querystring fields to extract for social referrers?

Am seeing some strange Yahoo data showing up as "search".

A client ran a homepage takeover on Yahoo a few weeks back and sent a lot of traffic from the yahoo homepage from hostname "au.yahoo.com".

I know this isn't search traffic, so when I queried it in Snowplow, this hostname had no real keywords.

In contrast, the hostname "au.search.yahoo.com" had quite a few search terms.

Is this a case of not provided?

Can someone with time create a phar version or do a pull request of one using box? This is above my head, but I cannot use without having a phar... would be extremely grateful if anyone has the time.

http://box-project.org/

Hi guys,

When I try to parce referers from that site I get exception. I use ASP.Net library.

Can you add the following source to your library?

http://search.tb.ask.com/search/GGmain.jhtml

Two reasons for this:

1. JSON seems faster to load than YAML in most languages (see tobie/ua-parser#117)
2. More languages have JSON handling built-in than YAML (e.g. Python)

This would involve updating the Java/Scala and Ruby ports, and making @donspaulding's build_json.py a standard part of the distribution for all ports.

Important note: we would still keep the master copy of the database in YAML for readability/editability purposes.