Instrumentation & Measurement Magazine 23-9 - 55

Fig. 2. Tracking Decision Diagram.

using this initial classification and the data collected by ORM.
The results match with most of the previous studies, showing Google as the main tracker company, followed by Facebook,
Yandex or Bing. Interestingly, comparing our results to some
of the previous works (e.g., [4]), there is a clear increase in the
domains being tracked by Google in any of its forms (e.g., analytics, ad services, tag services), reaching almost 55% of the
entire population. It is worth noting that Doubleclick.com, a
subsidiary of Google specialized in targeted advertisements,
is present in about 20,000 of the total websites, but that it has
been blocked more than 70,000 times. This proves that loading
multiples instances of the same third-party resource inside the
same website is a common practice, probably to get more revenues from targeted advertising.
Fig. 4 presents the cumulative distribution function (CDF)
of the total tracking percentage traffic within the collected
population (i.e., the fraction of traffic of the website devoted
to tracking). There are approximately 28,000 websites (out
of 100,000) that do not include any web tracking methods.
On the contrary, almost 72,000 include at least one or more
web tracking resources, with about a 10% including more
than 90% of resources dedicated to track the user or targeted
advertisement.
During the experiments we also observed that within the
top 100 websites (including www.google.com), third-party
tracking resources are usually not present or very rare. This
is because most of them perform tracking as first-party trackers, getting data directly introduced by the user. For instance,
Google does not need to look for the searches done by its users,

trackers or manually inspecting the code of the top most
used resources. Moreover, we instructed ORM to download,
unminify and store inside the database the script and document files, which are commonly used to execute web tracking
methods. The system also compares each resource URL with
EasyList and EasyPrivacy, the two most famous and used content-blocking lists, to check if it is an already known tracking/
advertisement resource. The resulting dataset has been made
publicly available at [2].
All of the information was taken for four different browser
configurations, one with the vanilla browser (no plug-
ins, no settings) and the
other three loading each of
them a different contentblocker plugin (AdBlock
Plus, Ghostery and uBlock
Origin). This configuration allowed us to generate
a partially-labeled dataset with the already known
web tracking methods by
the existing lists and content blockers, which can
be useful for training machine-learning algorithms.
The labeling process is
shown in Fig. 2. Intuitively,
if a resource present in the
vanilla browser is detected
by EasyList/EasyPrivacy
lists, or blocked by the content-blockers, that resource
can be considered as a
tracking system, either for
profile creation or for targeted advertisement.
Fig. 3 shows the top web
Fig. 3. Top trackers by number of websites and tracking resources.
tracking domains found
December 2020

IEEE Instrumentation & Measurement Magazine 55

http://Doubleclick.com http://www.google.com

Instrumentation & Measurement Magazine 23-9

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 23-9