=Paper=
{{Paper
|id=Vol-1225/full6
|storemode=property
|title=Personalisation of Web Search: Exploring Search Query Parameters and User Information Privacy Implications-The Case of Google
|pdfUrl=https://ceur-ws.org/Vol-1225/pir2014_submission_13.pdf
|volume=Vol-1225
|dblpUrl=https://dblp.org/rec/conf/sigir/FernandoDA14
}}
==Personalisation of Web Search: Exploring Search Query Parameters and User Information Privacy Implications-The Case of Google==
https://ceur-ws.org/Vol-1225/pir2014_submission_13.pdf
Personalisation of Web Search: Exploring Search Query
Parameters and User Information Privacy Implications –
The Case of Google
Anisha T. J. Fernando Jia Tina Du Helen Ashman
School of Information Technology and School of Information Technology School of Information Technology and
Mathematical Sciences. and Mathematical Sciences. Mathematical Sciences.
University of South Australia University of South Australia University of South Australia
anisha.fernando@mymail.unisa tina.du@unisa.edu.au helen.ashman@unisa.edu.au
.edu.au
ABSTRACT A brief background of search personalisation and the implications
Personalised search adapts search results to the needs and interests for personalised search is outlined. An initial experiment
of users. This is done through user data collected through various exploring the impact of search query parameters on
implicit and explicit methods and is used to build profiles of personalisation is discussed. Future research directions with the
information needs of users. This paper highlights the need to aim of exploring the impact search query parameters have on
explore search query parameters and determine their impact on personalisation and its impact on information privacy are also
personalisation. This is a first step in exploring the mechanisms of described.
personal data collection and how personalised search uses
personal data, which subsequently impacts the information 2. Search Personalisation
privacy of users. It was found that location parameters have more Traditional information retrieval techniques are limited by the use
impact on personalisation than the parameter ‘pws’ that switches of linguistic-based methods like keywords to express user
personalisation on or off. Hence, it is important to undertake information needs. Users have three broad categories of
further research that investigates the impact of other types of information needs according to their main goals, including
search query parameters, their contribution towards search navigational (to access a particular website), informational (to
personalisation and their impact on user information privacy. obtain relevant information on a specific area) and transactional
(to execute a web-based activity) [1]. The information needs drive
users to perform search queries and obtain relevant results. Users
Categories and Subject Descriptors expect an instant and relevant response to their queries, whilst
H.3.5 [Information Storage and Retrieval]: Online In-formation
facing a highly dynamic web environment which is in a constant
Services—Web-based services
state of flux and the issue of information overload, where a
countless amount of results are retrieved for a given search term
[2]. These challenges led to the development of personalised
Keywords search, which aims to provide relevant results to users based on
Search Personalisation, Information Privacy, User Privacy their information needs [3].
Concerns, Search Query Parameters
Personalised search can be content-based or collaborative-based
[4]. Content-based techniques use the content of items when
1. Introduction determining relevant results that match user queries. Content-
Personalisation of search aims to produce search results that based techniques include:
individual users are interested in because it caters to their specific • contextual search - where suggestions are provided based on
information needs. However, this raises privacy concerns given the user’s working context;
that personalisation collects personal user data to provide
personalisation functions. Hence, it is important to investigate • web search histories - where search results are based on
how personalisation occurs and what types of personal data is previous search history and selected results;
collected and used for the personalisation process. • hypertextual data extracted from web pages - where search
The aim of the paper is to discuss a range of key privacy issues results are modified by hypertextual algorithms to reflect
relating to web search personalisation. The scope of the paper is criteria important to users;
limited to discussing the initial progress of an experiment • rich user models - where feedback is provided to actively
investigating location-based search query parameters. The paper is build user models and store information about user
a first look at exploring the mechanisms of personal data preferences and search results;
collection and investigates the impact of selected search query
parameter on personalized search results. • adaptive result clustering - where search results are grouped
into clusters containing results on the same topic [3, 5].
Collaborative-based techniques use algorithms to produce results
that are based on models of different users and their needs [6].
Collaborative-based techniques include collaborative-based
search engines, which produce relevant results based on ratings of
prior users with similar preferences and collaborative-community In addressing public concern over privacy and personal
based recommendation systems which produces results based on information collected by organisations, governments all over the
an analysis of community based search [3, 7, 8]. Both content and world are changing privacy laws to reflect the ever-changing and
collaborative-based techniques collect information from users and dynamic digital world we live in. In Australia, the Australian
build a user profile modelling user information needs and Privacy Principles embody 13 key ideals that organisations and
interests. This information can be collected explicitly through government agencies that collect personal data have to abide by.
users providing ‘relevance feedback’ on search results or These include open and transparent management of personal
implicitly by capturing and processing click-through data [9]. information, anonymity and pseudonymity where possible,
guidelines regarding the collection of solicited personal
information, dealing with unsolicited personal information,
Personalisation may be social-based using information from social notification of the collection of personal information, use or
media, location–based which focuses on geographic location disclosure of personal information, direct marketing guidelines,
details of users or may involve behavioural profiling and data cross-border disclosure of personal information, adoption, use or
aggregation based on longitudinal data collected [10]. This paper disclosure of government related identifiers [15]. In addition,
will focus on location-based personalisation. quality and security of personal information must be upheld and
access to and correction of personal information should be
provided. The European Union’s Data Privacy Directive manifests
Personalisation provides users with many benefits such as
user data protection principles similar to that of the Australian
providing locally-relevant search results by catering to their user
Privacy Principles. Personal data must be collected and processed
needs. In addition, it helps overcome the problems of information
in a legitimate manner, with explicit user consent obtained and
overload and ambiguity associated with using keywords to
where the individual can refrain from providing personal data for
express user needs [11]. However, personalised search creates
processing in applicable situations [16]. Based on the Australian
implications for privacy because personalisation requires the
Privacy Principles, in the web search context Australian users
collection of user information to profile users and their
should be in control of their personal data. The user's personal
information needs [12].
details gathered through search queries may be used for purposes
other than the original reason for data collection and this is not in
accordance with national privacy laws.
3. Information Privacy and Implications for
Personalised Search 3.2 Implications for Personalised Search
To personalise search results, personal user information is
3.1 Information Privacy required which is routinely collected and profiled [17]. Significant
Privacy is a social construct with different people having different levels of personalisation can be created through processing basic
attitudes towards privacy. Information privacy is concerned with user information [18]. A search profile is a history of search
personal information of users, which can subsequently be used to queries where each query in the profile consists of the username,
identify them [13]. In the web search context, users may the time of the query, the query itself and when applicable, the
involuntarily reveal information about themselves whilst link the user followed after the query was submitted and each
searching and no longer have control over their data. Search query has context [19]. This enables a user profile to be built
engines collect large masses of this user data to profile user needs. containing valuable context information. With improved
User data may be public personal information, which is capabilities of technologies to capture information, easily transfer
confidential, and non-intimate in nature or it could be non-public and disseminate information and analyse information, personal
personal information [14]. data used in such a manner may have a serious impact on a
person’s privacy. Google, for example, is said to collect user data
automatically which are recorded in search logs when users type
It is pertinent to know the distinctions between non-public in search queries. These include information such as the search
personal data and public personal data in the web search context. query, IP address, browser information, date and time of request,
An example of non-public personal data would be a person’s cookies that identify the browser, hyperlinks clicked, operating
medical history that is held confidentially and stored securely and system, language, processor type, screen resolution and colour
could only be accessed by a user with authorized access such as a depth, active plug-ins etc. [20, 21]. Cookies alone may not be
medical practitioner. Public personal data would be a person’s effective in personalising as cookies are browser-dependent and
curriculum vitae containing work history on their personal more than one user may use a computer or a user may use
website. In both instances there are general rules regarding access multiple computers, thereby an inaccurate or incomplete user
and this helps protect the data stored and uphold its accuracy and profile may be created [20]. Alternatives such as 'flash cookies'
quality. Increasingly there are grey areas over what personal or (local stored objects) behave differently to normal cookies and
non-personal information is and whether users (subjects about can be managed independently of the browser through the Adobe
whom the data is stored) have enough access and control over flash security setting [22]. However the privacy implications are
what is stored or posted about them by others. Users may that these are harder to manage, being much less well-known and
willingly divulge personal details online in a context where they lacking control settings in most browser preferences.
feel comfortable to do so. However, most of what a user publishes
online is seemingly stored forever given the countless backups
and search result pages being indexed on a daily basis. Therefore, Personalisation may occur through account sign-in without
the ‘right to be forgotten’ in genuine instances is especially useful account sign-in methods. For instance, from Google’s perspective,
as it empowers users to have a stake or a claim if data posted an individual user’s search history is used for personalisation
about them is inaccurate or irrelevant. when logged into a Google account. For users who are not using
or logged into a Google account, personalisation still occurs sense of security in enticing users to opt in, or perhaps to not opt
through cookies connected to a web browser and may remain out.
there for a period of 180 days [21]. Therefore, it could be
plausible that users who do not wish to have personalised results
may still be given personalised results through use of such Many different types of data in isolation are not particularly
persistent cookies even if they proactively avoid personalisation personal, but taken together can reveal many details about a
through account sign-in by not logging into accounts. This person. In particular, when a collection of otherwise innocuous
increasingly difficult to avoid as users may use services from a data is focused around a single user, a comprehensive profile of
suite of products that an organisation has, such as Google Search the user can be built up - this is linkability of personal data. A
and Google+. It creates an exhaustive information source upon well-known example is that personal details about AOL user
which to profile user actions and preferences. This indicates a key Thelma Arnold were ascertained from seemingly random search
issue where users may be presented with search results based on queries like searches for people with the last name ‘Arnold’ and
what the personalisation algorithm determines suitable which is ‘landscapers in Lilburn, Georgia’ she had made, after AOL
manipulated from the data collected through use of the search released search keywords used by more than half a million users
engine’s services. This phenomenon named serendipity or the lack over a three-month duration. This was possible because AOL
of it called the ‘filter bubble’ has significant implications where released query data pseudonymising each user with a unique
users depend or trust the search results being presented to them numeric identifier assigned by the search engine, but neglected the
instead of actively looking at lower-ranked results that may linkability aspect by leaving the pseudonym unchanged for the
provide them with a more representative view [12, 23]. This is entire three months of search data [25]. These search queries
particularly pertinent given the low levels of user awareness of provide data that is used to build a user profile where decisions
what personalised results look like and how personalisation occurs can be detected or inferred from the context. If these search
and the use of personal user data in bringing about personalised queries are aggregated, then the search engine has the capability
search results. to identify and use personal details about people’s lives.
This collection of user data by search engines also brings about a It is also relatively easy to identify a person from information that
number of privacy problems such as: aggregation, where is already public, but is supposedly de-identified. For example,
information collected about a person over a time can be combined William Weld, a former governor of Massachusetts was identified
to find out details of the person; distortion, where information using only the ZIP code, birth date and gender from a combined
collected in search query records may be misleading and may not pool of ‘anonymised’ medical data sold by insurance companies
reflect the actual intent of the users; exclusion, low levels of and publicly available voter registration data [27]. Therefore, by
awareness by the public on what information is actually collected linking different sets of seemingly anonymised data, identities of
by search engines; secondary use of data which is not in line with people can be elicited and raises significant privacy concerns in
the original purpose of data collection; and political and social ensuring anonymity of user information, especially when this data
implications of searching sensitive topics of interest [24]. One of is publicly accessible.
the many benefits of search is that users can find out information
as and when they require it and with personalised search, it aims
The commoditisation of search to increase value-add for profit-
to provide results that users are interested in. However, privacy is
oriented search engines resulted in targeted advertising. This
usually traded off against the capability to use functionality such
behavioural advertising raises privacy concerns as it involves
as search [25]. This is because of the data collection that occurs
matching ads relevant to user needs based on the user profile,
and the ambiguity surrounding what exactly is collected. People
which captures user preferences and personal data from search
should be able to actively control personal information and know
history.
how non-personal information about them is being used. Some
key privacy requirements that should be upheld when data is
transmitted include transparency, openness, notice and consent,
where users are provided with options to control the level of 4. Exploring the Impact of Search Query
personal data being transferred and the level of personalisation
they prefer.
Parameters on Personalisation and its Privacy
Implications
A key concern is that users may not be aware that their personal
In an attempt to further understand what types of personal data are
user data is collected and how it is being used [19]. Users may
transferred when searching, a series of experiments focusing on
have also little control over how their personal data is being used
capturing and analysing HTTP requests and responses during
and how they may retrieve and eliminate personal information
these search requests are being conducted. An analysis of search
[14]. Users may not have an actual choice in using search engines,
query parameters is being undertaken because it is an effective
as opting not to use search engines results in the inability to search
method of examining the impact on personalisation and the
and retrieve information [26]. Also the privacy policies of search
ensuing impact on information privacy. It is important to
engines are usually concise to avoid lengthy details, but in doing
investigate at this basic level because it is a valid measure of
so may be vague and cover a broad spectrum of areas. However, it
ascertaining what information is transferred when searching.
is ascertained that in such a context, people do not know what
These experiments will consider the different types of
they are agreeing to and as such privacy policies fail in adequately
personalisation such as location-based, social-based, behavioural
communicating how user data is captured and used [26]. Also
profiling and data aggregation and will investigate the impact
search engines are careful to avoid using terms like user profiling
across both widely-used search engines like Google, Yahoo, Bing
in their privacy policies [20]. Therefore, this may create a false
and privacy-enhanced search engines like DuckDuckGo, Ixquick
and StartPage. As the investigations are in its early stages, the above scenario, an anonymising proxy, Tor1, was used to provide
scope of this paper will discuss preliminary investigations location anonymity by spoofing the IP address.
pertaining to location-based personalisation and will be limited to Interestingly, there were negligible visible variances between the
using the search engine Google as an experiment vehicle. Google personalisation parameter ‘pws’ being explicitly switched on and
was chosen as the search engine as it is a popular choice for off, or by default (i.e. ‘pws’ being absent from the search query)
search and a majority of web search tasks are carried out using it. even with account sign-in or without account sign in (Figures 1, 2
The overall aim of these evidence-based experiments is to explore and 3). Almost all of the search results were constrained to the
the impact of search query parameters on personalisation and its Australian context. However, when Tor was used and the same
subsequent impact on information privacy. Search queries were search scenario was repeated, there were visible differences in the
chosen from 5 different category types as queries are determined search results; for example, in one instance Google Germany was
to have varying degrees of personalisation based on the category used with a mixture of results from UK, US and Germany to name
of search query [28]. The search query categories include a few (Figure 4). Hence, location parameters are observed to have
technology, political news, entertainment, literature and science- more significant impact on personalisation than the
related topics. These queries will be similar to everyday queries personalisation parameter ‘pws’, which is designed to set
made by users and relate to topics that are popular and commonly personalisation on or off. This level of personalisation could vary
searched about such as music concerts and political news. depending on users with active logins and search history.
Personalisation has been shown to be more evident in specific Therefore, this initial exploration into whether there are
categories of queries such as political and less evident in other differences with how parameters influence personalisation opens
categories like health related queries [28]. Through the use of an up an avenue for further exploration into the importance of search
http-intercepting proxy tool, http and https requests and responses query parameters and identifying its influence on personalisation.
were captured when the search queries are performed and On-going and future experiments will be refined to control for
analysed to derive the impact of the search parameters. sources of noise such as search index updates – where search
indices are updated on a regular basis, distributed infrastructure –
When users search on the web, the HTTP protocol is used to send where results may differ to data centres being located in different
requests and receive responses [29]. These requests and responses geographic areas, geolocation – where a user’s IP address is used
consist of search query parameters like POST or GET parameters, to produce results that are locally relevant, a/b testing – which is
HTTP headers or cookies. These parameters may capture or store periodically conducted by search engine organisations to
user data that are transmitted when searching with the purpose of determine clickthrough preferences [28].
being relevant for personalisation and the potential for leaking
personal data. Hence, HTTP headers, POST or GET parameters
and cookies may act as potential sensitive data leakage points.
Most search query parameters are not officially documented.
Google’s Privacy Policy highlights some examples of personally
identifiable information, which may be used across the range of
Google’s services [30]. ‘pws’ is a GET HTTP parameter that can
be switched on or off to control personalisation as widely
described in the search engine optimisation community, but there
is some concern over the impact of the parameter as a control over
personalisation and little to evidence to prove it [31]. Hence, we
assume that different parameters may have more significant
impact on parameters than others.
An initial experiment was conducted to explore the impact
specific search query parameters have on other parameters. Using
a web debugging proxy tool to capture the search requests and
responses sent to the search engine, the personalisation parameter
‘pws’ was manipulated across the 5 different search categories.
This experiment was run across various scenarios involving all
possible combinations of pws/location and sign in parameters.
This included scenarios involving both types of personalisation
(i.e. personalisation with and without account sign-in, switching
the ‘pws’ parameter on and off and through location anonymity.
This was then repeated with a gap of 15 minutes to minimise the Figure 1- Example of a search result without account sign-in,
carry-over effect (where conducting similar search queries to personalisation parameter switched off and location spoofing
determine the effect a prior search has on the current search) off.
across the 5 distinct search query categories [28]. To account for
biases, each search query was done on a fresh instance of the web
browser after having cookies and other persistent web data cleared
after each web session. The test environment was constrained to
be Windows 7 OS and Firefox Therefore, personalisation through
account sign-in was through a Google account. In addition to the
1
Tor: https://www.torproject.org/
Figure 2- Example of a search result with account sign-in and
the personalisation parameter on without location spoofing
Figure 4- Example of a search result with location spoofing
using Tor and the personalisation parameter switched on
5. Conclusions and Future Research
Directions
It is important to find out the impact search query personalisation
parameters have on privacy as well. Future experiments will focus
on investigating the impact of other location-specific parameters,
and its purpose (such as if it is being used for personalisation or
for data collection), which would help examine the impact these
parameters have on the information privacy of users by
identifying what kinds of personal data is transferred. These future
experiments will be automated by using software that can analyse
and collect search engine results with the goal of investigating
different parameter sets. The initial investigations provided
familiarisation with how search query parameters work and
established that certain parameters like location impact
personalisation more than others. This impact could be due to a
lack of extensive user search history available to the search engine
in the experiment conducted. By identifying what types of user
data are affected based on the identified search query parameters
and its purposes, it would help recognise the significance of data
submitted by users. Additionally, another key area to examine
includes whether personalisation positively impacts the user
experience and the reasons underpinning its impact. User privacy
concerns could also be assessed in relation to personalised search
Figure 3 – Example of a search result with the account sign-in and information privacy through user-based experimental studies.
and the personalisation parameter and location spoofing
switched off With privacy being a major concern in the digital world, it is
important to understand how personalisation works and what
personal data are collected and used to perform personalisation.
Identification of search query parameters and its purposes is a first
step in determining how user information is used in the
personalisation process. There is a need to conduct further
research on the impact of the various types of search query
parameters and determining its importance in influencing
personalisation. Ascertaining what types of personal data is
transferred by analysing search query parameters would allow a
clearer idea of the level of personal information disclosed and [15] Office Of The Australian Information Commissioner. 2014.
inform and validate the privacy concerns of users. Therefore, Australian Privacy - Privacy Fact Sheet. Australian
future research work will continue exploring the impact of various Government, Canberra.
types of search query parameters, determine its purpose (if it is [16] European Union. 2011. Protection of Personal Data.
used for personalisation or data collection) and subsequently infer European Union. Retrieved 26 April 2013.
its impact on user information privacy. http://europa.eu/legislation_summaries/information_society/
data_protection/l14012_en.htm.
6. References [17] Berendt, B., & Teltzrow, M. (2005). Addressing users’
privacy concerns for improving personalization quality:
[1] Broder, A. 2002. A Taxonomy of Web Search. ACM SIGIR Towards an integration of user studies and algorithm
Forum. 36, 2, 3-10. evaluation. In Intelligent Techniques for Web
[2] Lincoln, A . 2011. FYI: TMI: Toward a holistic social theory Personalization (pp. 69-88). Springer Berlin Heidelberg.
of information overload. First Monday. 16, 3, 1-15. [18] Krause, A., & Horvitz, E.2010. A utility-theoretic approach
[3] Micarelli, A., Gasparetti, F., Sciarrone, F. and Gauch, S. to privacy and personalization. J. Artif. Intell. Resea. 39,
2007. Personalized Search on the World Wide Web. In P. 633–662.
Brusilovsky, A. Kobsa & W. Nejdl Ed. The Adaptive Web. [19] Brandi, W. and Olivier, M.S. 2010. In Search of Search
Springer-Verlag, 195-230. Privacy. In M. Soriano, S. Katsikas & J. Lopez Ed. Trust,
[4] Gao, M., Liu, K. and Wu, Z. 2010. Personalisation in Web Privacy and Security in Digital Business. 6264. Springer
Computing and Informatics: Theories, Techniques, Berlin Heidelberg, 102-116.
Applications, and Future Research. Inform. Syst. Front. 12, [20] Aljifri, H. and Navarro, D. 2004. Search Engines and
5, 607-629. Privacy. Comput. Secur. 23,5, 379-388.
[5] Teevan, J., Dumais, S. and Horvitz, E. 2010. Potential for [21] Google. 2013. Google Privacy Policy. Retrieved 25 June
Personalisation. ACM T Comput. Hum. Int. 17, 1, 1-31. 2013. http://www.google.com.au/policies/privacy/.
[6] Shapira, B. and Zabar, B. 2011. Personalized Search: [22] Adobe. 2013. Analytics and on-site personalisation services
Integrating Collaboration and Social Networks. J. Am. Soc. - Adobe - Privacy Policy Retrieved 7 September 2013.
Inform. Sci. 62, 1, 146-160. http://www.adobe.com/privacy/analytics.html.
[7] Reimer, K. and Brüggemann, F. 2006. Personalisation of [23] Pariser, E. 2011. The filter bubble: What the internet is
eSearch Services – Concepts, Techniques, and Market hiding from you. Penguin Press, New York.
Overview. In Proceedings of BLED 2006 -19th Bled
[24] Tene, O. 2008. What Google Knows: Privacy and Internet
eConference on eValues, pp. 1-16.
Search Engines. Utah Law Review. 2008, 4, 1433-1492.
[8] Smyth, B., Coyle., M & Briggs, P. 2011. Communities,
[25] Zimmer, M. 2008. The Gaze of the Perfect Search Engine:
Collaboration, and Recommender Systems in Personalized
Google as an Infrastructure of Dataveillance. In A. Spink &
Web Search. In F. Ricci et al. Ed. Recommender Systems
M. Zimmer Ed. Web Search, 14, Springer Berlin Heidelberg,
Handbook, Springer US, 579-614.
77-99.
[9] Steichen, B., Ashman, H. and Wade, V. 2012. A
[26] Nissenbaum, H. 2010. Privacy in Context: Technology,
Comparative Survey of Personalised Information Retrieval
Policy, and the Integrity of Social Life, Stanford University
and Adaptive Hypermedia Techniques. Inform. Process.
Press. Stanford, California.
Manag. 48, 4, 698-724.
[27] Sweeney, L. 2002. K-Anonymity: A Model for Protecting
[10] Toch, E. Wang, Y, Cranor, L.F.: Personalization and privacy:
Privacy. In International Journal on Uncertainty, Fuzziness
a survey of privacy risks and remedies in personalization-
and Knowledge-based Systems. 10,5,557-570.
based systems. User Model. User-Adapt. Inter. 221,2, 203–
220. [28] Hannak, A., Sapiezynski, P., Kakhki, A.M., Krishnamurthy,
B., Lazer, D., Mislove, A. and Wilson, C. 2013. Measuring
[11] Steichen, B., O'Connor, A. and Wade, V. 2011.
Personalisation of Web Search. In Proceedings of the 22nd
Personalisation in the Wild: Providing Personalisation across
International Conference on World Wide Web. International
Semantic, Social and Open-Web Resources. In Proceedings
World Wide Web Conferences Steering Committee, 527-
of the 22nd ACM Conference on Hypertext and Hypermedia,
538.
ACM, 73-82.
[29] Gourley, D., Totty, B., Sayer, M., Aggarwal, A., Reddy, S.
[12] Ashman, H., Brailsford, T., Cristea, A., Sheng, Qn.Z.,
2002. HTTP the definitive guide. O'Reilly, CA.
Stewart, C., Toms, E.G., Wade, V. 2014. The Ethical and
Social Implications of Personalisation Technologies for e- [30] Google. 2014. Google- Privacy & Terms: Privacy Policy.
Learning. Inform. & Manage. 1-23. Retrieved 10 June 2014.
DOI=http://dx.doi.org/10.1016/j.im.2014.04.003 http://www.google.com.au/policies/privacy/
[13] Smith, H.J., Dinev, T. and Xu, H. 2011. Information Privacy [31] Meyers, P.J. 2012. Face-off – 4 Ways to De-personalize
Research: An Interdisciplinary Review. MIS Quart. 35, 4, Google. The Moz Blog. Retrieved 1 May 2014.
989-1015. http://moz.com/blog/face-off-4-ways-to-de-personalize-
google
[14] Tavani, H. 2005. Search Engines, Personal Information and
the Problem of Privacy in Public. Int. Rev. Inform. Ethics. 3,
1, 39-45.