Internet o ers a number of various services that to maximise the user experience apply di erent personalisation techniques. An important resource of every personalisation method is a user pro le. The more information on the user is available in such pro le, the better. Therefore, together with maturing of these mechanisms, the notion of identity emerged. The identity exceeds the user pro le with information that is more detailed or enables bene ting from additional functionalities. The information stored within an identity needs to be understandable for di erent services to be easily reused. This can be achieved using the DBpedia. The goal of the article is to describe the design of a method that potentially enables providing data to build the user identity, based on his behaviour on the Web. The method is elaborated as well as an example of application is presented.
Most users leave a signi cant amount of information about themselves on the
Web. They abandon their anonymity freely (sometimes unconsciously), in order
to stay connected with their friends on the social networking sites, communicate
with their government or build their reputation [
One of the major challenges concerning the identity management systems is creation and maintenance of many perspectives on users identity, called virtual identities, most preferably without explicit actions of the user. Virtual identity is understood as a collection of topics concerning speci c interest of a user. In this paper we present a method that enables automatic identi cation of such topics using Wikipedia and information extraction techniques. The method is developed for the Polish language. It utilizes Wikipedia concepts but can easily be extended to DBpedia resources. As there is no Polish DBpedia yet, this will not be covered by this article. However, the work on Polish DBpedia is ongoing and this will be addressed in the future.
The remainder of the paper is structured as follows. Section 2 is devoted to a short summary of virtual identity de nitions. In the next section, we indicate current projects and existing approaches that raise the issue of users virtual identities and provide solutions in this area. Section 4 describes the method proposed to identify concepts building the users identity. Finally, in Section 5 we focus on a Use Case demonstrating the application of the method. The article concludes with the nal remarks. 2
The concept of an identity has been adopted by Information Science as a formal
representation of knowledge about a certain person, or any other (digital or
real-world) subject. Concerning an identity of a person, it is understood as a
set of attributes (permanent or temporary) characterizing a person [
More generally, a virtual identity can be de ned as a digital representation
of a set of claims made by one party about itself or another digital subject [
The identity of a digital subject can be established by combining both the
real-world attributes (for example name, address, social security number,
physical traits, etc.) and the digital ones (such as passwords, access rights, biometrics,
type of encoding, network address and so on) [
For the requirements of the "Ego - Virtual identity" (Ego) project1, presented in the paper, the identity is understood as an information structure describing the information needs of a user. This structure is grounded in the Wikipedia concepts' graph to ease its maintenance and assure usefulness while personalizing information content, especially from the information needs evolution point of view. The future work concerns extending the method towards DBpedia resources. 3
In the following sections, we present the state of the art analysis of the identity management systems on the Internet in terms of the business goals, that they pursue and the functionalities, that they provide. We identify the most important projects and solutions in the area of identity management systems, that may bene t from the approach we suggest. The main projects that we concentrated on are following: FIDIS2, SWIFT3, PICOS4, PRIME5, STORK6, ProjectVRM7. Moreover, there exist also frequently updated lists of identity-related e orts8.
In addition to the above-mentioned projects, a number of already
implemented solutions were analyzed. These solutions however mainly focus on the
authorisation aspect, leaving behind the notion of user representation e.g. the
OpenID protocol describes a user with a limited set of attributes only [
Some of the solutions are widely used in business, for example the OAuth protocol 9 or various OpenID implementations, while some of them are at earlier stages of development and adoption, e.g. WebID and Higgins.
Finally, its also very important to indicate, that several organizations have emerged and aim at consolidating and coordinating e orts in the area, of which 1 http://kie.ue.poznan.pl/en/project/ego-virtual-identity 2 http://www. dis.net/ 3 http://www.ist-swift.org/ 4 http://www.picos-project.eu/ 5 https://www.prime-project.eu/ 6 https://www.eid-stork.eu/ 7 http://projectvrm.org/ 8 For example: http://personaldataecosystem.org/2011/06/startup/, http://blogs.law.harvard.edu/vrm/development/, accessed on 15/10/2013 9 It is used for example by Facebook, Google and Last.FM the most important are probably Kantara Initiative10, Identity Commons11 and formerly Liberty Alliance12.
It can be easily noticed, that the concept of virtual identities is heavily studied. Nevertheless, as it is a wide eld to investigate, di erent areas of virtual identity creation, maintenance and usage can be explored by di erent projects. To the best of our knowledge, the approach focusing on automatic creation of users virtual identity that links experience from the elds of information extraction and Wikipedia does not exist. 4
This section presents details of the approach we apply to create the identity of a user. The phases of creating the users virtual identity are as follows:
Phase 1: Tracing user behavior. The rst step towards building a user's
identity concerns identi cation of topics of user's interest. Of course, these topics
can be entered manually by a user (a so-called explicit user modeling [
There is a lot of information about a user even before she or he starts using a given identity management system. Such information is often spread across multiple domains such as web portals, social networking sites, etc. Therefore, the identity management systems can try to somehow import and aggregate information about the user from such sources automatically. To make that feasible, the user's data export mechanisms must be made available by owners of such systems. An example of such initiatives are Data Liberation Front13 and Data Portability Project14.
We build the identity of a user based on a wide range of his activities on the
Web. Our goal is to engage the service providers in this process, as discussed
in [
Phase 2: Analysis of the visited Web sites. The content that is uploaded to the server is analyzed using the lexical extraction module to identify the di erentiating phrases and assign a topic. For the list of topics that are the most representative for the whole content of the network, we chose the Wikipedia categories and concepts list.
The extracted content of the website is analyzed using NLP to identify named entities, cross references, etc. and as a result provide a set of words (surfaces 10 http://kantarainitiative.org/ 11 http://www.identitycommons.net/ 12 http://projectliberty.org/ 13 http://www.dataliberation.org 14 http://dataportability.org/ existing in the text, further being referred to as phrases), that will be subject to further processing. What is important, that the approach works for the Polish language and is contextual.
Phase 3: Building a representation of a website for the needs of the identity building. The most crucial step, from the point of view of this paper, as well as for the user acceptance of the system being developed, is indication of a topic, the website mentions. This is done in the following steps.
Firstly (in the preparatory phase), all Wikipedia pages are processed in
order to identify concepts (Wikilinks) that appear on these pages in order to learn
a phrases-concepts mapping, similarly as it was done by [
The method of indication of a topic of a website assigns to each phrase from the text (f ) concepts from Wikipedia (c1 { c6 in Figure 1) obtained as indicated in the previous paragraph. Then, for these concepts (c1 { c6), the upper level categories of concepts are indicated (c11 { c51). The Wikipedia category structure enables to build a whole tree over the initial concepts that were assigned, e.g. for concept Peter Higgs, based on the Polish Wikipedia structure, we retrieve categories such as Scottish Physicists, Born in 1929, etc. Currently, we use only three levels within the tree (experimentally evaluated). Then, using the bottom-up propagation method the rst-level concepts (mapped from phrases extracted from the website content) vote for the upper level concepts. The bottom up propagation measure combines ve frequencies: { The number of times a phrase from the article text refers to a concept from the Wikipedia. { The number of times a phrase (surface form) appears in the Wikipedia. { The number of times a given concept is referenced within the Wikipedia. { The frequency of a word in the language (in our case the Polish language). { The number of sub-concepts of a concept.
As a result of the bottom-up propagation, we identify a concept (not necessarily the top-level one), that is the most probable topic of the website. Afterwards, the phrase from the website being the most strongly connected with the concept assigned as a topic, is removed from the initial list of phrases and the procedure is repeated for the remaining phrases. While experimenting, we identi ed that for most of the articles three iterations are enough to provide the most meaningful concepts describing the website's topic.
These concepts are then mapped on the user's virtual identity. Each new package of topics, changes the initial identity. The weights assigned to di erent topics within the identity, re ect also maturing in time. Also, user may support this process by manually extending the list of automatically assigned categories.
The user identity created by the system, may be then further used for the
needs of personalization of websites visited by the user. The Ego system is to
provide a number of functionalities enabling for sharing and encrypting the identity,
authorizing a service provider as well as enabling user to manage the identity
and to access it [
The presented approach is about to be validated with the real users, who committed to use Ego for a certain period of time and share their experiences. Up till now, the Use Case-based validation has been performed. For the sake of clarity, we present details based on one news article only. The article concerns the Noble Prize Winner Peter Higgs (it is in Polish and is available at Polskie Radio website15.
The content of the article was extracted and loaded in the database as a logical document (this concerns the topic and the content of the article; menus, comments etc. are not further analysed). Then, the lexical extraction rules extracted 44 di erent phrases from the article e.g. uroczystosci (celebration), professor, etc., out of which 31 were mapped on Wikipedia phrases.
For these Wikipedia phrases, 2052 Wikipedia concepts were retrieved (identi ed by di erent URLs) including three upper levels (2052 is a total number of concepts in the tree initially representing the topic of the article). The most frequent concepts in the rst level mapping were e.g. zyka (physics), konferencje miedzynarodowe (international conferences), mechanika kwantowa (quantum physics), II wojna swiatowa (second world war). 15 http://www.polskieradio.pl/23/266/Artykul/951564,
Profesor-Higgs-zapadl-sie-pod-ziemie
Then, the relations between di erent concept categories were exploited using the bottom up propagation method. After applying the method, the following concepts were identi ed as the most descriptive for the article (in the order of importance): { Urodzeni w XX wieku (born in XX century), { Popularnosc (Popularity), { Higgs, { Szkolnictwo wyz_sze (Higher education), { Nauki przyrodnicze (Natural sciences).
These concepts may be then further mapped on the Wikipedia category structure graph representing the users identity, but this issue is beyond the scope of this paper. 6
The goal of this paper was to present a method that enables for identi cation of topics that are of user's interest using Wikipedia and information extraction techniques, and based on the behavior of a user on the Web. Starting from a general summary of the Virtual Identity de nitions, we presented a method that may be used in order to create user identities using Wikipedia. We also demonstrated an application scenario.
The future work will especially be devoted to tuning of mechanisms developed as well as carrying out an extensive validation of the approach with the real users. The major issue that needs additional research is the bottom-up propagation method that should eliminate concepts being pointed from the multiple websites such as e.g. born in XX century.
Further research will also concern changing the Wikipedia to the DBpedia to allow for an extensive reasoning. This could also o er additional functionalities to an identity management system and service providers that will bene t from it. However, the work on the Polish DBpedia is still the ongoing e ort.
Acknowledgments. The work published in this article was supported by the Polish Ministry of Science and Higher Education (decision no. 0987/R/H03/2010/10), upon contract with the Polish National Centre of Research and Development (NCBiR) (contract no. NR11-0037-10/2011) on the project titled: \Ego { Virtual Identity" (http://kie.ue.poznan.pl/en/project/ego-virtual-identity).