Users of Social Network Sites (SNSs) spend considerable amounts of hours per day exchanging (consuming or sharing) information and using services provided by such platforms. However, nothing comes for free. SNSs survive at the expense of the information that users' upload to their pro les, and the knowledge derived from their on-line behavior. Discovering hidden knowledge in social networks is a centerpiece in many personalized on-line services and ad-targeting techniques, and helps to make a SNS pro table. However, users seem not to be aware of this common practice and keep sharing content compulsively. Nevertheless, self-disclosure and over-exposition can have severe consequences and can put users' integrity into risk. In order to develop better information control and awareness systems, we believe that it is important to take into account the users' on-line habits and behavior. In this work we introduce an initial assessment of the di erent factors that contribute to self-disclosure in Social Media, and discuss the elements that a selfadaptive solution should consider to address this issue.
Social Media has set new standards for our interpersonal relations, and has accelerated the dynamics of our lives. Many users are bridged through SNSs, and new sub-communities are built everyday based on common interests, likes or even mottos. The inhabitants of these virtual communities are spending considerable amounts of time exchanging (consuming or sharing) information, and using services provided by the SNSs. However, none of this is for free. SNSs survive at the expense of the information that users' place in their pro les, and the behavior they exhibit while using the di erent services provided by these platforms.
Discovering hidden knowledge in social networks is a centerpiece in many
personalized on-line services and ad-targeting techniques, and is basically what
makes a SNS pro table [
Although existing privacy-preserving mechanisms have been developed and improved over the years, they are still not helping users in distinguishing a selfexposition behavior that might put them into risk. It is very hard for a regular user to keep track of everything that he or she has shared through its \online life". Moreover, once the content has been shifted to the Internet, the user has no control over it anymore. This situation demands new mechanisms for tracking the sensitive information that a user has already shared, and the degree of sensitiveness that new information might have. Thus, users of SNSs can make a wiser decision before sharing content, and have a better vision of what they have shared (and would like to un-share) in the past.
In this work we present an analysis of the \self-disclosure" problem in Social
Media and provide insights towards a self-adaptive solution. Di erent dimensions
of the problem like the users' behavior and information sensitiveness are studied
from an inter-disciplinary perspective. Furthermore, initial guidelines for a
selfadaptive approach based on the MAPE-K model by IBM [
In the following section the fundamental bases and concepts involved in our proposal are initially introduced (Section 2). Section 3 covers the di erent aspects of the self-disclosure issue including: the diversity of information in SNSs, the so-called \privacy paradox", information sensitiveness, and an adapted version of the MAPE-K model. Next, Section 4 discusses alternative existing solutions, and nally Section 5 presents our conclusions and related future work. 2
This section introduces the fundamental concepts that form the bases of our proposal. Here, Autonomic Systems and run-time self-adaptation concepts are presented and analyzed for further application in a Social Media scenario.
In order to raise awareness of self-disclosure among the users of SNSs we
propose to develop an Autonomic Computing vision of this issue. The goal of
Autonomic Computing is to design and develop distributed and service-oriented
systems that can easily adapt to changes that a ect the system administration
and service delivery, while reducing some of the complexities associated with
the management of such systems [
The Monitor component provides the mechanisms to observe through
Sensors di erent events or changes that take place in the System (managed element).
It also lters and aggregates the data, and reports details or metrics [
The MAPE-K model is used as an architectural reference in cases where
a feedback loop is a distinctive characteristic of the system being built. Such
is the case of [
It is necessary to conduct an analysis of several factors that contribute to the problem and the solution of self-disclosure in SNSs. In this section we will go through the di erent types of information that can be found in a SNS (particularly on Facebook), and we will provide some insights for further sensitiveness classi cation. We will also discuss the in uence of users' on-line behavior and risk aversion. At the end of this section, an approach for addressing this issue based on the MAPE-K model will be introduced.
SNSs are a rich source of the most varied kinds of information. However, users
do not realize the importance that this information they \voluntarily" deposit in
these sites has. From a high level inspection, normally one can nd in a Facebook
pro le the following information: list of friends, personal information (e.g. rst
name, surname and profession), wall posts (public messages from other users),
messages, photos, and notes [
However, if one takes a closer look to the now improved \Facebook Security Centre", it is now possible for users to download a copy of the information that Facebook stores about them. Surprisingly, the list is way bigger than the one mentioned before, and includes (among other information)1: { Ads Clicked: Dates, times and titles of ads clicked by the user. { Ad Topics: A list of topics that the user is targeted against based on its likes, interests and other data included in its Timeline. { Check-ins: Places where the user has checked-in to. { Facial recognition data: A unique number based on a comparison of the photos the user has been targeted in. { IP Address. { Log-ins and Log-outs.
{ Deleted friends.
Clearly, users do not submit many of this information voluntarily to Facebook.
For someone familiar within SNSs and their privacy practices, it is not surprising
that Facebook (like many other SNSs) keeps all these records in their servers.
However, for many users (newcomers or advanced) this situation remains unclear,
even when the privacy settings of their Facebook accounts are public by default
[
Exposing personal information to other persons is referred as individuals'
selfdisclosure. Self-disclosure in on-line contexts like Social Media is, at least to a
certain extent, the precondition for a functional social network [
Self-disclosure is frequent among the users of SNSs. Furthermore, users seem
careless when providing sensitive information through SNSs. However, they
consider privacy protection an important issue that must be addressed. This
phenomenon of contradiction has been referred as the \privacy paradox" [
Several gaps and dilemmas have been identi ed when trying to de ne what
sensitive information is [
A SNS is a complex environment where multiple factors converge and (in many cases) are the ones that de ne the rules of interaction and contributions for the users. For instance, a post that can look trivial on Facebook can be totally inappropriate in another SNS like LinkedIn (e.g. a photo of you in a party might not look very professional). In other words, here the context a ects the degree of sensitiveness of the content. In this case the targeted audience of the SNS is a conditioning dimension of the context. 3.4
An awareness system like the one proposed in this work has a self-adaptive nature. Its purpose is to perform a constant monitoring over the user sharing activities and notify when a self-disclosure behavior is detected. This noti cation can be seen as an interaction with the user, where he or she will have the last word and control over the sharing act. In other words, the user should have the chance to accept or reject the recommendation of not to share potentially sensitive information. This sequence of detection-noti cation-acceptance de nes a feedback loop between the user and the awareness system.
As we have discussed in the previous sections, classifying information into categories of sensitiveness does not have a straightforward solution. However, as several legislations agree, it is possible to build categories of \personal" or \sensitive" data. Since the user's perception is also a determinant on the nal classi cation, it seems logical to perform a classi cation of the users based on their interpretation of particular pieces of information. Then, by combining these two approaches together with attributes of the SNS (e.g. the targeted audience and activity levels of the users), a better classi cation of the information can be performed (Fig. 2).
In Fig. 3 an adapted version of the MAPE-K loop is described. In this case, the Managed element corresponds to the representation of the user in a SNS, this is, the user's account. In this approach, the Monitor is sensing the activity of the user and responses when an information-sharing event takes place. As was previously mentioned, the goal of this system is to provide recommendations to the user when it attempts to publish content of sensitive nature. Therefore, what the Analysis unit should do is to analyze the information collected by the Monitor's sensors and classify it into sensitive or not sensitive. Here, the Knowledge base has a main role because it contains all what has been learned about sensitiveness and its in uencing factors. After this is done, the Plan will elaborate a recommendation for the user, and then the Execute module will proceed to deliver it to the user.
A privacy protection recommendation system must be able to adapt on users individual self-disclosing behavior without destructing the interactive nature of SNSs. Our approach takes this statement into account by asking the user \do you really want to share this?" instead of forbidding it to continue. By this, the autonomy of the user is ensured and its nal decision contributes to the feedback loop of the system.
Nevertheless, self-adaptation brings into account a fundamental reasoning
problem: decide which is the best course of action to follow based on the perceived
stimuli from the environment. In Arti cial Intelligence this type of reasoning is
usually called planning, where the condition to achieve is called goal and the
sequence of actions that will make the goal true is called a plan [
Many privacy breaches in SNSs have been identi ed and addressed through
di erent types of privacy-preserving software architectures (e.g. P2P). Many
researchers advocate particularly for decentralized architecture schemas unlike
predominant centralized approaches[
While these approaches focus mainly on the architectural elements that a privacy-preserving SNS must have, the solution presented in this work propose to contribute to privacy on the application level. This is, even with a centralized and non-privacy-preserving SNS architecture, it should be possible to arise user's awareness and hence prevent extensive self-disclosure. In this way, empowered users will take better control over their on-line acts and in consequence over their private data. This can be achieved since SNSs like Facebook provide APIs and extension points for including 3rd party applications, which would allow us to integrate our solution without forcing users to change into another SNS. 5
In o -line situations people's communication about sensitive topics take place
behind closed doors; whereas in SNSs users do not seem to lock their metaphorical
doors when they address sensitive topics [
It is true that some users are not much concerned about the consequences that self-disclosure in SNSs could bring to them, and are not willing to modify their behavior. However, this does not neglect the fact that it is necessary to support and empower them through better control and awareness systems. Instead, this raises the necessity of developing instruments that take into consideration users' distinctive characteristics that make them more or less adverse to the risks of over-exposition.
Self-disclosure and information sensitiveness analysis propose a number of challenges and opportunities for self-adaptive systems. This work has analyzed and summarized the requirements that a self-adaptive solution must cover for addressing self-disclosure in SNSs. Now this vision has to be put into practice and undoubtedly new challenges and research questions will arise. This is matter of our future work, together with an analysis of acceptance of such awareness system among the social network's community.
Acknowledgments. This work was supported by the Deutsche Forschungsgemeinschaft (DFG) under grant No. GRK 2167, Research Training Group "UserCentred Social Media".