The notion of "social networking" is known in sociology since 1954. Since the appearance of
online social networks (Facebook, iWiW, LinkedIn, etc.) a great ammount of information
is available for researchers of social networking.
However, this information can also be misused if obtained by unauthorized parties:
advertising agencies or spammers may send targeted advertisements or spam, the govern-
ment may observe the field "political view" on Facebook, and further fields like "religious
views" or "sexual orientation", etc. might reveal sensitive information. Thus the need for
examining privacy in online social networks.
In my master thesis I dealt with the question whether unknown user attribures could be
inferred from the information available in the social network. I pointed out that approxi-
mation or classification of the unknown attributes is possible, which implies user privacy
depends not only on the user himself, but also on other members of the social network,
I inferred the unknown attributes of a user from his public attributes and his friends’
attributes. This work covers the problem of approximation and classification also. I used
neural network in both cases. The selection of the input variables of the neural network
was based on the correlation with the target variable. This way the method can work in a
fully automated way.
The inferring method presented was tested with three attributes of iWiW users: "age",
"gender" and "marital status". The results showed that my method gave better results
than naïve methods, so it can be used for inferring unknown user attributes.