SOCIAL SENSING - Managing and Mining Sensor Data - page 256

The PrivStats system computes aggregate statistics on user loca-
tion information and guarantees location privacy even in the face
of side information about user location and movement patterns. It
is also resistant to large amounts of spurious data upload by users.
Many applications require the computation of a specific function on the
data, and therefore, it is critical to design methods for computing the
function accurately on the perturbed data. For example, the problem of
privacy-preserving regression modeling of sensor data has been discussed
in [3].
5. Trust in Social Sensing
At the broadest level, social sensing systems can be considered multi-
agent systems, that interact with one another and provide a variety of
data-centric services to one another. Therefore, a number of issues of
trust arise in the context of such large-scale social-centric applications,
which are common to many traditional peer-to-peer applications [138].
Such issues typically deal with the the aspect of designing trustworthy
protocols for interactions between different agents, both in terms of the
choice of interactions, and the time of these interactions. A detailed
survey of the (more traditional) literature along this direction may be
found in [138]. The more recent social sensing work has focussed on the
data-centric aspects of trust, rather than the interaction-centric aspects.
The openness of participatory sensing systems provides them with a
tremendous amount of power in collecting information from a wide vari-
ety of sources, and distilling this information for data mining purposes.
However, it is this very openness in data collection, which also leads to
numerous questions about the quality, credibility, integrity, and trust-
worthiness of the collected information [45, 51, 71, 72]. Furthermore, the
goals of privacy and trust would seem to be at odds with one another,
because all privacy-preservation mechanisms reduce the fidelity of the
data for the end-user, whereas the end-user trust is dependent on high
fidelity of the data. Numerous questions may arise in this respect:
How do we know that the information available to the end user is
correct, truthful and trustworthy?
When multiple sources provide conflicting information, how do we
know who to believe?
Have errors been generated in the process of data collection, be-
cause of inaccuracy or hardware errors?
The errors which arise during hardware collection are inherent to the
device used, and their effect can be ameliorated to some extent by care-