[Date Prev][Date Next][Date Index]

talk announcement Mon, 27 January 2014 - Alessandro Provetti "Analysis of heterogeneous networks of humans and cultural objects: first results"



Dear all,

the Institute for Information Systems, Database & Artificial
Intelligence Group cordially invites you to the following talk:


======================================================================

Speaker: Alessandro Provetti

Deptartment of Mathematics and Informatics,
University of Messina (Italy)

======================================================================

DATE:   Monday, January 27, 2014
TIME:   13:30 c.t.
VENUE:  Seminarroom 187/2
        (Favoritenstrasse 9-11, stairs 3, 2nd floor)

======================================================================

TITLE: Analysis of heterogeneous networks of humans and cultural
objects: first results

======================================================================

ABSTRACT:

With this seminar we would like to introduce you to the conceptual
framework and the research results we obtained in Messina on analysing
some of the user-generated content now available from Online Social
Networks (OSNs). We will describe how, starting from research in Web
data
extraction, we have become interested in different issues that are now
becoming of great interest, in view of the glory (so to speak) and
almost-ubiquity of OSNs and of their ever-increasing base of
content-generating users.

We will begin with the extraction and analysis of [snapshots of] the
Facebook friendship graph: what can (still) be done? How to study FB
friendship and its evolution? We will describe the main features of
two (large samples) we extracted from Facebook by applying two
different sampling strategies. Extracted samples have been studied by
applying methods which are largely accepted in the field of Complex
Network Analysis (vertex degree distribution, clustering coeffcient,
diameters and so on).

Second, we will cover the topic of community detection inside OSN, a
problem of obvious relevance and notorious computational complexity.
We briefly glance at our solution, the CONCLUDE algorithm, and argue
for its effectiveness and accuracy. Our results are twofold: on one
hand we designed randomized algorithms to weight network edges and
this tasks proves to be useful to improve the accuracy of the whole
community detection problem. On the other hand I will illustrate some
experimentas showing that our approach outperforms other, well-known
algorithms when applied on large, real-world OSN instances.

Finally, we will introduce our latest work on the aNobii network of
book-lovers (bibliophiles); we studied the intensity of a user's
participation to the SN in terms of i) joining groups (e.g., that on
French literature) or assigning tags to books they've read. We have
designed, implemented and validated a sampling algorithm that finds a
good approximation of the probability distribution of joint user
profiles. Our algorithm can be seen as an instantiation of the AA
meta-algorithm of Dagum, Karp et al. Its complexity is controlled by
the number of samples of a certain class it must find, even though the
number of iterations is not fixed a priori; the overall error is
bounded.

These results where obtained in a joint research effort with P. De
Meo, E. Ferrara, G. Fiumara and S. Catanese.

=====================================================================


With kind support of the Wolfgang Pauli Institut (WPI).