Turning Information Into
The Role of P2P Communication
*Science Dept. University of Siena
• Current simulation models of opinion dynamics (Duffuant et
el.,2001; 2002; Hegselmann and Krause, 2002) are based
on Social Impact Theory (Latané 1981; Nowak et al., 1990),
is said to depend on
distance, number, and strength (i.e.,
persuasiveness) of sources.
• Simulation-based studies of opinion dynamics observe how
opinions spread and aggregate as a function of the
distance among values assigned to them.
• But we know that social structures influence opinions,
more or less steadily
• Informational influence
(since Sherif, 1936) under
Agenda setting theory
(McCombs and Shaw,
1972): correlation between
frequency of information
delivered by media and
Let us see a recent
confirmation of this theory.
A case study.
Effect of media in the last Italian
(reproduced from Diamanti, 2008)
When speaking about opinion As in bounded confidence model
dynamics we must take agents’ (Deffuant et al),
• opinions are numerically defined
confidence into account (certainty)
• agents exchange opinions based
on the distance between their
values: they adjust opinions only
if preceding and received
+ information are close enough,
modelled by introducing a
parameter t for tolerance above
which opinions are resistant to
social scientists’ intuition and Agents are exposed to different
evidence gathered, that entities and force with variable
landscape of social influence influence
Receive inputs from distinct
is far from flat! sources of information
• In this paper we intend to explore the impact of different
interacting communication systems and information
sources on information quality
• Correlation between frequency of information delivered by
media and social perceptions?
• Does peer-to-peer communication amplify effects of old
media, or exercise independent influence
• Scale Free Network
• Agents are connected in a scale free network, in which
nodes are progressively added by introducing links to
the existing nodes on a “preferential attachment”
• The construction strategy of the algorithm aims at
maintaining the link probability between any couple of
nodes proportional to the number of existing links
already connected to the selected node.
• Bounded Confidence Model (BDM):
• Agents mix their opinions when differences is smaller
than threshold. More precisely…
Consider the set
• ml , mr represent values of events related to
welfare and security issues reported on by the
• V1 is the subset of agents that receive information
from the central media.
Agents’ preferences are set by a uniform random
Interacting peers, v V, are nodes of the Social
The more the agent’s opinions - with respect to
welfare and security - vl and vr approximate 1, the
more each issue is important for the agent.
• After broadcasting, each agent communicates with
neighbors within a distance set to 1.
• Following BCM convention if the difference between
two agents’ opinions - respectively represented by x
and xi - is lower than threshold (x−xi < t) these
opinions will be mixed by applying:
• To reproduce Italian central media in the
last political campaign, media are
assumed to deliver false information.
• The closer a reported information is to the
opposite information spread by central
networks, the higher its quality.
media and peers
• Peers acquire information from media according to a
passive protocol, by acquiring the values they send and
comparing them with their previous preferences.
• Information is accepted or not, based on bounded
• The agent’s preferences vl , vr and the information from the
media mld , mrd are transformed in two new agent’s
preferences. The function generates two new values for vl ,
• t stands for peer agents’ tolerance, i.e., subjective
disposition to accept others’ information. The higher the
value of t, the higher the agent’s disposition to accept
• Baseline experiment with nine scenarios, with
number of agents set to 100, no media
broadcasting and increasing levels of tolerance
(from 0.1 to 0.9 at step 0.1) for 100 turns.
Simulation is performed 10 times per scenario,
and results are averaged.
At beginning opinions
(welfare and security)
are set up randomly
interval ]0, 1[: both
around average value
of initial distribution,
meaning that, over a
scale free network,
leads to a ﬂat
How about mixed communicatiion?
• How do they interact? In
particular, does P2P
communication amplify or inhibit
the effect of central media?
• MB drives opinions by steering information among agents. It represents a
fundamental medium for knowledge diffusion.
• The wider the audience reached by the broadcasting system, the stronger
its inﬂuence especially when people are poorly self-conﬁdent and more
likely to accept incoming information. False information spreads fast and
• Is there any way to contrast such an inﬂuence?
• Peer-to-peer communication over a scale-free network can inhibit and
slower invasion of informational lemons.
• With reasonable level of tolerance, P2P communication inhibits
effects of MB until this has reached the 40% of the audience.
• P2P communication, thanks to reciprocation, allows self-conﬁdence
to act as an efficient ﬁlter of information.
Level of conﬁdence is not enough: effect of other types of
mentall states on opinion dynamics (beliefs, doctrines,
ideologies, ideals, etc.). What is an opinion? Whatʼs the
difference from beliefs?
More sophisticated mechanisms of belief/opinion formation,
revision and transmission, focusing on oneʼs representations of
other s beliefs;
Different types of P2P communication (for example, reporting
oneʼs Vs othersʼ beliefs);
Wise agents, i.e. a subset of agents having direct access to
Social structuresʼ properties, possibly implementing one real-
world network, and checking effects of speed in P2P