Transcript An Epidemiological Model for Semantics Dissemination

International Mobile Multimedia
Communications Conference 2007
An Epidemiological Model for
Semantics Dissemination
Christos Anagnostopoulos1,
Evangelos Zervas2,
Stathes Hadjiefthymiades1
1 National and Kapodistrian University of Athens, Department
of Informatics and Telecommunications, Pervasive Computing
Research Group
2 TEI of Athens, Department of Electronics
MobiMedia 2007, Nafpaktos, Greece
Collaborative Context-Aware
Computing

Context: the current values of specific ingredients that represent an activity of
an entity,

Awareness: understanding of the activities of an entity,

Context-Awareness: the ability of user applications (system) to discover (sense
and interpret) and react to changes in the environment they are situated in,

Collaborative Context-Awareness: an understanding of the activities /
conditions / environmental parameters of neighboring nodes that, consequently,
provides a more enhanced context for an individual.

Collaborative Context-Aware Applications:



generate inferred knowledge needed by the rest of the group,
adapt information dissemination algorithms, and,
exploit the ways in which users’ behavior coincides with their interests.
MobiMedia 2007, Nafpaktos, Greece
Contextual Information
analogous to Epidemic

Disseminated contextual information could match an epidemic:


a mobile node carrying a temporal valid piece of information
content becomes infectious;
otherwise it is susceptible.

Infectious node: disseminates information to its neighboring node
according to mobile context (e.g., network connectivity) and
interest (e.g., profile).

Epidemiological Model: Susceptible-Infected-Susceptible (SIS)
MobiMedia 2007, Nafpaktos, Greece
Semantics-based Dissemination

Epidemics are semantically related.

Pieces of context are hierarchically structured


from the most abstract to the most specific context [e.g., Soul is-a
Rhythm_and_Blues is-a Blues (music genres)]
Temporally valid pieces of context (e.g., recently sensed context
better interprets the depiction of a nodes' environment than least
sensed or obsolete context)


the knowledge derived from the most specific context implies also the
knowledge derived from the most abstract one
A node autonomously deduces whether the incoming epidemic
refers to context that adequately matches to a node’s interest or
not.
MobiMedia 2007, Nafpaktos, Greece
Epidemical Transmutation

More specific context refers to a stronger epidemic.

A node infers the context specificity through a semantic reasoning process:



The strongest epidemic infects a large portion of the group.
The weakest epidemic infects a small portion of the group.
Semantically-dependent epidemics through semantic relations in conceptual
hierarchies can transmute to stronger ones
(metallaxis in Greek).


Double-epidemical spreading:

Portion of the population is infected either with epidemics or with transmutations.

Corresponds to the heterogeneous need of each node, as required in the collaborative
context-aware systems, i.e., not all nodes is interested in the same context.
Proposed Epidemiological Model: Susceptible-aInfected-Susceptible (SaIS)


A node can be re-infected with a more stronger epidemic thus aggravating its condition
A node can be fully or partially cured
MobiMedia 2007, Nafpaktos, Greece
State Transitions in SaIS
A node transits between:
•Infection state p1,
•Infection state p2 (a transmutation of p1),
•Susceptible state p0.
infection
infection
p0
full cure δ10
aggravation
p1
partial cure δ21
full cure δ20
MobiMedia 2007, Nafpaktos, Greece
p2
Probability of aggravation
Given the status of the neighbors of node i at time instant
t and the fact that node i may be infectious at state pk, at
the next time instant t + 1, node i will be infectious at a
higher state pl with probability Qkl
•The probability that all the neighboring nodes being in a state
greater than l will not infect node i,
•The probability that one or more nodes will infect node i at infection
level l, and,
•The node i will not recover.
MobiMedia 2007, Nafpaktos, Greece
Analytical & Simulation Results

2D lattice – Homogeneous
network, M = 10,000,




β = 0.2,
δ10 = δ21= 0.1,
δ20 = 0.01
Since nodes reason about
more specific knowledge
then, they are re-infected
with the strongest epidemic
assuming that the latter
matches better to their
interests.
MobiMedia 2007, Nafpaktos, Greece
Analytical & Simulation Results



β = 0.2,
δ10 = δ21= 0.1,
δ20 = 0.6

If δ20 is relatively larger than β,
(e.g., a minor portion of nodes are
capable of reasoning) the
propagation process for the
strongest epidemic decays.

The weakest epidemic cannot
transmute to a stronger epidemic
due to the limited reasoning
capability of the majority of nodes

The infection of the strongest
epidemic p2 depends highly on
the fact that


at least one node is capable of
inferring p2 from p1, or,
at least a node is infected with p2
at the beginning of the process.
MobiMedia 2007, Nafpaktos, Greece
Thank you
Christos Anagnostopoulos
[email protected]
Pervasive Computing Research Group
http://p-comp.di.uoa.gr
MobiMedia 2007, Nafpaktos, Greece