Trustworthiness Management in
the Social Internet of Things
Michele Nitti, Roberto Girau, and Luigi Atzori
Presented by: Chris Morrell, 24 April 2014
Introduction & Background
The Proposed Solution
The Subjective Trust Model
The Objective Trust Model
Experimental Results
What is the SIoT
A social network where every node is an object
capable of establishing social relationships with
other things in an autonomous way according to
rules set by the owner.
Address the uncertainty of trust regarding other
nodes and to suggest strategies to establish
trustworthiness among nodes.
SIoT Relationships
• Relationships
– Parental Object Relationship (POR)
• Built in same period, by same manufacturer
– Co-Location Object Relationship (CLOR)
• Live in a common location (cohabitation)
– Co-Work Object Relationship (CWOR)
• Works in a common location (work)
– Ownership Object Relationship (OOR)
• Belong to the same owner
– Social Object Relationship (SOR)
• Objects that come into contact due to owner relationships
SIoT Architecture
• SIoT Architecture
– Relationship Management
• The intelligence that allows objects to start, update,
and terminate relationships
– Service Discovery
• Find the object that provides the desired service
– Service Composition
• Enables interaction among objects
– Trustworthiness Management
• How to know which services and objects to trust
Properties of Trustworthiness
Transitivity: Depends on the purpose
Composability: Combining recommendations
Personalization: Differing opinions matter
Asymmetric: See personalization
• The set of nodes:
P = {p1,..., pi ,..., pM }
• An undirected graph describing the network:
G = {P, E}
• A node i’s neighborhood
Ni = {p j Î P : pi , p j Î E}
• Common friends between pi and pj
Kij = {pk Î P : pk Î Ni Ç N j }
Notation (cont.)
• The set of services provided by pj
• The set of nodes that provides service h
Zh = {p j Î P : Sh Î S j }
• The set of edges that represents the path from
pi to pj
Rij = {pija pijb }
An Example Graph
• Nodes p1 – p10 where
node p1 is requesting
service S10
• Z10={p5}
• R1,5={p1p4,p4p8,p8p5}
• N1={p2,p3,p4} (Friends)
• K1,4={p2,p4} (Mutual
The Proposed Solution
Objects mimic human social behavior
Trust Models
• Subjective Trustworthiness
– Trust is local and based on experience of the local
node and its friends
– Trust is transitive, composable, personal, and
• Objective Trustworthiness
– Trust is network centric and is managed by PreTrusted Objects (PTOs) and based on experiences
of all nodes
– Trust is only composable (not transitive, personal,
or assymmetric)
Estimating Reputation
• Feedback (flij)
– Rate an experience from 0 to 1.
• Total Number of Transactions (Nij)
– Are the nodes artificially increasing ratings?
• Credibility (Subjective – Cji, Objective – Ci)
– Can we trust the ratings?
• Transaction Factor (wlij)
– Is the transaction relevant?
Estimating Reputation (cont.)
• Relationship Factor (Fij)
– How closely are the nodes connected
• Centrality (Subjective – Rij, Objective – Ri)
– How important is the node in the larger network?
• Computation Capability (Ij)
– How likely is it that the node will cheat?
Subjective Trustworthiness
Subjective Trustworthiness
• The trustworthiness of pj as seen by pi is:
– centrality of pj in pi’s “life”
– pi’s direct experience with pj
– Experience of pi and pj’s common friends
are weights (total weight must be 1)
Calculating Centrality
– the set of common friends between pi
and pj
• – the set of neighbors of pi
• Essentially, a ratio of common friends to
• Focuses centrality on the neighborhood,
rather than the entire network
Calculating Direct Opinion
Weighting of
Weighting of
Most Recent
Long Term
Remembering Opinions
The lengths of long and short term windows
Transaction weight factor
Transaction feedback
Calculating Indirect Opinions
• Sums the credibility of all of the K peers’ direct
opinions where
• Factors are weighted between peers’ direct
opinions and centrality
Quantifying Trustworthiness
• The trustworthiness of pj as seen by pi is:
• And remembering asymmetry, we know that
• This only works if pi have a direct social
relationship (neighbors)
Trustworthiness for non-Neighbors
• The product of Trustworthiness of all nodes
along the path to the service provider
Providing Feedback to Neighbors
• If peer node was correct in its advice, then its
opinion is reinforced
• Feedback on neighbors is stored locally and
used for future trust evaluations
Objective Trustworthiness Model
Trustworthiness Storage
• Trustworthiness is stored in a DHT which is
accessible by all nodes
• Only Pre-Trusted Objects are permitted to
store data in the DHT
Objective Trustworthiness
• The objective model removes direct and
indirect opinions
Calculating Centrality
• Q j – The number of times pj requested a service
• Aj – The number of times pj acted as an
intermediate node in a transaction
• Hj – The number of times pj provided a service
• A node is central if it is involved in many
transactions (not just as a requester)
• Centrality is now network wide
Remembering Opinions
Transaction weight factor
Transaction feedback
Considers feedback from all
Nodes that interacted with pj
Objective Credibility
• Considers Trustworthiness (Ti), Relationships
(Fij), Intelligence (Ij), and Number of
Transactions (Nij)
• Higher intelligence, stronger relationships, and
many transactions are assumed to be
indicators for collusive malicious behavior
Experimental Evaluation
Simulation Setup
• Small World In Motion model and a Brightkite
social network dataset
• Each human owns a set of things connected to
the SIoT. ½ of their things are with them as
they move
• Nodes may be benevolent
and cooperative or malicious
(depending on relationship
and intelligence)
Simulation Parameters
(and optimal configuration)
Varying Thing Types (SWIM)
Malicious Nodes are only
Class 2 (Sensor/RFID)
Malicious Nodes are only
Class 1 (Intelligent Devices)
Varying Thing Types (BrightKite)
Malicious Nodes are only
Class 2 (Sensor/RFID)
Malicious Nodes are only
Class 1 (Intelligent Devices)
Varying the Percentage of
Malicious Nodes (BrightKite)
Subjective Trustworthiness Model
Objective Trustworthiness Model
Dynamic Behavior
• Subjective model has a slower response to
• Subjective model is immune to malicious
actors who vary their behavior based on

similar documents