Page 166 - Proceedings of the 2017 ITU Kaleidoscope

2017 ITU Kaleidoscope Academic Conference

entity centric trust separately or in a collective manner. The
platform consists of several important modules such as Trust
Computation, Prediction and Decision Making (TCPD),
Trust Agent (TAg), Trust Data Access Object (TrustDAO),
Data Repository, Trust Computation and Decision making
module, Trust Service Enabler and API. Once the TCPD
identify a requirement of data, it ask the TAg via Trust DAO
to collect necessary information and preprocessed them for
trust evaluation. Then, these preprocessed data is stored in
the data repository to be used by other modules including
external platforms through TCPD API.

Afterwards, trust metric extraction module estimates the
necessary trust attributes based on the requirement. These
attributes can either be categorized as data centric attributes
as explained in Section 4.1 or traditional entity centric
attributes as described in Section 3. Next, all the attributes
are combined based on the REK model with the assistance
of trust computation module, which is capable of performing
the calculation based on either numerical methods or
artificial intelligence approach as described in sections 4.2
and 4.3 respectively. Finally, decision making and Fig. 3. A Data Trust Model
delegation module uses the predicted trust values in order to − Consistency ( ): the probability of valid,

complete the decision process perhaps with the support of accurate and unique records over total data records
service enabler who is actually perform the judgement made
by the decision module. In the following sections, we explain 4.2. Data Trust Computational Model
the data centric trust attribute estimation, data trust In this section, we extend our entity centric model in Fig. 1
computation and data trust prediction in detail.
to comply with the data centric trust as shown in Fig. 3 and
4.1. Data Trust Attributes explain how each DTA is combined to generate data centric
trust. For that we identify completeness, uniqueness,
Alongside with the REK model, we first consider a separate timeliness, validity, accuracy and consistency as DTAs
set of trust attributes which essentially define the properties which represent knowledge TM as it conveys trustworthiness
of data. Many research work on DQ shows that the six information about the trustee. On the other hand “success”
parameters (e.g., completeness, uniqueness, timeliness, DTA and “cost” DTA represent the experience DTM of the
validity, accuracy and consistency) provide prominent trustor after each task. Finally reputation DTM can be
insight for assessing the DQ matters as in [22], [25], [32]. considered by aggregating opinions of other trustees if there
With respect to trust notion, we can consider these properties are any. Based on this, basic data trust assessment towards B
as trustworthiness attributes. Further, we consider two by A ( ) over x DTM can be numerically modeled as

additional attributes, “success” and “cost”, which below:
characterize experience and reputation data trust metric
(DTM) calculation, in additional to aforementioned − Knowledge DTM ( )

attributes stated in Section 3. We consider these eight data

trust attributes (DTA) as the core dimensions in finding the = + + + +

trust between a data item and the trustor. Thus, we model + (1)

these properties as below:

− Success ( ) : the probability that B will where α,β,γ,δ,∈, and ε are weighting factors such

successfully execute the task that α+β+γ+δ+∈+ε=1. However, calculating these

− Cost ( ) : the probability that the cost of executing weighting factors are computationally costly and
not practical due to infinite number of possibilities.

the task by B is not more than expected Hence, we suggest to apply machine learning (ML)
− Completeness ( ): the probability of complete techniques to combine all TAs, which we have
data records over total data records discussed in our previous work [7].
− Uniqueness ( ): the probability of expected

records over total records noted − Experience DTM ( )
− Timeliness ( ): the difference between last = + (2)
update to the current one

− Validity ( ): the validity of data type, syntax and where σ and φ are weighting factors such that σ+φ

range =1 and >0. The ML method discussed in [7] is

− Accuracy ( ): the probability of accurate data preferable for TA combination in this case as well.

records over total data records
– 150 –