IEEE Technology and Society Magazine - March 2021 - 92

AI for a Fair, Just, and Equitable World

Stage I: Ontology definition to support
text mining
The first phase that we will face will consist of the
definition and development of an ontology that will
support the subsequent text mining process. This
ontology is an essential part of our research, so this
stage is crucial for the final results of our work.
Looking at existing indicator sets, such as the indicators collected by Bond for his Impact Builder tool7 or the
Global Impact Impact Reporting and Investment Standard (IRIS) database Investing Network,8 we observe several aspects that will help us define our ontology.
Although these indicators constitute a magnificent
compilation, their specification was based on a previous study of several reports and gave each of them a
wording that covers several of the indicators, so that
these indicators would never appear in an actual impact
measurement report. However, its study is of interest to
us to identify potential classes of our ontology.
For instance, looking at Bond set, we find indicators such as # and % of children enrolled in [primary/
secondary] school who are of [primary/secondary]
school age. First of all, usually the impact would be
measured using one of the two metrics, either percentage or number (or both, but separately), so in a
real measurement report, we would found indicators
such as number of children enrolled ... or percentage
of children enrolled. Another example, if we look at
7
8

https://www.bond.org.uk/resources/impact-builde
https://iris.thegiin.org/

the indicator # and % teachers who adapt lessons
to the needs and abilities of learners in their classes
(e.g., learners with disabilities, language minorities,
and traditionally excluded groups), we see that several groups are being included in the same indicator.
In a project, it is usual for us to focus on a specific
group, so the wording of the indicators in real reports
would be something like number of teachers who
[...] of learners with disabilities [...] of percentage of
teachers [...] of ethnic minorities learners [...].
Furthermore, we have based our ontology on existing ones that we consider most related to our research:
those described in [12] and [21]. Our particular interest are the dimensions highlighted in [12] as basic as
well as the stakeholder concept defined in both.
In view of what has been stated so far, for the
moment, we have chosen to define a generalist
ontology defining both original and based on the
cited existing ontologies classes. The structure of our
ontology and its relationships with the referenced
ontologies are shown in Figure 2.
In summary, as a first approach, this ontology
meets our basic requirements: It notably covers the
desired scope, it is based on the previous studies
most relevant for our research, it contains useful
information on cases to be able to recover them and,
finally, it contains the key elements: measurement
indicators and, if available, their proxies and associated data sources.

Figure 2. SIAMES ontology structure. Prefix siames: labels our original classes, prefix
sus: labels [20] classes, and prefix ene: labels [21] classes.

IEEE Technology and Society Magazine

https://www.bond.org.uk/resources/impact-builde https://iris.thegiin.org/

IEEE Technology and Society Magazine - March 2021

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - March 2021