IEEE Technology and Society Magazine - March 2021 - 90

AI for a Fair, Just, and Equitable World

standardization, verifiability, and accountability of
these metrics. SIAMES recommender system aims to
promote the standardization of social impact indicators relying on real impact measurement reports.
In this way, we seek, in the long term, to provide
the same indicators for similar social projects, facilitating comparisons between different projects and
approaches to solving social problems.

Text mining, CBR, and ontologies
The combination of text mining and CBR is not
new [25]. However, we will provide a new approach
by applying ontologies for text mining as well as for
information extraction, selection of previous cases,
and problem resolution. The role of ontologies is
fundamental in this research, since in the sector in
which we are working-impact measurement and
social entrepreneurship-hardly any work has been
done concerning ontologies.
Semantic text mining refers to data mining processes that systematically incorporate domain
knowledge into the process [26]. Regarding ontologies, they are formal representations of knowledge
defining domain concepts and the relationships
between them. The results of our research concerning the semiautomatic ontology population task are
of interest because of the heterogeneity of the existing impact measurement reports, which makes the
task of automatic extraction of information difficult.
Given the availability of a great number of quality
impact measurement reports, it seems that applying
CBR may be a good option. CBR systems are based
on recalling past experiences and using them to
resolve current situations that are similar to those
already resolved and stored [27].
The goal of using text mining and CBR combination is to attempt to maximize successes in the
selection of previous cases. On the other hand, ontology-based semantic analysis tries to take advantage
of this approach, as we can see in [26].
Finally, SIAMES ontology draws inspiration from
research [12] and [13]. Konys [12] proposes an
ontology focused on sustainability, understanding
this, like social entrepreneurship, in three main
areas: 1) social; 2) environmental; and 3) economic.
It is an excellent definition of structure, very complete and broad. However, fundamental aspects
relevant to our work (such as indicators or proxies)
do not exist in this proposal. On the other hand,
although the ontology proposed by Li et al. [13] is

not the domain that concerns us, it is an ontological
definition of quality in which the indicators take on
special importance.
We can distinguish different tasks related to the
extraction of information for the creation of ontologies. Some of these tasks affect the structure of
the ontology while, in other cases, they try to add
instances to a defined ontology.
Ontologies can be defined manually, through the
work of experts or by integrating existing ontologies,
or they can be built through automatic or semiautomatic processes. These processes are known as
ontology learning. On the other hand, when these
procedures, instead of creating the ontology from
scratch, modify the structure of an existing ontology
(for example, adding more classes and relations) we
speak of ontology enrichment. However, when the
automatic or semiautomatic processing of sources to
extract knowledge only adds instances to the ontology, without modifying its structure, this is when we
call it ontology population.
To carry out these tasks, we need one or more
data sources from which to extract the information.
These sources are composed of structured information (among others, databases), semistructured (e.g.,
in tables within HTML documents) or unstructured,
as in the case of corpora of texts in natural language.
In the bibliography, we find several approaches
when dealing with the population of ontologies using
as a source a set of unstructured texts. In some cases,
focused on solving the problem for a specific domain-
such as biomedicine [28], [29], tourism [30]-while
others contribute a more general vision and seek to be
applied in any field of knowledge [31], [32].
In the studies of Petasis et al. [33], Ruiz-Martinez
et al. [30], and Lubani et al. [34], we find reviews
of the methods most used to face the population
of ontologies. Both Ruiz-Martinez and Lubani distinguish subtasks within the town process. Both
coincide in defining some main tasks, called initial
requirements, by Ruiz-Martinez and information
extraction engine in the case of Lubani, which are in
charge of extracting the concepts and relationships
from the sources. Subsequently, tasks are performed
to ensure consistency, remove ambiguities, and
incorporate the final results into the ontology.
In most researches, the extraction of information
for the population of ontologies has traditionally
been separated into two tasks: On the one hand, the
extraction of the instances that will become part of

IEEE Technology and Society Magazine

IEEE Technology and Society Magazine - March 2021

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