IEEE Computational Intelligence Magazine - November 2020 - 58

MEDLINE dashboard and explorer tools, and further mined
through the available API.
One of the highly innovative technologies derived from the
research carried out to build the MIDAS platform is the MeSH
classifier [23]. It is an automated text classifier that has learned
over human hand annotation of MeSH classes from more than
25 million biomedical articles (part of the corpus was used for
evaluation) to perform the assignment of MeSH classes to any
given snippet of text. It uses advanced text mining algorithms
for this classification [19], and can classify any input text
(including news, reports and health records) with this wellaccepted health taxonomy. It is used to annotate scientific articles, news articles and reports relating to the COVID-19,
allowing for the utilization of the MeSH headings as search
topics over the corpus of these documents. It enhances the
searchable information and allows for data visualization modules where the user can see the health topics (based on the
most frequent MeSH classes) associated to the news over a
query. This classifier was evaluated over: (i) scientific articles,
part of the annotated MEDLINE dataset; and (ii) over news
articles, hand annotated by some of the health experts in
MIDAS on topics such as infectious diseases, diabetes, childhood obesity and mental health. The evaluation of the classifier
resulted in an F1 measure of 0.43 in the MeSH tree depth level
three for the classification of scientific articles, whilst F1 measures range between 0.55 to 0.85 for news articles in specific
health domains (including diabetes, mental health and infectious diseases). The details will be published in [27]. The MeSH
classifier offers a web portal and an API to enable a diversity of
usages and integrations in other solutions. The web portal
(accessible through the MIDAS COVID-19 toolset at the web
portal www.midasproject.eu/covid-19/) provides the positioning
of the MeSH categories that were assigned to the input text
snippet, their similarity percentage and the MeSH tree branches to which the class belongs. This classifier allows us to generate useful metadata (based on the MeSH categories assigned to
news articles, new research articles or medical reports) enabling
its usage in the MEDLINE explorer and dashboard described
previously. This explorer is served with an API that allows
access to the structured Coronavirus dataset, and that can be
enriched with other reports and annotated with the MeSH
classifier. This allows researchers to leverage the existing knowledge generated in the current research.
Worldwide News Monitoring

The MIDAS news monitoring dashboard is fed by Newsfeed
technology, collecting and analyzing more than 100 thousand
news articles daily in real-time through the Event Registry
technology, offering the MIDAS user insightful data visualizations to explore health-related news [17]. Since January 2020,
the MIDAS news engine collected more than 13 million
news articles on coronavirus-related topics across more than
60 languages. These included over 120 thousand articles on
COVID-19 and diabetes, more than eight thousand articles
on COVID-19 and retirement homes and elderly care, over

IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | NOVEMBER 2020

18 thousand articles on COVID-19 and obesity, more than
116 thousand articles on COVID-19 and mental health, and
approximately 191 thousand articles on COVID-19 related to
nursing. The news explorer within MIDAS allows the user to
explore the overall sentiment of the news and the categories
associated with it. From the total amount of collected coronavirus articles, approximately 36% have a positive sentiment and
0.69% relate to patient education whilst 0.71% relate to testing
facilities. Recently, IRCAI released a worldwide news monitoring dashboard dedicated to COVID-19 based on the same
news engine [30]. This general purpose health news monitoring dashboard exhibits the news on the epidemic outbreak in
real-time and allows the reader to explore the information provided by country. However, the user cannot customize the
news feed except using preset filters. MIDAS improves the
usability of that by providing a news stream (see the visualization module at the center of Figure 1) where the user can personalize the search query and even include blogs. It allows
further exploration of COVID-19 related news specific to topics on the user's own health policy priorities, such as home care
or childhood obesity. It includes a tag cloud to have a first grasp
over the main topics under discussion. Alongside this useful
tool, the MIDAS news dashboard [25] allows the user to further explore the news based on data visualization modules
including related concepts, entities and categories, or even the
sentiment of the news article selection. This analysis is particularly important to avoid bias in the health news search [26], and
to explore the several dimensions of misinformation caused by
the infodemia [37], in conjunction with the pandemic. The
search engine uses Wikipedia terms, to ensure the multilingual
potential of the dashboard. These include the following
COVID-19 related terms: "Coronavirus" (on the Coronavirus
virus family, available in 73 languages), "Coronavirus disease
2019" (corresponding to the specific COVID-19 sort, available
in 136 languages), and "2019-20 Coronavirus pandemic" (that
writes on the pandemic itself, available in 132 languages).
Moreover, we can backtrack the news articles about Coronavirus in Italy, discussing the triage of passengers commuting from
Wuhan, China, in the timeline exploration visualization module, to January 20th this year, at the beginning of the European
epidemic. We can further access these articles' entities to identify main actors and related topics. We can also explore the sentiment over these news articles and, in some cases, their impact
on social media through the number of times a particular news
article was shared. All these features are offered over comprehensive and well documented APIs.
The usefulness of the MeSH classifier, described in the latter
section, is extended in the MIDAS platform through its integration with the news dashboard. With this integration, the
user can use the MeSH heading terms together with keywords
in a query, when exploring a certain news topic, in a similar
fashion to the usage of the well-accepted PubMed workflow,
providing data visualization modules that include those classes.
A meaningful example is the visualization module "Article Categories" where a MIDAS user can see the distribution of news

http://www.midasproject.eu/covid-19/

IEEE Computational Intelligence Magazine - November 2020

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - November 2020