Biology Reference
In-Depth Information
the system may have missed, to make the final decision about vague
reports, and to link together disparate events. Processes 1 through 6
make human judgments quicker, cheaper, and more reliable through
data search and visualization of the database of mined facts.
8. Feedback and improvement is the last of the processes. After deploy-
ment, text mining systems require maintenance and updates in order
to keep up-to-date on new vocabulary, document types, and sources
of information, as well as to correct for misunderstandings.
15.2.1 gPHiN (Public Health Canada)
The Global Public Health Intelligence Network (GPHIN) is a secure, contin-
uously accessible (24 hours a day, 7 days a week) public health surveillance
system focused on civil emergency detection, risk assessment, and response
(Mawudeku et al. 2007). The system combines both automated and manual detec-
tion methods, and is run from within the Centre for Emergency Preparedness
and Response (CEPR) under the Public Health Agency of Canada (PHAC). In
operation 11 years, GPHIN has expertise in both the technical and managerial
aspects of emergency response, offering practical support for Canadian pro-
vincial and central government agencies in the area of first response and public
health security. Since its beginnings in the mid-1990s, GPHIN has developed a
close partnership with the WHO to evaluate unsubstantiated news reports of
public health significance (Mykhalovskiy and Weir 2006).
Language coverage in GHPIN aims to maximize global media coverage, and
includes the six official working languages of the United Nations—Arabic,
Simplified Chinese, English, French, Russian, and Spanish, in addition to
traditional Chinese, Portuguese, and Farsi. In terms of health threat cover-
age, GPHIN takes an all risks approach and has taxonomic coverage for most
human and animal diseases.
Data sourcing comes from two major commercial aggregators: Factiva
(most languages) and Al Bawaba (for Arabic). Additionally, Farsi is sourced
manually from the Web by human analysts. GPHIN analysts also review
output from MedISys and the EMM (Europe Media Monitor) in the European
Union regularly.
The GPHIN system combines both automatic and human analyses with
input from approximately 14 analysts whose backgrounds cover a range of
disciplines (e.g., the life sciences, journalism, and economics). The automatic
component incorporates a highly developed, multilingual taxonomic classifi-
cation system that performs document filtering and ranking. Human analysts
rank news items according to their domain expertise and understanding of
the regions they cover, as well as source additional news sources and edit
machine translation system output when quality falls below standard thresh-
olds. Output to the end user is provided by a secure Internet portal in the form
of alerts, biogeographic maps, and a searchable list of ranked news articles.
