Event Modeling from Amharic News Articles
No Thumbnail Available
Date
2/4/2018
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
With the increase of online information overload available to us, everyday tasks such as extracting,
searching, understanding and relating relevant data have become intractable. A large percentage
of these information are discussing past, current, and future real world events. Events are dynamic
data structures that play a key role in understanding phenomena happing in real world, which are
basically driven by the four āWāsā (what, who, when, and where). This natural progression of
questions is a classic example of what one might ask about an event. This results in natural way to
explain complicated relations between people, places, actions and objects. Event centered
modeling captures the dynamic aspects of an event along with semantic representation of event
facts.
In this research work, we have proposed and developed event extraction and representation model
from Amharic news article. Event modeling involves key event identification, event elements
extraction, and event semantic elements representation. Event triggers tells the action taking place
in news article. To identify the mention of event in news article we used manually collected event
trigger words and phrases from various news domains. For event elements extraction, we used
named entity recognizer and other local features like potential trigger, event extent, path from the
extent to head word of the trigger. Machine learning Maximum entropy classifier is trained using
event related news article collected from Fana Broadcast Corporate news archive. For event
representation, we designed ontology based event representation model that provides deeper
semantic through event information representation.
A prototype showing an event extraction and representation is developed using different
programming environment. Evaluation of trigger identification and event elements extraction is
carried out by comparing manually tagged news article with the automatically extracted event
information by the system. The evaluation result shows that the trigger identifier module obtain
precision (67.1%) of event correctly which contributes to the better event elements extraction. The
event elements extractor component shows greater obtaining precision (69.1%) while event
classification module classify about (72%) of event correctly. The representative ability of our
event representation model is evaluated with respect to requirements and event dimensions we
covered in this work.
Description
Keywords
Event, Event Modeling, Event Trigger, Event Elements, Event Representation