A semantic framework for discovering casual relationships
The explosive growth of information at a mind-boggling scale has become an emerging phenomenon of our times. Discovering knowledge from a vast pool of resources is expected to remain a major challenge. In this respect, the extraction of semantic relations then becomes an important research area....
محفوظ في:
| المؤلف الرئيسي: | |
|---|---|
| التنسيق: | أطروحة |
| اللغة: | English |
| منشور في: |
2015
|
| الموضوعات: | |
| الوصول للمادة أونلاين: | http://ir.unimas.my/id/eprint/9287/1/Amaal%20Saleh%20Hasan%20Al-Hashimy%20ft.pdf |
| الوسوم: |
إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
|
| الملخص: | The explosive growth of information at a mind-boggling scale has become an emerging
phenomenon of our times. Discovering knowledge from a vast pool of resources is expected to
remain a major challenge. In this respect, the extraction of semantic relations then becomes an
important research area. While the extraction of ontological relations has been widely explored,
the discovery of non-taxonomic relations is still a major bottleneck. Current approaches tend to
predominantly employ syntactic approaches and rely largely on extensive manual efforts in the
construction of linguistic resources. Our literature review has revealed major gaps in terms of the
extraction of non-taxonomic relationships, particularly when it comes to implicit relationships.
As a response to this problem, our research then explores a semantic approach for addressing the
discovery of non-taxonomic relations such as causal relationships.
Based on an empirical study of causality theory and related works, we have formulated a
semantic approach for extracting causal patterns in text. The proposed framework incorporates a
novel causality sense extraction method, “Purpose Based Word Sense Disambiguation”, together
with a context-specific approach, “Graph based Semantics”, for uncovering causality structural
patterns. Our approach has produced a set of causality features that is even able to highlight
implicit causality patterns. We have employed benchmark data sets of SemEval 2007 and
SemEval 2010 data sets together with standard linguistic resources such as WordNet, SemCore
and XWNGloss in producing a series of intermediary linguistic resources as building blocks of
the framework.
A new qualitative measure for determining causal patterns has been formulated and used in
conjunction with a gold standard for validating the significance of the findings. We have
employed the C5.0 classifier to evaluate the effectiveness of the causality patterns as derived
v
from the framework. We have demonstrated via the realization of the framework, a purely
semantic approach is possible without the need for extensive manual efforts. This research will
serve as a key milestone and basis for ensuing discovery of non-taxonomic semantic relations
such as causality. |
|---|