Figurative language detection using deep and contextual features
This thesis addresses the scarcity of research focused on deciphering the contextual meaning behind instances of Figurative Language (FL). Existing approaches often neglect the intricate contextual nuances by either relying solely on features extracted through deep learning architectures, abandon...
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2023
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/119806/ http://psasir.upm.edu.my/id/eprint/119806/1/119806.pdf |
| Summary: | This thesis addresses the scarcity of research focused on deciphering the contextual
meaning behind instances of Figurative Language (FL). Existing approaches often
neglect the intricate contextual nuances by either relying solely on features extracted
through deep learning architectures, abandoning the contextual essence, or resorting to
manually extracted features through rigorous processes, with limited exploration of
combinatory methods.
The research identifies a critical gap in the literature concerning the application of wellestablished
Machine Learning classification models, such as Support Vector Machine,
K-Nearest Neighbor, Logistic Regression, Decision Tree, and Linear Discriminant
Analysis, in the context of FL detection tasks. This study aims to bridge this gap by
conducting an in-depth exploration of the effectiveness of these models in discerning
Figurative Language instances.
Furthermore, the thesis critiques prior works employing manually crafted features for
Figurative Language detection, noting the lack of precision in identifying the most
crucial features. The research introduces a novel approach by combining features
extracted from a Convolutional Neural Network (CNN) model with manually extracted
features obtained from well-known lexicons. This integration aims to enhance the
robustness and accuracy of Figurative Language detection by leveraging the strengths of
both deep learning and traditional feature extraction methods.
The experimental design involves the use of a word-embedding technique, a CNN
model, and various well-known machine learning classification techniques. The study
not only investigates the efficiency of the proposed methodology but also delves into the
importance of individual features, providing precise insights and discussions on the
significance of lexicons used in the process. The findings of this research contribute to
the advancement of Figurative Language detection methods, offering a more nuanced
understanding of contextual meanings and paving the way for future research in this
domain. |
|---|