Bushra-KB/GMSATD

This repository contains the MSc thesis project titled "A Generic Multitask Summarizer for Amharic Text Documents". The project addresses the challenges of information overload and automatic text analysis by providing a versatile and parameterizable framework for extractive text summarization. # GMSATD 1.0 This is my MSc thesis project. Currently, the source code is private as it has yet to be presented for the final defense. I will make all files available once I have presented and received the necessary permissions from the school. Stay tuned, and thank you for your understanding! ## Title _"A GENERIC MULTITASK SUMMARIZER FOR AMHARIC TEXT DOCUMENTS"_ ## Abstract The thesis presents a generic multitask summarizer architecture for Amharic text documents, aiming to address the challenges of information overload and automatic text analysis in the digital era. The existing summarization methods are not able to adapt to the specific needs of the user or the document being summarized. They are limited to a single summarization approach to generate summaries, which can lead to suboptimal results. The architecture is designed to accommodate three main summarization tasks and incorporates multiple extractive-based approaches, providing a versatile framework for implementing various extractive text summarization methods. Additionally, the thesis introduces two novel graph-based summarization methods, extending the prominent algorithms TextRank and LexRank. The research investigates the impact of different term weighting methods, similarity measures, and feature weightings on the performance of the proposed summarization system, utilizing datasets collected from diverse Amharic news websites. Prototypes for frequency-based summarization (FBSumer), feature extraction-based summarization (FEBSumer), and graph-based summarization (GraphSumer) have been developed and evaluated using the Java version of the ROUGE toolkit (n-gram) to measure precision, recall, and F-measure. Comparative analysis with related methods in automatic text summarization demonstrates the substantial improvements achieved by the proposed methods. Overall, this work contributes a unique architecture and novel methodologies, advancing the field of automatic text summarization for Amharic text do …