Publication Date

2025

Document Type

Dissertation/Thesis

First Advisor

Alhoori, Hamed

Degree Name

M.S. (Master of Science)

Legacy Department

Department of Computer Science

Abstract

Traditional Topic Modeling approaches, as well as zero-shot, few-shot, and fine-tuned Large Language Models (LLMs), have struggled to generate topics alongside relevant text from diverse sources, particularly sections such as Limitations. This thesis investigates automated methods for analyzing and synthesizing key sections of scientific articles using LLMs, exploring multiple dimensions of scientific text analysis.

First, LimTopic is introduced as a method for extracting and modeling the limitations sections of research papers. By integrating LLM-based topic generation with BERTopic, the approach generates descriptive titles and concise summaries that highlight the boundaries and shortcomings of studies, ultimately guiding future research directions.

Second, visual-related limitations are addressed by generating detailed image descriptions from charts and graphs using multi-modal LLMs—including QWen, Llama, Llava, Pali-GEMMA, and GPT-4o. An evaluation framework employing an LLM-as-a-judge approach demonstrates that GPT-4o produces the most accurate and coherent descriptions compared to alternative models.

Finally, a novel method is proposed for generating suggestions for future work. Leveraging Retrieval-Augmented Generation (RAG) and an iterative LLM feedback mechanism, the method synthesizes meaningful research directions from key sections of scientific articles and related literature. Both qualitative and quantitative evaluations confirm that the RAG-based approach outperforms traditional techniques.

Overall, this work underscores the potential of LLM-driven methods to advance automated scientific exploration in the NLP domain, offering enhanced tools for understanding research limitations, improving visual content analysis, and inspiring future research initiatives.

Extent

108 pages

Language

en

Publisher

Northern Illinois University

Rights Statement

In Copyright

Rights Statement 2

NIU theses are protected by copyright. They may be viewed from Huskie Commons for any purpose, but reproduction or distribution in any format is prohibited without the written permission of the authors.

Media Type

Text

Share

COinS