Graduation Year

2024

Document Type

Thesis

Degree

M.S.C.S.

Degree Name

MS in Computer Science (M.S.C.S.)

Degree Granting Department

Computer Science and Engineering

Major Professor

Seungbae Kim, Ph.D.

Committee Member

John Licato, Ph.D.

Committee Member

Shaun Canavan, Ph.D.

Keywords

Custom Feedback, Fine-tuning, Prompting, Zero-Shot

Abstract

The advent of text data from social media, blogs, movie reviews, and other textual sources has opened new avenues for research, particularly in the domain of Author Profiling. Author Profiling helps in Capturing the Stylistic features and also useful for analyzing the required elements in the written text. This Study addresses one of the tasks in Author Profiling which is termed as gender detection or Classification of Gender from Text. The main goal of this research is to obtain valuable and relevant gender characteristics that will accurately classify the Author’s gender of a review extracted from an Anime Review website. This Research uses the current State of Art Large Language Models to Automatically Capture the Gender Differences. The data is processed through the proposed Method which uses both Custom Prompting along with Fine-Tuning in order to tweak some of the weights associated with Large Language Models(LLMs). Once the LLM gets Fine-Tuned, the Model is tested with unseen Review datapoints, subsequently the Testing prompt is modified over the testing process through feedback mechanism proposed in the testing phase. Also, the Error Analysis is demonstrated through the Feedback obtained from the LLMs. Furthermore, the model surpasses existing baseline methods in accuracy, as evidenced by comparative analysis. This study contributes to the broader field of author profiling by presenting an effective model for gender detection and a thorough error analysis, highlighting potential areas for future enhancements and practical applications.

Download

Share

COinS