Skripsi

FINE-TUNING MODEL ROBERTA UNTUK SISTEM TANYA JAWAB BERBASIS KONTEKS

---

The increasing amount of text-based information on the internet often requires users to read many long documents to find the answer information they need. Question Answering (QA) systems provide a solution to this problem by developing QA systems that can deliver direct answers from text without users having to read entire documents. This research focuses on developing an extractive question answering system by fine-tuning the pre-trained RoBERTa (Robustly Optimized BERT Pretraining Approach) model. RoBERTa was selected for its superiority in understanding contextual text meaning through the elimination of next sentence prediction tasks and the use of dynamic masking on large-scale data. The dataset used is the SQuAD v1 dataset containing question-context pairs with corresponding answers. The fine-tuning process is divided into 3 scenarios consisting of combinations of learning rate and batch size configurations. Based on testing results, the highest F1-score and Exact Match (EM) were obtained in scenario 3, namely the model trained with a learning rate configuration of 3e-5 and batch size of 16, achieving an F1-score of 92.24% and EM of 85,84%. This approach demonstrates that the utilization of RoBERTa in question answering systems is capable of understanding sentence context effectively.

---

Availability

Detail Information

Series Title

-

Call Number

T1898142025

Publisher

Indralaya : Prodi Teknik Informatika, Fakultas Ilmu Komputer Universitas Sriwijaya., 2025

Collation

xv, 101 hlm.; ilus.; tab.; 29 cm

Language

Indonesia

ISBN/ISSN

-

Classification

006.350 7

Content Type

Text

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Natural language processing
Prodi Teknik Informatika

Specific Detail Info

-

Statement of Responsibility

MI

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment