Skip to content

Latest commit

 

History

History
63 lines (41 loc) · 2.83 KB

README.md

File metadata and controls

63 lines (41 loc) · 2.83 KB

BART+BERT Ensemble ABSA Model for Indonesia Tourism Attraction Review

Overview

This repository contains the code and resources for final research project that combines BART and BERT models using the ensemble method. The main goal is to improve Aspect-based Sentiment Analysis (ABSA) model performance for the Indonesia’s Tourism Destinations dataset (Study case: Borobudur and Prambanan temples).

Proposed Model Architecture

Proposed Model Architecture

BART+BERT Ensemble Model

  1. Model Components: We utilize a pretrained model BARTLARGE model for BART and a BERTBASE Uncased model for BERT. These models are fine-tuned on the dataset for enhanced the performance.
  2. Ensemble Library: The ensemble transformers library is employed to seamlessly combine the two models.
  3. Dataset: The dataset used for fine-tuning can be found at this GitHub repository.

Fine-Tuning and Customization

  1. Fine-Tuning:
    • BART: Fine-tuned using the the dataset
    • BERT: Similarly, fine-tuned on the dataset
  2. Layer Modification:
    • For each fine-tuned model (BART and BERT), the final layer (usually known as head) is modified by adding a linear layer.
    • The ensemble model predicts by averaging the class probabilities from both models.

Model Interface

Take a peek at what the Streamlit app interface looks like: web-overview

Demo

Check out the video demo of the model below:

Watch the video

Model and Resources



🧞‍♂️ TASI-2324-118

1. 12S20024 - Sandro Sinaga
2. 12S20042 - Mastawila F. Simanjuntak
3. 12S20048 - Jevania

Supervisor I :  Samuel I. G. Situmeang, S.TI., M.Sc.
Supervisor II:  Sarah R. Tambunan, S.Kom, M.Sc.

© Information System Study Program, Faculty of Informatics and Electrical Engineering, Institut Teknologi Del, 2024

🧚🏼‍♀️ Spread the word!

If the information from this repo was useful to you in any way, make sure you give it a star 🌟, this way others can find it and benefit too!

Do you have any suggestions on how we could improve this project overall? Let us know! We'd love to hear your feedback!🙆🏼‍♀️