Analisis Sentimen Multimodal Berbasis Aspek untuk Ulasan Wisatawan Menggunakan Metode Deep Learning

Jannah, Shofa Wardatul (2025) Analisis Sentimen Multimodal Berbasis Aspek untuk Ulasan Wisatawan Menggunakan Metode Deep Learning. Masters thesis, Institut Teknologi Sepuluh Nopember Surabaya.

Text 6026231002-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2027. Download (4MB) | Request a copy

Abstract

Aspect-Based Sentiment Analysis (ABSA) telah menjadi salah satu metode penting untuk mengevaluasi opini pengguna di berbagai platform online. Seiring berkembangnya teknologi dan meningkatnya penggunaan konten visual, analisis sentimen berbasis aspek kini tidak hanya melibatkan teks, tetapi juga gambar. Penelitian ini mengusulkan sistem Aspect-Based Multimodal Sentiment Analysis (ABMSA) yang menggabungkan data teks dan gambar untuk menganalisis sentimen terkait tiga aspek utama dalam pariwisata Indonesia, yaitu lokasi, harga, dan suasana. Sistem ini terdiri dari lima tahap utama: pre-processing data, pengembangan ABSA teks, pengembangan ABSA gambar, integrasi model multimodal, dan klasifikasi sentimen. Pada tahap pre-processing, teks dibersihkan dan ditokenisasi, sedangkan gambar dinormalisasi dan diaugmentasi. Pada tahap kedua, fitur teks diekstraksi menggunakan IndoBERT dan diproses dengan model encoder seperti BiGRU, LSTM, atau CNN-BiLSTM. Tahap ketiga menggunakan model CNN seperti ResNet-152, InceptionV3, dan DenseNet-121 untuk ekstraksi fitur gambar. Pada tahap keempat, fitur teks dan gambar digabungkan menggunakan teknik late fusion. Terakhir, klasifikasi sentimen dilakukan dengan memproses gabungan fitur untuk memprediksi sentimen pada setiap aspek. Hasil eksperimen menunjukkan bahwa penggabungan CNN-BiLSTM dengan ResNet-152 menghasilkan akurasi yang baik dengan rata-rata 0.98 untuk semua aspek, berkat residual connections yang mendukung penyelarasan fitur teks dan gambar. InceptionV3, meski unggul dalam pemrosesan gambar, kurang optimal untuk multimodal, sementara kombinasi RNN dengan DenseNet-121 memiliki performa lebih rendah dibandingkan analisis teks mandiri karena arsitekturnya kompleks lebih sulit untuk digabungkan dengan fitur teks.
=================================================================================================================================
Aspect-Based Sentiment Analysis (ABSA) has become an important method for evaluating user opinions across various online platforms. With the development of technology and the growing use of visual content, ABSA now involves not only text but also images. This study proposes a Aspect-Based Multimodal Sentiment Analysis (ABMSA) system that combines text and image data to analyze sentiment related to three main aspects in Indonesian tourism: location, price, and atmosphere. The system consists of five main stages: data pre-processing, development of ABSA for text, development of ABSA for images, multimodal model integration, and sentiment classification. n the pre-processing stage, text is cleaned and tokenized, while images are normalized and augmented. In the second stage, text features are extracted using IndoBERT and processed with encoder models such as BiGRU, LSTM, or CNN-BiLSTM. The third stage uses CNN models like ResNet-152, InceptionV3, and DenseNet-121 for image feature extraction. In the fourth stage, text and image features are combined using late fusion techniques. Finally, sentiment classification is performed by processing the combined features to predict sentiment for each aspect. Experimental results show that the combination of CNN-BiLSTM with ResNet-152 yields high accuracy with an average of 0.98 for all aspects, due to residual connections that help align text and image features. InceptionV3, while strong in image processing, is less optimal for multimodal tasks, and the combination of RNN with DenseNet-121 performs lower than text-only analysis due to its complex architecture, making it harder to combine with text features.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Aspect-Based Sentiment Analysis (ABSA), Multimodal Learning, Deep Learning, Natural Language Processing (NLP), Pariwisata Indonesia Aspect-Based Sentiment Analysis (ABSA), Multimodal Learning, Deep Learning, Natural Language Processing (NLP), Indonesian Tourism
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 59101-(S2) Master Thesis
Depositing User:	Shofa Wardatul Jannah
Date Deposited:	01 Feb 2025 15:43
Last Modified:	01 Feb 2025 15:43
URI:	http://repository.its.ac.id/id/eprint/117461

Actions (login required)

View Item