Analisis Sentimen Berdasarkan Aspek Pada Review Hotel Menggunakan Probabilistic Latent Semantic Analysis, Word Embedding, Dan LSTM

Khotimah, Dewi Ayu Khusnul (2019) Analisis Sentimen Berdasarkan Aspek Pada Review Hotel Menggunakan Probabilistic Latent Semantic Analysis, Word Embedding, Dan LSTM. Masters thesis, Institut Teknologi Sepuluh Nopember.

Preview	Text 09211750053002_Dewi Ayu K.K.pdf Download (18MB) | Preview
	Text 09211750053002-Master_Thesis.pdf Download (18MB)

Abstract

Di era industri 5.0, ulasan produk sangat penting untuk keberlanjutan perusahaan. Ulasan produk adalah fitur User Generated Content (UGC) yang menggambarkan kepuasan pelanggan secara keseluruhan. Kepuasan pelanggan dapat diukur berdasarkan aspek yang mempengaruhinya. Penelitian ini menggunakan 3 pendekatan untuk kategorisasi aspek (AC) dan klasifikasi sentimen (SC) berdasarkan lima aspek hotel. Kelima aspek hotel diambil dari website Traveloka meliputi Location, Meal, Service, Comfort, dan Cleanliness. Penelitian ini hanya mengambil ulasan produk yang menggunakan bahasa inggris. Setiap ulasan produk, akan di preprocessing menjadi dokumen term list. Dokumen term list, digunakan sebagai input data awal untuk keseluruhan proses penelitian yang menggunakan bahasa pemrograman Python. Tiga pendekatan dalam kategorisasi aspek (AC) dan klasifikasi sentiment (SC) digunakan untuk menguji performa terbaik dari beberapa metode yang diusulkan. Penelitian ini mengusulkan metode Probabilistic Latent Semantic Analysis (PLSA) yang digunakan untuk menghasilkan hidden topic. Metode Semantic Similarity digunakan untuk klasifikasi topic ke dalam lima aspek hotel. Adapun perluasan term list pada saat pengukuran similarity menggunakan metode Term Frequency-Inverse Cluster Frequency (TF-ICF). Penelitian ini juga mengusulkan metode Word embedding untuk mendapatkan nilai vector pada metode deep learning dari Long Short-Term Memory (LSTM) untuk klasifikasi sentimen. Hasil penelitian menunjukkan bahwa pendekatan AC3 (metode PLSA + TF ICF 100% + Semantic Similarity) memiliki nilai yang lebih unggul yaitu 0,839 dalam kategorisasi aspek; Pendekatan SC1 (metode Word Embedding + LSTM) mengungguli dalam klasifikasi sentimen dengan nilai 0,940; aspek service memiliki nilai sentimen positive yang lebih tinggi yaitu 45,545 dibandingkan aspek yang lain; aspek comfort memiliki nilai sentimen negative yang lebih tinggi yaitu 12,871 dibandingkan aspek yang lain. Hasil penelitian dapat disimpulkan bahwa service hotel yang baik tidak menjamin kenyamanan pelanggan. Aspek hotel yang sama dapat memiliki nilai sentimen yang berbeda, hal ini dibuktikan dengan analisis sentimen yang dapat dipengaruhi oleh aspek.
==================================================================================================================
In the industrial era 5.0, product review is necessary for corporate sustainability. Product review is a User Generated Content (UGC) feature that describes overall customer satisfaction. Customer satisfaction can be measured based on a number of aspects. This study uses 3 approaches for aspect categorization (AC) and sentiment classification (SC) based on five hotel aspects of the hotel. Those five hotel aspects were obtained from the Traveloka website including Location, Meal, Service, Comfort, and Cleanliness. This study only took product reviews in English. Each product review underwent preprocessing to be a term list document. The term list document was used as the initial data input for the entire research process using the Python programming language. Three approaches of aspect categorization (AC) and sentiment classification (SC) were used to test the best performance of the proposed methods. This study proposed the Probabilistic Latent Semantic Analysis (PLSA) method to produce hidden topics. The Semantic Similarity method was used to classify the topic into five hotel aspects. The expansion of the term list during similarity measurement used the Term Frequency-Inverse Cluster Frequency (TF-ICF) method. This study also proposed the Word embedding method to obtain vector values in deep learning method of Long Short-Term Memory (LSTM) for sentiment classification. The result showed that the AC3 approach (ICF 100% + Semantic Similarity PLSA + TF) had a superior value of 0,839 in aspect categorization; The SC1 approach (Word Embedding + LSTM method) outperformed the sentiment classification with a value of 0,940; aspect service obtained higher positive sentiment value of 45.545 compared to the other aspects; the comfort aspect obtained a higher negative sentiment value of 12.871 compared to the other aspects. It can be concluded that good hotel service did not guarantee customer satisfaction. The same hotel aspect could obtain different sentiment values, evidenced by the sentiment analysis which could be affected by the aspects.

Item Type:	Thesis (Masters)
Additional Information:	RTMT 006.312 Kho a-1
Uncontrolled Keywords:	Ulasan Produk, Kepuasan Pelanggan, aspek Hotel, Dokumen term list, hidden topic, Semantic Similarity, Aspect categorization, Word Embedding, Deep Learning, Klasifikasi Sentimen
Subjects:	Q Science > QA Mathematics > QA278.55 Cluster analysis Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA76.9.D343 Data mining. Querying (Computer science)
Divisions:	Faculty of Creative Design and Digital Business (CREABIZ) > Technology Management > 61101-(S2) Master Thesis
Depositing User:	Khotimah Dewi Ayu Khusnul
Date Deposited:	26 Mar 2025 02:36
Last Modified:	26 Mar 2025 02:36
URI:	http://repository.its.ac.id/id/eprint/67024

Actions (login required)

View Item