Bi-LSTM untuk Peramalan Deret Waktu Harga Penutupan Saham BBCA dengan Memasukkan Sentimen Berbasis Topik dari Media Sosial X

Susanto, Muhammad Aflah Ghozi (2026) Bi-LSTM untuk Peramalan Deret Waktu Harga Penutupan Saham BBCA dengan Memasukkan Sentimen Berbasis Topik dari Media Sosial X. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221074-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Penelitian ini mengkaji kontribusi variabel sentimen berbasis topik dari media sosial X terhadap peramalan harga penutupan saham BBCA menggunakan model Bidirectional Long Short-Term Memory (Bi-LSTM). Pendekatan sentimen agregat dinilai kurang mampu menangkap konteks spesifik teks media sosial sehingga penelitian ini mengadopsi BERTopic berbasis transformer dengan embedding IndoSBERT-large sebagai metode topic modeling. Data yang digunakan mencakup 3.662 tweet terkait BBCA periode Januari 2024 hingga Desember 2025 dan data harga penutupan harian dari Yahoo Finance. Skor sentimen dihitung menggunakan hasil pelabelan validasi, diagregasi per topik per hari, kemudian menjadi skor sentimen harian sebagai variabel eksogen Bi-LSTM. Hasil evaluasi menunjukkan model dengan sentimen berbasis topik mengungguli model baseline pada seluruh metrik, dengan MAPE out-sample 5,080%, RMSE 450,484, dan MAE 395,589 yang dapat mengikuti pergerakan data actual lebih baik. Peramalan 5 hari ke depan menghasilkan prediksi harga Rp8.010–Rp8.139 dengan berfluktuasi naik dan menurun.
======================================================================================================================================
This study investigates the contribution of topic-based sentiment variables derived from X (Twitter) to forecasting the closing price of BBCA stock using a Bidirectional Long Short-Term Memory (Bi-LSTM) model. Aggregate sentiment approaches are considered less effective in capturing the specific context of social media texts. Therefore, this study employs BERTopic, a transformer-based topic modeling method, with IndoSBERT-large embeddings. The dataset consists of 3,662 tweets related to BBCA collected from January 2024 to December 2025, along with daily closing price data obtained from Yahoo Finance. Sentiment scores were calculated based on validated sentiment labels, aggregated by topic on a daily basis, and then used as daily sentiment scores to serve as exogenous variables in the Bi-LSTM model. The evaluation results show that the model incorporating topic-based sentiment outperformed the baseline model across all evaluation metrics, achieving an out-of-sample MAPE of 5.080%, RMSE of 450.484, and MAE of 395.589, while more closely following the actual stock price movements. The five-day-ahead forecast predicts BBCA closing prices ranging from IDR 8,010 to IDR 8,139, with prices expected to fluctuate through both upward and downward movements.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	BERTopic, Bi-LSTM, Media sosial X, Peramalan, Topic modeling, BERTopic, Bi-LSTM, Forecasting, Social Media X, Topic Modeling
Subjects:	H Social Sciences > HG Finance > HG4915 Stocks--Prices Q Science > Q Science (General) > Q325.78 Back propagation Q Science > QA Mathematics > QA336 Artificial Intelligence 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 Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	MUHAMMAD AFLAH GHOZI SUSANTO
Date Deposited:	01 Aug 2026 06:29
Last Modified:	01 Aug 2026 06:29
URI:	http://repository.its.ac.id/id/eprint/141385

Actions (login required)

View Item