TY  - THES
A1  - Fauzan Muhammad Fahrezi, -
A1  - Liptia Venica, -
A1  - Muhammad Rizalul Wahid, -
TI  - ANALISIS PENGARUH SENTIMEN MASYARAKAT INDONESIA DI MEDIA SOSIAL X TERHADAP MODEL PREDIKSI HARGA SAHAM ALFAMART MENGGUNAKAN ALGORITMA DEEP LEARNING
M1  - other
N1  - https://scholar.google.com/citations?hl=en&view_op=list_works&gmla=AH8HC4wUxvmpchMm52tsDOe639hV1R5kWnOawyIshHpWhlRTuiET5QqxacnPsEwdIG-VUSMzL24KXxI-wKn9Kw&user=osdmZEcAAAAJ

ID SINTA dosen pembimbing:
Liptia Venica: 6779029
Muhammad Rizalul Wahid: 6780434
Y1  - 2025/07/07/
UR  - https://repository.upi.edu/
EP  - 102
AV  - restricted
KW  - Prediksi Harga Saham
KW  -  Sentimen Publik
KW  -  Media Sosial X
KW  -  RNN
KW  -  LSTM
KW  -  ELM
KW  -  Deep Learning
KW  -  AMRT.

Stock Price Prediction
KW  -  Public Sentiment
KW  -  Social Media X
KW  -  RNN
KW  -  LSTM
KW  -  ELM
KW  -  Deep Learning
KW  -  AMRT.
PB  - Universitas Pendidikan Indonesia
ID  - repoupi134700
N2  - Pergerakan harga saham dipengaruhi oleh berbagai faktor, termasuk kondisi fundamental perusahaan, tren pasar, serta opini publik. Seiring meningkatnya penggunaan media sosial, analisis sentimen publik menjadi pendekatan yang semakin relevan dalam bidang prediksi harga saham. Penelitian ini bertujuan untuk menguji apakah data sentimen masyarakat Indonesia di media sosial X (sebelumnya Twitter) dapat mempengaruhi akurasi model prediksi harga saham PT Sumber Alfaria Trijaya Tbk (AMRT). Hal ini didasarkan pada temuan dari beberapa studi sebelumnya yang menunjukkan bahwa sentimen publik dapat meningkatkan performa prediksi harga saham perusahaan-perusahaan besar seperti Google, Apple, Tesla, dan Telkom. Penelitian ini membangun model prediksi harga saham menggunakan tiga algoritma deep learning yang berbeda, yaitu Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), dan Extreme Learning Machine (ELM). Model dikembangkan dalam dua skenario: pertama tanpa data sentimen, dan kedua dengan penambahan fitur sentimen publik sebagai variabel input. Data sentimen diperoleh melalui proses crawling dari media sosial X, kemudian diklasifikasi menggunakan tiga pendekatan, yakni Lexicon-Based, Neural Network, dan pre-trained BERT. Model BERT dipilih sebagai pendekatan terbaik berdasarkan evaluasi F1-score dan digunakan untuk memberi label pada seluruh dataset sentimen. Hasil pengujian menunjukkan bahwa meskipun integrasi data sentimen dapat menambah konteks dalam prediksi, namun pengaruhnya terhadap peningkatan akurasi model tidak selalu signifikan. Algoritma ELM menunjukkan performa terbaik secara keseluruhan dengan nilai R² mencapai 89,15% pada skenario tanpa sentimen. Penambahan fitur sentimen justru memberikan penurunan performa pada beberapa algoritma. Temuan ini menegaskan bahwa meskipun data sentimen berpotensi memperkaya model prediktif, penerapannya pada sektor ritel seperti AMRT membutuhkan pengolahan yang lebih selektif dan kontekstual.
-----
Stock price movements are influenced by various factors, including a company?s fundamentals, market trends, and public opinion. With the increasing use of social media, public sentiment analysis has become an increasingly relevant approach in stock price prediction. This study aims to examine whether Indonesian public sentiment on social media platform X (formerly Twitter) can affect the accuracy of stock price prediction models for PT Sumber Alfaria Trijaya Tbk (AMRT). The research is motivated by previous studies that demonstrate the use of sentiment data can enhance stock prediction performance for companies such as Google, Apple, Tesla, and Telkom. This research develops stock price prediction models using three different deep learning algorithms: Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Extreme Learning Machine (ELM). The models are evaluated in two scenarios: first, without sentiment data, and second, with the inclusion of public sentiment features as additional input variables. Sentiment data were collected through a crawling process on social media X and classified using three methods: Lexicon-Based, Neural Network, and a pre-trained BERT model. The BERT model was selected as the best classifier based on F1-score evaluation and was used to label the entire sentiment dataset. The results indicate that while the integration of sentiment data can add contextual value to the prediction, its impact on improving model accuracy is not always significant. The ELM algorithm achieved the highest performance overall, with an R² score of 89.15% in the scenario without sentiment. In some cases, the inclusion of sentiment features even reduced model performance. These findings suggest that although public sentiment data has potential to enrich predictive models, its application in the retail sector such as AMRT requires more selective and context-aware preprocessing.
ER  -