Penerapan Deep Learning dalam Menganalisis Sentimen di Media Sosial

Abstract

Abstract: This research aims to investigate the application of deep learning for sentiment analysis in social media. A systematic literature review was conducted to identify, assess, and interpret relevant research evidence on this topic. Journal searches were focused on publications from 2014 to 2024 using indexing sources such as Google Scholar, DOAJ, and Scopus. Inclusion criteria were applied to select studies directly related to deep learning and sentiment analysis in the context of social media. Subsequently, data from the selected studies were systematically extracted and organized for comprehensive analysis. The research findings demonstrate that deep learning models such as BERT, CNN, LSTM, and GRU outperform traditional methods like SVM in terms of accuracy, recall, and F1 score in social media sentiment analysis. The combination of BERT+CNN showed the best performance in accuracy and F1 score. The frequently used deep learning architecture is the Transformer model, particularly BERT, due to its high accuracy capabilities. The Hinglish dataset is commonly used to train deep learning models for sentiment analysis in social media. Future research directions may include further exploring the integration of deep learning with advanced natural language processing techniques such as RoBERTa in the context of social media sentiment analysis. Additionally, in-depth studies on factors influencing the success of deep learning models in sentiment analysis, including data volume, algorithms used, and effective NLP techniques, would contribute significantly to advancing the field of social media sentiment analysis.

Abstrak: Penelitian ini bertujuan untuk menyelidiki penerapan deep learning untuk menganalisis sentimen di media sosial. Penelitian ini melakukan tinjauan literatur sistematis untuk mengidentifikasi, menilai, dan menafsirkan bukti penelitian yang relevan mengenai topik ini. Pencarian jurnal dilakukan dengan fokus pada publikasi terbitan antara tahun 2014 hingga 2024 dari sumber indeks seperti Google Scholar, DOAJ, dan Scopus. Kriteria inklusi digunakan untuk memilih studi-studi yang secara langsung terkait dengan deep learning dan analisis sentimen dalam konteks media sosial. Setelah itu, data-data dari studi-studi yang terpilih diekstraksi dan disusun secara sistematis untuk dianalisis secara komprehensif. Hasil penelitian menunjukkan bahwa model-model pembelajaran mendalam seperti BERT, CNN, LSTM, dan GRU unggul dalam hal akurasi, ingatan, dan skor F1 dibandingkan dengan metode tradisional seperti SVM dalam analisis sentimen dalam media sosial. Kombinasi model BERT+CNN terbukti memiliki kinerja terbaik dalam hal akurasi dan skor F1. Arsitektur pembelajaran mendalam yang sering digunakan adalah model Transformer, khususnya BERT, karena kemampuannya mencapai akurasi tinggi. Dataset Hinglish umum digunakan untuk melatih model deep learning dalam analisis sentimen media sosial.