Perkembangan teknologi dan digitalisasi telah memudahkan akses informasi melalui internet, termasuk dalam jurnalistik. Data menunjukkan bahwa terdapat ribuan portal berita daring di Indonesia, dengan detik.com menjadi salah satu yang paling banyak diakses dan terverifikasi oleh Dewan Pers. Penelitian ini menganalisis tren dan sentimen berita pemilihan umum 2024 di detik.com menggunakan metode deep learning. Data dikumpulkan dari berita selama tiga bulan masa kampanye dan dibagi menjadi tiga dataset sesuai pasangan calon presiden. Metode yang digunakan mencakup Exploratory Data Analysis (EDA) dan analisis sentimen menggunakan model Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), dan Gated Recurrent Unit (GRU), serta advanced model-nya. Hasil penelitian menunjukkan pasangan calon nomor urut 02 memiliki frekuensi pemberitaan tertinggi yang fluktuatif, sementara pasangan calon nomor urut 01 dan 03 lebih stagnan. Sentimen positif mendominasi pemberitaan dengan topik berbeda pada setiap peak conversation. Analisis sentimen menunjukkan model CNN-LSTM memiliki akurasi tertinggi, namun model dasar CNN, LSTM, dan GRU juga menunjukkan performa baik dengan akurasi, F1-Score, precision, dan recall di atas 80%, serta waktu runtime yang lebih singkat, menjadikannya pilihan lebih optimal untuk dataset ini.

---

The rapid development of technology and digitalization has facilitated access to information, including journalism. Thousands of online news portals exist in Indonesia, with detik.com being one of the most accessed and verified by Dewan Pers. This study analyzes trends and sentiment in news about the 2024 general election reported by detik.com. Data were collected during the three-month campaign period and divided into three datasets for each presidential candidate pair. Methods used include Exploratory Data Analysis (EDA) to identify trends and news frequency, and sentiment analysis using Natural Language Processing (NLP) techniques like Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU), along with advanced models. Results show that candidate pair 02 has the highest and most fluctuating news frequency, while pairs 01 and 03 have more stable trends. Positive sentiment dominates coverage for all candidates, with different topics during peak discussions: pair 01 in campaign contexts, pair 02 during the third debate, and pair 03 early in the nomination period. The CNN-LSTM model shows the highest accuracy in sentiment analysis, but fundamental models—CNN, LSTM, and GRU—also perform well, achieving over 80% accuracy, F1-Score, precision, and recall, with shorter runtime.