Pemilihan presiden yang diadakan setiap lima tahun sekali merupakan cara Indonesia menerapkan demokrasi. Di era internet, masyarakat semakin aktif menggunakan media sosial untuk menyampaikan pendapat tentang pemilihan presiden. Media sosial, terutama Twitter, menjadi platform penting bagi tim sukses, pendukung, buzzer, dan partai politik untuk menyuarakan isi hati mereka, terutama pada saat hasil akhir perhitungan KPU pada pemilihan presiden 2024. Penelitian ini menerapkan metode Long Short-Term Memory (LSTM) untuk mengklasifikasikan sentimen opini publik di Twitter terkait hasil hitung KPU. Data dikumpulkan menggunakan teknik crawling, menghasilkan 4116 tweet yang kemudian dibersihkan melalui text cleaning, normalisasi, tokenisasi, stopword removal, dan stemming. Pelabelan menggunakan kamus Inset Lexicon Based menghasilkan 2114 tweet positif, 1309 tweet negatif, dan 663 tweet netral. Pembobotan kata dilakukan menggunakan metode Word2Vec dengan model Continuous Bag of Words (CBOW). Model LSTM digunakan untuk klasifikasi sentimen, menghasilkan akurasi sebesar 80% dan dievaluasi menggunakan Confusion Matrix, rata-rata nilai precision pada sentimen positif sebesar 82%, negatif 84%, dan netral 54%, sedangkan rata-rata recall pada sentimen positif sebesar 97%, negatif 72%, dan sentimen netral 32%. Hasil ini menunjukkan bahwa model LSTM berhasil diterapkan dalam mengklasifikasikan sentimen opini publik terhadap hasil hitung KPU.

Presidential elections, held every five years, are Indonesia's way of practicing democracy. In the internet era, people are increasingly using social media to express their opinions about the presidential election. Social media, especially Twitter, has become an important platform for success teams, supporters, buzzers, and political parties to voice their hearts, especially during the final results of the KPU's calculations in the 2024 presidential election. This research applies the Long Short-Term Memory (LSTM) method to classify the sentiment of public opinion on Twitter related to the KPU count results. Data was collected using crawling techniques, resulting in 4116 tweets which were then cleaned through text cleaning, normalization, tokenization, stopword removal, and stemming. Labeling using the Inset Lexicon Based dictionary resulted in 2114 positive tweets, 1309 negative tweets, and 663 neutral tweets. Word weighting is done using the Word2Vec method with the Continuous Bag of Words (CBOW) model. The LSTM model is used for sentiment classification, resulting in an accuracy of 80% and evaluated using Confusion Matrix, the average precision value on positive sentiment is 82%, negative 84%, and neutral 54%, while the average recall on positive sentiment is 97%, negative 72%, and neutral sentiment 32%. These results show that the LSTM model is successfully applied in classifying public opinion sentiments towards the KPU's count results.