Perkembangan teknologi di era digital mengubah paradigma penyiaran televisi dari analog ke digital. Penelitian ini mengeksplorasi analisis sentimen masyarakat terhadap migrasi televisi menggunakan model Bidirectional Long Short-Term Memory (BiLSTM). Dataset berisi 9.597 data tweet dari Twitter, diperoleh dengan library SNScrape melalui proses scraping. Model BiLSTM digunakan untuk mengekstraksi sentimen positif, negatif, dan netral dari data tweet. Dataset ini diperoleh melalui proses sampling Slovin dengan tingkat error 2%, untuk memastikan representativitas sampel yang optimal dengan total 3.098 data tweet, penelitian ini menerapkan rasio split data 80:20, dengan 2.478 data untuk pelatihan model dan 620 data untuk pengujian. Teknik validasi silang (crossvalidation) 10 kali memberikan reliabilitas hasil yang baik. Hasil eksperimen mencapai rata-rata akurasi 91.36%, menunjukkan kinerja yang impresif. Analisis sentimen dilakukan pada sejumlah besar data dengan distribusi yang signifikan, terdiri dari 1.698 sentimen positif, 5.208 sentimen negatif, dan 2.691 sentimen netral, memberikan wawasan yang tentang variasi sentimen dalam dataset. Selain itu, distribusi hasil prediksi sentimen teridentifikasi kata-kata kunci yang mempengaruhi sentimen negatif, seperti "stb," "analog," dan "beli," mencerminkan ketidakpuasan terhadap kebijakan migrasi. Di sisi lain, kata-kata positif seperti "digital" dan "siar" mencerminkan pandangan positif terhadap teknologi digital. Hasil eksperimen menunjukkan bahwa pendekatan ini mampu melakukan klasifikasi sentimen pada data teks dengan akurasi tinggi.


Deskripsi Lain

Technological developments in the digital era changed the paradigm of television broadcasting from analog to digital. This research explores the analysis of public sentiment towards television migration using a modelBidirectional Long Short-Term Memory (BiLSTM). The dataset contains 9,597 tweet data fromTwitter, obtained withlibrary SNScrape through the processscraping. The BiLSTM model is used to extract positive, negative and neutral sentiments from tweet data. This dataset was obtained through a sampling processSlovin with an error rate of 2%, to ensure optimal sample representativeness with a total of 3,098 tweet data, this study applied a data split ratio of 80:20, with 2,478 data for model training and 620 data for testing. Cross validation technique (cross-validation) 10 times provides good reliability of results. The experimental results achieved an average accuracy of 91.36%, showing impressive performance. Sentiment analysis was performed on a large amount of data with a significant distribution, consisting of 1,698 positive sentiments, 5,208 negative sentiments, and 2,691 neutral sentiments, providing insight into the variations in sentiment within the dataset. In addition, the distribution of sentiment prediction results identified key words that influence negative sentiment, such "stb," "analog," and "buy," reflecting dissatisfaction with migration policies. On the other hand, positive words such “digital” and “broadcast” reflect a positive view of digital technology. Experimental results show that this approach is able to perform sentiment classification on text data with high accuracy.