Informasi umumnya disampaikan kepada penerima melalui media cetak, televisi dan radio. Kini, informasi disampaikan menggunakan sistem komputer yaitu internet. Kebutuhan informasi terus mengalami peningkatan seiring meningkatnya pengguna internet. Akibatnya pertumbuhan berita menjadi sangat banyak sehingga diperlukan ringkasan teks berita untuk menampilkan informasi penting sebelum membaca lebih detail teks tersebut.Peringkasan teks otomatis dalam menentukan kalimat sebagai kalimat ringkasan pada penelitian ini, menggunakan metode neural network backpropagation dengan penggunaan fitur ekstraksi posisi kalimat, positive keyword, kemiripan kalimat, panjang kalimat dan bobot koneksi antar kalimat. Tahapan dari preprocessing yaitu tokenizing kalimat, case folding, filtering, tokenizing kata, dan stopword removal. Pengujian menggunakan single dokumen yang merupakan artikel berita Bahasa Indonesia, diambil dari media online kompas.com dengan bertemakan TEKNO terdiri dari 30 data training dan 20 data testing. Evaluasi hasil ringkasan teks menggunakan perhitungan recall, precision dan F-measure. Berdasarkan hasil pengujian peringkasan teks otomatis yang telah dilakukan dengan penggunaan epoch 700, learning rate 0.4 serta batas error 0.0001 mendapatkan nilai kurang cukup baik dengan rata – rata precision sebesar 53.14%, recall sebesar 67.07% dan F-measures sebesar 58.91%. Hal ini dikarenakan penggunaan data latih yang memiliki karakteristik tersendiri terutama dalam artikel berita TEKNO serta nilai variabel epoch, learning rate dan batas error dalam pembelajaran neural network.

Information is generally communicated to recipients through print, television and radio. Now, the information is delivered using a computer system that is the internet. Information needs continue to increase with increasing internet users. As a result the growth of news becomes so much that it takes a summary of the news text to display important information before reading more details of the text. Automatic text summary in determining sentences as summary sentences in this study, using neural network backpropagation method with the use of sentence position extraction features, positive keywords, sentence similarity, sentence length and connection weight between sentences. Stages of preprocessing are tokenizing sentences, case folding, filtering, word tokenizing, and stopword removal. Tests using a single document which is an Indonesian news article, taken from the online media kompas.com with the theme TEKNO consists of 30 training data and 20 data testing. Evaluation of text summary results using recall, precision and F-measure calculations. Based on the results of automatic text summary testing that has been done with the use of epoch 700, learning rate 0.4 and 0.0001 error limits get less good value with average precision of 53.14%, recall of 67.07% and F-measures of 58.91%. This is because the use of trainer data that has its own characteristics, especially in TEKNO news articles and the value of variable epoch, learning rate and error limit in learning neural network.

https://elib.unikom.ac.id/gdl.php?mod=browse&op=read&id=jbptunikompp-gdl-adesaepuln-39038