Irawati, Evia Nanda (2019) Analisis Sentimen Teks Bahasa Indonesia Pada Media Sosial Menggunakan Long Short Term Memory Dan Convolutional Neural Network (Studi Kasus: Operator Telekomunikasi). Other thesis, Institut Teknologi Sepuluh Nopember.
Preview |
Text
05211540000056-Undergraduate_Theses.pdf Download (2MB) | Preview |
Abstract
Analisis sentimen yang menggunakan algortima neural network mencapai hasil yang sangat baik dalam pemrosesan Bahasa alami dan menunjukkan kemampuan belajar yang baik. Kata dan frasa direpresentasikan menggunakan distributed representation, sehingga memperoleh karakteristik hubungan antara kata sebelum dan sesudah bahasa dan menghindari masalah dimensionalitas tinggi dan penyebaran data. Seiring dengan berjalan waktu, neural network semakin banyak mengalami perkembangan dan menghasilkan banyak macam arsitektur. Pada penelitian ini akan menggunakan Long Short Term Memory dan Convolutional Neural Network untuk menganalisis sentimen publik terhadap layanan operator telekomunikasi di Indonesia yang dimuat di media sosial. Penggunaan algoritma Long Short Term Memory dan Convolutional Neural Network melalui beberapa tahapan. Diawali dengan membagi dataset kedalam tiga subtask A, B dan C. Subtask A merupakan pengelompokan dataset yang memiliki dua macam label yaitu sentimen positif dan negatif. Kemudian untuk subtask B merupakan pengelompokan dataset yang memiliki tiga macam label yaitu positif, negatif dan netral sedangkan subtask C memiliki 5 macam label yaitu sangat positif, positif, sangat negatif, negatif dan netral. Selanjutnya dilakukan pengujian tiap subtask terhadap algoritma Long Short Term Memory dan Convolutional Neural Network yang sebelumnya diberi bobot awal dengan model Word2Vec dengan learning algorithm Skip-gram. Hasil dari pengujian tersebut yang memiliki performa terbaik menjadi masukan pada tahapan menggabungkan model Long Short Term Memory dan Convolutional Neural Network. Long Short Term Memory menggunakan 2 macam model yaitu bidirectional dan yang non-bidirectional atau lebih disebut dengan LSTM saja. Perbandingan hasil evaluasi pengukuran pada 3 subtask menunjukkan nilai performa model Bidirectional LSTM lebih baik pada 2 subtask yaitu A dan B dibandingkan dengan model LSTM. Proses pelatihan model Bidirectional LSTM pada 3 subtask berbeda didapatkan nilai optimizer paling baik adalah Adadelta dengan nilai learning rate 1. Sedangkan untuk model LSTM pada subtask A dan C didapatkan nilai optimizer paling baik adalah Adadelta dengan nilai learning rate 1, pda subtask B nilai optimizer paling baik adalah Adam dengan learning rate 0.001. Kemudian untuk model Convolutional Neural Network yang menggunakan multi filter region size performanya mengungguli Long Short Term Memory pada semua subtask. Penggabungan model Convolutional Neural Network dan Long Short Term Memory tidak dapat mengungguli performa model individu dalam melakukan proses klasifikasi.
==================================================================================================================================
Sentiment analysis using neural network algorithms achieves excellent results in natural language processing and shows good learning skills. Words and phrases are represented using distributed representations, thus obtaining the characteristics of the relationship between the words before and after language and avoiding high dimensionality problems and data dissemination. As time goes on, more and more neural networks experience development and produce many kinds of architecture. In this study, we will use Long Short Term Memory and Convolutional Neural Network to analyze public sentiment towards telecommunication operator services in Indonesia that are published on social media. The use of Long Short Term Memory and Convolutional Neural Network algorithms through several stages. Beginning by dividing the dataset into three subtasks A, B and C. Subtask A is a grouping of datasets that have two kinds of labels, namely positive and negative sentiments. Then for subtask B is a grouping of datasets that have three kinds of labels, namely positive, negative and neutral while subtask C has 5 kinds of labels, namely very positive, positive, very negative, negative and neutral. Then testing each subtask on the Long Short Term Memory algorithm and Convolutional Neural Network algorithm which was previously given the initial weight with the Word2Vec model with the Skip-gram learning algorithm. The results of these tests which have the best performance are input to the stages of combining the Long Short Term Memory and Convolutional Neural Network models. Long Short Term Memory uses 2 types of models, namely bidirectional and non-bidirectional or more so-called LSTM. Comparison of the results of evaluation of measurements in 3 subtasks shows the performance value of the Bidirectional LSTM model is better in 2 subtasks, namely A and B compared to the LSTM model. The training process of the Bidirectional LSTM model in 3 different subtasks found that the best optimizer value was Adadelta with a learning rate 1. As for the LSTM model in the A and C subtask, the best optimizer value was Adadelta with the learning rate 1, in the B subtask the optimizer value was good is Adam with a learning rate of 0.001. Then for the Convolutional Neural Network model that uses a multi filter region the size performance outperforms the Long Short Term Memory in all subtasks. Combining the Convolutional Neural Network and Long Short Term Memory models cannot outperform the individual models in carrying out the classification process.
| Item Type: | Thesis (Other) |
|---|---|
| Additional Information: | RSSI 006.312 Ira a-1 2019 |
| Uncontrolled Keywords: | Analisis sentimen, Media Sosial, Long Short Term Memory, Convolutional Neural Network |
| Subjects: | T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T58.62 Decision support systems |
| Divisions: | Faculty of Information Technology > Information System > 57201-(S1) Undergraduate Thesis |
| Depositing User: | Irawati Evia Nanda |
| Date Deposited: | 25 Jan 2024 07:22 |
| Last Modified: | 25 Jan 2024 07:22 |
| URI: | http://repository.its.ac.id/id/eprint/64387 |
Actions (login required)
 |
View Item |