Aplikasi deteksi kejadian darurat berdasarkan data media sosial menggunakan paradigma pembelajaran dalam = Emergency event detection application based on social media data using deep learning paradigm

Wiyadi, Petrus Damianus Sammy (2018) Aplikasi deteksi kejadian darurat berdasarkan data media sosial menggunakan paradigma pembelajaran dalam = Emergency event detection application based on social media data using deep learning paradigm. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.

[img] Text
05111440000117-Undergraduate_Theses.pdf - Published Version
Restricted to Repository staff only

Download (4MB) | Request a copy

Abstract

Emergency Event digunakan sebagai istilah untuk menggambarkan kejadian yang mewajibkan tingkat kewaspadaan tinggi dan tanggapan cepat. Paradigma pembelajaran representasi data pada deep learning yang mampu memanfaatkan data yang berjumlah masif dapat terutilisasi secara maksimal berkat adanya perkembangan teknologi GPU Computation dalam beberapa tahun terakhir. Tugas Akhir ini mengimplementasikan variasi metode pembelajaran dalam, khususnya Convolutional Neural Network (CNN) dan Long Short-Term Memory (LSTM) pada data kejadian darurat. Dataset diambil dari Twitter API yang diketahui memiliki gaya bahasa pada tweet yang beragam serta citra yang variatif. Model dilatih menggunakan framework Keras yang telah mampu melakukan utilisasi teknologi komputasi GPU. Penyusunan Tugas Akhir ini melibatkan beberapa evaluasi terhadap optimasi arsitektur meliputi augmentasi data, optimasi hyperparameter, dan regularisasi. Arsitektur model citra yang dibandingkan meliputi AlexNet, VGG16, VGG19, dan SqueezeNet. Pada data teks, dilakukan pengujian antara arsitektur CNN, LSTM, dan C-LSTM. Didapat nilai akurasi terbesar pada data citra sebesar 99,08% dengan waktu 64 detik tiap epoch. Akurasi terbaik didapatkan pada data teks, sebesar 99,08% dengan waktu training 32 detik per epoch. ================= Emergency event are used as a term to describe events that require high levels of alertness and quick responses. The learning representation exists in deep learning paradigm that able to utilize massive amount of data now can be maximized optimally thanks to the development of computation GPU technology in recent years. This undergraduate theses implements variations of deep learning methods, especially the Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) in emergency event data. The dataset is taken from the Twitter API which is known to have a style of tweets that vary as well as containing varied imagery. The model is trained using the Keras framework that has been able to utilize GPU computing technology. The preparation of this undergraduate theses involves several evaluations on architecture optimization including data augmentation, hyperparameter optimization, and regularization. Compared image model architectures include AlexNet, VGG16, VGG19, and SqueezeNet. In the text data, the greatest accuracy obtained value in image data of 99.08% with time of 64 seconds per epoch. The best accuracy is found in text data, 99.08% with training time of 32 seconds per epoch.

Item Type: Thesis (Undergraduate)
Additional Information: RSIf 005.133 Wiy a-1
Uncontrolled Keywords: Emergency Event, Convolutional Neural Network, Long Short-Term Memory, Deep Learning, Image Classification, Text Classification
Subjects: Q Science > QA Mathematics > QA336 Artificial Intelligence
Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Q Science > QA Mathematics > QA76.9.D343 Data mining
T Technology > T Technology (General) > T57.5 Data Processing
Divisions: Faculty of Information Technology > Informatics Engineering > (S1) Undergraduate Theses
Depositing User: S W Petrus Damianus
Date Deposited: 08 Jan 2019 04:53
Last Modified: 08 Jan 2019 04:53
URI: http://repository.its.ac.id/id/eprint/53058

Actions (login required)

View Item View Item