Doctor of Fasilkom UI Conducts Reasearch in Utilization of Deep Learning through Argumentation Mining

Fakultas Ilmu Komputer Universitas Indonesia > E-News > Doctor of Fasilkom UI Conducts Reasearch in Utilization of Deep Learning through Argumentation Mining


Fasilkom UI presented computer science doctoral dissertation defense session to Derwin Suhartono with a dissertation title “Studi Mengenai Mekanisme Atensi pada Deep Learning untuk Anotasi dan Analisa Kalimat Argumentasi/ Study about Attention Mechanism in Deep Learning to Annotate and Analyse Argumentative Sentences.” The oral defense was held on Monday, 16 July 2018 in Fasilkom UI Auditorium with chair of oral defense session, Mirna Adriani, Ph.D. and Prof. Dr. Ir. Aniati Murni Arymurthy, M.Sc. as a promotor and Dr. Eng. Mohamad Ivan Fanany S.Si., M. Kom. as a co-promotor. Other attendees were examiners: Dr. Indra Budi S. Kom., M. Kom; Ir. Wahyu Catur Wibowo M.Sc., Ph.D.; Dr. Ir. Erdefi Rakun M.Sc.; Prof. Dr. Agus Buono; and Ir. Suryana Setiawan M.Sc., Ph.D.

Argumentation mining is a research study focusing on sentence and argumentative type, which often used in daily basis communication and benefits to a decision making or drawing conclusion. Argumentative Annotation explained in the research was grouped into several classes, they were major claim, claim, premise and non-argumentation. The argumentative analysis alone led to characteristics and validities arranged in particular topics. Looking into the research, there were 402 persuasive essays harnessed as datasets which beforehand been translated to Indonesian. The translated datasets drew pictures on how model would work in different language.

Research validity was conducted in combinations of conventional shallow and deep learning, as well as combinations of deep learning with XGBoot as the classification part. Meanwhile, Glove was one of the outputs from the deep learning observed as a feature. The research pointed out many valuable outcomes that would be beneficial for the next research in the same field, that are; pre-trained word vector does not possess any significance in the argumentative annotation and is not able to classify which as major claim and claim. Despite, the use of whole deep learning model either by utilizing attention or not, when it is required to test with non-English dataset then researcher has to weigh data size to generate word embedding as feature extractor.

Dr. Derwin Suhartono received his degree as the 69th Doctoral Computer Science graduate in Fasilkom UI. Predicate given to his dissertation by the committees is very satisfactory.