KLASIFIKASI KOMENTAR BERACUN MENGGUNAKAN METODE LONG SHORT TERM MEMORY (LSTM)
Total View This Week0

Institusion

Universitas Sriwijaya

Author

AMRULLAH, IDHAM ATTA
Abdiansah, Abdiansah
Alfarissi, Alfarissi

Subject

T1-995 Technology (General) 

Datestamp

2024-05-21 02:16:02 

Abstract :
The Kaggle platform, known as a hub for data and analytics competitions, provides a comprehensive dataset encompassing a range of comments, including toxic ones. Poisonous comments, often containing abusive, disrespectful, and demeaning language, impact the psychological well-being of individuals, particularly in the context of mental health. The presence of these comments on social media poses a serious challenge, as they not only disrupt healthy discussion but can also exacerbate mental health conditions. This study aims to classify toxic comments using the Long Short Term Memory. A total of 2,100 labeled data points were used, divided into two categories: toxic and non-toxi. The best LSTM model for classifying toxic comments had the optimal configuration with a learning rate of 0.0001, batch size of 8, 10 epochs, 32 neurons in the LSTM layer without LSTM dropout, and a dropout layer value of 0.2. With an accuracy of 85%, precision of 87.38%, recall of 82.95%, and f-measure of 85.11%, the model's effectiveness in classifying toxic comments is demonstrated. 

File :
RAMA_55201_09021381924145.pdf
RAMA_55201_09021381924145_TURNITIN.pdf
RAMA_55201_09021381924145_0001108401_0215128501_01_front_ref.pdf
RAMA_55201_09021381924145_0001108401_0215128501_02.pdf
RAMA_55201_09021381924145_0001108401_0215128501_03.pdf
RAMA_55201_09021381924145_0001108401_0215128501_04.pdf
RAMA_55201_09021381924145_0001108401_0215128501_05.pdf
RAMA_55201_09021381924145_0001108401_0215128501_06.pdf
RAMA_55201_09021381924145_0001108401_0215128501_07_ref.pdf
RAMA_55201_09021381924145_0001108401_0215128501_08_lamp.pdf