OPTIMIZATION FOR MACHINE LEARNING BASED EXTRACTIVE SUMMARIZATION USING REINFORCEMENT LEARNING ON THE INDONESIAN NEWS ARTICLES

• Main Author: Zia Davida, Bethea
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/43653
• Institution: Institut Teknologi Bandung

Automatic text summarization is generally done by building models without optimization. Optimization in an automatic text summation model can be done using reinforcement learning. Reinforcement learning is a learning technique based on reward from the environment. In automatic text summarizing, the reward value is obtained from the calculation of the performance metrics, Recall-Oriented Understanding for Gisting Evaluation (ROUGE). One type of reinforcement learning that can be used is Actor Critic. To add reinforcement learning to extractive summarization, an extractive summation model is added to the reinforcement learning module. Automatic text summarization with extractive methods that were built using the reinforcement learning module has the best F1 values for ROUGE-1, ROUGE-2, and ROUGE-L of 0.689; 0.619; and 0.681. From this final project, it can be concluded that reinforcement learning can improve the performance of extractive summation with a ROUGE difference of around 0.02. The best model of this final project consists of a convolutional model with 8 kernel sizes, a Bi-LSTM (Bidirectional LSTM) model, and a Pointer Network, and optimization using reinforcement learning.