Automatic Story Text Generation Using Recurrent Neural Network Algorithm

In the process of writing stories it takes time to bestow feelings and pour imagination into stories, making the writers run out of ideas with stories to be written which makes the writing process inhibiting. Previous research has demonstrated that algorithms such as Natural Language Generation (NLG), Generative Adversarial Networks (GAN), Generate and Rank Methodology, and Recurrent Neural Networks can generate sentences based on input words. However, the datasets used and the results obtained have been limited to the English language. Therefore, this research will focus on developing a model for generating novel text using the Recurrent Neural Network algorithm, specifically with GRU and LSTM architectures, and employing a dataset in Indonesian. This research aims to implement deep learning in creating automatic story text based on word input using RNN with GRU and LSTM architectures. There are several stages of research, namely data collection, data preprocessing, modeling, data testing and evaluation of relevant levels between words. Data taken from the Z-Library site with a character length of 13.535.514, the comparison of train and testing data used is 8:2. The results of the GRU architecture training have the best model precision with a bias loss of 0.7371. Based on the calculation of the average relevant word and standard deviation, the output text generated by the LSTM architectural model obtained values of 3.9 and 1.7288 respectively.