Implementasi Model Berbasis Transformer untuk Membuat Game Fiksi Interaktif Cerita Anak Berbahasa Indonesia

Abstract

Interactive Fiction is a text-based game that provides a certain narrative and can interact with players regarding the continuation of the narrative. This interaction can take the form of text input or options. However, traditional interactive fiction like this has several shortcomings such as less varied narratives, limited interaction that can be provided to players, and also the lack of availability of Indonesian. This research was created to answer these shortcomings by creating an interactive fiction that implements a transformer-based model that is capable of carrying out text generation (one of the tasks in NLP) to produce stories in Indonesian and process input in Indonesian. The author also uses children's stories as a type of narrative because interactive fiction itself can be used as a new form of literacy, and can have a moral message for learning. The creation of interactive fiction in this research will use a model called Mistral 7B Indo, which is a derivative model of Mistral 7B which has been fine-tuned to Indonesian. This research produces an interactive fiction game that is provided in web form so that it can be accessed online. Based on the results of the quality evaluation carried out in this research, it was found that this interactive fiction game can produce varied children's stories. However, there are several shortcomings, such as 70% of players assessing that the diction is complex, 60% of players assessing that the story sequence is not coherent and only 20% of players assessing that the resulting moral message is appropriate to the story.