SEMANTIC RETRIEVAL FOR INDONESIAN QURAN AUTOCOMPLETION

Abstract

Attending lectures is a common way to learn Islamic knowledge. The speaker talks in front of the forum, and participants take notes on the lecture material. Many participants listen to the lecture while taking notes either in books or on other digital devices to avoid forgetting the discussed topics. However, note-taking during the lecture can be challenging, with no complementing module from the speaker. Lecturers have different paces and varying ways of delivering. In addition, sometimes, participants cannot always focus during the lecture. Those factors can cause problems in the note-taking process: some details can be lost or even shift the meaning. For note-taking on sensitive topics, such as verses from the Quran, the note-taking process must be done carefully and avoid mistakes. In this study, we proposed an autocomplete system for the Indonesian translation of the Quran that will help the user in note-taking Islamic lectures. The user writes down words, the parts of the Quran verse that he hears, and the system will retrieve the most similar verse. With semantic retrieval, the user does not need to write down the exact words of the verses he heard. The system can also handle typographical-error that usually occur in note-taking. We use Fasttext and calculate the cosine distance between the query and verses for the retrieval process. We also performed several optimization steps to create a robust system for the production stage. The system is evaluated by comparing how close the returned verse is with the ground truth. The proposed method's result accuracy reached 79.41% for the top 5 retrieved verse and 85.29% for the top 10 retrieved verse.