Automatic Text Categorization to Standard Classification of Indonesian Business Fields (KBLI) 2020

Indonesia used a Standard Classification of Indonesian Business Fields (KBLI) to classify and categorize economic activities and businesses based on their primary functions and outputs. Practically, to determine the economic activity of a worker, a questionnaire is designed using open-ended questions that generate free-form text responses, and then the enumerator manually selects the appropriate category of KBLI. In this study, we develop text classification models using machine learning and transfer learning approach to automatically assign the category of KBLI from respondents' economic activity descriptions through free-form text responses. The dataset consists of a pair of respondents' job descriptions and the category of KBLI derived from the tourism survey. Based on the evaluation, our transfer learning model from pre-trained IndoBERT trained on a large monolingual Indonesian corpus outperformed machine learning models but with slight differences.