<![CDATA[Introduction/Main Objectives: This research aims to conduct a preliminary study into the use of LLMs for extracting information to fill out questionnaires in survey interviews. Background Problems: BPS-Statistics Indonesia used paper-based questionnaires for interviews and is recently utilizing the Computer Assisted Personal Interviewing (CAPI) method. However, the CAPI method has some drawbacks. Enumerators must input data into the device, which can be burdensome and prone to errors. Novelty: This study uses a large language model (LLM) to extract information from survey interviews. Research Methods: This study utilizes a text-to-speech application to translate interview results into text. Translation accuracy is measured by the Word Error Rate (WER). Then the text was extracted using the ChatGPT 3.5 Turbo model. GPT-3.5 Turbo is part of the GPT family of algorithms developed by OpenAI. Finding/Results: The extraction results are formatted into a JSON file, which is intended to be used for automatic filling into the database and then evaluated using precision, recall, and F1-score. Based on research conducted by utilizing the Speech Recognition API by Google and the ChatGPT 3.5 Turbo model, an average WER of 10% was obtained in speech recognition and an average accuracy of 76.16% in automatic data extraction.]]>
Copyrights © 2025
