Opinion Mining from YouTube Captions Using ChatGPT: A Case Study of Street Interviews Polling the 2023 Turkish Elections

Abstract: Opinion mining plays a critical role in understanding public sentiment and preferences, particularly in the context of political elections. Traditional polling methods, while useful, can be expensive and less scalable. Social media offers an alternative source of data for opinion mining but presents challenges such as noise, biases, and platform limitations in data collection. In this paper, we propose a novel approach for opinion mining, utilizing YouTube's auto-generated captions from public interviews as a data source, specifically focusing on the 2023 Turkish elections as a case study. We introduce an opinion mining framework using ChatGPT to mass-annotate voting intentions and motivations that represent the stance and frames prior to the election. We report that ChatGPT can predict the preferred candidate with 97\% accuracy and identify the correct voting motivation out of 13 possible choices with 71\% accuracy based on the data collected from 325 interviews. We conclude by discussing the robustness of our approach, accounting for factors such as captions quality, interview length, and channels. This new method will offer a less noisy and cost-effective alternative for opinion mining using social media data.