Design and Build an Instagram Content Sentiment Analysis Application Using the Bidirectional Encoder Representation From Transformer Model

Abstract

The high activity of Instagram users in Indonesia, with the number of users reaching 89.15 million in January 2023, reflects the great potential for sentiment analysis on social media. This research aims to develop a web-based application that can analyze the sentiment of Instagram comments in real-time regarding government policies, using the BERT (Bidirectional Encoder Representation from Transformer) model. BERT is a deep learning model developed by Google in 2018, which has the ability to understand word context bidirectionally. This research uses a method of processing comment text from Instagram with NLP techniques and trains the IndoBERT model, which is specially adapted for Indonesian. The dataset used includes labeled data from Instagram and Kaggle scraping, as well as real-time data captured using Instaloader. The results showed that the BERT model achieved 91.57% accuracy in classifying sentiment as positive or negative. In conclusion, the developed application successfully provides an effective platform for sentiment analysis of Instagram content, providing valuable insights into public opinion towards government policies. This research recommends further exploration of various BERT models in Indonesian language social media sentiment analysis.