Dynamic YouTube Comment Sentiment Analysis with Supervised Fine-Tuned BERT

In the present digital landscape, platforms such as YouTube have become central hubs for sharing content and engaging audiences, leading to enormous comments that reflect the sentiment of the audiences. The major challenge faced by content creators is to manually analyze the vast number of comments to understand the feedback of the audiences. This process is cumbersome in real time and leads to approximated results. Thus, it is highly required to address the challenges of manual computation and lack of accuracy of the results in YouTube comment analysis. This proposed work is an attempt to address this challenge and aims to provide an automated sentiment analyzer that performs sentiment analysis on YouTube comments and outputs real-time actionable insights. The methodology includes web scraping techniques to dynamically extract real-time YouTube comments and utilizing the pre-trained Bidirectional Encoder Representations from Transformers (BERT) model for sentiment analysis which was originally designed to classify the text by sentiments. The model requires fine-tuning to recognize patterns of social media comments. Such tuning has been achieved by training the model using the annotated Twitter comments dataset. This adaptation enables the model to improve the probability of classifying the sentiment patterns in dynamic YouTube comments. The proposed approach achieved an evaluation accuracy of 80.69% on testing data, enhancing real-time sentiment analysis for YouTube comments. The derived outcome supports content creators and organizations to better understand the sentiments of the audiences and address their feedback. Also, real-time insights provide support to businesses and organizations to make faster and smarter decisions toward improvement.