Sentiment analysis on Bangladeshi media streaming platform: using Bidirectional Encoder Representations from Transformers (BERT)

Abstract

This thesis delves into the intricate landscape of user sentiments within Bangladeshi media streaming platforms, employing a comprehensive sentiment analysis methodology. With a focal point on the evolutionary trajectory of the media streaming industry in Bangladesh, particular attention is given to the pivotal role of user-generated content and the imperative need for sophisticated sentiment analysis tools. Leveraging the Bidirectional Encoder Representations from Transformers (BERT) architecture for its contextual comprehension, coupled with Aspect-Based Sentiment Analysis for granular insights, the study seeks to elevate user experiences, steer strategic platform enhancements, integrate cultural sensitivity into sentiment analysis, propel natural language processing forward, and establish industry benchmarks. The investigation unveils nuanced insights derived from three advanced models—BERT, RoBERTa, and GPT—contributing valuable inputs for strategic decision-making and content curation. Acknowledging notable achievements, the study transparently recognizes its limitations and proposes future research avenues, aiming to catalyze continuous advancements in sentiment analysis methodologies within the dynamic realm of digital user interactions.