A Benchmark Dataset for Cricket Sentiment Analysis in Bangla Social Media Text

This study introduces a novel benchmark dataset designed for Cricket Sentiment Analysis on Bangla social media posts, emphasizing a low-resource setting. The dataset was meticulously curated through manual collection across diverse social media platforms, ensuring comprehensive representation of user sentiments. Annotations validated dataset quality, achieving a remarkable Cohen Kappa score of 0.97. Experimentation with machine learning (ML) models revealed challenges, with traditional approaches yielding modest RNN accuracy of 0.5239. However, deep learning (DL) models showcased significant performance enhancements. The LSTM model achieved 0.897 accuracy, while the BiLSTM model surpassed expectations at 0.952. These findings highlight DL’s efficacy in capturing nuanced sentiments in Bangla cricket-related social media posts, contributing a high-quality benchmark dataset and insights into DL’s suitability for sentiment analysis in low-resource linguistic contexts.

Перевод пока недоступен