The RBE_BengaliPattern dataset is a large-scale Roman Bengali-English code-mixed e-commerce dataset, derived from the Amazon English Sentiment dataset. With 18,086 text samples and a total of 1,303,474 words.The dataset provides a rich and diverse resource for analyzing code-mixing patterns in low-resource language contexts.
Here Roman Bengali words dominate the dataset, accounting for approximately 72.47% of all tokens (944,904 words), while English words comprise only 27.53% (358,570 words). This strong emphasizes the prominence of Romanized Bengali in user-generated content within the dataset.
The high volume of data also provides sufficient coverage to train robust machine learning models for downstream tasks like code-mixed language identification, sentiment classification, and token-level tagging.
In summary, the RBE_BengaliPattern dataset reflects a real-world linguistic trend of heavy Roman Bengali usage with embedded English, offering valuable insights for research in multilingual NLP, especially in low-resource and code-mixed language settings.