Enhancing Bangla-English Code-Mixed Sentiment Analysis with Cross-Lingual Word Replacement and Data Augmentation

Sentiment analysis employs natural language processing, computational linguistics, and other textual analysis approaches to locate and extract subjective information to assess the text's conveyed attitude or emotional state. The prevalence of internet-based media and online businesses encourages extensive user engagement. Individuals frequently express their opinions, thoughts, ideas, attitudes, feelings, and more using familiar languages on various social media platforms. The majority of individuals tend to articulate their opinions using a combination of languages, often incorporating English with their native tongue. The lack of annotated code-mixed data poses a significant problem for automated sentiment analysis in languages with limited resources such as Bangla. To tackle this challenge, we aim to gather code-mixed content in Bangla and English. Subsequently, we plan to utilize data augmentation techniques to expand the dataset. Following this, we aim to conduct a comparative analysis by employing a variety of machine-learning techniques. This research delves into four common machine learning techniques—Support Vector Machine (SVM), Decision Tree (DT), Stochastic Gradient Descent (SGD), and Random Forest (RF)—employing feature extraction named TF-IDF methods. The experimental results indicate that Random Forest with TF-IDF gained the highest accuracy, reaching 83%, outperforming other techniques.