Article: XLM-R-BiLSTM-CNN: ensemble deep learning algorithm for code-mixed sentiment systems analysis Journal: International Journal of System of Systems Engineering (IJSSE) 2026 Vol.16 No.3 pp.267 - 289 Abstract: This research addresses the challenge of analysing sentiment in text that combines different languages, known as code-mixed sentiment analysis. The study presents a specialised deep-learning algorithm designed for code-mixed sentiment analysis in English and Tamil. The approach combines language identification and sentiment analysis models to improve accuracy in classifying sentiments in code-mixed text. The process involves preprocessing and tokenising the code-mixed data and using pre-trained word embeddings specific to each language. The XLM-R-BiLSTM-CNN architecture is used to create separate sentiment analysis models for English and Tamil. The ensemble method utilises language identification to select the appropriate sentiment analysis model for each token. The sentiment predictions from both models are combined, considering language-specific weights, to produce the final sentiment prediction. Experimental assessments on a code-mixed dataset demonstrate that our ensemble approach outperforms baseline models in terms of accuracy, precision, recall, and F1 score. This proposed technique significantly enhances the accuracy of sentiment analysis in multilingual code-mixed data, making it a valuable tool for understanding sentiments in various language contexts. Inderscience Publishers - linking academia, business and industry through research

Title: XLM-R-BiLSTM-CNN: ensemble deep learning algorithm for code-mixed sentiment systems analysis

Authors: C. Kumaresan; P. Thangaraju

Addresses: Department of Computer Science, Bishop Heber College (Autonomous), Affiliated to Bharathidasan University, Tiruchirappalli-620 017, Tamil Nadu, India ' Department of Computer Science, Bishop Heber College (Autonomous), Affiliated to Bharathidasan University, Tiruchirappalli-620 017, Tamil Nadu, India

Abstract: This research addresses the challenge of analysing sentiment in text that combines different languages, known as code-mixed sentiment analysis. The study presents a specialised deep-learning algorithm designed for code-mixed sentiment analysis in English and Tamil. The approach combines language identification and sentiment analysis models to improve accuracy in classifying sentiments in code-mixed text. The process involves preprocessing and tokenising the code-mixed data and using pre-trained word embeddings specific to each language. The XLM-R-BiLSTM-CNN architecture is used to create separate sentiment analysis models for English and Tamil. The ensemble method utilises language identification to select the appropriate sentiment analysis model for each token. The sentiment predictions from both models are combined, considering language-specific weights, to produce the final sentiment prediction. Experimental assessments on a code-mixed dataset demonstrate that our ensemble approach outperforms baseline models in terms of accuracy, precision, recall, and F1 score. This proposed technique significantly enhances the accuracy of sentiment analysis in multilingual code-mixed data, making it a valuable tool for understanding sentiments in various language contexts.

Keywords: code-mixed sentiment analysis; ensemble deep learning; language identification; BiLSTM; bi-directional long short-term memory; CNN; convolutional neural network; XLM-R; mixed sentiment systems analysis.

DOI: 10.1504/IJSSE.2026.154879

International Journal of System of Systems Engineering, 2026 Vol.16 No.3, pp.267 - 289

Received: 30 Oct 2023
Accepted: 01 Feb 2024
Published online: 17 Jul 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: XLM-R-BiLSTM-CNN: ensemble deep learning algorithm for code-mixed sentiment systems analysis

Keep up-to-date