Bonfring International Journal of Data Mining

Online ISSN: 2277-5048 | Print ISSN: 2250-107X | Frequency: 4 Issues/Year

Impact Factor: 0.245 | International Scientific Indexing(ISI) calculate based on International Citation Report(ICR)

Comparative Analysis of BiLSTM and KNN Algorithms for Sentiment Analysis in Tamil-English Code-Mixed Text: A Comprehensive Evaluation Framework

Aakash Ashok Kumar, Poundoss Chellamuthu, Denceli Dennis and Aathithya Sharan

---

Abstract:

Sentiment analysis in Tamil-Code Mixed data is challenging mainly due to its linguistic challenges. This paper analyzes several methodologies like deep learning (DL), machine learning (ML) and transformer-driven approaches and at the same time proposing a BiLSTM-driven approach for evaluating the sentiment of a Code-Mixed Tamil sentence. This research compares two approaches to analyze sentiment polarity in Code-Mixed Tamil sentence, BiLSTM and KNN, out of which the BiLSTM algorithm came out as the better algorithm for sentiment classification for Code-Mixed Tamil Text. The methodology achieves an F1 score of 0.435563, an accuracy of 0.746383, a precision of 0.332918, a recall of 0.629717, and a loss of 0.01 for the BiLSTM algorithm, and for the KNN algorithm, the methodology achieves an F1 score of 0.235714, a recall of 0.155660, a precision of 0.485294, and an accuracy of 0.828731. This study highlights several methods such from TF-IDF-based models to transformer architectures, highlighting the challenges faced during sentiment analysis in Code-Mixed Tamil sentences.

Keywords: Tamil-English Code-Mixed, NLP, BiLSTM, KNN, Deep Learning, Text Classification, Sentiment Analysis, Data Imbalance, Code-Mixed Text.

Volume: 16 | Issue: 1

Pages: 1-11

Issue Date: June , 2026

DOI: 10.9756/BIJDM/V16I1/BIJ26007

Full Text