Offensive Sentiment Detection with Chat GPT and Other Transformers in Kannada

Social media has provided us the privilege of free speech. This liberty however is often subject to misappropriation by users. Active detection of the presence of offensive and foul language in online communication to prevent defamation of individuals and targeted groups thus has become imperative....

Full description

Saved in:
Bibliographic Details
Published in2023 IEEE 2nd International Conference on Data, Decision and Systems (ICDDS) pp. 1 - 6
Main Authors Garani, Yogita, Joshi, Shreya, Kulkarni, Savitri
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.12.2023
Subjects
Bidirectional LSTM
ChatGPT
Computational modeling
Data models
Ensembling
Feature extraction
Large Language Models (LLMs)
Offensive Sentiment
Sentiment analysis
Social networking (online)
Training
Transfer Learning
Transformers
Online AccessGet full text
DOI10.1109/ICDDS59137.2023.10434684

Cover

More Information
Summary:Social media has provided us the privilege of free speech. This liberty however is often subject to misappropriation by users. Active detection of the presence of offensive and foul language in online communication to prevent defamation of individuals and targeted groups thus has become imperative. This paper aims to improve offensive sentiment detection in social media comments in Kannada. We introduce a novel ensemble that combines the power of 4 large language models - XLM RoBERTa, mBERT, indicBERT and kannadaBERT to understand the complexity in the language and achieve a noticeable improvement in performance for detecting offensive sentiment in Kannada comments. We introduce a novel comparison with pre - trained Chat GPT that presents a remarkable lift in performance in comparison with the fine - tuned models - mBERT, XLM RoBERTa and the proposed ensemble.
DOI:10.1109/ICDDS59137.2023.10434684