Multilingual Hate Speech and Offensive Content Detection using Modified Cross-entropy Loss
This work addresses the problem of efficiently monitoring harmful content on social media for platforms and users, but it is incremental as it applies existing fine-tuning techniques with a loss modification to specific datasets.
The paper tackled automated detection of hate speech and offensive content in multilingual social media posts by fine-tuning large language models with a modified cross-entropy loss to address data imbalance, achieving macro F1-scores up to 0.808 in English and 0.737 in Hindi tasks.
The number of increased social media users has led to a lot of people misusing these platforms to spread offensive content and use hate speech. Manual tracking the vast amount of posts is impractical so it is necessary to devise automated methods to identify them quickly. Large language models are trained on a lot of data and they also make use of contextual embeddings. We fine-tune the large language models to help in our task. The data is also quite unbalanced; so we used a modified cross-entropy loss to tackle the issue. We observed that using a model which is fine-tuned in hindi corpora performs better. Our team (HNLP) achieved the macro F1-scores of 0.808, 0.639 in English Subtask A and English Subtask B respectively. For Hindi Subtask A, Hindi Subtask B our team achieved macro F1-scores of 0.737, 0.443 respectively in HASOC 2021.