Development of Code-Mixed Marathi-English Dataset for Hate Speech Detection

In India, people commonly use a mix of English and their regional languages on social media, resulting in a substantial amount of code-mixed content. As of the present date, there is only a limited number of hate speech code-mixed datasets available for Indian languages, including Hindi-English, Bengali-Hindi, Malayalam-English, and Tamil-English. However, there are no resources or datasets available for Marathi-English code-mixed data with hate speech labels. This paper introduces a novel gold standard corpus for sentiment analysis of code-mixed text in Marathi-English, annotated by voluntary annotators. The gold standard corpus achieved an impressive Krippendorff's alpha score of above 0.9 for the dataset. Utilizing this new corpus, the paper establishes a benchmark for sentiment analysis in Marathi-English code-mixed texts.