Detecting Homophobic and Transphobic Comments on Social media in Malayalam and English Languages

Nowadays people are using social media on a daily basis. Everyone is free to share their views and opinions to a wide audience from the convenience of their own homes. Amidst the advantages and benefits, the negativity that is spreading day by day has no limits. It could make one’s life physically and mentally challenging. This brief study presents the initial findings from experiments by utilizing transformer based models to classify whether a comment in social media platform is homophobic, transphobic, or non-anti-LGBT+ (Lesbian, Gay, Bisexual Transgender community) content in English and Malayalam languages. The dataset utilized for this work has been taken from a shared task conducted by LT-EDI-EACL 2024. At the beginning of the paper, we have introduced how social media toxic comments affect the mental health of people. Further, we discussed the methodology and models used for classification along with their F1 scores. The results were as follows: we achieved F1 scores of 0.45 with BERT (achieving fourth rank) and and 0.44 with RoBERTa respectively for English dataset, and for Malayalam dataset, we achieved F1 score of 0.95 with MuRIL and secured first rank and and F1 score of 0.91 with mBERT respectively.