Multilingual hope speech detection in English and Dravidian languages

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Social Media Analysis

Background:

• Current language technologies focus on detecting and removing negative content like hate speech and cyberbullying.
• Existing models often rely on machine learning with labeled datasets for content moderation.
• There's a growing need to balance content moderation with promoting freedom of speech and positive online interactions.

Purpose of the Study:

• To develop a multilingual dataset for identifying "hope speech" to foster positivity, equality, diversity, and inclusion (EDI) on social media.
• To shift the focus from deleting offensive content to promoting supportive and inclusive language.
• To create a benchmark for evaluating systems designed to detect hope speech.

Main Methods:

• Collection of a novel multilingual dataset in English, Tamil, Malayalam, and Kannada.
• Inclusion of data from underrepresented communities such as LGBTQIA+, persons with disabilities, and women in STEM.
• Experimentation with various state-of-the-art machine learning and deep learning models to establish benchmark results.

Main Results:

• A unique multilingual dataset (Hope Speech dataset for Equality, Diversity and Inclusion - HopeEDI) was created.
• Benchmark results were established using diverse machine learning and deep learning models.
• The dataset facilitates research into promoting positive and inclusive language online.

Conclusions:

• The developed HopeEDI dataset is a valuable resource for advancing research in hope speech detection.
• This work contributes to creating more inclusive and positive online environments through language technology.
• Future work can leverage this dataset to build more sophisticated systems for promoting EDI in digital communication.