Hope speech detection in YouTube comments.

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Social Media Analysis

Background:

• Social media moderation often focuses on detecting and removing abusive language like hate speech and cyberbullying.
• Existing machine learning models rely on tagged datasets for negativity identification.
• There's a growing need to balance content moderation with the enhancement of free expression online.

Purpose of the Study:

• To develop a system that recognizes and encourages positive language in social media comments, rather than solely focusing on eliminating negative content.
• To create a multilingual dataset for training and evaluating models on positivity detection.
• To propose a novel deep network architecture for enhanced positivity identification.

Main Methods:

• Creation of a multilingual dataset for recognizing and encouraging positive online comments.
• Development of a novel custom deep network architecture using T5-Sentence embeddings.
• Experimentation with various machine learning models including Support Vector Machines (SVM), logistic regression, K-nearest neighbor, decision tree, and a new Convolutional Neural Network (CNN) based model.

Main Results:

• The proposed CNN-based model demonstrated superior performance compared to traditional machine learning models.
• The model achieved a macro F1-score of 0.75 for English, 0.62 for Tamil, and 0.67 for Malayalam in positivity detection.
• The study successfully established a multilingual approach to identifying and promoting positive online discourse.

Conclusions:

• The novel deep network architecture effectively identifies and encourages positive language on social media platforms.
• The findings suggest a shift in focus from solely eradicating negativity to actively promoting positivity for a healthier online environment.
• The developed multilingual dataset and model offer a valuable resource for future research in computational social science and language technology.