Doctoral Candidate at GESIS

Connect

Follow

Rabiraj Bandyopadhyay

DC 8 at GESIS

I want to use NLP to mitigate harmful content and make social media safe and inclusive for everyone.

I am passionate about leveraging NLP to help people and my PhD topic will deal with developing socially aware NLP systems that can help in mitigating harmful online content. I have experience in building NLP systems to solve various business problems, and using NLP and Deep Learning to model online user behavior in social networks (Twitter) and text repositories (Wikipedia) while pursuing my Masters from University at Buffalo (State University of New York).

DC8 Individual Project

Hybrid data generation and retrieval for the detection of online harassment

Research Objectives:

Designing Inclusive Datasets for Harassment Detection

To design a hybrid strategy to ensure that datasets that are used to train and test harassment detection systems cover different groups of victims, different styles and languages, as well as different aspects of online harassment that are described in the social sciences and have been measured with survey instruments in the past;

Innovative Data Generation for Harassment Detection

To explore data collection and data generation strategies (such as language generation models and style transfer models) to create high quality datasets for online harassment detection;

Metrics for Assessing Harassment Detection Datasets

To define metrics to evaluate the quality of datasets for online harassment detection;

Automated Detection and Generation of Online Harassment Examples

To develop algorithms to detect online harassment in texts from different domains and use those algorithms to detect and generate new examples automatically.

Expected Results:

Comprehensive Datasets for Harassment Detection Systems

High-quality training and test data for online harassment detection systems.

Automated Tools for Enhanced Harassment Detection

Development of automated tools to detect online harassment and generate/collect data that is similar and/or complementary.

Validating Methodology and Tools in Diverse Use Cases

Methodology and tool validation in use cases on women, LGBTQ and migrants

Supervisors

In our pursuit of academic excellence, HYBRIDS Doctoral Candidates are guided by a dedicated team of supervisors. Comprising the Main Supervisor, Co-Supervisor, and Inter-sectoral Supervisor, this team of professionals offers a wealth of knowledge, mentorship, and interdisciplinary insights.