Reproducibility Package for DyLex-Net

This repository provides the reproducibility package and implementation of the DyLex-Net framework proposed in the study:

“Automated Hate Speech Profiling via Lexicon-Enriched Ensemble Learning and Ego-Network Analysis”.

DyLex-Net is a digital forensic-oriented framework designed for profiling hate speech actors on social media platforms, particularly Twitter/X, by integrating lexicon-based detection, machine learning validation, and ego-network analysis. The system emphasizes explainability, traceability, and actor-centric investigation in multilingual and code-mixed environments.

The implementation includes a Flask-based application that supports end-to-end investigative workflows, including data acquisition, hate speech detection, network analysis, and model evaluation.

The dataset used in this study is constructed by integrating multiple publicly available hate speech datasets. Due to privacy and platform restrictions, raw data redistribution is limited; however, the repository includes configuration details and sample data to support reproducibility.

This package is intended to ensure transparency and reproducibility of the proposed framework by providing detailed implementation workflows, model configurations, and evaluation procedures.

Ethical considerations are acknowledged, particularly regarding the risks of false positives, bias in automated classification, and the implications of assigning user-level risk labels such as “SUSPICIOUS” and “DANGEROUS”. The system is designed for research and analytical purposes and should not be used as a standalone decision-making tool without human validation.