CyberbullyX-63K: A Large-Scale Twitter/X Multilingual Hinglish Tweets Dataset for Cyberbullying Detection

Published: 4 June 2026| Version 1 | DOI: 10.17632/s2t8khj9ht.1
Contributors:
,
,
,
,

Description

CyberBullyX-63K is a large-scale Twitter/X dataset developed for cyberbullying detection and harmful content classification research. The dataset contains 63,145 publicly available tweets collected through the official Twitter/X Developer API between January 2020 and May 2026. Data collection was conducted in multiple phases across different time periods to ensure temporal diversity and broad representation of online interactions, abusive language, toxic communication, harassment, and non-cyberbullying content. The dataset was created to support research in cyberbullying detection using machine learning, deep learning, and large language model (LLM)-based approaches. In the final_label column, 0 represents not cyberbullying whereas 1 represents the text is classified as cyberbullying. CyberBullyX-63K can be used for cyberbullying detection, abusive language identification, online harassment analysis, automated moderation systems, comparative evaluation of LLMs, and NLP classification research. All tweets were collected in compliance with Twitter/X developer policies and include only publicly accessible content intended for academic and non-commercial research purposes.

Files

Steps to reproduce

Create a Twitter/X Developer account and obtain API access credentials from the official Twitter/X Developer Portal. Use the Twitter/X API to collect publicly available tweets between January 2020 and May 2026. Data collection should be performed in multiple phases using keyword-based queries related to cyberbullying, abusive language, harassment, toxic communication, and general conversational content. Store the collected tweets in CSV format containing tweet text and metadata fields required for analysis. Use multiple annotators to independently classify each tweet into: Label 0: Non-cyberbullying, Label 1: Cyberbullying Apply majority voting across model predictions to generate the final annotation stored in the final_label column.

Categories

Social Media, Natural Language Processing, Machine Learning, Deep Learning, Twitter

Funders

Licence