Corpus of Human-Labeled Indirect Harassment

In an age of information and digital communication, social media platforms are becoming essential spaces where individuals can voice their thoughts, exchange ideas, and take part in discussions on a variety of subjects. High quality and high value datasets of such ideas and discussions in textual and formatted form are fundamental to the research aspects of Artificial Intelligence and data-science. In this paper, we present a dataset of indirect harassment. There are approximately 10,700 tweets with binary labels. The labels were assigned by a team of researchers collectively. The dataset contains approximately 19 percent positive indirect harassment labels and 81 percent negative indirect harassment labels. The data is useful for training and running machine and deep learning models to detect indirect and direct harassment. The ease in understanding the data for researchers and other peers is also very crucial. The corpus mentioned in this paper is easy to understand via the use of a binary label. The data is simplistic in nature because there are only two columns with one being the text and the other being the indirect label. This work was necessary because it was a component of a larger project that also needed this kind of dataset.