When the Bad Is Good and the Good Is Bad: Understanding Cyber Social Health Through Online Behavioral Change

Following the Special Issue on Cyber Social Health: Part 1 in the January/February 2021 issue, in this issue, we highlight another five papers that were accepted based on the quality of the analysis, results, and presentation. In “Towards Hate Speech Detection at Large via Deep Generative Modeling,” the authors developed an approach to improve supervised hate speech detection on social media by creating a large dataset of hate speech from a small seed set. They introduced a big ground truth dataset and assessed the generalizability of models to the variability in communications with hate speech. This work attempts to overcome the lack of diversity and improve coverage in the input dataset, and the data imbalance. The authors employ GPT-2 fine-tuned on the existing labeled datasets, to generate a larger diverse hate speech dataset. They also perform a comparative analysis on the inductive biases of DL methods during training on individual hate-speech datasets.