Abstract: Past studies have contributed quite significantly to
the detection of hate speech on social media platforms.
However, most of such models stop at evaluating the models‟
efficacy. This has to be taken further with the provisions of the
new ML UI tools to visually design an explainable predictive
model with the plausibility of real-world usefulness beyond just
accuracy. With this, classifying, scoring, and visualizing hate
speeches from its mildest to aggressive form would be essential for
the regulators of social networks and moderators of social platforms
to be well-informed about the consequential scale of societal hate
polarization flowing in the social space. UI visualization of hate
speech provides high level warning signal that could go miles for
actions that assuages the risk of potential violence. This study
implements a user interface for determing the intensity of hate
speech using Gradio and the well-known BERT model. BERT
leverages the power of word contextualization and parallelization to
reveal the most useful information from its key predictions. The
proposed model was contextually efficient in differentiating between
related hate comments through a simple interface. This study
demonstrates a practically usable system to reliably rank the
intensity of Hate-speech by their statistical probabilities.
Keywords: Explainable Hate Speech, Hate Speech User Interface,
Hate Speech Intensity, Multi-label Classification, BERT, Gradio
Publication Date: 2024-11-26