Bias in hate speech detection poses significant challenges, including the risk of disproportionately flagging content from certain demographic groups, misclassifying non-hateful content as hate speech, or failing to identify hate speech due to cultural or linguistic nuances. At the core of this problem is the training data, and its imbalanced representation of target groups. Target group identification has become critical for bias mitigation, particularly challenged by the use of slang, non-standard, and domain-specific language. In this presentation, I will discuss the use of semantic resources, such as knowledge graphs and linguistic tools, to improve both human and machine understanding of hate speech language. I will share findings from a four-year investigation on integrating semantics for target group identification, with a focus on enhancing deep learning models and human annotations. The discussion will underscore the role of semantics in enriching training data to better identify hate speech directed at specific groups. 
Bio:Paula Reyero Lobo is a KTP Associate at Manchester Metropolitan University and TVS Supply Chain Solutions. She previously worked as Marie Skłodowska-Curie Fellow within the "NoBIAS-Artificial Intelligence without Bias" at the Knowledge Media Institute (KMi) of the Open University, UK. Her PhD research focused on mitigating bias in hate speech detection, particularly enhancing target group identification using semantic resources. Paula is dedicated to reducing harm in AI-based decision making by integrating ethical and legal considerations in the design of technical solutions.