2026-05-04T19:54:08.579Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:1:11702 2025-11-12T00:07:00Z 1

Tuning XGBoost for Twitter Racism Classification Using Word2Vec and Adapted Botox Optimization Algorithm 2025 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/1/11702 https://ieeexplore.ieee.org/document/11212193 V. Zeljkovic D. Bulaja M. Sarac M. Pavkovic L. Babic M. Zivkovic T. Zivkovic N. Bacanin The rise of digital communication has brought new opportunities for connection, but it has also intensified exposure to harmful content - particularly racial harassment. Such abuse is increasingly prevalent on social media platforms and has been shown to negatively impact the mental health of marginalized communities. While natural language processing (NLP) and artificial intelligence (AI) offer promising tools for identifying and mitigating online harassment, their effectiveness is limited by the challenges of model selection and hyperparameter tuning. This paper proposes a novel AI-driven approach for detecting racial harassment in tweets, leveraging Word2Vec embeddings for semantic representation of textual data and the XGBoost algorithm for classification. In order to enhance model performance, a modified implementation of the botox optimization algorithm is introduced for efficient hyperparameter tuning. The proposed method is evaluated against alternative metaheuristic techniques using key classification metrics. Experimental results demonstrate the effectiveness of this approach, with the top-performing model achieving an accuracy of approximately 92.6 %. These findings underscore the potential of combining NLP with metaheuristic optimization for reliable detection of racially abusive content in online environments. conferenceObject 1 8 10.1109/AIC66080.2025.11212193 2025 IEEE 4th World Conference on Applied Intelligence and Computing (AIC), IEEE, GB Nagar, Gwalior, India